gtk/gsk/ngl/gskngldriver.c

1342 lines
46 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
/* gskngldriver.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/gdktextureprivate.h>
#include <gsk/gskdebugprivate.h>
#include <gsk/gskglshaderprivate.h>
#include <gsk/gskrendererprivate.h>
#include "gsknglcommandqueueprivate.h"
#include "gsknglcompilerprivate.h"
#include "gskngldriverprivate.h"
#include "gsknglglyphlibraryprivate.h"
#include "gskngliconlibraryprivate.h"
#include "gsknglprogramprivate.h"
#include "gsknglshadowlibraryprivate.h"
#include "gskngltexturepoolprivate.h"
#define ATLAS_SIZE 512
#define MAX_OLD_RATIO 0.5
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
typedef struct _GskNglTextureState
{
GdkGLContext *context;
GLuint texture_id;
} GskNglTextureState;
G_DEFINE_TYPE (GskNglDriver, gsk_ngl_driver, G_TYPE_OBJECT)
static guint
texture_key_hash (gconstpointer v)
{
const GskTextureKey *k = (const GskTextureKey *)v;
/* Optimize for 0..3 where 0 is the scaled out case. Usually
* we'll be squarely on 1 or 2 for standard vs HiDPI. When rendering
* to a texture scaled out like in node-editor, we might be < 1.
*/
guint scale_x = floorf (k->scale_x);
guint scale_y = floorf (k->scale_y);
return GPOINTER_TO_SIZE (k->pointer) ^
((scale_x << 8) |
(scale_y << 6) |
(k->filter << 1) |
k->pointer_is_child);
}
static gboolean
texture_key_equal (gconstpointer v1,
gconstpointer v2)
{
const GskTextureKey *k1 = (const GskTextureKey *)v1;
const GskTextureKey *k2 = (const GskTextureKey *)v2;
return k1->pointer == k2->pointer &&
k1->scale_x == k2->scale_x &&
k1->scale_y == k2->scale_y &&
k1->filter == k2->filter &&
k1->pointer_is_child == k2->pointer_is_child &&
(!k1->pointer_is_child || memcmp (&k1->parent_rect, &k2->parent_rect, sizeof k1->parent_rect) == 0);
}
static void
remove_texture_key_for_id (GskNglDriver *self,
guint texture_id)
{
GskTextureKey *key;
g_assert (GSK_IS_NGL_DRIVER (self));
g_assert (texture_id > 0);
/* g_hash_table_remove() will cause @key to be freed */
if (g_hash_table_steal_extended (self->texture_id_to_key,
GUINT_TO_POINTER (texture_id),
NULL,
(gpointer *)&key))
g_hash_table_remove (self->key_to_texture_id, key);
}
static void
gsk_ngl_texture_destroyed (gpointer data)
{
((GskNglTexture *)data)->user = NULL;
}
static guint
gsk_ngl_driver_collect_unused_textures (GskNglDriver *self,
gint64 watermark)
{
GHashTableIter iter;
gpointer k, v;
guint old_size;
guint collected;
g_assert (GSK_IS_NGL_DRIVER (self));
old_size = g_hash_table_size (self->textures);
g_hash_table_iter_init (&iter, self->textures);
while (g_hash_table_iter_next (&iter, &k, &v))
{
GskNglTexture *t = v;
if (t->user || t->permanent)
continue;
if (t->last_used_in_frame <= watermark)
{
g_hash_table_iter_steal (&iter);
g_assert (t->link.prev == NULL);
g_assert (t->link.next == NULL);
g_assert (t->link.data == t);
/* Steal this texture and put it back into the pool */
remove_texture_key_for_id (self, t->texture_id);
gsk_ngl_texture_pool_put (&self->texture_pool, t);
}
}
collected = old_size - g_hash_table_size (self->textures);
return collected;
}
static void
gsk_ngl_texture_atlas_free (GskNglTextureAtlas *atlas)
{
if (atlas->texture_id != 0)
{
glDeleteTextures (1, &atlas->texture_id);
atlas->texture_id = 0;
}
g_clear_pointer (&atlas->nodes, g_free);
g_slice_free (GskNglTextureAtlas, atlas);
}
GskNglTextureAtlas *
gsk_ngl_driver_create_atlas (GskNglDriver *self)
{
GskNglTextureAtlas *atlas;
g_return_val_if_fail (GSK_IS_NGL_DRIVER (self), NULL);
atlas = g_slice_new0 (GskNglTextureAtlas);
atlas->width = ATLAS_SIZE;
atlas->height = ATLAS_SIZE;
/* TODO: We might want to change the strategy about the amount of
* nodes here? stb_rect_pack.h says width is optimal. */
atlas->nodes = g_malloc0_n (atlas->width, sizeof (struct stbrp_node));
stbrp_init_target (&atlas->context, atlas->width, atlas->height, atlas->nodes, atlas->width);
atlas->texture_id = gsk_ngl_command_queue_create_texture (self->command_queue,
atlas->width,
atlas->height,
GL_LINEAR,
GL_LINEAR);
gdk_gl_context_label_object_printf (gdk_gl_context_get_current (),
GL_TEXTURE, atlas->texture_id,
"Texture atlas %d",
atlas->texture_id);
g_ptr_array_add (self->atlases, atlas);
return atlas;
}
static void
remove_program (gpointer data)
{
GskNglProgram *program = data;
g_assert (!program || GSK_IS_NGL_PROGRAM (program));
if (program != NULL)
{
gsk_ngl_program_delete (program);
g_object_unref (program);
}
}
static void
gsk_ngl_driver_shader_weak_cb (gpointer data,
GObject *where_object_was)
{
GskNglDriver *self = data;
g_assert (GSK_IS_NGL_DRIVER (self));
