AuroraMiddleware/gtk2
1
0
Fork 0
forked from AuroraMiddleware/gtk
Code Issues Pull Requests Projects Releases Wiki Activity
809299f9e4
gtk2/gsk/ngl/gsknglrenderjob.c

3978 lines
154 KiB
C
Raw Normal View History

gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
/* gsknglrenderjob.c
*
* Copyright 2017 Timm Bäder <mail@baedert.org>
* Copyright 2018 Matthias Clasen <mclasen@redhat.com>
* Copyright 2018 Alexander Larsson <alexl@redhat.com>
* Copyright 2020 Christian Hergert <chergert@redhat.com>
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this program. If not, see <http://www.gnu.org/licenses/>.
*
* SPDX-License-Identifier: LGPL-2.1-or-later
*/
#include "config.h"
#include <gdk/gdkglcontextprivate.h>
#include <gdk/gdkprofilerprivate.h>
#include <gdk/gdkrgbaprivate.h>
#include <gsk/gskrendernodeprivate.h>
#include <gsk/gskglshaderprivate.h>
#include <gdk/gdktextureprivate.h>
#include <gsk/gsktransformprivate.h>
#include <gsk/gskroundedrectprivate.h>
#include <math.h>
#include <string.h>
#include "gsknglcommandqueueprivate.h"
#include "gskngldriverprivate.h"
#include "gsknglglyphlibraryprivate.h"
#include "gskngliconlibraryprivate.h"
#include "gsknglprogramprivate.h"
#include "gsknglrenderjobprivate.h"
#include "gsknglshadowlibraryprivate.h"
#include "ninesliceprivate.h"
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
#include "fp16private.h"
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
#define ORTHO_NEAR_PLANE -10000
#define ORTHO_FAR_PLANE 10000
#define MAX_GRADIENT_STOPS 6
#define SHADOW_EXTRA_SIZE 4
/* Make sure gradient stops fits in packed array_count */
G_STATIC_ASSERT ((MAX_GRADIENT_STOPS * 5) < (1 << GSK_NGL_UNIFORM_ARRAY_BITS));
#define rounded_rect_top_left(r) \
(GRAPHENE_RECT_INIT(r->bounds.origin.x, \
r->bounds.origin.y, \
r->corner[0].width, r->corner[0].height))
#define rounded_rect_top_right(r) \
(GRAPHENE_RECT_INIT(r->bounds.origin.x + r->bounds.size.width - r->corner[1].width, \
r->bounds.origin.y, \
r->corner[1].width, r->corner[1].height))
#define rounded_rect_bottom_right(r) \
(GRAPHENE_RECT_INIT(r->bounds.origin.x + r->bounds.size.width - r->corner[2].width, \
r->bounds.origin.y + r->bounds.size.height - r->corner[2].height, \
r->corner[2].width, r->corner[2].height))
#define rounded_rect_bottom_left(r) \
(GRAPHENE_RECT_INIT(r->bounds.origin.x, \
r->bounds.origin.y + r->bounds.size.height - r->corner[2].height, \
r->corner[3].width, r->corner[3].height))
#define rounded_rect_corner0(r) rounded_rect_top_left(r)
#define rounded_rect_corner1(r) rounded_rect_top_right(r)
#define rounded_rect_corner2(r) rounded_rect_bottom_right(r)
#define rounded_rect_corner3(r) rounded_rect_bottom_left(r)
#define rounded_rect_corner(r, i) (rounded_rect_corner##i(r))
#define ALPHA_IS_CLEAR(alpha) ((alpha) < ((float) 0x00ff / (float) 0xffff))
#define RGBA_IS_CLEAR(rgba) ALPHA_IS_CLEAR((rgba)->alpha)
typedef struct _GskNglRenderClip
{
GskRoundedRect rect;
guint is_rectilinear : 1;
ngl: Classify clips Maintain the 'fully contained' information in the clip stack. This will let us pick simpler shaders in the future.
2021-03-07 15:02:46 +00:00
guint is_fully_contained : 1;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
} GskNglRenderClip;
typedef struct _GskNglRenderModelview
{
GskTransform *transform;
float scale_x;
float scale_y;
float offset_x_before;
float offset_y_before;
graphene_matrix_t matrix;
} GskNglRenderModelview;
struct _GskNglRenderJob
{
/* The context containing the framebuffer we are drawing to. Generally this
* is the context of the surface but may be a shared context if rendering to
* an offscreen texture such as gsk_ngl_renderer_render_texture().
*/
GdkGLContext *context;
/* The driver to be used. This is shared among all the renderers on a given
* GdkDisplay and uses the shared GL context to send commands.
*/
GskNglDriver *driver;
/* The command queue (which is just a faster pointer to the driver's
* command queue.
*/
GskNglCommandQueue *command_queue;
/* The region that we are clipping. Normalized to a single rectangle region. */
cairo_region_t *region;
/* The framebuffer to draw to in the @context GL context. So 0 would be the
* default framebuffer of @context. This is important to note as many other
* operations could be done using objects shared from the command queues
* GL context.
*/
guint framebuffer;
/* The viewport we are using. This state is updated as we process render
* nodes in the specific visitor callbacks.
*/
graphene_rect_t viewport;
/* The current projection, updated as we process nodes */
graphene_matrix_t projection;
/* An array of GskNglRenderModelview updated as nodes are processed. The
* current modelview is the last element.
*/
GArray *modelview;
/* An array of GskNglRenderClip updated as nodes are processed. The
* current clip is the last element.
*/
GArray *clip;
/* Our current alpha state as we process nodes */
float alpha;
/* Offset (delta x,y) as we process nodes. Occasionally this is merged into
* a transform that is referenced from child transform nodes.
*/
float offset_x;
float offset_y;
/* The scale we are processing, possibly updated by transforms */
float scale_x;
float scale_y;
/* Cached pointers */
const GskNglRenderClip *current_clip;
const GskNglRenderModelview *current_modelview;
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
GskNglProgram *current_program;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
/* If we should be rendering red zones over fallback nodes */
guint debug_fallback : 1;
};
typedef struct _GskNglRenderOffscreen
{
ngl: Add sonme comments The GskNglRenderOffscreen struct is a mix of in and out arguments. Better annotate that a little bit.
2021-03-31 18:32:47 +00:00
/* The bounds to render */
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
const graphene_rect_t *bounds;
ngl: Add sonme comments The GskNglRenderOffscreen struct is a mix of in and out arguments. Better annotate that a little bit.
2021-03-31 18:32:47 +00:00
/* Return location for texture coordinates */
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
struct {
float x;
float y;
float x2;
float y2;
} area;
ngl: Add sonme comments The GskNglRenderOffscreen struct is a mix of in and out arguments. Better annotate that a little bit.
2021-03-31 18:32:47 +00:00
/* Return location for texture ID */
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
guint texture_id;
ngl: Add sonme comments The GskNglRenderOffscreen struct is a mix of in and out arguments. Better annotate that a little bit.
2021-03-31 18:32:47 +00:00
/* Whether to force creating a new texture, even if the
* input already is a texture
*/
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
guint force_offscreen : 1;
guint reset_clip : 1;
guint do_not_cache : 1;
guint linear_filter : 1;
ngl: Add sonme comments The GskNglRenderOffscreen struct is a mix of in and out arguments. Better annotate that a little bit.
2021-03-31 18:32:47 +00:00
/* Return location for whether we created a texture */
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
guint was_offscreen : 1;
} GskNglRenderOffscreen;
static void gsk_ngl_render_job_visit_node (GskNglRenderJob *job,
const GskRenderNode *node);
static gboolean gsk_ngl_render_job_visit_node_with_offscreen (GskNglRenderJob *job,
const GskRenderNode *node,
GskNglRenderOffscreen *offscreen);
static inline void
init_full_texture_region (GskNglRenderOffscreen *offscreen)
{
offscreen->area.x = 0;
offscreen->area.y = 0;
offscreen->area.x2 = 1;
offscreen->area.y2 = 1;
}
static inline gboolean G_GNUC_PURE
node_is_invisible (const GskRenderNode *node)
{
return node->bounds.size.width == 0.0f ||
node->bounds.size.height == 0.0f;
}
ngl: Small optimization We really want equality here.
2021-04-11 00:26:45 +00:00
static inline gboolean G_GNUC_PURE
rounded_rect_equal (const GskRoundedRect *r1,
const GskRoundedRect *r2)
{
return memcmp (r1, r2, sizeof (GskRoundedRect)) == 0;
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
static inline void
gsk_rounded_rect_shrink_to_minimum (GskRoundedRect *self)
{
self->bounds.size.width = MAX (self->corner[0].width + self->corner[1].width,
self->corner[3].width + self->corner[2].width);
self->bounds.size.height = MAX (self->corner[0].height + self->corner[3].height,
self->corner[1].height + self->corner[2].height);
}
static inline gboolean G_GNUC_PURE
node_supports_transform (const GskRenderNode *node)
{
/* Some nodes can't handle non-trivial transforms without being
* rendered to a texture (e.g. rotated clips, etc.). Some however work
* just fine, mostly because they already draw their child to a
* texture and just render the texture manipulated in some way, think
* opacity or color matrix.
*/
switch ((int)gsk_render_node_get_node_type (node))
{
case GSK_COLOR_NODE:
case GSK_OPACITY_NODE:
case GSK_COLOR_MATRIX_NODE:
case GSK_TEXTURE_NODE:
case GSK_CROSS_FADE_NODE:
case GSK_LINEAR_GRADIENT_NODE:
case GSK_DEBUG_NODE:
case GSK_TEXT_NODE:
return TRUE;
case GSK_TRANSFORM_NODE:
return node_supports_transform (gsk_transform_node_get_child (node));
default:
return FALSE;
}
}
static inline gboolean G_GNUC_PURE
color_matrix_modifies_alpha (const GskRenderNode *node)
{
const graphene_matrix_t *matrix = gsk_color_matrix_node_get_color_matrix (node);
const graphene_vec4_t *offset = gsk_color_matrix_node_get_color_offset (node);
graphene_vec4_t row3;
if (graphene_vec4_get_w (offset) != 0.0f)
return TRUE;
graphene_matrix_get_row (matrix, 3, &row3);
return !graphene_vec4_equal (graphene_vec4_w_axis (), &row3);
}
static inline gboolean G_GNUC_PURE
rect_contains_rect (const graphene_rect_t *r1,
const graphene_rect_t *r2)
{
return r2->origin.x >= r1->origin.x &&
(r2->origin.x + r2->size.width) <= (r1->origin.x + r1->size.width) &&
r2->origin.y >= r1->origin.y &&
(r2->origin.y + r2->size.height) <= (r1->origin.y + r1->size.height);
}
static inline gboolean
rounded_inner_rect_contains_rect (const GskRoundedRect *rounded,
const graphene_rect_t *rect)
{
const graphene_rect_t *rounded_bounds = &rounded->bounds;
graphene_rect_t inner;
float offset_x;
float offset_y;
/* TODO: This is pretty conservative and we could go further,
* more fine-grained checks to avoid offscreen drawing.
*/
offset_x = MAX (rounded->corner[GSK_CORNER_TOP_LEFT].width,
rounded->corner[GSK_CORNER_BOTTOM_LEFT].width);
offset_y = MAX (rounded->corner[GSK_CORNER_TOP_LEFT].height,
rounded->corner[GSK_CORNER_TOP_RIGHT].height);
inner.origin.x = rounded_bounds->origin.x + offset_x;
inner.origin.y = rounded_bounds->origin.y + offset_y;
inner.size.width = rounded_bounds->size.width - offset_x -
MAX (rounded->corner[GSK_CORNER_TOP_RIGHT].width,
rounded->corner[GSK_CORNER_BOTTOM_RIGHT].width);
inner.size.height = rounded_bounds->size.height - offset_y -
MAX (rounded->corner[GSK_CORNER_BOTTOM_LEFT].height,
rounded->corner[GSK_CORNER_BOTTOM_RIGHT].height);
return rect_contains_rect (&inner, rect);
}
static inline gboolean G_GNUC_PURE
rect_intersects (const graphene_rect_t *r1,
const graphene_rect_t *r2)
{
/* Assume both rects are already normalized, as they usually are */
if (r1->origin.x > (r2->origin.x + r2->size.width) ||
(r1->origin.x + r1->size.width) < r2->origin.x)
return FALSE;
else if (r1->origin.y > (r2->origin.y + r2->size.height) ||
(r1->origin.y + r1->size.height) < r2->origin.y)
return FALSE;
else
return TRUE;
}
static inline gboolean
rounded_rect_has_corner (const GskRoundedRect *r,
guint i)
{
return r->corner[i].width > 0 && r->corner[i].height > 0;
}
/* Current clip is NOT rounded but new one is definitely! */
static inline gboolean
intersect_rounded_rectilinear (const graphene_rect_t *non_rounded,
const GskRoundedRect *rounded,
GskRoundedRect *result)
{
gboolean corners[4];
/* Intersects with top left corner? */
corners[0] = rounded_rect_has_corner (rounded, 0) &&
rect_intersects (non_rounded,
&rounded_rect_corner (rounded, 0));
/* top right? */
corners[1] = rounded_rect_has_corner (rounded, 1) &&
rect_intersects (non_rounded,
&rounded_rect_corner (rounded, 1));
/* bottom right? */
corners[2] = rounded_rect_has_corner (rounded, 2) &&
rect_intersects (non_rounded,
&rounded_rect_corner (rounded, 2));
/* bottom left */
corners[3] = rounded_rect_has_corner (rounded, 3) &&
rect_intersects (non_rounded,
&rounded_rect_corner (rounded, 3));
if (corners[0] && !rect_contains_rect (non_rounded, &rounded_rect_corner (rounded, 0)))
return FALSE;
if (corners[1] && !rect_contains_rect (non_rounded, &rounded_rect_corner (rounded, 1)))
return FALSE;
if (corners[2] && !rect_contains_rect (non_rounded, &rounded_rect_corner (rounded, 2)))
return FALSE;
if (corners[3] && !rect_contains_rect (non_rounded, &rounded_rect_corner (rounded, 3)))
return FALSE;
/* We do intersect with at least one of the corners, but in such a way that the
* intersection between the two clips can still be represented by a single rounded
* rect in a trivial way. do that. */
graphene_rect_intersection (non_rounded, &rounded->bounds, &result->bounds);
for (guint i = 0; i < 4; i++)
{
if (corners[i])
result->corner[i] = rounded->corner[i];
else
result->corner[i].width = result->corner[i].height = 0;
}
return TRUE;
}
static inline void
init_projection_matrix (graphene_matrix_t *projection,
const graphene_rect_t *viewport)
{
graphene_matrix_init_ortho (projection,
viewport->origin.x,
viewport->origin.x + viewport->size.width,
viewport->origin.y,
viewport->origin.y + viewport->size.height,
ORTHO_NEAR_PLANE,
ORTHO_FAR_PLANE);
graphene_matrix_scale (projection, 1, -1, 1);
}
static inline float
gsk_ngl_render_job_set_alpha (GskNglRenderJob *job,
float alpha)
{
if (job->alpha != alpha)
{
float ret = job->alpha;
job->alpha = alpha;
job->driver->stamps[UNIFORM_SHARED_ALPHA]++;
return ret;
}
return alpha;
}
static void
extract_matrix_metadata (GskNglRenderModelview *modelview)
{
float dummy;
gsk_transform_to_matrix (modelview->transform, &modelview->matrix);
switch (gsk_transform_get_category (modelview->transform))
{
case GSK_TRANSFORM_CATEGORY_IDENTITY:
case GSK_TRANSFORM_CATEGORY_2D_TRANSLATE:
modelview->scale_x = 1;
modelview->scale_y = 1;
break;
case GSK_TRANSFORM_CATEGORY_2D_AFFINE:
gsk_transform_to_affine (modelview->transform,
&modelview->scale_x, &modelview->scale_y,
&dummy, &dummy);
break;
case GSK_TRANSFORM_CATEGORY_UNKNOWN:
case GSK_TRANSFORM_CATEGORY_ANY:
case GSK_TRANSFORM_CATEGORY_3D:
case GSK_TRANSFORM_CATEGORY_2D:
{
graphene_vec3_t col1;
graphene_vec3_t col2;
/* TODO: 90% sure this is incorrect. But we should never hit this code
* path anyway. */
graphene_vec3_init (&col1,
ngl: Fix an uninitialized read We were using the wrong matrix here. valgrind pointed this point.
2021-03-27 20:23:40 +00:00
graphene_matrix_get_value (&modelview->matrix, 0, 0),
graphene_matrix_get_value (&modelview->matrix, 1, 0),
graphene_matrix_get_value (&modelview->matrix, 2, 0));
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
graphene_vec3_init (&col2,
ngl: Fix an uninitialized read We were using the wrong matrix here. valgrind pointed this point.
2021-03-27 20:23:40 +00:00
graphene_matrix_get_value (&modelview->matrix, 0, 1),
graphene_matrix_get_value (&modelview->matrix, 1, 1),
graphene_matrix_get_value (&modelview->matrix, 2, 1));
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
modelview->scale_x = graphene_vec3_length (&col1);
modelview->scale_y = graphene_vec3_length (&col2);
}
break;
default:
break;
}
}
static void
gsk_ngl_render_job_set_modelview (GskNglRenderJob *job,
GskTransform *transform)
{
GskNglRenderModelview *modelview;
g_assert (job != NULL);
g_assert (job->modelview != NULL);
job->driver->stamps[UNIFORM_SHARED_MODELVIEW]++;
g_array_set_size (job->modelview, job->modelview->len + 1);
modelview = &g_array_index (job->modelview,
GskNglRenderModelview,
job->modelview->len - 1);
modelview->transform = transform;
modelview->offset_x_before = job->offset_x;
modelview->offset_y_before = job->offset_y;
extract_matrix_metadata (modelview);
job->offset_x = 0;
job->offset_y = 0;
job->scale_x = modelview->scale_x;
job->scale_y = modelview->scale_y;
job->current_modelview = modelview;
}
static void
gsk_ngl_render_job_push_modelview (GskNglRenderJob *job,
GskTransform *transform)
{
GskNglRenderModelview *modelview;
g_assert (job != NULL);
g_assert (job->modelview != NULL);
g_assert (transform != NULL);
job->driver->stamps[UNIFORM_SHARED_MODELVIEW]++;
g_array_set_size (job->modelview, job->modelview->len + 1);
modelview = &g_array_index (job->modelview,
GskNglRenderModelview,
job->modelview->len - 1);
if G_LIKELY (job->modelview->len > 1)
{
GskNglRenderModelview *last;
GskTransform *t = NULL;
last = &g_array_index (job->modelview,
GskNglRenderModelview,
job->modelview->len - 2);
/* Multiply given matrix with our previews modelview */
t = gsk_transform_translate (gsk_transform_ref (last->transform),
&(graphene_point_t) {
job->offset_x,
job->offset_y
});
t = gsk_transform_transform (t, transform);
modelview->transform = t;
}
else
{
modelview->transform = gsk_transform_ref (transform);
}
modelview->offset_x_before = job->offset_x;
modelview->offset_y_before = job->offset_y;
extract_matrix_metadata (modelview);
job->offset_x = 0;
job->offset_y = 0;
job->scale_x = modelview->scale_x;
job->scale_y = modelview->scale_y;
job->current_modelview = modelview;
}
static void
gsk_ngl_render_job_pop_modelview (GskNglRenderJob *job)
{
const GskNglRenderModelview *head;
g_assert (job != NULL);
g_assert (job->modelview);
g_assert (job->modelview->len > 0);
job->driver->stamps[UNIFORM_SHARED_MODELVIEW]++;
head = job->current_modelview;
job->offset_x = head->offset_x_before;
job->offset_y = head->offset_y_before;
gsk_transform_unref (head->transform);
job->modelview->len--;
if (job->modelview->len >= 1)
{
head = &g_array_index (job->modelview, GskNglRenderModelview, job->modelview->len - 1);
job->scale_x = head->scale_x;
job->scale_y = head->scale_y;
job->current_modelview = head;
}
else
{
job->current_modelview = NULL;
}
}
static void
gsk_ngl_render_job_push_clip (GskNglRenderJob *job,
const GskRoundedRect *rect)
{
GskNglRenderClip *clip;
g_assert (job != NULL);
g_assert (job->clip != NULL);
g_assert (rect != NULL);
job->driver->stamps[UNIFORM_SHARED_CLIP_RECT]++;
g_array_set_size (job->clip, job->clip->len + 1);
clip = &g_array_index (job->clip, GskNglRenderClip, job->clip->len - 1);
memcpy (&clip->rect, rect, sizeof *rect);
clip->is_rectilinear = gsk_rounded_rect_is_rectilinear (rect);
ngl: Classify clips Maintain the 'fully contained' information in the clip stack. This will let us pick simpler shaders in the future.
2021-03-07 15:02:46 +00:00
clip->is_fully_contained = FALSE;
job->current_clip = clip;
}
static void
gsk_ngl_render_job_push_contained_clip (GskNglRenderJob *job)
{
GskNglRenderClip *clip;
GskNglRenderClip *old_clip;
g_assert (job != NULL);
g_assert (job->clip != NULL);
g_assert (job->clip->len > 0);
job->driver->stamps[UNIFORM_SHARED_CLIP_RECT]++;
old_clip = &g_array_index (job->clip, GskNglRenderClip, job->clip->len - 1);
g_array_set_size (job->clip, job->clip->len + 1);
clip = &g_array_index (job->clip, GskNglRenderClip, job->clip->len - 1);
memcpy (&clip->rect.bounds, &old_clip->rect.bounds, sizeof (graphene_rect_t));
memset (clip->rect.corner, 0, sizeof clip->rect.corner);
clip->is_rectilinear = TRUE;
clip->is_fully_contained = TRUE;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
job->current_clip = clip;
}
static void
gsk_ngl_render_job_pop_clip (GskNglRenderJob *job)
{
g_assert (job != NULL);
g_assert (job->clip != NULL);
g_assert (job->clip->len > 0);
job->driver->stamps[UNIFORM_SHARED_CLIP_RECT]++;
job->current_clip--;
job->clip->len--;
}
static inline void
gsk_ngl_render_job_offset (GskNglRenderJob *job,
float offset_x,
float offset_y)
{
if (offset_x || offset_y)
{
job->offset_x += offset_x;
job->offset_y += offset_y;
}
}
static inline void
gsk_ngl_render_job_set_projection (GskNglRenderJob *job,
const graphene_matrix_t *projection)
{
memcpy (&job->projection, projection, sizeof job->projection);
job->driver->stamps[UNIFORM_SHARED_PROJECTION]++;
}
static inline void
gsk_ngl_render_job_set_projection_from_rect (GskNglRenderJob *job,
const graphene_rect_t *rect,
graphene_matrix_t *prev_projection)
{
if (prev_projection)
memcpy (prev_projection, &job->projection, sizeof *prev_projection);
init_projection_matrix (&job->projection, rect);
job->driver->stamps[UNIFORM_SHARED_PROJECTION]++;
}
static inline void
gsk_ngl_render_job_set_projection_for_size (GskNglRenderJob *job,
float width,
float height,
graphene_matrix_t *prev_projection)
{
if (prev_projection)
memcpy (prev_projection, &job->projection, sizeof *prev_projection);
graphene_matrix_init_ortho (&job->projection, 0, width, 0, height, ORTHO_NEAR_PLANE, ORTHO_FAR_PLANE);
graphene_matrix_scale (&job->projection, 1, -1, 1);
job->driver->stamps[UNIFORM_SHARED_PROJECTION]++;
}
static inline void
gsk_ngl_render_job_set_viewport (GskNglRenderJob *job,
const graphene_rect_t *viewport,
graphene_rect_t *prev_viewport)
{
if (prev_viewport)
memcpy (prev_viewport, &job->viewport, sizeof *prev_viewport);
memcpy (&job->viewport, viewport, sizeof job->viewport);
job->driver->stamps[UNIFORM_SHARED_VIEWPORT]++;
}
static inline void
gsk_ngl_render_job_set_viewport_for_size (GskNglRenderJob *job,
float width,
float height,
graphene_rect_t *prev_viewport)
{
if (prev_viewport)
memcpy (prev_viewport, &job->viewport, sizeof *prev_viewport);
job->viewport.origin.x = 0;
job->viewport.origin.y = 0;
job->viewport.size.width = width;
job->viewport.size.height = height;
job->driver->stamps[UNIFORM_SHARED_VIEWPORT]++;
}
static inline void
gsk_ngl_render_job_transform_bounds (GskNglRenderJob *job,
const graphene_rect_t *rect,
graphene_rect_t *out_rect)
{
GskTransform *transform;
GskTransformCategory category;
g_assert (job != NULL);
g_assert (job->modelview->len > 0);
g_assert (rect != NULL);
g_assert (out_rect != NULL);
transform = job->current_modelview->transform;
category = gsk_transform_get_category (transform);
/* Our most common transform is 2d-affine, so inline it.
* Both identity and 2d-translate are virtually unseen here.
*/
ngl: Fix an oversight We were special-casing 2D affine transforms, but overlooked even simpler transforms.
2021-04-05 03:44:26 +00:00
if G_LIKELY (category >= GSK_TRANSFORM_CATEGORY_2D_AFFINE)
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
{
float dx, dy, scale_x, scale_y;
gsk_transform_to_affine (transform, &scale_x, &scale_y, &dx, &dy);
/* Init directly into out rect */
out_rect->origin.x = ((rect->origin.x + job->offset_x) * scale_x) + dx;
out_rect->origin.y = ((rect->origin.y + job->offset_y) * scale_y) + dy;
out_rect->size.width = rect->size.width * scale_x;
out_rect->size.height = rect->size.height * scale_y;
ngl: Typo fix
2021-03-28 17:27:09 +00:00
/* Normalize in place */
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (out_rect->size.width < 0.f)
{
float size = fabsf (out_rect->size.width);
out_rect->origin.x -= size;
out_rect->size.width = size;
}
if (out_rect->size.height < 0.f)
{
float size = fabsf (out_rect->size.height);
out_rect->origin.y -= size;
out_rect->size.height = size;
}
}
else
{
graphene_rect_t r;
r.origin.x = rect->origin.x + job->offset_x;
r.origin.y = rect->origin.y + job->offset_y;
r.size.width = rect->size.width;
r.size.height = rect->size.height;
gsk_transform_transform_bounds (transform, &r, out_rect);
}
}
static inline void
gsk_ngl_render_job_transform_rounded_rect (GskNglRenderJob *job,
const GskRoundedRect *rect,
GskRoundedRect *out_rect)
{
out_rect->bounds.origin.x = job->offset_x + rect->bounds.origin.x;
out_rect->bounds.origin.y = job->offset_y + rect->bounds.origin.y;
out_rect->bounds.size.width = rect->bounds.size.width;
out_rect->bounds.size.height = rect->bounds.size.height;
memcpy (out_rect->corner, rect->corner, sizeof rect->corner);
}
ngl: Classify clips Maintain the 'fully contained' information in the clip stack. This will let us pick simpler shaders in the future.
2021-03-07 15:02:46 +00:00
static inline void
rounded_rect_get_inner (const GskRoundedRect *rect,
graphene_rect_t *inner)
{
float left = MAX (rect->corner[GSK_CORNER_TOP_LEFT].width, rect->corner[GSK_CORNER_BOTTOM_LEFT].width);
float right = MAX (rect->corner[GSK_CORNER_TOP_RIGHT].width, rect->corner[GSK_CORNER_BOTTOM_RIGHT].width);
float top = MAX (rect->corner[GSK_CORNER_TOP_LEFT].height, rect->corner[GSK_CORNER_TOP_RIGHT].height);
float bottom = MAX (rect->corner[GSK_CORNER_BOTTOM_LEFT].height, rect->corner[GSK_CORNER_BOTTOM_RIGHT].height);
inner->origin.x = rect->bounds.origin.x + left;
inner->size.width = rect->bounds.size.width - (left + right);
inner->origin.y = rect->bounds.origin.y + top;
inner->size.height = rect->bounds.size.height - (top + bottom);
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
static inline gboolean
ngl: Classify clips Maintain the 'fully contained' information in the clip stack. This will let us pick simpler shaders in the future.
2021-03-07 15:02:46 +00:00
interval_contains (float p1, float w1,
float p2, float w2)
{
if (p2 < p1)
return FALSE;
if (p2 + w2 > p1 + w1)
return FALSE;
return TRUE;
}
static inline gboolean
ngl: Allow passing bounds to update_clip update_clip only uses the bounds of the node that we are currently passing. This opens the door to updating the clip while handling a single node.
2021-04-11 15:57:22 +00:00
gsk_ngl_render_job_update_clip (GskNglRenderJob *job,
const graphene_rect_t *bounds,
gboolean *pushed_clip)
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
{
graphene_rect_t transformed_bounds;
ngl: Classify clips Maintain the 'fully contained' information in the clip stack. This will let us pick simpler shaders in the future.
2021-03-07 15:02:46 +00:00
gboolean no_clip = FALSE;
gboolean rect_clip = FALSE;
*pushed_clip = FALSE;
if (job->current_clip->is_fully_contained)
{
/* Already fully contained - no further checks needed */
return TRUE;
}
ngl: Allow passing bounds to update_clip update_clip only uses the bounds of the node that we are currently passing. This opens the door to updating the clip while handling a single node.
2021-04-11 15:57:22 +00:00
gsk_ngl_render_job_transform_bounds (job, bounds, &transformed_bounds);
ngl: Classify clips Maintain the 'fully contained' information in the clip stack. This will let us pick simpler shaders in the future.
2021-03-07 15:02:46 +00:00
if (!rect_intersects (&job->current_clip->rect.bounds, &transformed_bounds))
{
/* Completely clipped away */
return FALSE;
}
if (job->current_clip->is_rectilinear)
{
if (rect_contains_rect (&job->current_clip->rect.bounds, &transformed_bounds))
no_clip = TRUE;
else
rect_clip = TRUE;
}
else if (gsk_rounded_rect_contains_rect (&job->current_clip->rect, &transformed_bounds))
{
no_clip = TRUE;
}
else
{
graphene_rect_t inner;
rounded_rect_get_inner (&job->current_clip->rect, &inner);
if (interval_contains (inner.origin.x, inner.size.width,
transformed_bounds.origin.x, transformed_bounds.size.width) ||
interval_contains (inner.origin.y, inner.size.height,
transformed_bounds.origin.y, transformed_bounds.size.height))
rect_clip = TRUE;
}
if (no_clip)
{
/* This node is completely contained inside the clip.
* Record this fact on the clip stack, so we don't do more work
* for child nodes.
*/
gsk_ngl_render_job_push_contained_clip (job);
*pushed_clip = TRUE;
}
else if (rect_clip && !job->current_clip->is_rectilinear)
{
graphene_rect_t rect;
/* The clip gets simpler for this node */
graphene_rect_intersection (&job->current_clip->rect.bounds, &transformed_bounds, &rect);
gsk_ngl_render_job_push_clip (job, &GSK_ROUNDED_RECT_INIT_FROM_RECT (rect));
*pushed_clip = TRUE;
}
return TRUE;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
static inline void
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (const GdkRGBA *rgba,
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
guint16 h[4])
{
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
float_to_half4 ((const float *)rgba, h);
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
}
ngl: Set color as vertex attribute Instead of using uniforms for color used in multiple programs, pass it as vertex attributes. This will let us batch more draw calls, since we don't have to change uniforms so often. In particular, for syntax-highlighted text.
2021-03-12 00:46:15 +00:00
/* fill_vertex_data */
static void
gsk_ngl_render_job_draw_coords (GskNglRenderJob *job,
float min_x,
float min_y,
float max_x,
float max_y,
float min_u,
float min_v,
float max_u,
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
float max_v,
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 c[4])
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
{
GskNglDrawVertex *vertices = gsk_ngl_command_queue_add_vertices (job->command_queue);
ngl: Reduce vertex storage We only have one shader that uses the color2 attribute, and it doesn't use the uv attribute, so save vertex memory by putting those in the same space. This reduce the per vertex space from 32 to 24 bytes.
2021-04-07 02:24:37 +00:00
vertices[0] = (GskNglDrawVertex) { .position = { min_x, min_y }, .uv = { min_u, min_v }, .color = { c[0], c[1], c[2], c[3] } };
vertices[1] = (GskNglDrawVertex) { .position = { min_x, max_y }, .uv = { min_u, max_v }, .color = { c[0], c[1], c[2], c[3] } };
vertices[2] = (GskNglDrawVertex) { .position = { max_x, min_y }, .uv = { max_u, min_v }, .color = { c[0], c[1], c[2], c[3] } };
vertices[3] = (GskNglDrawVertex) { .position = { max_x, max_y }, .uv = { max_u, max_v }, .color = { c[0], c[1], c[2], c[3] } };
vertices[4] = (GskNglDrawVertex) { .position = { min_x, max_y }, .uv = { min_u, max_v }, .color = { c[0], c[1], c[2], c[3] } };
vertices[5] = (GskNglDrawVertex) { .position = { max_x, min_y }, .uv = { max_u, min_v }, .color = { c[0], c[1], c[2], c[3] } };
ngl: Set color as vertex attribute Instead of using uniforms for color used in multiple programs, pass it as vertex attributes. This will let us batch more draw calls, since we don't have to change uniforms so often. In particular, for syntax-highlighted text.
2021-03-12 00:46:15 +00:00
}
/* load_vertex_data_with_region */
static inline void
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (GskNglRenderJob *job,
const graphene_rect_t *bounds,
const GskNglRenderOffscreen *offscreen,
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4])
ngl: Set color as vertex attribute Instead of using uniforms for color used in multiple programs, pass it as vertex attributes. This will let us batch more draw calls, since we don't have to change uniforms so often. In particular, for syntax-highlighted text.