if (self->shader_cache != NULL)
g_hash_table_remove (self->shader_cache, where_object_was);
}
static void
gsk_ngl_driver_dispose (GObject *object)
{
GskNglDriver *self = (GskNglDriver *)object;
g_assert (GSK_IS_NGL_DRIVER (self));
g_assert (self->in_frame == FALSE);
#define GSK_NGL_NO_UNIFORMS
#define GSK_NGL_ADD_UNIFORM(pos, KEY, name)
#define GSK_NGL_DEFINE_PROGRAM(name, resource, uniforms) \
GSK_NGL_DELETE_PROGRAM(name); \
GSK_NGL_DELETE_PROGRAM(name ## _no_clip); \
GSK_NGL_DELETE_PROGRAM(name ## _rect_clip);
#define GSK_NGL_DELETE_PROGRAM(name) \
G_STMT_START { \
if (self->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
gsk_ngl_program_delete (self->name); \
g_clear_object (&self->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
} G_STMT_END;
# include "gsknglprograms.defs"
#undef GSK_NGL_NO_UNIFORMS
#undef GSK_NGL_ADD_UNIFORM
#undef GSK_NGL_DEFINE_PROGRAM
if (self->shader_cache != NULL)
{
GHashTableIter iter;
gpointer k, v;
g_hash_table_iter_init (&iter, self->shader_cache);
while (g_hash_table_iter_next (&iter, &k, &v))
{
GskGLShader *shader = k;
g_object_weak_unref (G_OBJECT (shader),
gsk_ngl_driver_shader_weak_cb,
self);
g_hash_table_iter_remove (&iter);
}
g_clear_pointer (&self->shader_cache, g_hash_table_unref);
}
if (self->command_queue != NULL)
{
gsk_ngl_command_queue_make_current (self->command_queue);
gsk_ngl_driver_collect_unused_textures (self, 0);
g_clear_object (&self->command_queue);
}
if (self->autorelease_framebuffers->len > 0)
{
glDeleteFramebuffers (self->autorelease_framebuffers->len,
(GLuint *)(gpointer)self->autorelease_framebuffers->data);
self->autorelease_framebuffers->len = 0;
}
gsk_ngl_texture_pool_clear (&self->texture_pool);
g_assert (!self->textures || g_hash_table_size (self->textures) == 0);
g_assert (!self->texture_id_to_key || g_hash_table_size (self->texture_id_to_key) == 0);
g_assert (!self->key_to_texture_id|| g_hash_table_size (self->key_to_texture_id) == 0);
g_clear_object (&self->glyphs);
g_clear_object (&self->icons);
g_clear_object (&self->shadows);
g_clear_pointer (&self->atlases, g_ptr_array_unref);
g_clear_pointer (&self->autorelease_framebuffers, g_array_unref);
g_clear_pointer (&self->key_to_texture_id, g_hash_table_unref);
g_clear_pointer (&self->textures, g_hash_table_unref);
g_clear_pointer (&self->key_to_texture_id, g_hash_table_unref);
g_clear_pointer (&self->texture_id_to_key, g_hash_table_unref);
g_clear_pointer (&self->render_targets, g_ptr_array_unref);
g_clear_pointer (&self->shader_cache, g_hash_table_unref);
g_clear_object (&self->command_queue);
g_clear_object (&self->shared_command_queue);
G_OBJECT_CLASS (gsk_ngl_driver_parent_class)->dispose (object);
}
static void
gsk_ngl_driver_class_init (GskNglDriverClass *klass)
{
GObjectClass *object_class = G_OBJECT_CLASS (klass);
object_class->dispose = gsk_ngl_driver_dispose;
}
static void
gsk_ngl_driver_init (GskNglDriver *self)
{
self->autorelease_framebuffers = g_array_new (FALSE, FALSE, sizeof (guint));
self->textures = g_hash_table_new_full (NULL, NULL, NULL,
(GDestroyNotify)gsk_ngl_texture_free);
self->texture_id_to_key = g_hash_table_new (NULL, NULL);
self->key_to_texture_id = g_hash_table_new_full (texture_key_hash,
texture_key_equal,
g_free,
NULL);
self->shader_cache = g_hash_table_new_full (NULL, NULL, NULL, remove_program);
gsk_ngl_texture_pool_init (&self->texture_pool);
self->render_targets = g_ptr_array_new ();
self->atlases = g_ptr_array_new_with_free_func ((GDestroyNotify)gsk_ngl_texture_atlas_free);
}
static gboolean
gsk_ngl_driver_load_programs (GskNglDriver *self,
GError **error)
{
GskNglCompiler *compiler;
gboolean ret = FALSE;
g_assert (GSK_IS_NGL_DRIVER (self));
g_assert (GSK_IS_NGL_COMMAND_QUEUE (self->command_queue));
compiler = gsk_ngl_compiler_new (self, self->debug);
/* Setup preambles that are shared by all shaders */
gsk_ngl_compiler_set_preamble_from_resource (compiler,
GSK_NGL_COMPILER_ALL,
"/org/gtk/libgsk/ngl/preamble.glsl");
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_compiler_set_preamble_from_resource (compiler,
GSK_NGL_COMPILER_VERTEX,
"/org/gtk/libgsk/ngl/preamble.vs.glsl");
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_compiler_set_preamble_from_resource (compiler,
GSK_NGL_COMPILER_FRAGMENT,
"/org/gtk/libgsk/ngl/preamble.fs.glsl");
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
/* Setup attributes that are provided via VBO */
gsk_ngl_compiler_bind_attribute (compiler, "aPosition", 0);
gsk_ngl_compiler_bind_attribute (compiler, "aUv", 1);
gsk_ngl_compiler_bind_attribute (compiler, "aColor", 2);