2021-03-12 00:46:15 +00:00
{
float min_x = job->offset_x + bounds->origin.x;
float min_y = job->offset_y + bounds->origin.y;
float max_x = min_x + bounds->size.width;
float max_y = min_y + bounds->size.height;
float y1 = offscreen->was_offscreen ? offscreen->area.y2 : offscreen->area.y;
float y2 = offscreen->was_offscreen ? offscreen->area.y : offscreen->area.y2;
gsk_ngl_render_job_draw_coords (job,
min_x, min_y, max_x, max_y,
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
offscreen->area.x, y1, offscreen->area.x2, y2,
color);
}
static inline void
gsk_ngl_render_job_draw_offscreen (GskNglRenderJob *job,
const graphene_rect_t *bounds,
const GskNglRenderOffscreen *offscreen)
{
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (job, bounds, offscreen,
(guint16[]) { FP16_ZERO, FP16_ZERO, FP16_ZERO, FP16_ZERO });
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
/* load_float_vertex_data */
static inline void
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_with_color (GskNglRenderJob *job,
float x,
float y,
float width,
float height,
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4])
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
{
float min_x = job->offset_x + x;
float min_y = job->offset_y + y;
float max_x = min_x + width;
float max_y = min_y + height;
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_coords (job, min_x, min_y, max_x, max_y, 0, 0, 1, 1, color);
}
static inline void
gsk_ngl_render_job_draw (GskNglRenderJob *job,
float x,
float y,
float width,
float height)
{
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
gsk_ngl_render_job_draw_with_color (job, x, y, width, height,
(guint16[]) { FP_ZERO, FP_ZERO, FP_ZERO, FP_ZERO });
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
/* load_vertex_data */
static inline void
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_rect_with_color (GskNglRenderJob *job,
const graphene_rect_t *bounds,
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4])
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
{
gsk_ngl_render_job_draw_with_color (job,
bounds->origin.x,
bounds->origin.y,
bounds->size.width,
bounds->size.height,
color);
}
static inline void
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_draw_rect (GskNglRenderJob *job,
const graphene_rect_t *bounds)
{
gsk_ngl_render_job_draw (job,
bounds->origin.x,
bounds->origin.y,
bounds->size.width,
bounds->size.height);
}
/* load_offscreen_vertex_data */
static inline void
gsk_ngl_render_job_draw_offscreen_rect (GskNglRenderJob *job,
const graphene_rect_t *bounds)
{
float min_x = job->offset_x + bounds->origin.x;
float min_y = job->offset_y + bounds->origin.y;
float max_x = min_x + bounds->size.width;
float max_y = min_y + bounds->size.height;
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4] = { FP16_ZERO, FP16_ZERO, FP16_ZERO, FP16_ZERO };
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_coords (job,
min_x, min_y, max_x, max_y,
0, 1, 1, 0,
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
static inline void
gsk_ngl_render_job_begin_draw (GskNglRenderJob *job,
GskNglProgram *program)
{
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
job->current_program = program;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_command_queue_begin_draw (job->command_queue,
program->program_info,
job->viewport.size.width,
job->viewport.size.height);
ngl: move uniform key mapping into uniform state Previously, we translated the uniform key (an enum) into a location within the shader program in GskNglProgram. A number of performance improvements were focused around having low nubers for the uniform locations. Generally this is the case, but some drivers such as old Intel drivers on Windows may use rather large numbers for those. To combat this, we can push the translation of uniform keys into locations at the GskNglUniformState level so that we work with unranslated keys through the process until applying them. Fixes #3780
2021-03-23 19:10:12 +00:00
gsk_ngl_uniform_state_set4fv (program->uniforms,
program->program_info,
UNIFORM_SHARED_VIEWPORT,
job->driver->stamps[UNIFORM_SHARED_VIEWPORT],
1,
(const float *)&job->viewport);
gsk_ngl_uniform_state_set_matrix (program->uniforms,
program->program_info,
UNIFORM_SHARED_MODELVIEW,
job->driver->stamps[UNIFORM_SHARED_MODELVIEW],
&job->current_modelview->matrix);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: move uniform key mapping into uniform state Previously, we translated the uniform key (an enum) into a location within the shader program in GskNglProgram. A number of performance improvements were focused around having low nubers for the uniform locations. Generally this is the case, but some drivers such as old Intel drivers on Windows may use rather large numbers for those. To combat this, we can push the translation of uniform keys into locations at the GskNglUniformState level so that we work with unranslated keys through the process until applying them. Fixes #3780
2021-03-23 19:10:12 +00:00
gsk_ngl_uniform_state_set_matrix (program->uniforms,
program->program_info,
UNIFORM_SHARED_PROJECTION,
job->driver->stamps[UNIFORM_SHARED_PROJECTION],
&job->projection);
gsk_ngl_uniform_state_set_rounded_rect (program->uniforms,
program->program_info,
UNIFORM_SHARED_CLIP_RECT,
job->driver->stamps[UNIFORM_SHARED_CLIP_RECT],
&job->current_clip->rect);
gsk_ngl_uniform_state_set1f (program->uniforms,
program->program_info,
UNIFORM_SHARED_ALPHA,
job->driver->stamps[UNIFORM_SHARED_ALPHA],
job->alpha);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
#define CHOOSE_PROGRAM(job,name) \
(job->current_clip->is_fully_contained \
? job->driver->name ## _no_clip \
: (job->current_clip->is_rectilinear \
? job->driver->name ## _rect_clip \
: job->driver->name))
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
static inline void
gsk_ngl_render_job_split_draw (GskNglRenderJob *job)
{
gsk_ngl_command_queue_split_draw (job->command_queue);
}
static inline void
gsk_ngl_render_job_end_draw (GskNglRenderJob *job)
{
gsk_ngl_command_queue_end_draw (job->command_queue);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
job->current_program = NULL;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
static inline void
gsk_ngl_render_job_visit_as_fallback (GskNglRenderJob *job,
const GskRenderNode *node)
{
float scale_x = job->scale_x;
float scale_y = job->scale_y;
int surface_width = ceilf (node->bounds.size.width * scale_x);
int surface_height = ceilf (node->bounds.size.height * scale_y);
GdkTexture *texture;
cairo_surface_t *surface;
cairo_surface_t *rendered_surface;
cairo_t *cr;
int cached_id;
int texture_id;
GskTextureKey key;
if (surface_width <= 0 || surface_height <= 0)
return;
key.pointer = node;
key.pointer_is_child = FALSE;
key.scale_x = scale_x;
key.scale_y = scale_y;
key.filter = GL_NEAREST;
cached_id = gsk_ngl_driver_lookup_texture (job->driver, &key);
if (cached_id != 0)
{
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D, GL_TEXTURE0, cached_id);
gsk_ngl_render_job_draw_offscreen_rect (job, &node->bounds);
gsk_ngl_render_job_end_draw (job);
return;
}
/* We first draw the recording surface on an image surface,
* just because the scaleY(-1) later otherwise screws up the
* rendering... */
{
rendered_surface = cairo_image_surface_create (CAIRO_FORMAT_ARGB32,
surface_width,
surface_height);
cairo_surface_set_device_scale (rendered_surface, scale_x, scale_y);
cr = cairo_create (rendered_surface);
cairo_save (cr);
cairo_translate (cr, - floorf (node->bounds.origin.x), - floorf (node->bounds.origin.y));
/* Render nodes don't modify state, so casting away the const is fine here */
gsk_render_node_draw ((GskRenderNode *)node, cr);
cairo_restore (cr);
cairo_destroy (cr);
}
surface = cairo_image_surface_create (CAIRO_FORMAT_ARGB32,
surface_width,
surface_height);
cairo_surface_set_device_scale (surface, scale_x, scale_y);
cr = cairo_create (surface);
/* We draw upside down here, so it matches what GL does. */
cairo_save (cr);
cairo_scale (cr, 1, -1);
cairo_translate (cr, 0, - surface_height / scale_y);
cairo_set_source_surface (cr, rendered_surface, 0, 0);
cairo_rectangle (cr, 0, 0, surface_width / scale_x, surface_height / scale_y);
cairo_fill (cr);
cairo_restore (cr);
#ifdef G_ENABLE_DEBUG
if (job->debug_fallback)
{
cairo_move_to (cr, 0, 0);
cairo_rectangle (cr, 0, 0, node->bounds.size.width, node->bounds.size.height);
if (gsk_render_node_get_node_type (node) == GSK_CAIRO_NODE)
cairo_set_source_rgba (cr, 0.3, 0, 1, 0.25);
else
cairo_set_source_rgba (cr, 1, 0, 0, 0.25);
cairo_fill_preserve (cr);
if (gsk_render_node_get_node_type (node) == GSK_CAIRO_NODE)
cairo_set_source_rgba (cr, 0.3, 0, 1, 1);
else
cairo_set_source_rgba (cr, 1, 0, 0, 1);
cairo_stroke (cr);
}
#endif
cairo_destroy (cr);
/* Create texture to upload */
texture = gdk_texture_new_for_surface (surface);
texture_id = gsk_ngl_driver_load_texture (job->driver, texture,
GL_NEAREST, GL_NEAREST);
if (gdk_gl_context_has_debug (job->command_queue->context))
gdk_gl_context_label_object_printf (job->command_queue->context, GL_TEXTURE, texture_id,
"Fallback %s %d",
g_type_name_from_instance ((GTypeInstance *) node),
texture_id);
g_object_unref (texture);
cairo_surface_destroy (surface);
cairo_surface_destroy (rendered_surface);
gsk_ngl_driver_cache_texture (job->driver, &key, texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
texture_id);
gsk_ngl_render_job_draw_offscreen_rect (job, &node->bounds);
gsk_ngl_render_job_end_draw (job);
}
static guint
blur_offscreen (GskNglRenderJob *job,
GskNglRenderOffscreen *offscreen,
int texture_to_blur_width,
int texture_to_blur_height,
float blur_radius_x,
float blur_radius_y)
{
const GskRoundedRect new_clip = GSK_ROUNDED_RECT_INIT (0, 0, texture_to_blur_width, texture_to_blur_height);
GskNglRenderTarget *pass1;
GskNglRenderTarget *pass2;
graphene_matrix_t prev_projection;
graphene_rect_t prev_viewport;
guint prev_fbo;
g_assert (blur_radius_x > 0);
g_assert (blur_radius_y > 0);
g_assert (offscreen->texture_id > 0);
g_assert (offscreen->area.x2 > offscreen->area.x);
g_assert (offscreen->area.y2 > offscreen->area.y);
if (!gsk_ngl_driver_create_render_target (job->driver,
MAX (texture_to_blur_width, 1),
MAX (texture_to_blur_height, 1),
GL_NEAREST, GL_NEAREST,
&pass1))
return 0;
if (texture_to_blur_width <= 0 || texture_to_blur_height <= 0)
return gsk_ngl_driver_release_render_target (job->driver, pass1, FALSE);
if (!gsk_ngl_driver_create_render_target (job->driver,
texture_to_blur_width,
texture_to_blur_height,
GL_NEAREST, GL_NEAREST,
&pass2))
return gsk_ngl_driver_release_render_target (job->driver, pass1, FALSE);
gsk_ngl_render_job_set_viewport (job, &new_clip.bounds, &prev_viewport);
gsk_ngl_render_job_set_projection_from_rect (job, &new_clip.bounds, &prev_projection);
gsk_ngl_render_job_set_modelview (job, NULL);
gsk_ngl_render_job_push_clip (job, &new_clip);
/* Bind new framebuffer and clear it */
prev_fbo = gsk_ngl_command_queue_bind_framebuffer (job->command_queue, pass1->framebuffer_id);
gsk_ngl_command_queue_clear (job->command_queue, 0, &job->viewport);
/* Begin drawing the first horizontal pass, using offscreen as the
* source texture for the program.
*/
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blur));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen->texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_BLUR_RADIUS, 0,
blur_radius_x);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform2f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_BLUR_SIZE, 0,
texture_to_blur_width,
texture_to_blur_height);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform2f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_BLUR_DIR, 0,
1, 0);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_coords (job,
0, 0, texture_to_blur_width, texture_to_blur_height,
0, 1, 1, 0,
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
(guint16[]) { FP16_ZERO, FP16_ZERO, FP16_ZERO, FP16_ZERO });
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
/* Bind second pass framebuffer and clear it */
gsk_ngl_command_queue_bind_framebuffer (job->command_queue, pass2->framebuffer_id);
gsk_ngl_command_queue_clear (job->command_queue, 0, &job->viewport);
/* Draw using blur program with first pass as source texture */
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blur));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
pass1->texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_BLUR_RADIUS, 0,
blur_radius_y);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform2f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_BLUR_SIZE, 0,
texture_to_blur_width,
texture_to_blur_height);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform2f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_BLUR_DIR, 0,
0, 1);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_coords (job,
0, 0, texture_to_blur_width, texture_to_blur_height,
0, 1, 1, 0,
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
(guint16[]) { FP16_ZERO, FP16_ZERO, FP16_ZERO, FP16_ZERO });
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
gsk_ngl_render_job_pop_modelview (job);
gsk_ngl_render_job_pop_clip (job);
gsk_ngl_render_job_set_viewport (job, &prev_viewport, NULL);
gsk_ngl_render_job_set_projection (job, &prev_projection);
gsk_ngl_command_queue_bind_framebuffer (job->command_queue, prev_fbo);
gsk_ngl_driver_release_render_target (job->driver, pass1, TRUE);
return gsk_ngl_driver_release_render_target (job->driver, pass2, FALSE);
}
static void
blur_node (GskNglRenderJob *job,
GskNglRenderOffscreen *offscreen,
const GskRenderNode *node,
float blur_radius,
float *min_x,
float *max_x,
float *min_y,
float *max_y)
{
const float blur_extra = blur_radius * 2.0; /* 2.0 = shader radius_multiplier */
const float half_blur_extra = (blur_extra / 2.0);
float scale_x = job->scale_x;
float scale_y = job->scale_y;
float texture_width;
float texture_height;
g_assert (blur_radius > 0);
/* Increase texture size for the given blur radius and scale it */
texture_width = ceilf ((node->bounds.size.width + blur_extra));
texture_height = ceilf ((node->bounds.size.height + blur_extra));
/* Only blur this if the out region has no texture id yet */
if (offscreen->texture_id == 0)
{
const graphene_rect_t bounds = GRAPHENE_RECT_INIT (node->bounds.origin.x - half_blur_extra,
node->bounds.origin.y - half_blur_extra,
texture_width, texture_height);
offscreen->bounds = &bounds;
offscreen->reset_clip = TRUE;
offscreen->force_offscreen = TRUE;
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, node, offscreen))
g_assert_not_reached ();
/* Ensure that we actually got a real texture_id */
g_assert (offscreen->texture_id != 0);
offscreen->texture_id = blur_offscreen (job,
offscreen,
texture_width * scale_x,
texture_height * scale_y,
blur_radius * scale_x,
blur_radius * scale_y);
init_full_texture_region (offscreen);
}
*min_x = job->offset_x + node->bounds.origin.x - half_blur_extra;
*max_x = job->offset_x + node->bounds.origin.x + node->bounds.size.width + half_blur_extra;
*min_y = job->offset_y + node->bounds.origin.y - half_blur_extra;
*max_y = job->offset_y + node->bounds.origin.y + node->bounds.size.height + half_blur_extra;
}
ngl: Optimize underlines in text Opportunistically use the coloring program for drawing underlines instead of the color program. This avoids program changes in the middle of text. For the Emoji text scrolling benchmark, this reduces the program changes per frame from > 1000 to around 100.
2021-04-09 03:11:06 +00:00
#define ATLAS_SIZE 512
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
static inline void
gsk_ngl_render_job_visit_color_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4];
ngl: Optimize underlines in text Opportunistically use the coloring program for drawing underlines instead of the color program. This avoids program changes in the middle of text. For the Emoji text scrolling benchmark, this reduces the program changes per frame from > 1000 to around 100.
2021-04-09 03:11:06 +00:00
GskNglProgram *program;
GskNglCommandBatch *batch;
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (gsk_color_node_get_color (node), color);
ngl: Optimize underlines in text Opportunistically use the coloring program for drawing underlines instead of the color program. This avoids program changes in the middle of text. For the Emoji text scrolling benchmark, this reduces the program changes per frame from > 1000 to around 100.
2021-04-09 03:11:06 +00:00
/* Avoid switching away from the coloring program for
* rendering a solid color.
*/
program = CHOOSE_PROGRAM (job, coloring);
batch = gsk_ngl_command_queue_get_batch (job->command_queue);
if (batch->any.kind == GSK_NGL_COMMAND_KIND_DRAW &&
batch->any.program == program->id)
{
GskNglRenderOffscreen offscreen = {0};
gsk_ngl_render_job_begin_draw (job, program);
/* The top left few pixels in our atlases are always
* solid white, so we can use it here, without
* having to choose any particular atlas texture.
*/
offscreen.was_offscreen = FALSE;
offscreen.area.x = 1.f / ATLAS_SIZE;
offscreen.area.y = 1.f / ATLAS_SIZE;
offscreen.area.x2 = 2.f / ATLAS_SIZE;
offscreen.area.y2 = 2.f / ATLAS_SIZE;
gsk_ngl_render_job_draw_offscreen_with_color (job,
&node->bounds,
&offscreen,
color);
gsk_ngl_render_job_end_draw (job);
}
else
{
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, color));
gsk_ngl_render_job_draw_rect_with_color (job, &node->bounds, color);
gsk_ngl_render_job_end_draw (job);
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
static inline void
gsk_ngl_render_job_visit_linear_gradient_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskColorStop *stops = gsk_linear_gradient_node_get_color_stops (node, NULL);
const graphene_point_t *start = gsk_linear_gradient_node_get_start (node);
const graphene_point_t *end = gsk_linear_gradient_node_get_end (node);
int n_color_stops = gsk_linear_gradient_node_get_n_color_stops (node);
gboolean repeat = gsk_render_node_get_node_type (node) == GSK_REPEATING_LINEAR_GRADIENT_NODE;
float x1 = job->offset_x + start->x;
float x2 = job->offset_x + end->x;
float y1 = job->offset_y + start->y;
float y2 = job->offset_y + end->y;
g_assert (n_color_stops < MAX_GRADIENT_STOPS);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, linear_gradient));
gsk_ngl_program_set_uniform1i (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_LINEAR_GRADIENT_NUM_COLOR_STOPS, 0,
n_color_stops);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1fv (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_LINEAR_GRADIENT_COLOR_STOPS, 0,
n_color_stops * 5,
(const float *)stops);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform4f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_LINEAR_GRADIENT_POINTS, 0,
x1, y1, x2 - x1, y2 - y1);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1i (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_LINEAR_GRADIENT_REPEAT, 0,
repeat);
gsk_ngl_render_job_draw_rect (job, &node->bounds);
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_conic_gradient_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
static const float scale = 0.5f * M_1_PI;
const GskColorStop *stops = gsk_conic_gradient_node_get_color_stops (node, NULL);
const graphene_point_t *center = gsk_conic_gradient_node_get_center (node);
int n_color_stops = gsk_conic_gradient_node_get_n_color_stops (node);
float angle = gsk_conic_gradient_node_get_angle (node);
float bias = angle * scale + 2.0f;
g_assert (n_color_stops < MAX_GRADIENT_STOPS);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, conic_gradient));
gsk_ngl_program_set_uniform1i (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_CONIC_GRADIENT_NUM_COLOR_STOPS, 0,
n_color_stops);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1fv (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_CONIC_GRADIENT_COLOR_STOPS, 0,
n_color_stops * 5,
(const float *)stops);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform4f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_CONIC_GRADIENT_GEOMETRY, 0,
job->offset_x + center->x,
job->offset_y + center->y,
scale,
bias);
gsk_ngl_render_job_draw_rect (job, &node->bounds);
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_radial_gradient_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
int n_color_stops = gsk_radial_gradient_node_get_n_color_stops (node);
const GskColorStop *stops = gsk_radial_gradient_node_get_color_stops (node, NULL);
const graphene_point_t *center = gsk_radial_gradient_node_get_center (node);
float start = gsk_radial_gradient_node_get_start (node);
float end = gsk_radial_gradient_node_get_end (node);
float hradius = gsk_radial_gradient_node_get_hradius (node);
float vradius = gsk_radial_gradient_node_get_vradius (node);
gboolean repeat = gsk_render_node_get_node_type (node) == GSK_REPEATING_RADIAL_GRADIENT_NODE;
float scale = 1.0f / (end - start);
float bias = -start * scale;
g_assert (n_color_stops < MAX_GRADIENT_STOPS);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, radial_gradient));
gsk_ngl_program_set_uniform1i (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_RADIAL_GRADIENT_NUM_COLOR_STOPS, 0,
n_color_stops);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1fv (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_RADIAL_GRADIENT_COLOR_STOPS, 0,
n_color_stops * 5,
(const float *)stops);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1i (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_RADIAL_GRADIENT_REPEAT, 0,
repeat);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform2f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_RADIAL_GRADIENT_RANGE, 0,
scale, bias);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform4f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_RADIAL_GRADIENT_GEOMETRY, 0,
job->offset_x + center->x,
job->offset_y + center->y,
1.0f / (hradius * job->scale_x),
1.0f / (vradius * job->scale_y));
gsk_ngl_render_job_draw_rect (job, &node->bounds);
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_clipped_child (GskNglRenderJob *job,
const GskRenderNode *child,
const graphene_rect_t *clip)
{
graphene_rect_t transformed_clip;
GskRoundedRect intersection;
gsk_ngl_render_job_transform_bounds (job, clip, &transformed_clip);
if (job->current_clip->is_rectilinear)
{
memset (&intersection.corner, 0, sizeof intersection.corner);
graphene_rect_intersection (&transformed_clip,
&job->current_clip->rect.bounds,
&intersection.bounds);
gsk_ngl_render_job_push_clip (job, &intersection);
gsk_ngl_render_job_visit_node (job, child);
gsk_ngl_render_job_pop_clip (job);
}
else if (intersect_rounded_rectilinear (&transformed_clip,
&job->current_clip->rect,
&intersection))
{
gsk_ngl_render_job_push_clip (job, &intersection);
gsk_ngl_render_job_visit_node (job, child);
gsk_ngl_render_job_pop_clip (job);
}
else
{
GskNglRenderOffscreen offscreen = {0};
ngl: Avoid huge intermediate textures Instead of rendering the unclipped child to a texture (and risking blowing the texture size limit, and bad downscaling), just render the clipped region, and live with the fact that we can't cache the rendered texture. This avoid bad artifacts when scrolling long textviews in rounded clips.
2021-04-02 23:33:44 +00:00
offscreen.bounds = clip;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
offscreen.force_offscreen = TRUE;
ngl: Avoid huge intermediate textures Instead of rendering the unclipped child to a texture (and risking blowing the texture size limit, and bad downscaling), just render the clipped region, and live with the fact that we can't cache the rendered texture. This avoid bad artifacts when scrolling long textviews in rounded clips.
2021-04-02 23:33:44 +00:00
offscreen.reset_clip = TRUE;
offscreen.do_not_cache = TRUE;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_visit_node_with_offscreen (job, child, &offscreen);
g_assert (offscreen.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen.texture_id);
ngl: Avoid huge intermediate textures Instead of rendering the unclipped child to a texture (and risking blowing the texture size limit, and bad downscaling), just render the clipped region, and live with the fact that we can't cache the rendered texture. This avoid bad artifacts when scrolling long textviews in rounded clips.
2021-04-02 23:33:44 +00:00
gsk_ngl_render_job_draw_offscreen_rect (job, clip);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
}
static inline void
gsk_ngl_render_job_visit_clip_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const graphene_rect_t *clip = gsk_clip_node_get_clip (node);
const GskRenderNode *child = gsk_clip_node_get_child (node);
gsk_ngl_render_job_visit_clipped_child (job, child, clip);
}
static inline void
gsk_ngl_render_job_visit_rounded_clip_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRenderNode *child = gsk_rounded_clip_node_get_child (node);
const GskRoundedRect *clip = gsk_rounded_clip_node_get_clip (node);
GskRoundedRect transformed_clip;
float scale_x = job->scale_x;
float scale_y = job->scale_y;
gboolean need_offscreen;
if (node_is_invisible (child))
return;
gsk_ngl_render_job_transform_bounds (job, &clip->bounds, &transformed_clip.bounds);
for (guint i = 0; i < G_N_ELEMENTS (transformed_clip.corner); i++)
{
transformed_clip.corner[i].width = clip->corner[i].width * scale_x;
transformed_clip.corner[i].height = clip->corner[i].height * scale_y;
}
if (job->current_clip->is_rectilinear)
{
GskRoundedRect intersected_clip;
if (intersect_rounded_rectilinear (&job->current_clip->rect.bounds,
&transformed_clip,
&intersected_clip))
{
gsk_ngl_render_job_push_clip (job, &intersected_clip);
gsk_ngl_render_job_visit_node (job, child);
gsk_ngl_render_job_pop_clip (job);
return;
}
}
/* After this point we are really working with a new and a current clip
* which both have rounded corners.