gsk_ngl_compiler_bind_attribute (compiler, "aColor2", 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
/* Use XMacros to register all of our programs and their uniforms */
#define GSK_NGL_NO_UNIFORMS
#define GSK_NGL_ADD_UNIFORM(pos, KEY, name) \
gsk_ngl_program_add_uniform (program, #name, UNIFORM_##KEY);
#define GSK_NGL_DEFINE_PROGRAM(name, resource, uniforms) \
gsk_ngl_compiler_set_source_from_resource (compiler, GSK_NGL_COMPILER_ALL, resource); \
GSK_NGL_COMPILE_PROGRAM(name ## _no_clip, uniforms, "#define NO_CLIP 1\n"); \
GSK_NGL_COMPILE_PROGRAM(name ## _rect_clip, uniforms, "#define RECT_CLIP 1\n"); \
GSK_NGL_COMPILE_PROGRAM(name, uniforms, "");
#define GSK_NGL_COMPILE_PROGRAM(name, uniforms, clip) \
G_STMT_START { \
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
GskNglProgram *program; \
gboolean have_alpha; \
gboolean have_source; \
\
if (!(program = gsk_ngl_compiler_compile (compiler, #name, clip, error))) \
goto failure; \
\
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
have_alpha = gsk_ngl_program_add_uniform (program, "u_alpha", UNIFORM_SHARED_ALPHA); \
have_source = gsk_ngl_program_add_uniform (program, "u_source", UNIFORM_SHARED_SOURCE); \
gsk_ngl_program_add_uniform (program, "u_clip_rect", UNIFORM_SHARED_CLIP_RECT); \
gsk_ngl_program_add_uniform (program, "u_viewport", UNIFORM_SHARED_VIEWPORT); \
gsk_ngl_program_add_uniform (program, "u_projection", UNIFORM_SHARED_PROJECTION); \
gsk_ngl_program_add_uniform (program, "u_modelview", UNIFORM_SHARED_MODELVIEW); \
\
uniforms \
\
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_program_uniforms_added (program, have_source); \
if (have_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
gsk_ngl_program_set_uniform1f (program, UNIFORM_SHARED_ALPHA, 0, 1.0f); \
\
*(GskNglProgram **)(((guint8 *)self) + G_STRUCT_OFFSET (GskNglDriver, name)) = \
g_steal_pointer (&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
} G_STMT_END;
# include "gsknglprograms.defs"
#undef GSK_NGL_DEFINE_PROGRAM_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
#undef GSK_NGL_DEFINE_PROGRAM
#undef GSK_NGL_ADD_UNIFORM
ret = TRUE;
failure:
g_clear_object (&compiler);
return ret;
}
/**
* gsk_ngl_driver_autorelease_framebuffer:
* @self: a #GskNglDriver
* @framebuffer_id: the id of the OpenGL framebuffer
*
* Marks @framebuffer_id to be deleted when the current frame has cmopleted.
*/
static void
gsk_ngl_driver_autorelease_framebuffer (GskNglDriver *self,
guint framebuffer_id)
{
g_assert (GSK_IS_NGL_DRIVER (self));
g_array_append_val (self->autorelease_framebuffers, framebuffer_id);
}
static GskNglDriver *
gsk_ngl_driver_new (GskNglCommandQueue *command_queue,
gboolean debug_shaders,
GError **error)
{
GskNglDriver *self;
GdkGLContext *context;
g_return_val_if_fail (GSK_IS_NGL_COMMAND_QUEUE (command_queue), NULL);
context = gsk_ngl_command_queue_get_context (command_queue);
gdk_gl_context_make_current (context);
self = g_object_new (GSK_TYPE_NGL_DRIVER, NULL);
self->command_queue = g_object_ref (command_queue);
self->shared_command_queue = g_object_ref (command_queue);
self->debug = !!debug_shaders;
if (!gsk_ngl_driver_load_programs (self, error))
{
g_object_unref (self);
return NULL;
}
self->glyphs = gsk_ngl_glyph_library_new (self);
self->icons = gsk_ngl_icon_library_new (self);
self->shadows = gsk_ngl_shadow_library_new (self);
return g_steal_pointer (&self);
}
/**
* gsk_ngl_driver_from_shared_context:
* @context: a shared #GdkGLContext retrieved with gdk_gl_context_get_shared_context()
* @debug_shaders: if debug information for shaders should be displayed
* @error: location for error information
*
* Retrieves a driver for a shared context. Generally this is shared across all GL
* contexts for a display so that fewer programs are necessary for driving output.
*
* Returns: (transfer full): a #GskNglDriver if successful; otherwise %NULL and
* @error is set.
*/
GskNglDriver *
gsk_ngl_driver_from_shared_context (GdkGLContext *context,
gboolean debug_shaders,
GError **error)
{
GskNglCommandQueue *command_queue = NULL;
GskNglDriver *driver;
g_return_val_if_fail (GDK_IS_GL_CONTEXT (context), NULL);
if ((driver = g_object_get_data (G_OBJECT (context), "GSK_NGL_DRIVER")))
return g_object_ref (driver);
gdk_gl_context_make_current (context);
/* Initially we create a command queue using the shared context. However,
* as frames are processed this will be replaced with the command queue
* for a given renderer. But since the programs are compiled into the
* shared context, all other contexts sharing with it will have access
* to those programs.