*/
if (job->clip->len <= 1)
need_offscreen = FALSE;
else if (rounded_inner_rect_contains_rect (&job->current_clip->rect, &transformed_clip.bounds))
need_offscreen = FALSE;
else
need_offscreen = TRUE;
if (!need_offscreen)
{
/* If the new clip entirely contains the current clip, the intersection is simply
* the current clip, so we can ignore the new one.
*/
if (rounded_inner_rect_contains_rect (&transformed_clip, &job->current_clip->rect.bounds))
{
gsk_ngl_render_job_visit_node (job, child);
return;
}
gsk_ngl_render_job_push_clip (job, &transformed_clip);
gsk_ngl_render_job_visit_node (job, child);
gsk_ngl_render_job_pop_clip (job);
}
else
{
GskNglRenderOffscreen offscreen = {0};
offscreen.bounds = &node->bounds;
offscreen.force_offscreen = TRUE;
ngl: Be explicit about offscreen clipping Explicitly set all the input arguments. This makes it clearer what is going on.
2021-03-31 18:34:16 +00:00
offscreen.reset_clip = FALSE;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_push_clip (job, &transformed_clip);
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, child, &offscreen))
g_assert_not_reached ();
gsk_ngl_render_job_pop_clip (job);
g_assert (offscreen.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen.texture_id);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen (job, &node->bounds, &offscreen);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
}
ngl: Render simple borders simply Use 4 color commands for uniform, rectilinear borders. Most of our borders are rounded with Adwaita, but there are still some of these.
2021-03-09 02:37:28 +00:00
static inline void
gsk_ngl_render_job_visit_rect_border_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GdkRGBA *colors = gsk_border_node_get_colors (node);
const float *widths = gsk_border_node_get_widths (node);
const graphene_point_t *origin = &node->bounds.origin;
const graphene_size_t *size = &node->bounds.size;
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4];
ngl: Render simple borders simply Use 4 color commands for uniform, rectilinear borders. Most of our borders are rounded with Adwaita, but there are still some of these.
2021-03-09 02:37:28 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, color));
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
if (widths[0] > 0)
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
{
rgba_to_half (&colors[0], color);
gsk_ngl_render_job_draw_rect_with_color (job,
&GRAPHENE_RECT_INIT (origin->x, origin->y, size->width - widths[1], widths[0]),
color);
}
ngl: Render simple borders simply Use 4 color commands for uniform, rectilinear borders. Most of our borders are rounded with Adwaita, but there are still some of these.
2021-03-09 02:37:28 +00:00
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
if (widths[1] > 0)
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
{
rgba_to_half (&colors[1], color);
gsk_ngl_render_job_draw_rect_with_color (job,
&GRAPHENE_RECT_INIT (origin->x + size->width - widths[1], origin->y, widths[1], size->height - widths[2]),
color);
}
ngl: Render simple borders simply Use 4 color commands for uniform, rectilinear borders. Most of our borders are rounded with Adwaita, but there are still some of these.
2021-03-09 02:37:28 +00:00
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
if (widths[2] > 0)
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
{
rgba_to_half (&colors[2], color);
gsk_ngl_render_job_draw_rect_with_color (job,
&GRAPHENE_RECT_INIT (origin->x + widths[3], origin->y + size->height - widths[2], size->width - widths[1], widths[2]),
color);
}
ngl: Render simple borders simply Use 4 color commands for uniform, rectilinear borders. Most of our borders are rounded with Adwaita, but there are still some of these.
2021-03-09 02:37:28 +00:00
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
if (widths[3] > 0)
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
{
rgba_to_half (&colors[3], color);
gsk_ngl_render_job_draw_rect_with_color (job,
&GRAPHENE_RECT_INIT (origin->x, origin->y + widths[0], widths[3], size->height - widths[0]),
color);
}
ngl: Render simple borders simply Use 4 color commands for uniform, rectilinear borders. Most of our borders are rounded with Adwaita, but there are still some of these.
2021-03-09 02:37:28 +00:00
gsk_ngl_render_job_end_draw (job);
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
static inline void
gsk_ngl_render_job_visit_border_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRoundedRect *rounded_outline = gsk_border_node_get_outline (node);
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
const GdkRGBA *colors = gsk_border_node_get_colors (node);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
const float *widths = gsk_border_node_get_widths (node);
struct {
float w;
float h;
} sizes[4];
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
float min_x = job->offset_x + node->bounds.origin.x;
float min_y = job->offset_y + node->bounds.origin.y;
float max_x = min_x + node->bounds.size.width;
float max_y = min_y + node->bounds.size.height;
GskRoundedRect outline;
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4];
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
memset (sizes, 0, sizeof sizes);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (widths[0] > 0)
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
{
sizes[0].h = MAX (widths[0], rounded_outline->corner[0].height);
sizes[1].h = MAX (widths[0], rounded_outline->corner[1].height);
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (widths[1] > 0)
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
{
sizes[1].w = MAX (widths[1], rounded_outline->corner[1].width);
sizes[2].w = MAX (widths[1], rounded_outline->corner[2].width);
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
if (widths[2] > 0)
{
sizes[2].h = MAX (widths[2], rounded_outline->corner[2].height);
sizes[3].h = MAX (widths[2], rounded_outline->corner[3].height);
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
if (widths[3] > 0)
{
sizes[0].w = MAX (widths[3], rounded_outline->corner[0].width);
sizes[3].w = MAX (widths[3], rounded_outline->corner[3].width);
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
gsk_ngl_render_job_transform_rounded_rect (job, rounded_outline, &outline);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, border));
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
gsk_ngl_program_set_uniform4fv (job->current_program,
UNIFORM_BORDER_WIDTHS, 0,
1,
widths);
gsk_ngl_program_set_uniform_rounded_rect (job->current_program,
UNIFORM_BORDER_OUTLINE_RECT, 0,
&outline);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
if (widths[0] > 0)
{
GskNglDrawVertex *vertices = gsk_ngl_command_queue_add_vertices (job->command_queue);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (&colors[0], color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
vertices[0] = (GskNglDrawVertex) { .position = { min_x, min_y }, .uv = { 0, 1 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[1] = (GskNglDrawVertex) { .position = { min_x + sizes[0].w, min_y + sizes[0].h }, .uv = { 0, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[2] = (GskNglDrawVertex) { .position = { max_x, min_y }, .uv = { 1, 1 }, .color = { color[0], color[1], color[2], color[3] } };
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
vertices[3] = (GskNglDrawVertex) { .position = { max_x - sizes[1].w, min_y + sizes[1].h }, .uv = { 1, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[4] = (GskNglDrawVertex) { .position = { min_x + sizes[0].w, min_y + sizes[0].h }, .uv = { 0, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[5] = (GskNglDrawVertex) { .position = { max_x, min_y }, .uv = { 1, 1 }, .color = { color[0], color[1], color[2], color[3] } };
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
if (widths[1] > 0)
{
GskNglDrawVertex *vertices = gsk_ngl_command_queue_add_vertices (job->command_queue);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (&colors[1], color);
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
vertices[0] = (GskNglDrawVertex) { .position = { max_x - sizes[1].w, min_y + sizes[1].h }, .uv = { 0, 1 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[1] = (GskNglDrawVertex) { .position = { max_x - sizes[2].w, max_y - sizes[2].h }, .uv = { 0, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[2] = (GskNglDrawVertex) { .position = { max_x, min_y }, .uv = { 1, 1 }, .color = { color[0], color[1], color[2], color[3] } };
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
vertices[3] = (GskNglDrawVertex) { .position = { max_x, max_y }, .uv = { 1, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[4] = (GskNglDrawVertex) { .position = { max_x - sizes[2].w, max_y - sizes[2].h }, .uv = { 0, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[5] = (GskNglDrawVertex) { .position = { max_x, min_y }, .uv = { 1, 1 }, .color = { color[0], color[1], color[2], color[3] } };
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
if (widths[2] > 0)
{
GskNglDrawVertex *vertices = gsk_ngl_command_queue_add_vertices (job->command_queue);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (&colors[2], color);
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
vertices[0] = (GskNglDrawVertex) { .position = { min_x + sizes[3].w, max_y - sizes[3].h }, .uv = { 0, 1 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[1] = (GskNglDrawVertex) { .position = { min_x, max_y }, .uv = { 0, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[2] = (GskNglDrawVertex) { .position = { max_x - sizes[2].w, max_y - sizes[2].h }, .uv = { 1, 1 }, .color = { color[0], color[1], color[2], color[3] } };
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
vertices[3] = (GskNglDrawVertex) { .position = { max_x, max_y }, .uv = { 1, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[4] = (GskNglDrawVertex) { .position = { min_x , max_y }, .uv = { 0, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[5] = (GskNglDrawVertex) { .position = { max_x - sizes[2].w, max_y - sizes[2].h }, .uv = { 1, 1 }, .color = { color[0], color[1], color[2], color[3] } };
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
}
if (widths[3] > 0)
{
GskNglDrawVertex *vertices = gsk_ngl_command_queue_add_vertices (job->command_queue);
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (&colors[3], color);
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
vertices[0] = (GskNglDrawVertex) { .position = { min_x, min_y }, .uv = { 0, 1 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[1] = (GskNglDrawVertex) { .position = { min_x, max_y }, .uv = { 0, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[2] = (GskNglDrawVertex) { .position = { min_x + sizes[0].w, min_y + sizes[0].h }, .uv = { 1, 1 }, .color = { color[0], color[1], color[2], color[3] } };
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
vertices[3] = (GskNglDrawVertex) { .position = { min_x + sizes[3].w, max_y - sizes[3].h }, .uv = { 1, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[4] = (GskNglDrawVertex) { .position = { min_x, max_y }, .uv = { 0, 0 }, .color = { color[0], color[1], color[2], color[3] } };
vertices[5] = (GskNglDrawVertex) { .position = { min_x + sizes[0].w, min_y + sizes[0].h }, .uv = { 1, 1 }, .color = { color[0], color[1], color[2], color[3] } };
ngl: Refactor gsk_ngl_render_job_visit_border_node With color sorting out of the picture, this can be much simpler.
2021-03-13 18:40:04 +00:00
}
gsk_ngl_render_job_end_draw (job);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
ngl: Special-case css backgrounds Recognize a common pattern: A rounded clip with a color node, followed by a border node, with the same outline. This is what CSS backgrounds frequently produce, and we can render it more efficiently with a combined shader.
2021-03-14 04:30:18 +00:00
/* A special case for a pattern that occurs frequently with CSS
* backgrounds: two sibling nodes, the first of which is a rounded
* clip node with a color node as child, and the second one is a
* border node, with the same outline as the clip node. We render
* this using the filled_border shader.
*/
static void
gsk_ngl_render_job_visit_css_background (GskNglRenderJob *job,
const GskRenderNode *node,
const GskRenderNode *node2)
{
const GskRenderNode *child = gsk_rounded_clip_node_get_child (node);
const GskRoundedRect *rounded_outline = gsk_border_node_get_outline (node2);
const float *widths = gsk_border_node_get_widths (node2);
float min_x = job->offset_x + node2->bounds.origin.x;
float min_y = job->offset_y + node2->bounds.origin.y;
float max_x = min_x + node2->bounds.size.width;
float max_y = min_y + node2->bounds.size.height;
GskRoundedRect outline;
GskNglDrawVertex *vertices;
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4];
guint16 color2[4];
ngl: Special-case css backgrounds Recognize a common pattern: A rounded clip with a color node, followed by a border node, with the same outline. This is what CSS backgrounds frequently produce, and we can render it more efficiently with a combined shader.
2021-03-14 04:30:18 +00:00
if (node_is_invisible (node2))
return;
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (&gsk_border_node_get_colors (node2)[0], color);
rgba_to_half (gsk_color_node_get_color (child), color2);
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
ngl: Special-case css backgrounds Recognize a common pattern: A rounded clip with a color node, followed by a border node, with the same outline. This is what CSS backgrounds frequently produce, and we can render it more efficiently with a combined shader.
2021-03-14 04:30:18 +00:00
gsk_ngl_render_job_transform_rounded_rect (job, rounded_outline, &outline);
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, filled_border));
gsk_ngl_program_set_uniform4fv (job->current_program,
UNIFORM_FILLED_BORDER_WIDTHS, 0,
1,
widths);
gsk_ngl_program_set_uniform_rounded_rect (job->current_program,
UNIFORM_FILLED_BORDER_OUTLINE_RECT, 0,
&outline);
vertices = gsk_ngl_command_queue_add_vertices (job->command_queue);
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
vertices[0] = (GskNglDrawVertex) { .position = { min_x, min_y }, .color = { color[0], color[1], color[2], color[3] }, .color2 = { color2[0], color2[1], color2[2], color2[3] } };
vertices[1] = (GskNglDrawVertex) { .position = { min_x, max_y }, .color = { color[0], color[1], color[2], color[3] }, .color2 = { color2[0], color2[1], color2[2], color2[3] } };
vertices[2] = (GskNglDrawVertex) { .position = { max_x, min_y }, .color = { color[0], color[1], color[2], color[3] }, .color2 = { color2[0], color2[1], color2[2], color2[3] } };
vertices[3] = (GskNglDrawVertex) { .position = { max_x, max_y }, .color = { color[0], color[1], color[2], color[3] }, .color2 = { color2[0], color2[1], color2[2], color2[3] } };
vertices[4] = (GskNglDrawVertex) { .position = { min_x, max_y }, .color = { color[0], color[1], color[2], color[3] }, .color2 = { color2[0], color2[1], color2[2], color2[3] } };
vertices[5] = (GskNglDrawVertex) { .position = { max_x, min_y }, .color = { color[0], color[1], color[2], color[3] }, .color2 = { color2[0], color2[1], color2[2], color2[3] } };
ngl: Special-case css backgrounds Recognize a common pattern: A rounded clip with a color node, followed by a border node, with the same outline. This is what CSS backgrounds frequently produce, and we can render it more efficiently with a combined shader.
2021-03-14 04:30:18 +00:00
gsk_ngl_render_job_end_draw (job);
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
/* Returns TRUE if applying @transform to @bounds
* yields an axis-aligned rectangle
*/
static gboolean
result_is_axis_aligned (GskTransform *transform,
const graphene_rect_t *bounds)
{
graphene_matrix_t m;
graphene_quad_t q;
graphene_rect_t b;
graphene_point_t b1, b2;
const graphene_point_t *p;
gsk_transform_to_matrix (transform, &m);
gsk_matrix_transform_rect (&m, bounds, &q);
graphene_quad_bounds (&q, &b);
graphene_rect_get_top_left (&b, &b1);
graphene_rect_get_bottom_right (&b, &b2);
for (guint i = 0; i < 4; i++)
{
p = graphene_quad_get_point (&q, i);
if (fabs (p->x - b1.x) > FLT_EPSILON && fabs (p->x - b2.x) > FLT_EPSILON)
return FALSE;
if (fabs (p->y - b1.y) > FLT_EPSILON && fabs (p->y - b2.y) > FLT_EPSILON)
return FALSE;
}
return TRUE;
}
static inline void
gsk_ngl_render_job_visit_transform_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
GskTransform *transform = gsk_transform_node_get_transform (node);
const GskTransformCategory category = gsk_transform_get_category (transform);
const GskRenderNode *child = gsk_transform_node_get_child (node);
switch (category)
{
case GSK_TRANSFORM_CATEGORY_IDENTITY:
gsk_ngl_render_job_visit_node (job, child);
break;
case GSK_TRANSFORM_CATEGORY_2D_TRANSLATE:
{
float dx, dy;
gsk_transform_to_translate (transform, &dx, &dy);
gsk_ngl_render_job_offset (job, dx, dy);
gsk_ngl_render_job_visit_node (job, child);
gsk_ngl_render_job_offset (job, -dx, -dy);
}
break;
case GSK_TRANSFORM_CATEGORY_2D_AFFINE:
{
gsk_ngl_render_job_push_modelview (job, transform);
gsk_ngl_render_job_visit_node (job, child);
gsk_ngl_render_job_pop_modelview (job);
}
break;
case GSK_TRANSFORM_CATEGORY_2D:
case GSK_TRANSFORM_CATEGORY_3D:
case GSK_TRANSFORM_CATEGORY_ANY:
case GSK_TRANSFORM_CATEGORY_UNKNOWN:
if (node_supports_transform (child))
{
gsk_ngl_render_job_push_modelview (job, transform);
gsk_ngl_render_job_visit_node (job, child);
gsk_ngl_render_job_pop_modelview (job);
}
else
{
GskNglRenderOffscreen offscreen = {0};
offscreen.bounds = &child->bounds;
ngl: Plug a memory leak This was introduced in f9457af128012fd3935a69fdcefb.
2021-04-02 22:41:53 +00:00
offscreen.force_offscreen = FALSE;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
offscreen.reset_clip = TRUE;
if (!result_is_axis_aligned (transform, &child->bounds))
offscreen.linear_filter = TRUE;
if (gsk_ngl_render_job_visit_node_with_offscreen (job, child, &offscreen))
{
/* For non-trivial transforms, we draw everything on a texture and then
* draw the texture transformed. */
/* TODO: We should compute a modelview containing only the "non-trivial"
* part (e.g. the rotation) and use that. We want to keep the scale
* for the texture.
*/
gsk_ngl_render_job_push_modelview (job, transform);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen.texture_id);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen (job, &child->bounds, &offscreen);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
gsk_ngl_render_job_pop_modelview (job);
}
}
break;
default:
g_assert_not_reached ();
}
}
static inline void
gsk_ngl_render_job_visit_unblurred_inset_shadow_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRoundedRect *outline = gsk_inset_shadow_node_get_outline (node);
GskRoundedRect transformed_outline;
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4];
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_transform_rounded_rect (job, outline, &transformed_outline);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, inset_shadow));
gsk_ngl_program_set_uniform_rounded_rect (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_INSET_SHADOW_OUTLINE_RECT, 0,
&transformed_outline);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_INSET_SHADOW_SPREAD, 0,
gsk_inset_shadow_node_get_spread (node));
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform2f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_INSET_SHADOW_OFFSET, 0,
gsk_inset_shadow_node_get_dx (node),
gsk_inset_shadow_node_get_dy (node));
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (gsk_inset_shadow_node_get_color (node), color);
gsk_ngl_render_job_draw_rect_with_color (job, &node->bounds, color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_blurred_inset_shadow_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRoundedRect *node_outline = gsk_inset_shadow_node_get_outline (node);
float blur_radius = gsk_inset_shadow_node_get_blur_radius (node);
float offset_x = gsk_inset_shadow_node_get_dx (node);
float offset_y = gsk_inset_shadow_node_get_dy (node);
float scale_x = job->scale_x;
float scale_y = job->scale_y;
float blur_extra = blur_radius * 2.0; /* 2.0 = shader radius_multiplier */
float half_blur_extra = blur_radius;
float texture_width;
float texture_height;
int blurred_texture_id;
GskTextureKey key;
GskNglRenderOffscreen offscreen = {0};
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4];
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
g_assert (blur_radius > 0);
texture_width = ceilf ((node_outline->bounds.size.width + blur_extra) * scale_x);
texture_height = ceilf ((node_outline->bounds.size.height + blur_extra) * scale_y);
key.pointer = node;
key.pointer_is_child = FALSE;
key.scale_x = scale_x;
key.scale_y = scale_y;
key.filter = GL_NEAREST;
blurred_texture_id = gsk_ngl_driver_lookup_texture (job->driver, &key);
if (blurred_texture_id == 0)
{
float spread = gsk_inset_shadow_node_get_spread (node) + half_blur_extra;
GskRoundedRect transformed_outline;
GskRoundedRect outline_to_blur;
GskNglRenderTarget *render_target;
graphene_matrix_t prev_projection;
graphene_rect_t prev_viewport;
guint prev_fbo;
/* TODO: In the following code, we have to be careful about where we apply the scale.
* We're manually scaling stuff (e.g. the outline) so we can later use texture_width
* and texture_height (which are already scaled) as the geometry and keep the modelview
* at a scale of 1. That's kinda complicated though... */
/* Outline of what we actually want to blur later.
* Spread grows inside, so we don't need to account for that. But the blur will need
* to read outside of the inset shadow, so we need to draw some color in there. */
outline_to_blur = *node_outline;
gsk_rounded_rect_shrink (&outline_to_blur,
-half_blur_extra,
-half_blur_extra,
-half_blur_extra,
-half_blur_extra);
/* Fit to our texture */
outline_to_blur.bounds.origin.x = 0;
outline_to_blur.bounds.origin.y = 0;
outline_to_blur.bounds.size.width *= scale_x;
outline_to_blur.bounds.size.height *= scale_y;
for (guint i = 0; i < 4; i ++)
{
outline_to_blur.corner[i].width *= scale_x;
outline_to_blur.corner[i].height *= scale_y;
}
if (!gsk_ngl_driver_create_render_target (job->driver,
texture_width, texture_height,
GL_NEAREST, GL_NEAREST,
&render_target))
g_assert_not_reached ();
gsk_ngl_render_job_set_viewport_for_size (job, texture_width, texture_height, &prev_viewport);
gsk_ngl_render_job_set_projection_for_size (job, texture_width, texture_height, &prev_projection);
gsk_ngl_render_job_set_modelview (job, NULL);
gsk_ngl_render_job_push_clip (job, &GSK_ROUNDED_RECT_INIT (0, 0, texture_width, texture_height));
prev_fbo = gsk_ngl_command_queue_bind_framebuffer (job->command_queue, render_target->framebuffer_id);
gsk_ngl_command_queue_clear (job->command_queue, 0, &job->viewport);
gsk_ngl_render_job_transform_rounded_rect (job, &outline_to_blur, &transformed_outline);
/* Actual inset shadow outline drawing */
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, inset_shadow));
gsk_ngl_program_set_uniform_rounded_rect (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_INSET_SHADOW_OUTLINE_RECT, 0,
&transformed_outline);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_INSET_SHADOW_SPREAD, 0,
spread * MAX (scale_x, scale_y));
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform2f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_INSET_SHADOW_OFFSET, 0,
offset_x * scale_x,
offset_y * scale_y);
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (gsk_inset_shadow_node_get_color (node), color);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_with_color (job,
0, 0, texture_width, texture_height,
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
gsk_ngl_render_job_pop_modelview (job);
gsk_ngl_render_job_pop_clip (job);
gsk_ngl_render_job_set_projection (job, &prev_projection);
gsk_ngl_render_job_set_viewport (job, &prev_viewport, NULL);
gsk_ngl_command_queue_bind_framebuffer (job->command_queue, prev_fbo);
offscreen.texture_id = render_target->texture_id;
init_full_texture_region (&offscreen);
blurred_texture_id = blur_offscreen (job,
&offscreen,
texture_width,
texture_height,
blur_radius * scale_x,
blur_radius * scale_y);
gsk_ngl_driver_release_render_target (job->driver, render_target, TRUE);
gsk_ngl_driver_cache_texture (job->driver, &key, blurred_texture_id);
}
g_assert (blurred_texture_id != 0);
/* Blur the rendered unblurred inset shadow */
/* Use a clip to cut away the unwanted parts outside of the original outline */
{
const gboolean needs_clip = !gsk_rounded_rect_is_rectilinear (node_outline);
const float tx1 = half_blur_extra * scale_x / texture_width;
const float tx2 = 1.0 - tx1;
const float ty1 = half_blur_extra * scale_y / texture_height;
const float ty2 = 1.0 - ty1;
if (needs_clip)
{
GskRoundedRect node_clip;
gsk_ngl_render_job_transform_bounds (job, &node_outline->bounds, &node_clip.bounds);
for (guint i = 0; i < 4; i ++)
{
node_clip.corner[i].width = node_outline->corner[i].width * scale_x;
node_clip.corner[i].height = node_outline->corner[i].height * scale_y;
}
gsk_ngl_render_job_push_clip (job, &node_clip);
}
offscreen.was_offscreen = TRUE;
offscreen.area.x = tx1;
offscreen.area.y = ty1;
offscreen.area.x2 = tx2;
offscreen.area.y2 = ty2;
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
blurred_texture_id);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen (job, &node->bounds, &offscreen);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
if (needs_clip)
gsk_ngl_render_job_pop_clip (job);
}
}
static inline void
gsk_ngl_render_job_visit_unblurred_outset_shadow_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRoundedRect *outline = gsk_outset_shadow_node_get_outline (node);
GskRoundedRect transformed_outline;
float x = node->bounds.origin.x;
float y = node->bounds.origin.y;
float w = node->bounds.size.width;
float h = node->bounds.size.height;
float spread = gsk_outset_shadow_node_get_spread (node);
float dx = gsk_outset_shadow_node_get_dx (node);
float dy = gsk_outset_shadow_node_get_dy (node);
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4];
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
const float edge_sizes[] = { // Top, right, bottom, left
spread - dy, spread + dx, spread + dy, spread - dx
};
const float corner_sizes[][2] = { // top left, top right, bottom right, bottom left
{ outline->corner[0].width + spread - dx, outline->corner[0].height + spread - dy },
{ outline->corner[1].width + spread + dx, outline->corner[1].height + spread - dy },
{ outline->corner[2].width + spread + dx, outline->corner[2].height + spread + dy },
{ outline->corner[3].width + spread - dx, outline->corner[3].height + spread + dy },
};
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (gsk_outset_shadow_node_get_color (node), color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_transform_rounded_rect (job, outline, &transformed_outline);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, unblurred_outset_shadow));
gsk_ngl_program_set_uniform_rounded_rect (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_UNBLURRED_OUTSET_SHADOW_OUTLINE_RECT, 0,
&transformed_outline);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_UNBLURRED_OUTSET_SHADOW_SPREAD, 0,
spread);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform2f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_UNBLURRED_OUTSET_SHADOW_OFFSET, 0,
dx, dy);
/* Corners... */
if (corner_sizes[0][0] > 0 && corner_sizes[0][1] > 0) /* Top left */
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_with_color (job,
x, y, corner_sizes[0][0], corner_sizes[0][1],
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (corner_sizes[1][0] > 0 && corner_sizes[1][1] > 0) /* Top right */
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_with_color (job,
x + w - corner_sizes[1][0], y,
corner_sizes[1][0], corner_sizes[1][1],
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (corner_sizes[2][0] > 0 && corner_sizes[2][1] > 0) /* Bottom right */
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_with_color (job,
x + w - corner_sizes[2][0], y + h - corner_sizes[2][1],
corner_sizes[2][0], corner_sizes[2][1],
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (corner_sizes[3][0] > 0 && corner_sizes[3][1] > 0) /* Bottom left */
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_with_color (job,
x, y + h - corner_sizes[3][1],
corner_sizes[3][0], corner_sizes[3][1],
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
/* Edges... */;
if (edge_sizes[0] > 0) /* Top */
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_with_color (job,
x + corner_sizes[0][0], y,
w - corner_sizes[0][0] - corner_sizes[1][0], edge_sizes[0],
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (edge_sizes[1] > 0) /* Right */
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_with_color (job,
x + w - edge_sizes[1], y + corner_sizes[1][1],
edge_sizes[1], h - corner_sizes[1][1] - corner_sizes[2][1],
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (edge_sizes[2] > 0) /* Bottom */
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_with_color (job,
x + corner_sizes[3][0], y + h - edge_sizes[2],
w - corner_sizes[3][0] - corner_sizes[2][0], edge_sizes[2],
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (edge_sizes[3] > 0) /* Left */
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_with_color (job,
x, y + corner_sizes[0][1],
edge_sizes[3], h - corner_sizes[0][1] - corner_sizes[3][1],
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_blurred_outset_shadow_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRoundedRect *outline = gsk_outset_shadow_node_get_outline (node);
float scale_x = job->scale_x;
float scale_y = job->scale_y;
float blur_radius = gsk_outset_shadow_node_get_blur_radius (node);
float blur_extra = blur_radius * 2.0f; /* 2.0 = shader radius_multiplier */
float half_blur_extra = blur_extra / 2.0f;
ngl: Fix unevenly scaled shadows Ensure that we don't cut them off at the edges.
2021-04-12 02:26:36 +00:00
int extra_blur_pixels_x = ceilf (half_blur_extra * scale_x);
int extra_blur_pixels_y = ceilf (half_blur_extra * scale_y);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
float spread = gsk_outset_shadow_node_get_spread (node);
float dx = gsk_outset_shadow_node_get_dx (node);
float dy = gsk_outset_shadow_node_get_dy (node);
GskRoundedRect scaled_outline;
GskRoundedRect transformed_outline;
GskNglRenderOffscreen offscreen = {0};
int texture_width, texture_height;
int blurred_texture_id;
int cached_tid;
gboolean do_slicing;
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4];
ngl: Don't slice unsliceable shadows When the corners are too big to make slicing work, don't do it, since it leads to broken results.
2021-04-12 02:09:45 +00:00
float half_width = outline->bounds.size.width / 2;
float half_height = outline->bounds.size.height / 2;
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (gsk_outset_shadow_node_get_color (node), color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
/* scaled_outline is the minimal outline we need to draw the given drop shadow,
* enlarged by the spread and offset by the blur radius. */
scaled_outline = *outline;
if (outline->bounds.size.width < blur_extra ||
ngl: Don't slice unsliceable shadows When the corners are too big to make slicing work, don't do it, since it leads to broken results.
2021-04-12 02:09:45 +00:00
outline->bounds.size.height < blur_extra ||
outline->corner[0].width >= half_width ||
outline->corner[1].width >= half_width ||
outline->corner[2].width >= half_width ||
outline->corner[3].width >= half_width ||
outline->corner[0].height >= half_height ||
outline->corner[1].height >= half_height ||
outline->corner[2].height >= half_height ||
outline->corner[3].height >= half_height)
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
{
do_slicing = FALSE;
gsk_rounded_rect_shrink (&scaled_outline, -spread, -spread, -spread, -spread);
}
else
{
/* Shrink our outline to the minimum size that can still hold all the border radii */
gsk_rounded_rect_shrink_to_minimum (&scaled_outline);
/* Increase by the spread */
gsk_rounded_rect_shrink (&scaled_outline, -spread, -spread, -spread, -spread);
/* Grow bounds but don't grow corners */
graphene_rect_inset (&scaled_outline.bounds, - blur_extra / 2.0, - blur_extra / 2.0);
/* For the center part, we add a few pixels */
scaled_outline.bounds.size.width += SHADOW_EXTRA_SIZE;
scaled_outline.bounds.size.height += SHADOW_EXTRA_SIZE;
do_slicing = TRUE;
}
texture_width = (int)ceil ((scaled_outline.bounds.size.width + blur_extra) * scale_x);
texture_height = (int)ceil ((scaled_outline.bounds.size.height + blur_extra) * scale_y);
ngl: Fix unevenly scaled shadows Ensure that we don't cut them off at the edges.
2021-04-12 02:26:36 +00:00
scaled_outline.bounds.origin.x = extra_blur_pixels_x;
scaled_outline.bounds.origin.y = extra_blur_pixels_y;
scaled_outline.bounds.size.width = texture_width - (extra_blur_pixels_x * 2);
scaled_outline.bounds.size.height = texture_height - (extra_blur_pixels_y * 2);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
for (guint i = 0; i < G_N_ELEMENTS (scaled_outline.corner); i++)
{
scaled_outline.corner[i].width *= scale_x;
scaled_outline.corner[i].height *= scale_y;
}
cached_tid = gsk_ngl_shadow_library_lookup (job->driver->shadows, &scaled_outline, blur_radius);
if (cached_tid == 0)
{
GdkGLContext *context = job->command_queue->context;
GskNglRenderTarget *render_target;
graphene_matrix_t prev_projection;
graphene_rect_t prev_viewport;
guint prev_fbo;
gsk_ngl_driver_create_render_target (job->driver,
texture_width, texture_height,
GL_NEAREST, GL_NEAREST,
&render_target);
if (gdk_gl_context_has_debug (context))
{
gdk_gl_context_label_object_printf (context,
GL_TEXTURE,
render_target->texture_id,
"Outset Shadow Temp %d",
render_target->texture_id);
gdk_gl_context_label_object_printf (context,
GL_FRAMEBUFFER,
render_target->framebuffer_id,
"Outset Shadow FB Temp %d",
render_target->framebuffer_id);
}
/* Change state for offscreen */
gsk_ngl_render_job_set_projection_for_size (job, texture_width, texture_height, &prev_projection);
gsk_ngl_render_job_set_viewport_for_size (job, texture_width, texture_height, &prev_viewport);
gsk_ngl_render_job_set_modelview (job, NULL);
gsk_ngl_render_job_push_clip (job, &scaled_outline);
/* Bind render target and clear it */
prev_fbo = gsk_ngl_command_queue_bind_framebuffer (job->command_queue, render_target->framebuffer_id);
gsk_ngl_command_queue_clear (job->command_queue, 0, &job->viewport);
/* Draw the outline using color program */
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, color));
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
gsk_ngl_render_job_draw_with_color (job, 0, 0, texture_width, texture_height,
(guint16[]){ FP16_ONE, FP16_ONE, FP16_ONE, FP16_ONE });
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
/* Reset state from offscreen */
gsk_ngl_render_job_pop_clip (job);
gsk_ngl_render_job_pop_modelview (job);
gsk_ngl_render_job_set_viewport (job, &prev_viewport, NULL);
gsk_ngl_render_job_set_projection (job, &prev_projection);
/* Now blur the outline */
init_full_texture_region (&offscreen);
offscreen.texture_id = gsk_ngl_driver_release_render_target (job->driver, render_target, FALSE);
blurred_texture_id = blur_offscreen (job,
&offscreen,
texture_width,
texture_height,
blur_radius * scale_x,
blur_radius * scale_y);
gsk_ngl_shadow_library_insert (job->driver->shadows,
&scaled_outline,
blur_radius,
blurred_texture_id);
gsk_ngl_command_queue_bind_framebuffer (job->command_queue, prev_fbo);
}
else
{
blurred_texture_id = cached_tid;
}
gsk_ngl_render_job_transform_rounded_rect (job, outline, &transformed_outline);
if (!do_slicing)
{
float min_x = floorf (outline->bounds.origin.x - spread - half_blur_extra + dx);
float min_y = floorf (outline->bounds.origin.y - spread - half_blur_extra + dy);
ngl: Fix a case of flipped shadow In the non-sliced case, we were rendering the shadow upside down.
2021-04-12 02:02:44 +00:00
offscreen.was_offscreen = TRUE;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
offscreen.texture_id = blurred_texture_id;
init_full_texture_region (&offscreen);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, outset_shadow));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
blurred_texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform_rounded_rect (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_OUTSET_SHADOW_OUTLINE_RECT, 0,
&transformed_outline);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (job,
&GRAPHENE_RECT_INIT (min_x,
min_y,
texture_width / scale_x,
texture_height / scale_y),
&offscreen,
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
return;
}
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
/* slicing */
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, outset_shadow));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
blurred_texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform_rounded_rect (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_OUTSET_SHADOW_OUTLINE_RECT, 0,
&transformed_outline);
{
float min_x = floorf (outline->bounds.origin.x - spread - half_blur_extra + dx);
float min_y = floorf (outline->bounds.origin.y - spread - half_blur_extra + dy);
float max_x = ceilf (outline->bounds.origin.x + outline->bounds.size.width +
half_blur_extra + dx + spread);
float max_y = ceilf (outline->bounds.origin.y + outline->bounds.size.height +
half_blur_extra + dy + spread);
const GskNglTextureNineSlice *slices;
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
float left_width, center_width, right_width;
float top_height, center_height, bottom_height;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
GskNglTexture *texture;
texture = gsk_ngl_driver_get_texture_by_id (job->driver, blurred_texture_id);
ngl: Fix unevenly scaled shadows Ensure that we don't cut them off at the edges.
2021-04-12 02:26:36 +00:00
slices = gsk_ngl_texture_get_nine_slice (texture, &scaled_outline, extra_blur_pixels_x, extra_blur_pixels_y);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
offscreen.was_offscreen = TRUE;
/* Our texture coordinates MUST be scaled, while the actual vertex coords
ngl: Fix unevenly scaled shadows Ensure that we don't cut them off at the edges.
2021-04-12 02:26:36 +00:00
* MUST NOT be scaled.
*/
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
left_width = slices[NINE_SLICE_TOP_LEFT].rect.width / scale_x;
right_width = slices[NINE_SLICE_TOP_RIGHT].rect.width / scale_x;
center_width = (max_x - min_x) - (left_width + right_width);
top_height = slices[NINE_SLICE_TOP_LEFT].rect.height / scale_y;
bottom_height = slices[NINE_SLICE_BOTTOM_LEFT].rect.height / scale_y;
center_height = (max_y - min_y) - (top_height + bottom_height);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
/* Top left */
if (nine_slice_is_visible (&slices[NINE_SLICE_TOP_LEFT]))
{
memcpy (&offscreen.area, &slices[NINE_SLICE_TOP_LEFT].area, sizeof offscreen.area);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (job,
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
&GRAPHENE_RECT_INIT (min_x,
min_y,
left_width,
top_height),
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
&offscreen,
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
/* Top center */
if (nine_slice_is_visible (&slices[NINE_SLICE_TOP_CENTER]))
{
memcpy (&offscreen.area, &slices[NINE_SLICE_TOP_CENTER].area, sizeof offscreen.area);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (job,
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
&GRAPHENE_RECT_INIT (min_x + left_width,
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
min_y,
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
center_width,
top_height),
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
&offscreen,
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
/* Top right */
if (nine_slice_is_visible (&slices[NINE_SLICE_TOP_RIGHT]))
{
memcpy (&offscreen.area, &slices[NINE_SLICE_TOP_RIGHT].area, sizeof offscreen.area);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (job,
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
&GRAPHENE_RECT_INIT (max_x - right_width,
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
min_y,
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
right_width,
top_height),
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
&offscreen,
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
/* Bottom right */
if (nine_slice_is_visible (&slices[NINE_SLICE_BOTTOM_RIGHT]))
{
memcpy (&offscreen.area, &slices[NINE_SLICE_BOTTOM_RIGHT].area, sizeof offscreen.area);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (job,
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
&GRAPHENE_RECT_INIT (max_x - right_width,
max_y - bottom_height,
right_width,
bottom_height),
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
&offscreen,
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
/* Bottom left */
if (nine_slice_is_visible (&slices[NINE_SLICE_BOTTOM_LEFT]))
{
memcpy (&offscreen.area, &slices[NINE_SLICE_BOTTOM_LEFT].area, sizeof offscreen.area);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (job,
&GRAPHENE_RECT_INIT (min_x,
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
max_y - bottom_height,
left_width,
bottom_height),
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
&offscreen,
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
/* Left side */
if (nine_slice_is_visible (&slices[NINE_SLICE_LEFT_CENTER]))
{
memcpy (&offscreen.area, &slices[NINE_SLICE_LEFT_CENTER].area, sizeof offscreen.area);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (job,
&GRAPHENE_RECT_INIT (min_x,
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
min_y + top_height,
left_width,
center_height),
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
&offscreen,
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
/* Right side */
if (nine_slice_is_visible (&slices[NINE_SLICE_RIGHT_CENTER]))
{
memcpy (&offscreen.area, &slices[NINE_SLICE_RIGHT_CENTER].area, sizeof offscreen.area);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (job,
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
&GRAPHENE_RECT_INIT (max_x - right_width,
min_y + top_height,
right_width,
center_height),
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
&offscreen,
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
/* Bottom side */
if (nine_slice_is_visible (&slices[NINE_SLICE_BOTTOM_CENTER]))
{
memcpy (&offscreen.area, &slices[NINE_SLICE_BOTTOM_CENTER].area, sizeof offscreen.area);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen_with_color (job,
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
&GRAPHENE_RECT_INIT (min_x + left_width,
max_y - bottom_height,
center_width,
bottom_height),
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
&offscreen,
color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
/* Middle */
if (nine_slice_is_visible (&slices[NINE_SLICE_CENTER]))
{
ngl: Small optimization for shadows When 9-slicing shadows, omit the center tile when it is entirely contained in the outline (that is not always the case, depending on corners and offsets).