*/
command_queue = gsk_ngl_command_queue_new (context, NULL);
if (!(driver = gsk_ngl_driver_new (command_queue, debug_shaders, error)))
goto failure;
g_object_set_data_full (G_OBJECT (context),
"GSK_NGL_DRIVER",
g_object_ref (driver),
g_object_unref);
failure:
g_clear_object (&command_queue);
return g_steal_pointer (&driver);
}
static GPtrArray *
gsk_ngl_driver_compact_atlases (GskNglDriver *self)
{
GPtrArray *removed = NULL;
g_assert (GSK_IS_NGL_DRIVER (self));
for (guint i = self->atlases->len; i > 0; i--)
{
GskNglTextureAtlas *atlas = g_ptr_array_index (self->atlases, i - 1);
if (gsk_ngl_texture_atlas_get_unused_ratio (atlas) > MAX_OLD_RATIO)
{
GSK_NOTE (GLYPH_CACHE,
g_message ("Dropping atlas %d (%g.2%% old)", i,
100.0 * gsk_ngl_texture_atlas_get_unused_ratio (atlas)));
if (removed == NULL)
removed = g_ptr_array_new_with_free_func ((GDestroyNotify)gsk_ngl_texture_atlas_free);
g_ptr_array_add (removed, g_ptr_array_steal_index (self->atlases, i - 1));
}
}
GSK_NOTE (GLYPH_CACHE, {
static guint timestamp;
if (timestamp++ % 60 == 0)
g_message ("%d atlases", self->atlases->len);
});
return removed;
}
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_driver_begin_frame:
* @self: a #GskNglDriver
* @command_queue: A #GskNglCommandQueue from the renderer
*
* Begin a new frame.
*
* Texture atlases, pools, and other resources will be prepared to draw the
* next frame. The command queue should be one that was created for the
* target context to be drawn into (the context of the renderer's surface).
*/
void
gsk_ngl_driver_begin_frame (GskNglDriver *self,
GskNglCommandQueue *command_queue)
{
gint64 last_frame_id;
GPtrArray *removed;
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_return_if_fail (GSK_IS_NGL_DRIVER (self));
g_return_if_fail (GSK_IS_NGL_COMMAND_QUEUE (command_queue));
g_return_if_fail (self->in_frame == FALSE);
last_frame_id = self->current_frame_id;
self->in_frame = TRUE;
self->current_frame_id++;
g_set_object (&self->command_queue, command_queue);
gsk_ngl_command_queue_begin_frame (self->command_queue);
/* Compact atlases with too many freed pixels */
removed = gsk_ngl_driver_compact_atlases (self);
/* Mark unused pixel regions of the atlases */
gsk_ngl_texture_library_begin_frame (GSK_NGL_TEXTURE_LIBRARY (self->icons),
self->current_frame_id,
removed);
gsk_ngl_texture_library_begin_frame (GSK_NGL_TEXTURE_LIBRARY (self->glyphs),
self->current_frame_id,
removed);
/* Cleanup old shadows */
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_shadow_library_begin_frame (self->shadows);
/* Remove all textures that are from a previous frame or are no
* longer used by linked GdkTexture. We do this at the beginning
* of the following frame instead of the end so that we reduce chances
* we block on any resources while delivering our frames.
*/
gsk_ngl_driver_collect_unused_textures (self, last_frame_id - 1);
/* Now free atlas textures */
g_clear_pointer (&removed, g_ptr_array_unref);
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_driver_end_frame:
* @self: a #GskNglDriver
*
* Clean up resources from drawing the current frame.
*
* Temporary resources used while drawing will be released.
*/
void
gsk_ngl_driver_end_frame (GskNglDriver *self)
{
g_return_if_fail (GSK_IS_NGL_DRIVER (self));
g_return_if_fail (self->in_frame == TRUE);
gsk_ngl_command_queue_make_current (self->command_queue);
gsk_ngl_command_queue_end_frame (self->command_queue);
self->in_frame = FALSE;
}
/**
* gsk_ngl_driver_after_frame:
* @self: a #GskNglDriver
*
* This function does post-frame cleanup operations.
*
* To reduce the chances of blocking on the driver it is performed
* after the frame has swapped buffers.
*/
void
gsk_ngl_driver_after_frame (GskNglDriver *self)
{
g_return_if_fail (GSK_IS_NGL_DRIVER (self));
g_return_if_fail (self->in_frame == FALSE);
/* Release any render targets (possibly adding them to
* self->autorelease_framebuffers) so we can release the FBOs immediately
* afterwards.
*/
while (self->render_targets->len > 0)
{
GskNglRenderTarget *render_target = g_ptr_array_index (self->render_targets, self->render_targets->len - 1);
gsk_ngl_driver_autorelease_framebuffer (self, render_target->framebuffer_id);
glDeleteTextures (1, &render_target->texture_id);
g_slice_free (GskNglRenderTarget, render_target);
self->render_targets->len--;
}
/* Now that we have collected render targets, release all the FBOs */
if (self->autorelease_framebuffers->len > 0)
{
glDeleteFramebuffers (self->autorelease_framebuffers->len,
(GLuint *)(gpointer)self->autorelease_framebuffers->data);
self->autorelease_framebuffers->len = 0;
}
/* Release any cached textures we used during the frame */
gsk_ngl_texture_pool_clear (&self->texture_pool);
/* Reset command queue to our shared queue incase we have operations
* that need to be processed outside of a frame (such as callbacks
* from external systems such as GDK).