2021-04-10 15:23:06 +00:00
if (!gsk_rounded_rect_contains_rect (outline, &GRAPHENE_RECT_INIT (min_x + left_width,
min_y + top_height,
center_width,
center_height)))
{
memcpy (&offscreen.area, &slices[NINE_SLICE_CENTER].area, sizeof offscreen.area);
gsk_ngl_render_job_draw_offscreen_with_color (job,
&GRAPHENE_RECT_INIT (min_x + left_width,
min_y + top_height,
center_width,
center_height),
&offscreen,
color);
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
}
gsk_ngl_render_job_end_draw (job);
}
static inline gboolean G_GNUC_PURE
equal_texture_nodes (const GskRenderNode *node1,
const GskRenderNode *node2)
{
if (gsk_render_node_get_node_type (node1) != GSK_TEXTURE_NODE ||
gsk_render_node_get_node_type (node2) != GSK_TEXTURE_NODE)
return FALSE;
if (gsk_texture_node_get_texture (node1) !=
gsk_texture_node_get_texture (node2))
return FALSE;
return graphene_rect_equal (&node1->bounds, &node2->bounds);
}
static inline void
gsk_ngl_render_job_visit_cross_fade_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRenderNode *start_node = gsk_cross_fade_node_get_start_child (node);
const GskRenderNode *end_node = gsk_cross_fade_node_get_end_child (node);
float progress = gsk_cross_fade_node_get_progress (node);
GskNglRenderOffscreen offscreen_start = {0};
GskNglRenderOffscreen offscreen_end = {0};
g_assert (progress > 0.0);
g_assert (progress < 1.0);
offscreen_start.force_offscreen = TRUE;
offscreen_start.reset_clip = TRUE;
offscreen_start.bounds = &node->bounds;
offscreen_end.force_offscreen = TRUE;
offscreen_end.reset_clip = TRUE;
offscreen_end.bounds = &node->bounds;
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, start_node, &offscreen_start))
{
gsk_ngl_render_job_visit_node (job, end_node);
return;
}
g_assert (offscreen_start.texture_id);
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, end_node, &offscreen_end))
{
float prev_alpha = gsk_ngl_render_job_set_alpha (job, job->alpha * progress);
gsk_ngl_render_job_visit_node (job, start_node);
gsk_ngl_render_job_set_alpha (job, prev_alpha);
return;
}
g_assert (offscreen_end.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, cross_fade));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen_start.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_CROSS_FADE_SOURCE2, 0,
GL_TEXTURE_2D,
GL_TEXTURE1,
offscreen_end.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_CROSS_FADE_PROGRESS, 0,
progress);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen (job, &node->bounds, &offscreen_end);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_opacity_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRenderNode *child = gsk_opacity_node_get_child (node);
float opacity = gsk_opacity_node_get_opacity (node);
float new_alpha = job->alpha * opacity;
if (!ALPHA_IS_CLEAR (new_alpha))
{
float prev_alpha = gsk_ngl_render_job_set_alpha (job, new_alpha);
if (gsk_render_node_get_node_type (child) == GSK_CONTAINER_NODE)
{
GskNglRenderOffscreen offscreen = {0};
offscreen.bounds = &child->bounds;
offscreen.force_offscreen = TRUE;
offscreen.reset_clip = TRUE;
/* The semantics of an opacity node mandate that when, e.g., two
* color nodes overlap, there may not be any blending between them.
*/
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, child, &offscreen))
return;
g_assert (offscreen.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen.texture_id);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen (job, &node->bounds, &offscreen);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
else
{
gsk_ngl_render_job_visit_node (job, child);
}
gsk_ngl_render_job_set_alpha (job, prev_alpha);
}
}
ngl: Speed up glyph loop Move some work out of the loop in visit_text_node. This takes advantage of the fact that the yoffset of most glyphs is zero, so yphase generally does not change in a line of text.
2021-03-29 00:20:50 +00:00
static inline int
compute_phase_and_pos (float value, float *pos)
{
float v;
*pos = floorf (value);
v = value - *pos;
if (v < 0.125)
return 0;
else if (v < 0.375)
return 1;
else if (v < 0.625)
return 2;
else if (v < 0.875)
return 3;
else
{
*pos += 1;
return 0;
}
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
static inline void
gsk_ngl_render_job_visit_text_node (GskNglRenderJob *job,
const GskRenderNode *node,
const GdkRGBA *color,
gboolean force_color)
{
const PangoFont *font = gsk_text_node_get_font (node);
const PangoGlyphInfo *glyphs = gsk_text_node_get_glyphs (node, NULL);
const graphene_point_t *offset = gsk_text_node_get_offset (node);
float text_scale = MAX (job->scale_x, job->scale_y); /* TODO: Fix for uneven scales? */
guint num_glyphs = gsk_text_node_get_num_glyphs (node);
float x = offset->x + job->offset_x;
float y = offset->y + job->offset_y;
GskNglGlyphLibrary *library = job->driver->glyphs;
GskNglCommandBatch *batch;
int x_position = 0;
GskNglGlyphKey lookup;
guint last_texture = 0;
GskNglDrawVertex *vertices;
guint used = 0;
ngl: Use a constant for (fp16)-1 No need to convert that again for every glyph.
2021-04-07 03:11:27 +00:00
guint16 c[4] = { FP16_MINUS_ONE, FP16_MINUS_ONE, FP16_MINUS_ONE, FP16_MINUS_ONE };
ngl: Cosmetics Remove a local variable in gsk_render_job_visit_text_node.
2021-03-28 20:41:59 +00:00
const PangoGlyphInfo *gi;
guint i;
ngl: Speed up glyph loop Move some work out of the loop in visit_text_node. This takes advantage of the fact that the yoffset of most glyphs is zero, so yphase generally does not change in a line of text.
2021-03-29 00:20:50 +00:00
int yshift;
float ypos;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (num_glyphs == 0)
return;
ngl: Make the coloring shader more versatile When the color passed is transparent black, use the color from the texture as source, instead of as mask. This lets use use the coloring program both for regular and color glyphs, avoiding program changes in text with Emoji.
2021-03-12 06:27:19 +00:00
/* If the font has color glyphs, we don't need to recolor anything.
* We tell the shader by setting the color to vec4(-1).
*/
ngl: Use a constant for (fp16)-1 No need to convert that again for every glyph.
2021-04-07 03:11:27 +00:00
if (force_color || !gsk_text_node_has_color_glyphs (node))
ngl: Use fp16 for colors This reduces the size of our Vertex struct from 48 to 32 bytes. It would be nicer if we could store the colors in fp16 format in the rendernodes, and avoid conversion here. But this is still good.
2021-04-07 01:55:50 +00:00
rgba_to_half (color, c);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
lookup.font = (PangoFont *)font;
lookup.scale = (guint) (text_scale * 1024);
ngl: Speed up glyph loop Move some work out of the loop in visit_text_node. This takes advantage of the fact that the yoffset of most glyphs is zero, so yphase generally does not change in a line of text.
2021-03-29 00:20:50 +00:00
yshift = compute_phase_and_pos (y, &ypos);
ngl: Rewrite gsk_ngl_render_job_visit_text_node Make this more compact, and thus easier to read.
2021-03-13 19:36:03 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, coloring));
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
batch = gsk_ngl_command_queue_get_batch (job->command_queue);
vertices = gsk_ngl_command_queue_add_n_vertices (job->command_queue, num_glyphs);
/* We use one quad per character */
ngl: Cosmetics Remove a local variable in gsk_render_job_visit_text_node.
2021-03-28 20:41:59 +00:00
for (i = 0, gi = glyphs; i < num_glyphs; i++, gi++)
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
{
const GskNglGlyphValue *glyph;
float glyph_x, glyph_y, glyph_x2, glyph_y2;
float tx, ty, tx2, ty2;
float cx;
float cy;
guint texture_id;
ngl: Speed up glyph loop Move some work out of the loop in visit_text_node. This takes advantage of the fact that the yoffset of most glyphs is zero, so yphase generally does not change in a line of text.
2021-03-29 00:20:50 +00:00
lookup.glyph = gi->glyph;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
cx = (float)(x_position + gi->geometry.x_offset) / PANGO_SCALE;
ngl: Speed up glyph loop Move some work out of the loop in visit_text_node. This takes advantage of the fact that the yoffset of most glyphs is zero, so yphase generally does not change in a line of text.
2021-03-29 00:20:50 +00:00
lookup.xshift = compute_phase_and_pos (x + cx, &cx);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Speed up glyph loop Move some work out of the loop in visit_text_node. This takes advantage of the fact that the yoffset of most glyphs is zero, so yphase generally does not change in a line of text.
2021-03-29 00:20:50 +00:00
if G_UNLIKELY (gi->geometry.y_offset != 0)
{
cy = (float)(gi->geometry.y_offset) / PANGO_SCALE;
lookup.yshift = compute_phase_and_pos (y + cy, &cy);
}
else
{
lookup.yshift = yshift;
cy = ypos;
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Avoid a goto Reshuffle the code slightly, so we can use a continue instead.
2021-03-29 00:53:46 +00:00
x_position += gi->geometry.width;
ngl: Speed up glyph loop Move some work out of the loop in visit_text_node. This takes advantage of the fact that the yoffset of most glyphs is zero, so yphase generally does not change in a line of text.
2021-03-29 00:20:50 +00:00
texture_id = gsk_ngl_glyph_library_lookup_or_add (library, &lookup, &glyph);
if G_UNLIKELY (texture_id == 0)
ngl: Avoid a goto Reshuffle the code slightly, so we can use a continue instead.
2021-03-29 00:53:46 +00:00
continue;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if G_UNLIKELY (last_texture != texture_id)
{
if G_LIKELY (last_texture != 0)
{
guint vbo_offset = batch->draw.vbo_offset + batch->draw.vbo_count;
/* Since we have batched added our VBO vertices to avoid repeated
* calls to the buffer, we need to manually tweak the vbo offset
* of the new batch as otherwise it will point at the end of our
* vbo array.
*/
gsk_ngl_render_job_split_draw (job);
batch = gsk_ngl_command_queue_get_batch (job->command_queue);
batch->draw.vbo_offset = vbo_offset;
}
ngl: Rewrite gsk_ngl_render_job_visit_text_node Make this more compact, and thus easier to read.
2021-03-13 19:36:03 +00:00
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
texture_id);
last_texture = texture_id;
}
tx = glyph->entry.area.x;
ty = glyph->entry.area.y;
tx2 = glyph->entry.area.x2;
ty2 = glyph->entry.area.y2;
ngl: Speed up glyph loop Move some work out of the loop in visit_text_node. This takes advantage of the fact that the yoffset of most glyphs is zero, so yphase generally does not change in a line of text.
2021-03-29 00:20:50 +00:00
glyph_x = cx + glyph->ink_rect.x;
glyph_y = cy + glyph->ink_rect.y;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
glyph_x2 = glyph_x + glyph->ink_rect.width;
glyph_y2 = glyph_y + glyph->ink_rect.height;
ngl: Reduce vertex storage We only have one shader that uses the color2 attribute, and it doesn't use the uv attribute, so save vertex memory by putting those in the same space. This reduce the per vertex space from 32 to 24 bytes.
2021-04-07 02:24:37 +00:00
*(vertices++) = (GskNglDrawVertex) { .position = { glyph_x, glyph_y }, .uv = { tx, ty }, .color = { c[0], c[1], c[2], c[3] } };
*(vertices++) = (GskNglDrawVertex) { .position = { glyph_x, glyph_y2 }, .uv = { tx, ty2 }, .color = { c[0], c[1], c[2], c[3] } };
*(vertices++) = (GskNglDrawVertex) { .position = { glyph_x2, glyph_y }, .uv = { tx2, ty }, .color = { c[0], c[1], c[2], c[3] } };
ngl: Rewrite gsk_ngl_render_job_visit_text_node Make this more compact, and thus easier to read.
2021-03-13 19:36:03 +00:00
ngl: Reduce vertex storage We only have one shader that uses the color2 attribute, and it doesn't use the uv attribute, so save vertex memory by putting those in the same space. This reduce the per vertex space from 32 to 24 bytes.
2021-04-07 02:24:37 +00:00
*(vertices++) = (GskNglDrawVertex) { .position = { glyph_x2, glyph_y2 }, .uv = { tx2, ty2 }, .color = { c[0], c[1], c[2], c[3] } };
*(vertices++) = (GskNglDrawVertex) { .position = { glyph_x, glyph_y2 }, .uv = { tx, ty2 }, .color = { c[0], c[1], c[2], c[3] } };
*(vertices++) = (GskNglDrawVertex) { .position = { glyph_x2, glyph_y }, .uv = { tx2, ty }, .color = { c[0], c[1], c[2], c[3] } };
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
batch->draw.vbo_count += GSK_NGL_N_VERTICES;
used++;
}
if (used != num_glyphs)
gsk_ngl_command_queue_retract_n_vertices (job->command_queue, num_glyphs - used);
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_shadow_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const gsize n_shadows = gsk_shadow_node_get_n_shadows (node);
const GskRenderNode *original_child = gsk_shadow_node_get_child (node);
const GskRenderNode *shadow_child = original_child;
/* Shadow nodes recolor every pixel of the source texture, but leave the alpha in tact.
* If the child is a color matrix node that doesn't touch the alpha, we can throw that away. */
if (gsk_render_node_get_node_type (shadow_child) == GSK_COLOR_MATRIX_NODE &&
!color_matrix_modifies_alpha (shadow_child))
shadow_child = gsk_color_matrix_node_get_child (shadow_child);
for (guint i = 0; i < n_shadows; i++)
{
const GskShadow *shadow = gsk_shadow_node_get_shadow (node, i);
const float dx = shadow->dx;
const float dy = shadow->dy;
GskNglRenderOffscreen offscreen = {0};
graphene_rect_t bounds;
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 color[4];
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (shadow->radius == 0 &&
gsk_render_node_get_node_type (shadow_child) == GSK_TEXT_NODE)
{
gsk_ngl_render_job_offset (job, dx, dy);
gsk_ngl_render_job_visit_text_node (job, shadow_child, &shadow->color, TRUE);
gsk_ngl_render_job_offset (job, -dx, -dy);
continue;
}
if (RGBA_IS_CLEAR (&shadow->color))
continue;
if (node_is_invisible (shadow_child))
continue;
if (shadow->radius > 0)
{
float min_x;
float min_y;
float max_x;
float max_y;
offscreen.do_not_cache = TRUE;
blur_node (job,
&offscreen,
shadow_child,
shadow->radius,
&min_x, &max_x,
&min_y, &max_y);
bounds.origin.x = min_x - job->offset_x;
bounds.origin.y = min_y - job->offset_y;
bounds.size.width = max_x - min_x;
bounds.size.height = max_y - min_y;
offscreen.was_offscreen = TRUE;
}
else if (dx == 0 && dy == 0)
{
continue; /* Invisible anyway */
}
else
{
offscreen.bounds = &shadow_child->bounds;
offscreen.reset_clip = TRUE;
offscreen.do_not_cache = TRUE;
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, shadow_child, &offscreen))
g_assert_not_reached ();
bounds = shadow_child->bounds;
}
gsk_ngl_render_job_offset (job, dx, dy);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, coloring));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen.texture_id);
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
rgba_to_half (&shadow->color, color);
gsk_ngl_render_job_draw_offscreen_with_color (job, &bounds, &offscreen, color);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
gsk_ngl_render_job_offset (job, -dx, -dy);
}
/* Now draw the child normally */
gsk_ngl_render_job_visit_node (job, original_child);
}
static inline void
gsk_ngl_render_job_visit_blur_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRenderNode *child = gsk_blur_node_get_child (node);
float blur_radius = gsk_blur_node_get_radius (node);
GskNglRenderOffscreen offscreen = {0};
GskTextureKey key;
gboolean cache_texture;
float min_x;
float max_x;
float min_y;
float max_y;
g_assert (blur_radius > 0);
Revert "ngl: Remove a few redundant checks" This reverts commit 4375f8e382ef035d970dd255f56d1a7470c1a78c. These were not redundant after all, as pointed out by Timm.
2021-03-08 13:10:25 +00:00
if (node_is_invisible (child))
return;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
key.pointer = node;
key.pointer_is_child = FALSE;
key.scale_x = job->scale_x;
key.scale_y = job->scale_y;
key.filter = GL_NEAREST;
offscreen.texture_id = gsk_ngl_driver_lookup_texture (job->driver, &key);
cache_texture = offscreen.texture_id == 0;
blur_node (job,
&offscreen,
child,
blur_radius,
&min_x, &max_x, &min_y, &max_y);
g_assert (offscreen.texture_id != 0);
if (cache_texture)
gsk_ngl_driver_cache_texture (job->driver, &key, offscreen.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen.texture_id);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_coords (job,
min_x, min_y, max_x, max_y,
0, 1, 1, 0,
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
(guint16[]) { FP16_ZERO, FP16_ZERO, FP16_ZERO, FP16_ZERO } );
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_blend_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRenderNode *top_child = gsk_blend_node_get_top_child (node);
const GskRenderNode *bottom_child = gsk_blend_node_get_bottom_child (node);
GskNglRenderOffscreen top_offscreen = {0};
GskNglRenderOffscreen bottom_offscreen = {0};
top_offscreen.bounds = &node->bounds;
top_offscreen.force_offscreen = TRUE;
top_offscreen.reset_clip = TRUE;
bottom_offscreen.bounds = &node->bounds;
bottom_offscreen.force_offscreen = TRUE;
bottom_offscreen.reset_clip = TRUE;
/* TODO: We create 2 textures here as big as the blend node, but both the
* start and the end node might be a lot smaller than that. */
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, bottom_child, &bottom_offscreen))
{
gsk_ngl_render_job_visit_node (job, top_child);
return;
}
g_assert (bottom_offscreen.was_offscreen);
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, top_child, &top_offscreen))
{
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
bottom_offscreen.texture_id);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen (job, &node->bounds, &bottom_offscreen);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
return;
}
g_assert (top_offscreen.was_offscreen);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blend));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
bottom_offscreen.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_BLEND_SOURCE2, 0,
GL_TEXTURE_2D,
GL_TEXTURE1,
top_offscreen.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform1i (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_BLEND_MODE, 0,
gsk_blend_node_get_blend_mode (node));
gsk_ngl_render_job_draw_offscreen_rect (job, &node->bounds);
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_color_matrix_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRenderNode *child = gsk_color_matrix_node_get_child (node);
GskNglRenderOffscreen offscreen = {0};
float offset[4];
Revert "ngl: Remove a few redundant checks" This reverts commit 4375f8e382ef035d970dd255f56d1a7470c1a78c. These were not redundant after all, as pointed out by Timm.