*/
g_set_object (&self->command_queue, self->shared_command_queue);
}
GdkGLContext *
gsk_ngl_driver_get_context (GskNglDriver *self)
{
g_return_val_if_fail (GSK_IS_NGL_DRIVER (self), NULL);
g_return_val_if_fail (GSK_IS_NGL_COMMAND_QUEUE (self->command_queue), NULL);
return gsk_ngl_command_queue_get_context (self->command_queue);
}
/**
* gsk_ngl_driver_cache_texture:
* @self: a #GskNglDriver
* @key: the key for the texture
* @texture_id: the id of the texture to be cached
*
* Inserts @texture_id into the texture cache using @key.
*
* Textures can be looked up by @key after calling this function using
* gsk_ngl_driver_lookup_texture().
*
* Textures that have not been used within a number of frames will be
* purged from the texture cache automatically.
*/
void
gsk_ngl_driver_cache_texture (GskNglDriver *self,
const GskTextureKey *key,
guint texture_id)
{
GskTextureKey *k;
g_assert (GSK_IS_NGL_DRIVER (self));
g_assert (key != NULL);
g_assert (texture_id > 0);
g_assert (g_hash_table_contains (self->textures, GUINT_TO_POINTER (texture_id)));
k = g_memdup (key, sizeof *key);
g_hash_table_insert (self->key_to_texture_id, k, GUINT_TO_POINTER (texture_id));
g_hash_table_insert (self->texture_id_to_key, GUINT_TO_POINTER (texture_id), k);
}
/**
* gsk_ngl_driver_load_texture:
* @self: a #GdkTexture
* @texture: a #GdkTexture
* @min_filter: GL_NEAREST or GL_LINEAR
* @mag_filter: GL_NEAREST or GL_LINEAR
*
* Loads a #GdkTexture by uploading the contents to the GPU when
* necessary. If @texture is a #GdkGLTexture, it can be used without
* uploading contents to the GPU.
*
* If the texture has already been uploaded and not yet released
* from cache, this function returns that texture id without further
* work.
*
* If the texture has not been used for a number of frames, it will
* be removed from cache.
*
* There is no need to release the resulting texture identifier after
* using it. It will be released automatically.
*
* Returns: a texture identifier
*/
guint
gsk_ngl_driver_load_texture (GskNglDriver *self,
GdkTexture *texture,
int min_filter,
int mag_filter)
{
GdkGLContext *context;
GdkTexture *downloaded_texture = NULL;
GdkTexture *source_texture;
GskNglTexture *t;
guint texture_id;
int height;
int width;
g_return_val_if_fail (GSK_IS_NGL_DRIVER (self), 0);
g_return_val_if_fail (GDK_IS_TEXTURE (texture), 0);
g_return_val_if_fail (GSK_IS_NGL_COMMAND_QUEUE (self->command_queue), 0);
context = self->command_queue->context;
if (GDK_IS_GL_TEXTURE (texture))
{
GdkGLContext *texture_context = gdk_gl_texture_get_context ((GdkGLTexture *)texture);
GdkGLContext *shared_context = gdk_gl_context_get_shared_context (context);
if (texture_context == context ||
(shared_context != NULL &&
shared_context == gdk_gl_context_get_shared_context (texture_context)))
{
/* A GL texture from the same GL context is a simple task... */
return gdk_gl_texture_get_id ((GdkGLTexture *)texture);
}
else
{
cairo_surface_t *surface;
/* In this case, we have to temporarily make the texture's
* context the current one, download its data into our context
* and then create a texture from it. */
if (texture_context != NULL)
gdk_gl_context_make_current (texture_context);
surface = gdk_texture_download_surface (texture);
downloaded_texture = gdk_texture_new_for_surface (surface);
cairo_surface_destroy (surface);
gdk_gl_context_make_current (context);
source_texture = downloaded_texture;
}
}
else
{
if ((t = gdk_texture_get_render_data (texture, self)))
{
if (t->min_filter == min_filter && t->mag_filter == mag_filter)
return t->texture_id;
}
source_texture = texture;
}
width = gdk_texture_get_width (texture);
height = gdk_texture_get_height (texture);
texture_id = gsk_ngl_command_queue_upload_texture (self->command_queue,
source_texture,
0,
0,
width,
height,
min_filter,
mag_filter);
t = gsk_ngl_texture_new (texture_id,
width, height, min_filter, mag_filter,
self->current_frame_id);
g_hash_table_insert (self->textures, GUINT_TO_POINTER (texture_id), t);
if (gdk_texture_set_render_data (texture, self, t, gsk_ngl_texture_destroyed))
t->user = texture;
gdk_gl_context_label_object_printf (context, GL_TEXTURE, t->texture_id,
"GdkTexture<%p> %d", texture, t->texture_id);
g_clear_object (&downloaded_texture);
return texture_id;
}
/**
* gsk_ngl_driver_create_texture:
* @self: a #GskNglDriver
* @width: the width of the texture
* @height: the height of the texture
* @min_filter: GL_NEAREST or GL_LINEAR
* @mag_filter: GL_NEAREST or GL_FILTER
*
* Creates a new texture immediately that can be used by the caller
* to upload data, map to a framebuffer, or other uses which may
* modify the texture immediately.
*
* Use gsk_ngl_driver_release_texture() to release this texture back into
* the pool so it may be reused later in the pipeline.
*
* Returns: a #GskNglTexture which can be returned to the pool with
* gsk_ngl_driver_release_texture().