2021-03-08 13:10:25 +00:00
if (node_is_invisible (child))
return;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
offscreen.bounds = &node->bounds;
offscreen.reset_clip = TRUE;
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, child, &offscreen))
g_assert_not_reached ();
g_assert (offscreen.texture_id > 0);
graphene_vec4_to_float (gsk_color_matrix_node_get_color_offset (node), offset);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, color_matrix));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform_matrix (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_COLOR_MATRIX_COLOR_MATRIX, 0,
gsk_color_matrix_node_get_color_matrix (node));
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform4fv (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_COLOR_MATRIX_COLOR_OFFSET, 0,
1,
offset);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen (job, &node->bounds, &offscreen);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_gl_shader_node_fallback (GskNglRenderJob *job,
const GskRenderNode *node)
{
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
guint16 pink[4] = { 15360, 13975, 14758, 15360 }; /* 255 105 180 */
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, color));
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
gsk_ngl_render_job_draw_rect_with_color (job, &node->bounds, pink);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
static inline void
gsk_ngl_render_job_visit_gl_shader_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
GError *error = NULL;
GskGLShader *shader;
GskNglProgram *program;
int n_children;
shader = gsk_gl_shader_node_get_shader (node);
program = gsk_ngl_driver_lookup_shader (job->driver, shader, &error);
n_children = gsk_gl_shader_node_get_n_children (node);
if G_UNLIKELY (program == NULL)
{
if (g_object_get_data (G_OBJECT (shader), "gsk-did-warn") == NULL)
{
g_object_set_data (G_OBJECT (shader), "gsk-did-warn", GUINT_TO_POINTER (1));
g_warning ("Failed to compile gl shader: %s", error->message);
}
gsk_ngl_render_job_visit_gl_shader_node_fallback (job, node);
g_clear_error (&error);
}
else
{
GskNglRenderOffscreen offscreens[4] = {{0}};
const GskGLUniform *uniforms;
const guint8 *base;
GBytes *args;
int n_uniforms;
g_assert (n_children < G_N_ELEMENTS (offscreens));
for (guint i = 0; i < n_children; i++)
{
const GskRenderNode *child = gsk_gl_shader_node_get_child (node, i);
offscreens[i].bounds = &node->bounds;
offscreens[i].force_offscreen = TRUE;
offscreens[i].reset_clip = TRUE;
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, child, &offscreens[i]))
return;
}
args = gsk_gl_shader_node_get_args (node);
base = g_bytes_get_data (args, NULL);
uniforms = gsk_gl_shader_get_uniforms (shader, &n_uniforms);
gsk_ngl_render_job_begin_draw (job, program);
for (guint i = 0; i < n_children; i++)
gsk_ngl_program_set_uniform_texture (program,
UNIFORM_CUSTOM_TEXTURE1 + i, 0,
GL_TEXTURE_2D,
GL_TEXTURE0 + i,
offscreens[i].texture_id);
gsk_ngl_program_set_uniform2f (program,
UNIFORM_CUSTOM_SIZE, 0,
node->bounds.size.width,
node->bounds.size.height);
for (guint i = 0; i < n_uniforms; i++)
{
const GskGLUniform *u = &uniforms[i];
const guint8 *data = base + u->offset;
switch (u->type)
{
default:
case GSK_GL_UNIFORM_TYPE_NONE:
break;
case GSK_GL_UNIFORM_TYPE_FLOAT:
gsk_ngl_uniform_state_set1fv (job->command_queue->uniforms,
program->program_info,
ngl: move uniform key mapping into uniform state Previously, we translated the uniform key (an enum) into a location within the shader program in GskNglProgram. A number of performance improvements were focused around having low nubers for the uniform locations. Generally this is the case, but some drivers such as old Intel drivers on Windows may use rather large numbers for those. To combat this, we can push the translation of uniform keys into locations at the GskNglUniformState level so that we work with unranslated keys through the process until applying them. Fixes #3780
2021-03-23 19:10:12 +00:00
UNIFORM_CUSTOM_ARG0 + i,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
0, 1, (const float *)data);
break;
case GSK_GL_UNIFORM_TYPE_INT:
gsk_ngl_uniform_state_set1i (job->command_queue->uniforms,
program->program_info,
ngl: move uniform key mapping into uniform state Previously, we translated the uniform key (an enum) into a location within the shader program in GskNglProgram. A number of performance improvements were focused around having low nubers for the uniform locations. Generally this is the case, but some drivers such as old Intel drivers on Windows may use rather large numbers for those. To combat this, we can push the translation of uniform keys into locations at the GskNglUniformState level so that we work with unranslated keys through the process until applying them. Fixes #3780
2021-03-23 19:10:12 +00:00
UNIFORM_CUSTOM_ARG0 + i,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
0, *(const gint32 *)data);
break;
case GSK_GL_UNIFORM_TYPE_UINT:
case GSK_GL_UNIFORM_TYPE_BOOL:
gsk_ngl_uniform_state_set1ui (job->command_queue->uniforms,
program->program_info,
ngl: move uniform key mapping into uniform state Previously, we translated the uniform key (an enum) into a location within the shader program in GskNglProgram. A number of performance improvements were focused around having low nubers for the uniform locations. Generally this is the case, but some drivers such as old Intel drivers on Windows may use rather large numbers for those. To combat this, we can push the translation of uniform keys into locations at the GskNglUniformState level so that we work with unranslated keys through the process until applying them. Fixes #3780
2021-03-23 19:10:12 +00:00
UNIFORM_CUSTOM_ARG0 + i,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
0, *(const guint32 *)data);
break;
case GSK_GL_UNIFORM_TYPE_VEC2:
gsk_ngl_uniform_state_set2fv (job->command_queue->uniforms,
program->program_info,
ngl: move uniform key mapping into uniform state Previously, we translated the uniform key (an enum) into a location within the shader program in GskNglProgram. A number of performance improvements were focused around having low nubers for the uniform locations. Generally this is the case, but some drivers such as old Intel drivers on Windows may use rather large numbers for those. To combat this, we can push the translation of uniform keys into locations at the GskNglUniformState level so that we work with unranslated keys through the process until applying them. Fixes #3780
2021-03-23 19:10:12 +00:00
UNIFORM_CUSTOM_ARG0 + i,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
0, 1, (const float *)data);
break;
case GSK_GL_UNIFORM_TYPE_VEC3:
gsk_ngl_uniform_state_set3fv (job->command_queue->uniforms,
program->program_info,
ngl: move uniform key mapping into uniform state Previously, we translated the uniform key (an enum) into a location within the shader program in GskNglProgram. A number of performance improvements were focused around having low nubers for the uniform locations. Generally this is the case, but some drivers such as old Intel drivers on Windows may use rather large numbers for those. To combat this, we can push the translation of uniform keys into locations at the GskNglUniformState level so that we work with unranslated keys through the process until applying them. Fixes #3780
2021-03-23 19:10:12 +00:00
UNIFORM_CUSTOM_ARG0 + i,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
0, 1, (const float *)data);
break;
case GSK_GL_UNIFORM_TYPE_VEC4:
gsk_ngl_uniform_state_set4fv (job->command_queue->uniforms,
program->program_info,
ngl: move uniform key mapping into uniform state Previously, we translated the uniform key (an enum) into a location within the shader program in GskNglProgram. A number of performance improvements were focused around having low nubers for the uniform locations. Generally this is the case, but some drivers such as old Intel drivers on Windows may use rather large numbers for those. To combat this, we can push the translation of uniform keys into locations at the GskNglUniformState level so that we work with unranslated keys through the process until applying them. Fixes #3780
2021-03-23 19:10:12 +00:00
UNIFORM_CUSTOM_ARG0 + i,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
0, 1, (const float *)data);
break;
}
}
gsk_ngl_render_job_draw_offscreen_rect (job, &node->bounds);
gsk_ngl_render_job_end_draw (job);
}
}
static void
gsk_ngl_render_job_upload_texture (GskNglRenderJob *job,
GdkTexture *texture,
GskNglRenderOffscreen *offscreen)
{
if (gsk_ngl_texture_library_can_cache (GSK_NGL_TEXTURE_LIBRARY (job->driver->icons),
texture->width,
texture->height) &&
!GDK_IS_GL_TEXTURE (texture))
{
const GskNglIconData *icon_data;
gsk_ngl_icon_library_lookup_or_add (job->driver->icons, texture, &icon_data);
offscreen->texture_id = GSK_NGL_TEXTURE_ATLAS_ENTRY_TEXTURE (icon_data);
memcpy (&offscreen->area, &icon_data->entry.area, sizeof offscreen->area);
}
else
{
offscreen->texture_id = gsk_ngl_driver_load_texture (job->driver, texture, GL_LINEAR, GL_LINEAR);
init_full_texture_region (offscreen);
}
}
static inline void
gsk_ngl_render_job_visit_texture_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
GdkTexture *texture = gsk_texture_node_get_texture (node);
int max_texture_size = job->command_queue->max_texture_size;
if G_LIKELY (texture->width <= max_texture_size &&
texture->height <= max_texture_size)
{
GskNglRenderOffscreen offscreen = {0};
gsk_ngl_render_job_upload_texture (job, texture, &offscreen);
g_assert (offscreen.texture_id);
g_assert (offscreen.was_offscreen == FALSE);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen.texture_id);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen (job, &node->bounds, &offscreen);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
else
{
float min_x = job->offset_x + node->bounds.origin.x;
float min_y = job->offset_y + node->bounds.origin.y;
float max_x = min_x + node->bounds.size.width;
float max_y = min_y + node->bounds.size.height;
float scale_x = (max_x - min_x) / texture->width;
float scale_y = (max_y - min_y) / texture->height;
GskNglTextureSlice *slices = NULL;
guint n_slices = 0;
gsk_ngl_driver_slice_texture (job->driver, texture, &slices, &n_slices);
g_assert (slices != NULL);
g_assert (n_slices > 0);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
for (guint i = 0; i < n_slices; i ++)
{
const GskNglTextureSlice *slice = &slices[i];
float x1, x2, y1, y2;
x1 = min_x + (scale_x * slice->rect.x);
x2 = x1 + (slice->rect.width * scale_x);
y1 = min_y + (scale_y * slice->rect.y);
y2 = y1 + (slice->rect.height * scale_y);
if (i > 0)
gsk_ngl_render_job_split_draw (job);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
slice->texture_id);
ngl: Set color as vertex attribute Instead of using uniforms for color used in multiple programs, pass it as vertex attributes. This will let us batch more draw calls, since we don't have to change uniforms so often. In particular, for syntax-highlighted text.
2021-03-12 00:46:15 +00:00
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_coords (job,
x1, y1, x2, y2,
0, 0, 1, 1,
Move color conversion up Doing the color conversion early lets us reuse the results whenever we do more than one draw per node.
2021-04-07 03:52:33 +00:00
(guint16[]) { FP16_ZERO, FP16_ZERO, FP16_ZERO, FP16_ZERO });
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
gsk_ngl_render_job_end_draw (job);
}
}
static inline void
gsk_ngl_render_job_visit_repeat_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
const GskRenderNode *child = gsk_repeat_node_get_child (node);
const graphene_rect_t *child_bounds = gsk_repeat_node_get_child_bounds (node);
GskNglRenderOffscreen offscreen = {0};
if (node_is_invisible (child))
return;
if (!graphene_rect_equal (child_bounds, &child->bounds))
{
/* TODO: implement these repeat nodes. */
gsk_ngl_render_job_visit_as_fallback (job, node);
return;
}
/* If the size of the repeat node is smaller than the size of the
* child node, we don't repeat at all and can just draw that part
* of the child texture... */
if (rect_contains_rect (child_bounds, &node->bounds))
{
gsk_ngl_render_job_visit_clipped_child (job, child, &node->bounds);
return;
}
offscreen.bounds = &child->bounds;
offscreen.reset_clip = TRUE;
if (!gsk_ngl_render_job_visit_node_with_offscreen (job, child, &offscreen))
g_assert_not_reached ();
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, repeat));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
offscreen.texture_id);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform4f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_REPEAT_CHILD_BOUNDS, 0,
(node->bounds.origin.x - child_bounds->origin.x) / child_bounds->size.width,
(node->bounds.origin.y - child_bounds->origin.y) / child_bounds->size.height,
node->bounds.size.width / child_bounds->size.width,
node->bounds.size.height / child_bounds->size.height);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_program_set_uniform4f (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_REPEAT_TEXTURE_RECT, 0,
offscreen.area.x,
offscreen.was_offscreen ? offscreen.area.y2 : offscreen.area.y,
offscreen.area.x2,
offscreen.was_offscreen ? offscreen.area.y : offscreen.area.y2);
ngl: Drop gsk_ngl_render_job_set_color Colors are not state that we carry across draw ops, so setting the color on the render job doesn't make much sense. Instead, pass the color to the various draw calls. Add a few new ones for that purpose. Also, shorten the names of some by going from 'load_vertices_from_offscreen' to 'draw_offscreen'.
2021-03-13 19:29:34 +00:00
gsk_ngl_render_job_draw_offscreen (job, &node->bounds, &offscreen);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_end_draw (job);
}
static void
gsk_ngl_render_job_visit_node (GskNglRenderJob *job,
const GskRenderNode *node)
{
ngl: Classify clips Maintain the 'fully contained' information in the clip stack. This will let us pick simpler shaders in the future.
2021-03-07 15:02:46 +00:00
gboolean has_clip;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
g_assert (job != NULL);
g_assert (node != NULL);
g_assert (GSK_IS_NGL_DRIVER (job->driver));
g_assert (GSK_IS_NGL_COMMAND_QUEUE (job->command_queue));
ngl: Classify clips Maintain the 'fully contained' information in the clip stack. This will let us pick simpler shaders in the future.
2021-03-07 15:02:46 +00:00
if (node_is_invisible (node))
return;
ngl: Allow passing bounds to update_clip update_clip only uses the bounds of the node that we are currently passing. This opens the door to updating the clip while handling a single node.
2021-04-11 15:57:22 +00:00
if (!gsk_ngl_render_job_update_clip (job, &node->bounds, &has_clip))
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
return;
switch (gsk_render_node_get_node_type (node))
{
case GSK_BLEND_NODE:
gsk_ngl_render_job_visit_blend_node (job, node);
break;
case GSK_BLUR_NODE:
if (gsk_blur_node_get_radius (node) > 0)
gsk_ngl_render_job_visit_blur_node (job, node);
else
gsk_ngl_render_job_visit_node (job, gsk_blur_node_get_child (node));
break;
case GSK_BORDER_NODE:
ngl: Render simple borders simply Use 4 color commands for uniform, rectilinear borders. Most of our borders are rounded with Adwaita, but there are still some of these.
2021-03-09 02:37:28 +00:00
if (gsk_border_node_get_uniform_color (node) &&
gsk_rounded_rect_is_rectilinear (gsk_border_node_get_outline (node)))
gsk_ngl_render_job_visit_rect_border_node (job, node);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
else
gsk_ngl_render_job_visit_border_node (job, node);
break;
case GSK_CLIP_NODE:
gsk_ngl_render_job_visit_clip_node (job, node);
break;
case GSK_COLOR_NODE:
gsk_ngl_render_job_visit_color_node (job, node);
break;
case GSK_COLOR_MATRIX_NODE:
gsk_ngl_render_job_visit_color_matrix_node (job, node);
break;
case GSK_CONIC_GRADIENT_NODE:
if (gsk_conic_gradient_node_get_n_color_stops (node) < MAX_GRADIENT_STOPS)
gsk_ngl_render_job_visit_conic_gradient_node (job, node);
else
gsk_ngl_render_job_visit_as_fallback (job, node);
break;
case GSK_CONTAINER_NODE:
{
guint n_children = gsk_container_node_get_n_children (node);
for (guint i = 0; i < n_children; i++)
{
const GskRenderNode *child = gsk_container_node_get_child (node, i);
ngl: Special-case css backgrounds Recognize a common pattern: A rounded clip with a color node, followed by a border node, with the same outline. This is what CSS backgrounds frequently produce, and we can render it more efficiently with a combined shader.
2021-03-14 04:30:18 +00:00
if (i + 1 < n_children &&
job->current_clip->is_fully_contained &&
gsk_render_node_get_node_type (child) == GSK_ROUNDED_CLIP_NODE)
{
const GskRenderNode *grandchild = gsk_rounded_clip_node_get_child (child);
const GskRenderNode *child2 = gsk_container_node_get_child (node, i + 1);
if (gsk_render_node_get_node_type (grandchild) == GSK_COLOR_NODE &&
gsk_render_node_get_node_type (child2) == GSK_BORDER_NODE &&
gsk_border_node_get_uniform_color (child2) &&
ngl: Small optimization We really want equality here.
2021-04-11 00:26:45 +00:00
rounded_rect_equal (gsk_rounded_clip_node_get_clip (child),
gsk_border_node_get_outline (child2)))
ngl: Special-case css backgrounds Recognize a common pattern: A rounded clip with a color node, followed by a border node, with the same outline. This is what CSS backgrounds frequently produce, and we can render it more efficiently with a combined shader.
2021-03-14 04:30:18 +00:00
{
gsk_ngl_render_job_visit_css_background (job, child, child2);
i++; /* skip the border node */
continue;
}
}
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_visit_node (job, child);
}
}
break;
case GSK_CROSS_FADE_NODE:
{
const GskRenderNode *start_node = gsk_cross_fade_node_get_start_child (node);
const GskRenderNode *end_node = gsk_cross_fade_node_get_end_child (node);
float progress = gsk_cross_fade_node_get_progress (node);
if (progress <= 0.0f)
gsk_ngl_render_job_visit_node (job, gsk_cross_fade_node_get_start_child (node));
else if (progress >= 1.0f || equal_texture_nodes (start_node, end_node))
gsk_ngl_render_job_visit_node (job, gsk_cross_fade_node_get_end_child (node));
else
gsk_ngl_render_job_visit_cross_fade_node (job, node);
}
break;
case GSK_DEBUG_NODE:
/* Debug nodes are ignored because draws get reordered anyway */
gsk_ngl_render_job_visit_node (job, gsk_debug_node_get_child (node));
break;
case GSK_GL_SHADER_NODE:
gsk_ngl_render_job_visit_gl_shader_node (job, node);
break;
case GSK_INSET_SHADOW_NODE:
if (gsk_inset_shadow_node_get_blur_radius (node) > 0)
gsk_ngl_render_job_visit_blurred_inset_shadow_node (job, node);
else
gsk_ngl_render_job_visit_unblurred_inset_shadow_node (job, node);
break;
case GSK_LINEAR_GRADIENT_NODE:
case GSK_REPEATING_LINEAR_GRADIENT_NODE:
if (gsk_linear_gradient_node_get_n_color_stops (node) < MAX_GRADIENT_STOPS)
gsk_ngl_render_job_visit_linear_gradient_node (job, node);
else
gsk_ngl_render_job_visit_as_fallback (job, node);
break;
case GSK_OPACITY_NODE:
gsk_ngl_render_job_visit_opacity_node (job, node);
break;
case GSK_OUTSET_SHADOW_NODE:
if (gsk_outset_shadow_node_get_blur_radius (node) > 0)
gsk_ngl_render_job_visit_blurred_outset_shadow_node (job, node);
else
gsk_ngl_render_job_visit_unblurred_outset_shadow_node (job, node);
break;
case GSK_RADIAL_GRADIENT_NODE:
case GSK_REPEATING_RADIAL_GRADIENT_NODE:
ngl: Treat radial gradients like others Handle colorstop overflow for radial gradients the same way we do for linear and conic gradients.