*/
GskNglTexture *
gsk_ngl_driver_create_texture (GskNglDriver *self,
float width,
float height,
int min_filter,
int mag_filter)
{
GskNglTexture *texture;
g_return_val_if_fail (GSK_IS_NGL_DRIVER (self), NULL);
texture = gsk_ngl_texture_pool_get (&self->texture_pool,
width, height,
min_filter, mag_filter);
g_hash_table_insert (self->textures,
GUINT_TO_POINTER (texture->texture_id),
texture);
texture->last_used_in_frame = self->current_frame_id;
return texture;
}
/**
* gsk_ngl_driver_release_texture:
* @self: a #GskNglDriver
* @texture: a #GskNglTexture
*
* Releases @texture back into the pool so that it can be used later
* in the command stream by future batches. This helps reduce VRAM
* usage on the GPU.
*
* When the frame has completed, pooled textures will be released
* to free additional VRAM back to the system.
*/
void
gsk_ngl_driver_release_texture (GskNglDriver *self,
GskNglTexture *texture)
{
guint texture_id;
g_assert (GSK_IS_NGL_DRIVER (self));
g_assert (texture != NULL);
texture_id = texture->texture_id;
if (texture_id > 0)
remove_texture_key_for_id (self, texture_id);
g_hash_table_steal (self->textures, GUINT_TO_POINTER (texture_id));
gsk_ngl_texture_pool_put (&self->texture_pool, texture);
}
/**
* gsk_ngl_driver_create_render_target:
* @self: a #GskNglDriver
* @width: the width for the render target
* @height: the height for the render target
* @min_filter: the min filter to use for the texture
* @mag_filter: the mag filter to use for the texture
* @out_render_target: (out): a location for the render target
*
* Creates a new render target which contains a framebuffer and a texture
* bound to that framebuffer of the size @width x @height and using the
* appropriate filters.
*
* Use gsk_ngl_driver_release_render_target() when you are finished with
* the render target to release it. You may steal the texture from the
* render target when releasing it.
*
* Returns: %TRUE if successful; otherwise %FALSE and @out_fbo_id and
* @out_texture_id are undefined.
*/
gboolean
gsk_ngl_driver_create_render_target (GskNglDriver *self,
int width,
int height,
int min_filter,
int mag_filter,
GskNglRenderTarget **out_render_target)
{
guint framebuffer_id;
guint texture_id;
g_return_val_if_fail (GSK_IS_NGL_DRIVER (self), FALSE);
g_return_val_if_fail (GSK_IS_NGL_COMMAND_QUEUE (self->command_queue), FALSE);
g_return_val_if_fail (out_render_target != NULL, FALSE);
#if 0
if (self->render_targets->len > 0)
{
for (guint i = self->render_targets->len; i > 0; i--)
{
GskNglRenderTarget *render_target = g_ptr_array_index (self->render_targets, i-1);
if (render_target->width == width &&
render_target->height == height &&
render_target->min_filter == min_filter &&
render_target->mag_filter == mag_filter)
{
*out_render_target = g_ptr_array_steal_index_fast (self->render_targets, i-1);
return TRUE;
}
}
}
#endif
if (gsk_ngl_command_queue_create_render_target (self->command_queue,
width, height,
min_filter, mag_filter,
&framebuffer_id, &texture_id))
{
GskNglRenderTarget *render_target;
render_target = g_slice_new0 (GskNglRenderTarget);
render_target->min_filter = min_filter;
render_target->mag_filter = mag_filter;
render_target->width = width;
render_target->height = height;
render_target->framebuffer_id = framebuffer_id;
render_target->texture_id = texture_id;
*out_render_target = render_target;
return TRUE;
}
*out_render_target = NULL;
return FALSE;
}
/**
* gsk_ngl_driver_release_render_target:
* @self: a #GskNglDriver
* @render_target: a #GskNglRenderTarget created with
* gsk_ngl_driver_create_render_target().
* @release_texture: if the texture should also be released
*
* Releases a render target that was previously created. An attempt may
* be made to cache the render target so that future creations of render
* targets are performed faster.
*
* If @release_texture is %FALSE, the backing texture id is returned and
* the framebuffer is released. Otherwise, both the texture and framebuffer
* are released or cached until the end of the frame.
*
* This may be called when building the render job as the texture or
* framebuffer will not be removed immediately.
*
* Returns: a texture id if @release_texture is %FALSE, otherwise zero.
*/
guint
gsk_ngl_driver_release_render_target (GskNglDriver *self,
GskNglRenderTarget *render_target,
gboolean release_texture)
{
guint texture_id;
g_return_val_if_fail (GSK_IS_NGL_DRIVER (self), 0);
g_return_val_if_fail (render_target != NULL, 0);
if (release_texture)
{
texture_id = 0;
g_ptr_array_add (self->render_targets, render_target);
}
else
{
GskNglTexture *texture;
texture_id = render_target->texture_id;
texture = gsk_ngl_texture_new (render_target->texture_id,
render_target->width,
render_target->height,
render_target->min_filter,
render_target->mag_filter,
self->current_frame_id);
g_hash_table_insert (self->textures,
GUINT_TO_POINTER (texture_id),
g_steal_pointer (&texture));
gsk_ngl_driver_autorelease_framebuffer (self, render_target->framebuffer_id);
g_slice_free (GskNglRenderTarget, render_target);
}
return texture_id;
}
/**
* gsk_ngl_driver_lookup_shader:
* @self: a #GskNglDriver
* @shader: the shader to lookup or load
* @error: a location for a #GError
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
*
* Attepts to load @shader from the shader cache.
*
* If it has not been loaded, then it will compile the shader on demand.
*
* Returns: (nullable) (transfer none): a #GskGLShader
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
*/
GskNglProgram *
gsk_ngl_driver_lookup_shader (GskNglDriver *self,
GskGLShader *shader,
GError **error)
{
GskNglProgram *program;
g_return_val_if_fail (self != NULL, NULL);
g_return_val_if_fail (shader != NULL, NULL);
program = g_hash_table_lookup (self->shader_cache, shader);
if (program == NULL)
{
const GskGLUniform *uniforms;
GskNglCompiler *compiler;
GBytes *suffix;
int n_required_textures;
int n_uniforms;
uniforms = gsk_gl_shader_get_uniforms (shader, &n_uniforms);
if (n_uniforms > GSK_NGL_PROGRAM_MAX_CUSTOM_ARGS)
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_set_error (error,
GDK_GL_ERROR,
GDK_GL_ERROR_UNSUPPORTED_FORMAT,
"Tried to use %d uniforms, while only %d is supported",
n_uniforms,
GSK_NGL_PROGRAM_MAX_CUSTOM_ARGS);
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 NULL;
}
n_required_textures = gsk_gl_shader_get_n_textures (shader);
if (n_required_textures > GSK_NGL_PROGRAM_MAX_CUSTOM_TEXTURES)
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_set_error (error,
GDK_GL_ERROR,
GDK_GL_ERROR_UNSUPPORTED_FORMAT,
"Tried to use %d textures, while only %d is supported",
n_required_textures,
GSK_NGL_PROGRAM_MAX_CUSTOM_TEXTURES);
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 NULL;
}
compiler = gsk_ngl_compiler_new (self, FALSE);
suffix = gsk_gl_shader_get_source (shader);
gsk_ngl_compiler_set_preamble_from_resource (compiler,
GSK_NGL_COMPILER_ALL,
"/org/gtk/libgsk/ngl/preamble.glsl");
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_compiler_set_preamble_from_resource (compiler,
GSK_NGL_COMPILER_VERTEX,
"/org/gtk/libgsk/ngl/preamble.vs.glsl");
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_compiler_set_preamble_from_resource (compiler,
GSK_NGL_COMPILER_FRAGMENT,
"/org/gtk/libgsk/ngl/preamble.fs.glsl");
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_compiler_set_source_from_resource (compiler,
GSK_NGL_COMPILER_ALL,
"/org/gtk/libgsk/ngl/custom.glsl");
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_compiler_set_suffix (compiler, GSK_NGL_COMPILER_FRAGMENT, suffix);
/* Setup attributes that are provided via VBO */
gsk_ngl_compiler_bind_attribute (compiler, "aPosition", 0);
gsk_ngl_compiler_bind_attribute (compiler, "aUv", 1);
gsk_ngl_compiler_bind_attribute (compiler, "aColor", 2);
gsk_ngl_compiler_bind_attribute (compiler, "aColor2", 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
if ((program = gsk_ngl_compiler_compile (compiler, NULL, "", error)))
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
{
gboolean have_alpha;
gsk_ngl_program_add_uniform (program, "u_source", UNIFORM_SHARED_SOURCE);
gsk_ngl_program_add_uniform (program, "u_clip_rect", UNIFORM_SHARED_CLIP_RECT);
gsk_ngl_program_add_uniform (program, "u_viewport", UNIFORM_SHARED_VIEWPORT);
gsk_ngl_program_add_uniform (program, "u_projection", UNIFORM_SHARED_PROJECTION);
gsk_ngl_program_add_uniform (program, "u_modelview", UNIFORM_SHARED_MODELVIEW);
have_alpha = gsk_ngl_program_add_uniform (program, "u_alpha", UNIFORM_SHARED_ALPHA);
gsk_ngl_program_add_uniform (program, "u_size", UNIFORM_CUSTOM_SIZE);
gsk_ngl_program_add_uniform (program, "u_texture1", UNIFORM_CUSTOM_TEXTURE1);
gsk_ngl_program_add_uniform (program, "u_texture2", UNIFORM_CUSTOM_TEXTURE2);
gsk_ngl_program_add_uniform (program, "u_texture3", UNIFORM_CUSTOM_TEXTURE3);
gsk_ngl_program_add_uniform (program, "u_texture4", UNIFORM_CUSTOM_TEXTURE4);
/* Custom arguments (max is 8) */
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_uniforms; i++)
gsk_ngl_program_add_uniform (program, uniforms[i].name, 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
gsk_ngl_program_uniforms_added (program, TRUE);
if (have_alpha)
gsk_ngl_program_set_uniform1f (program, UNIFORM_SHARED_ALPHA, 0, 1.0f);
g_hash_table_insert (self->shader_cache, shader, program);
g_object_weak_ref (G_OBJECT (shader),
gsk_ngl_driver_shader_weak_cb,
self);
}
g_object_unref (compiler);
}
return program;
}
#ifdef G_ENABLE_DEBUG
static void
write_atlas_to_png (GskNglTextureAtlas *atlas,
const char *filename)
{
int stride = cairo_format_stride_for_width (CAIRO_FORMAT_ARGB32, atlas->width);
guchar *data = g_malloc (atlas->height * stride);
cairo_surface_t *s;
glBindTexture (GL_TEXTURE_2D, atlas->texture_id);
glGetTexImage (GL_TEXTURE_2D, 0, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, data);
s = cairo_image_surface_create_for_data (data, CAIRO_FORMAT_ARGB32, atlas->width, atlas->height, stride);
cairo_surface_write_to_png (s, filename);
cairo_surface_destroy (s);
g_free (data);
}
void
gsk_ngl_driver_save_atlases_to_png (GskNglDriver *self,
const char *directory)
{