2021-03-07 15:17:34 +00:00
if (gsk_radial_gradient_node_get_n_color_stops (node) < MAX_GRADIENT_STOPS)
gsk_ngl_render_job_visit_radial_gradient_node (job, node);
else
gsk_ngl_render_job_visit_as_fallback (job, node);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
break;
case GSK_REPEAT_NODE:
gsk_ngl_render_job_visit_repeat_node (job, node);
break;
case GSK_ROUNDED_CLIP_NODE:
gsk_ngl_render_job_visit_rounded_clip_node (job, node);
break;
case GSK_SHADOW_NODE:
gsk_ngl_render_job_visit_shadow_node (job, node);
break;
case GSK_TEXT_NODE:
gsk_ngl_render_job_visit_text_node (job,
node,
gsk_text_node_get_color (node),
FALSE);
break;
case GSK_TEXTURE_NODE:
gsk_ngl_render_job_visit_texture_node (job, node);
break;
case GSK_TRANSFORM_NODE:
gsk_ngl_render_job_visit_transform_node (job, node);
break;
case GSK_CAIRO_NODE:
gsk_ngl_render_job_visit_as_fallback (job, node);
break;
case GSK_NOT_A_RENDER_NODE:
default:
g_assert_not_reached ();
break;
}
ngl: Classify clips Maintain the 'fully contained' information in the clip stack. This will let us pick simpler shaders in the future.
2021-03-07 15:02:46 +00:00
if (has_clip)
gsk_ngl_render_job_pop_clip (job);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
}
static gboolean
gsk_ngl_render_job_visit_node_with_offscreen (GskNglRenderJob *job,
const GskRenderNode *node,
GskNglRenderOffscreen *offscreen)
{
GskTextureKey key;
guint cached_id;
int filter;
g_assert (job != NULL);
g_assert (node != NULL);
g_assert (offscreen != NULL);
g_assert (offscreen->texture_id == 0);
g_assert (offscreen->bounds != NULL);
if (node_is_invisible (node))
{
/* Just to be safe. */
offscreen->texture_id = 0;
init_full_texture_region (offscreen);
offscreen->was_offscreen = FALSE;
return FALSE;
}
if (gsk_render_node_get_node_type (node) == GSK_TEXTURE_NODE &&
offscreen->force_offscreen == FALSE)
{
GdkTexture *texture = gsk_texture_node_get_texture (node);
gsk_ngl_render_job_upload_texture (job, texture, offscreen);
g_assert (offscreen->was_offscreen == FALSE);
return TRUE;
}
filter = offscreen->linear_filter ? GL_LINEAR : GL_NEAREST;
/* Check if we've already cached the drawn texture. */
key.pointer = node;
key.pointer_is_child = TRUE; /* Don't conflict with the child using the cache too */
key.parent_rect = *offscreen->bounds;
key.scale_x = job->scale_x;
key.scale_y = job->scale_y;
key.filter = filter;
cached_id = gsk_ngl_driver_lookup_texture (job->driver, &key);
if (cached_id != 0)
{
offscreen->texture_id = cached_id;
init_full_texture_region (offscreen);
/* We didn't render it offscreen, but hand out an offscreen texture id */
offscreen->was_offscreen = TRUE;
return TRUE;
}
float scaled_width;
float scaled_height;
ngl: Fix offscreen rendering with transforms There was confusion here about the handling of the modelview transform. The modelview transform we are getting is already set up for rendering the node we are given, so keep it - except for possible adding an extra scale on top when the texture would otherwise be too big.
2021-03-31 18:36:24 +00:00
float downscale_x = 1;
float downscale_y = 1;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
g_assert (job->command_queue->max_texture_size > 0);
/* Tweak the scale factor so that the required texture doesn't
* exceed the max texture limit. This will render with a lower
* resolution, but this is better than clipping.
*/
{
int max_texture_size = job->command_queue->max_texture_size;
ngl: Fix offscreen rendering with transforms There was confusion here about the handling of the modelview transform. The modelview transform we are getting is already set up for rendering the node we are given, so keep it - except for possible adding an extra scale on top when the texture would otherwise be too big.
2021-03-31 18:36:24 +00:00
scaled_width = ceilf (offscreen->bounds->size.width * job->scale_x);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (scaled_width > max_texture_size)
{
ngl: Fix offscreen rendering with transforms There was confusion here about the handling of the modelview transform. The modelview transform we are getting is already set up for rendering the node we are given, so keep it - except for possible adding an extra scale on top when the texture would otherwise be too big.
2021-03-31 18:36:24 +00:00
downscale_x = (float)max_texture_size / scaled_width;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
scaled_width = max_texture_size;
}
ngl: Fix offscreen rendering with transforms There was confusion here about the handling of the modelview transform. The modelview transform we are getting is already set up for rendering the node we are given, so keep it - except for possible adding an extra scale on top when the texture would otherwise be too big.
2021-03-31 18:36:24 +00:00
scaled_height = ceilf (offscreen->bounds->size.height * job->scale_y);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
if (scaled_height > max_texture_size)
{
ngl: Fix offscreen rendering with transforms There was confusion here about the handling of the modelview transform. The modelview transform we are getting is already set up for rendering the node we are given, so keep it - except for possible adding an extra scale on top when the texture would otherwise be too big.
2021-03-31 18:36:24 +00:00
downscale_y = (float)max_texture_size / scaled_height;
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
scaled_height = max_texture_size;
}
}
GskNglRenderTarget *render_target;
graphene_matrix_t prev_projection;
graphene_rect_t prev_viewport;
graphene_rect_t viewport;
float offset_x = job->offset_x;
float offset_y = job->offset_y;
float prev_alpha;
guint prev_fbo;
if (!gsk_ngl_driver_create_render_target (job->driver,
scaled_width, scaled_height,
filter, filter,
&render_target))
g_assert_not_reached ();
if (gdk_gl_context_has_debug (job->command_queue->context))
{
gdk_gl_context_label_object_printf (job->command_queue->context,
GL_TEXTURE,
render_target->texture_id,
"Offscreen<%s> %d",
g_type_name_from_instance ((GTypeInstance *) node),
render_target->texture_id);
gdk_gl_context_label_object_printf (job->command_queue->context,
GL_FRAMEBUFFER,
render_target->framebuffer_id,
"Offscreen<%s> FB %d",
g_type_name_from_instance ((GTypeInstance *) node),
render_target->framebuffer_id);
}
ngl: Fix downscaled textures It is not pretty, but at least it works now.
2021-04-02 22:33:01 +00:00
if (downscale_x != 1 || downscale_y != 1)
ngl: Plug a memory leak This was introduced in f9457af128012fd3935a69fdcefb.
2021-04-02 22:41:53 +00:00
{
GskTransform *transform = gsk_transform_scale (NULL, downscale_x, downscale_y);
gsk_ngl_render_job_push_modelview (job, transform);
gsk_transform_unref (transform);
}
ngl: Fix downscaled textures It is not pretty, but at least it works now.
2021-04-02 22:33:01 +00:00
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_transform_bounds (job, offscreen->bounds, &viewport);
/* Code above will scale the size with the scale we use in the render ops,
* but for the viewport size, we need our own size limited by the texture size */
viewport.size.width = scaled_width;
viewport.size.height = scaled_height;
gsk_ngl_render_job_set_viewport (job, &viewport, &prev_viewport);
gsk_ngl_render_job_set_projection_from_rect (job, &job->viewport, &prev_projection);
prev_alpha = gsk_ngl_render_job_set_alpha (job, 1.0f);
prev_fbo = gsk_ngl_command_queue_bind_framebuffer (job->command_queue, render_target->framebuffer_id);
gsk_ngl_command_queue_clear (job->command_queue, 0, &job->viewport);
if (offscreen->reset_clip)
gsk_ngl_render_job_push_clip (job, &GSK_ROUNDED_RECT_INIT_FROM_RECT (job->viewport));
gsk_ngl_render_job_visit_node (job, node);
if (offscreen->reset_clip)
gsk_ngl_render_job_pop_clip (job);
ngl: Fix offscreen rendering with transforms There was confusion here about the handling of the modelview transform. The modelview transform we are getting is already set up for rendering the node we are given, so keep it - except for possible adding an extra scale on top when the texture would otherwise be too big.
2021-03-31 18:36:24 +00:00
if (downscale_x != 1 || downscale_y != 1)
gsk_ngl_render_job_pop_modelview (job);
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
gsk_ngl_render_job_set_viewport (job, &prev_viewport, NULL);
gsk_ngl_render_job_set_projection (job, &prev_projection);
gsk_ngl_render_job_set_alpha (job, prev_alpha);
gsk_ngl_command_queue_bind_framebuffer (job->command_queue, prev_fbo);
job->offset_x = offset_x;
job->offset_y = offset_y;
offscreen->was_offscreen = TRUE;
offscreen->texture_id = gsk_ngl_driver_release_render_target (job->driver,
render_target,
FALSE);
init_full_texture_region (offscreen);
if (!offscreen->do_not_cache)
gsk_ngl_driver_cache_texture (job->driver, &key, offscreen->texture_id);
return TRUE;
}
void
gsk_ngl_render_job_render_flipped (GskNglRenderJob *job,
GskRenderNode *root)
{
graphene_matrix_t proj;
guint framebuffer_id;
guint texture_id;
guint surface_height;
g_return_if_fail (job != NULL);
g_return_if_fail (root != NULL);
g_return_if_fail (GSK_IS_NGL_DRIVER (job->driver));
surface_height = job->viewport.size.height;
graphene_matrix_init_ortho (&proj,
job->viewport.origin.x,
job->viewport.origin.x + job->viewport.size.width,
job->viewport.origin.y,
job->viewport.origin.y + job->viewport.size.height,
ORTHO_NEAR_PLANE,
ORTHO_FAR_PLANE);
graphene_matrix_scale (&proj, 1, -1, 1);
if (!gsk_ngl_command_queue_create_render_target (job->command_queue,
MAX (1, job->viewport.size.width),
MAX (1, job->viewport.size.height),
GL_NEAREST, GL_NEAREST,
&framebuffer_id, &texture_id))
return;
/* Setup drawing to our offscreen texture/framebuffer which is flipped */
gsk_ngl_command_queue_bind_framebuffer (job->command_queue, framebuffer_id);
gsk_ngl_command_queue_clear (job->command_queue, 0, &job->viewport);
/* Visit all nodes creating batches */
gdk_gl_context_push_debug_group (job->command_queue->context, "Building command queue");
gsk_ngl_render_job_visit_node (job, root);
gdk_gl_context_pop_debug_group (job->command_queue->context);
/* Now draw to our real destination, but flipped */
gsk_ngl_render_job_set_alpha (job, 1.0f);
gsk_ngl_command_queue_bind_framebuffer (job->command_queue, job->framebuffer);
gsk_ngl_command_queue_clear (job->command_queue, 0, &job->viewport);
ngl: Add infrastructure for clip-variants of shaders Make ngl compile each shader three times. Once with #define NO_CLIP 1 once with #define RECT_CLIP 1 and once as before.
2021-03-07 22:50:19 +00:00
gsk_ngl_render_job_begin_draw (job, CHOOSE_PROGRAM (job, blit));
gsk_ngl_program_set_uniform_texture (job->current_program,
gsk: add OpenGL based GskNglRenderer The primary goal here was to cleanup the current GL renderer to make maintenance easier going forward. Furthermore, it tracks state to allow us to implement more advanced renderer features going forward. Reordering This renderer will reorder batches by render target to reduce the number of times render targets are changed. In the future, we could also reorder by program within the render target if we can determine that vertices do not overlap. Uniform Snapshots To allow for reordering of batches all uniforms need to be tracked for the programs. This allows us to create the full uniform state when the batch has been moved into a new position. Some care was taken as it can be performance sensitive. Attachment Snapshots Similar to uniform snapshots, we need to know all of the texture attachments so that we can rebind them when necessary. Render Jobs To help isolate the process of creating GL commands from the renderer abstraction a render job abstraction was added. This could be extended in the future if we decided to do tiling. Command Queue Render jobs create batches using the command queue. The command queue will snapshot uniform and attachment state so that it can reorder batches right before executing them. Currently, the only reordering done is to ensure that we only visit each render target once. We could extend this by tracking vertices, attachments, and others. This code currently uses an inline array helper to reduce overhead from GArray which was showing up on profiles. It could be changed to use GdkArray without too much work, but had roughly double the instructions. Cycle counts have not yet been determined. GLSL Programs This was simplified to use XMACROS so that we can just extend one file (gskglprograms.defs) instead of multiple places. The programs are added as fields in the driver for easy access. Driver The driver manages textures, render targets, access to atlases, programs, and more. There is one driver per display, by using the shared GL context. Some work could be done here to batch uploads so that we make fewer calls to upload when sending icon theme data to the GPU. We'd need to keep a copy of the atlas data for such purposes.
2020-12-19 01:36:59 +00:00
UNIFORM_SHARED_SOURCE, 0,
GL_TEXTURE_2D,
GL_TEXTURE0,
texture_id);
gsk_ngl_render_job_draw_rect (job, &job->viewport);
gsk_ngl_render_job_end_draw (job);
gdk_gl_context_push_debug_group (job->command_queue->context, "Executing command queue");
gsk_ngl_command_queue_execute (job->command_queue, surface_height, 1, NULL);
gdk_gl_context_pop_debug_group (job->command_queue->context);
glDeleteFramebuffers (1, &framebuffer_id);
glDeleteTextures (1, &texture_id);
}
void
gsk_ngl_render_job_render (GskNglRenderJob *job,
GskRenderNode *root)
{
G_GNUC_UNUSED gint64 start_time;
guint scale_factor;
guint surface_height;
g_return_if_fail (job != NULL);
g_return_if_fail (root != NULL);
g_return_if_fail (GSK_IS_NGL_DRIVER (job->driver));
scale_factor = MAX (job->scale_x, job->scale_y);
surface_height = job->viewport.size.height;
gsk_ngl_command_queue_make_current (job->command_queue);
/* Build the command queue using the shared GL context for all renderers
* on the same display.
*/
start_time = GDK_PROFILER_CURRENT_TIME;
gdk_gl_context_push_debug_group (job->command_queue->context, "Building command queue");
gsk_ngl_command_queue_bind_framebuffer (job->command_queue, job->framebuffer);
gsk_ngl_command_queue_clear (job->command_queue, 0, &job->viewport);
gsk_ngl_render_job_visit_node (job, root);
gdk_gl_context_pop_debug_group (job->command_queue->context);
gdk_profiler_add_mark (start_time, GDK_PROFILER_CURRENT_TIME-start_time, "Build GL command queue", "");
#if 0
/* At this point the atlases have uploaded content while we processed
* nodes but have not necessarily been used by the commands in the queue.
*/
gsk_ngl_driver_save_atlases_to_png (job->driver, NULL);
#endif
/* But now for executing the command queue, we want to use the context
* that was provided to us when creating the render job as framebuffer 0
* is bound to that context.
*/
start_time = GDK_PROFILER_CURRENT_TIME;
gsk_ngl_command_queue_make_current (job->command_queue);
gdk_gl_context_push_debug_group (job->command_queue->context, "Executing command queue");
gsk_ngl_command_queue_execute (job->command_queue, surface_height, scale_factor, job->region);
gdk_gl_context_pop_debug_group (job->command_queue->context);
gdk_profiler_add_mark (start_time, GDK_PROFILER_CURRENT_TIME-start_time, "Execute GL command queue", "");
}
void
gsk_ngl_render_job_set_debug_fallback (GskNglRenderJob *job,
gboolean debug_fallback)
{
g_return_if_fail (job != NULL);
job->debug_fallback = !!debug_fallback;
}
GskNglRenderJob *
gsk_ngl_render_job_new (GskNglDriver *driver,
const graphene_rect_t *viewport,
float scale_factor,
const cairo_region_t *region,
guint framebuffer)
{
const graphene_rect_t *clip_rect = viewport;
graphene_rect_t transformed_extents;
GskNglRenderJob *job;
g_return_val_if_fail (GSK_IS_NGL_DRIVER (driver), NULL);
g_return_val_if_fail (viewport != NULL, NULL);
g_return_val_if_fail (scale_factor > 0, NULL);
job = g_slice_new0 (GskNglRenderJob);
job->driver = g_object_ref (driver);
job->command_queue = job->driver->command_queue;
job->clip = g_array_sized_new (FALSE, FALSE, sizeof (GskNglRenderClip), 16);
job->modelview = g_array_sized_new (FALSE, FALSE, sizeof (GskNglRenderModelview), 16);
job->framebuffer = framebuffer;
job->offset_x = 0;
job->offset_y = 0;
job->scale_x = scale_factor;
job->scale_y = scale_factor;
job->viewport = *viewport;
gsk_ngl_render_job_set_alpha (job, 1.0);
gsk_ngl_render_job_set_projection_from_rect (job, viewport, NULL);
gsk_ngl_render_job_set_modelview (job, gsk_transform_scale (NULL, scale_factor, scale_factor));
/* Setup our initial clip. If region is NULL then we are drawing the
* whole viewport. Otherwise, we need to convert the region to a
* bounding box and clip based on that.
*/
if (region != NULL)
{
cairo_rectangle_int_t extents;
cairo_region_get_extents (region, &extents);
gsk_ngl_render_job_transform_bounds (job,
&GRAPHENE_RECT_INIT (extents.x,
extents.y,
extents.width,
extents.height),
&transformed_extents);
clip_rect = &transformed_extents;
job->region = cairo_region_create_rectangle (&extents);
}
gsk_ngl_render_job_push_clip (job,
&GSK_ROUNDED_RECT_INIT (clip_rect->origin.x,
clip_rect->origin.y,
clip_rect->size.width,
clip_rect->size.height));
return job;
}
void
gsk_ngl_render_job_free (GskNglRenderJob *job)
{
job->current_modelview = NULL;
job->current_clip = NULL;
while (job->modelview->len > 0)
{
GskNglRenderModelview *modelview = &g_array_index (job->modelview, GskNglRenderModelview, job->modelview->len-1);
g_clear_pointer (&modelview->transform, gsk_transform_unref);
job->modelview->len--;
}
g_clear_object (&job->driver);
g_clear_pointer (&job->region, cairo_region_destroy);
g_clear_pointer (&job->modelview, g_array_unref);
g_clear_pointer (&job->clip, g_array_unref);
g_slice_free (GskNglRenderJob, job);
}
Reference in New Issue Copy Permalink