g_return_if_fail (GSK_IS_NGL_DRIVER (self));
if (directory == NULL)
directory = ".";
for (guint i = 0; i < self->atlases->len; i++)
{
GskNglTextureAtlas *atlas = g_ptr_array_index (self->atlases, i);
char *filename = g_strdup_printf ("%s%sframe-%d-atlas-%d.png",
directory,
G_DIR_SEPARATOR_S,
(int)self->current_frame_id,
atlas->texture_id);
write_atlas_to_png (atlas, filename);
g_free (filename);
}
}
#endif
GskNglCommandQueue *
gsk_ngl_driver_create_command_queue (GskNglDriver *self,
GdkGLContext *context)
{
g_return_val_if_fail (GSK_IS_NGL_DRIVER (self), NULL);
g_return_val_if_fail (GDK_IS_GL_CONTEXT (context), NULL);
return gsk_ngl_command_queue_new (context, self->shared_command_queue->uniforms);
}
void
gsk_ngl_driver_add_texture_slices (GskNglDriver *self,
GdkTexture *texture,
GskNglTextureSlice **out_slices,
guint *out_n_slices)
{
int max_texture_size;
GskNglTextureSlice *slices;
GskNglTexture *t;
guint n_slices;
guint cols;
guint rows;
int tex_width;
int tex_height;
int x = 0, y = 0;
g_assert (GSK_IS_NGL_DRIVER (self));
g_assert (GDK_IS_TEXTURE (texture));
g_assert (out_slices != NULL);
g_assert (out_n_slices != NULL);
/* XXX: Too much? */
max_texture_size = self->command_queue->max_texture_size / 4;
tex_width = texture->width;
tex_height = texture->height;
cols = (texture->width / max_texture_size) + 1;
rows = (texture->height / max_texture_size) + 1;
if ((t = gdk_texture_get_render_data (texture, self)))
{
*out_slices = t->slices;
*out_n_slices = t->n_slices;
return;
}
n_slices = cols * rows;
slices = g_new0 (GskNglTextureSlice, n_slices);
for (guint col = 0; col < cols; col ++)
{
int slice_width = MIN (max_texture_size, texture->width - x);
for (guint row = 0; row < rows; row ++)
{
int slice_height = MIN (max_texture_size, texture->height - y);
int slice_index = (col * rows) + row;
guint texture_id;
texture_id = gsk_ngl_command_queue_upload_texture (self->command_queue,
texture,
x, y,
slice_width, slice_height,
GL_NEAREST, GL_NEAREST);
slices[slice_index].rect.x = x;
slices[slice_index].rect.y = y;
slices[slice_index].rect.width = slice_width;
slices[slice_index].rect.height = slice_height;
slices[slice_index].texture_id = texture_id;
y += slice_height;
}
y = 0;
x += slice_width;
}
/* Allocate one Texture for the entire thing. */
t = gsk_ngl_texture_new (0,
tex_width, tex_height,
GL_NEAREST, GL_NEAREST,
self->current_frame_id);
/* Use gsk_ngl_texture_free() as destroy notify here since we are
* not inserting this GskNglTexture into self->textures!
*/
gdk_texture_set_render_data (texture, self, t,
(GDestroyNotify)gsk_ngl_texture_free);
t->slices = *out_slices = slices;
t->n_slices = *out_n_slices = n_slices;
}
GskNglTexture *
gsk_ngl_driver_mark_texture_permanent (GskNglDriver *self,
guint texture_id)
{
GskNglTexture *t;
g_return_val_if_fail (GSK_IS_NGL_DRIVER (self), NULL);
g_return_val_if_fail (texture_id > 0, NULL);
if ((t = g_hash_table_lookup (self->textures, GUINT_TO_POINTER (texture_id))))
t->permanent = TRUE;
return t;
}
void
gsk_ngl_driver_release_texture_by_id (GskNglDriver *self,
guint texture_id)
{
GskNglTexture *texture;
g_return_if_fail (GSK_IS_NGL_DRIVER (self));
g_return_if_fail (texture_id > 0);
remove_texture_key_for_id (self, texture_id);
if ((texture = g_hash_table_lookup (self->textures, GUINT_TO_POINTER (texture_id))))
gsk_ngl_driver_release_texture (self, texture);
}
static void
create_texture_from_texture_destroy (gpointer data)
{
GskNglTextureState *state = data;
g_assert (state != NULL);
g_assert (GDK_IS_GL_CONTEXT (state->context));
gdk_gl_context_make_current (state->context);
glDeleteTextures (1, &state->texture_id);
g_clear_object (&state->context);
g_slice_free (GskNglTextureState, state);
}
GdkTexture *
gsk_ngl_driver_create_gdk_texture (GskNglDriver *self,
guint texture_id)
{
GskNglTextureState *state;
GskNglTexture *texture;
g_return_val_if_fail (GSK_IS_NGL_DRIVER (self), NULL);
g_return_val_if_fail (self->command_queue != NULL, NULL);
g_return_val_if_fail (GDK_IS_GL_CONTEXT (self->command_queue->context), NULL);
g_return_val_if_fail (texture_id > 0, NULL);
g_return_val_if_fail (!g_hash_table_contains (self->texture_id_to_key, GUINT_TO_POINTER (texture_id)), NULL);
/* We must be tracking this texture_id already to use it */
if (!(texture = g_hash_table_lookup (self->textures, GUINT_TO_POINTER (texture_id))))
g_return_val_if_reached (NULL);
state = g_slice_new0 (GskNglTextureState);
state->texture_id = texture_id;
state->context = g_object_ref (self->command_queue->context);
g_hash_table_steal (self->textures, GUINT_TO_POINTER (texture_id));
return gdk_gl_texture_new (self->command_queue->context,
texture_id,
texture->width,
texture->height,
create_texture_from_texture_destroy,
state);
}