gtk/gsk/gpu/gskvulkanframe.c
Benjamin Otte 30dddf2412 gpu: Refactor waiting for frames
Instead of having renderer API to wait for any number of frames, just
have gsk_gpu_frame_wait() to wait for a single frame.

This unifies behavior on Vulkan and GL, because unlike Vulkan, GL does
not allow waiting for multiple fences.

To make up for it, we replace waiting for multiple frames with finding
the frame with the earliest timestamp and waiting for that one.

Also implement wait() for GL.
2024-03-14 06:06:33 +01:00

449 lines
16 KiB
C

#include "config.h"
#include "gskvulkanframeprivate.h"
#include "gskgpuopprivate.h"
#include "gskvulkanbufferprivate.h"
#include "gskvulkandescriptorsprivate.h"
#include "gskvulkandeviceprivate.h"
#include "gskvulkanimageprivate.h"
#include "gskvulkanrealdescriptorsprivate.h"
#include "gskvulkansubdescriptorsprivate.h"
#include "gdk/gdkdisplayprivate.h"
#include "gdk/gdkdmabuftextureprivate.h"
#define GDK_ARRAY_NAME gsk_descriptors
#define GDK_ARRAY_TYPE_NAME GskDescriptors
#define GDK_ARRAY_ELEMENT_TYPE GskVulkanRealDescriptors *
#define GDK_ARRAY_FREE_FUNC g_object_unref
#define GDK_ARRAY_NO_MEMSET 1
#include "gdk/gdkarrayimpl.c"
#define GDK_ARRAY_NAME gsk_semaphores
#define GDK_ARRAY_TYPE_NAME GskSemaphores
#define GDK_ARRAY_ELEMENT_TYPE VkSemaphore
#define GDK_ARRAY_PREALLOC 16
#define GDK_ARRAY_NO_MEMSET 1
#include "gdk/gdkarrayimpl.c"
#define GDK_ARRAY_NAME gsk_pipeline_stages
#define GDK_ARRAY_TYPE_NAME GskPipelineStages
#define GDK_ARRAY_ELEMENT_TYPE VkPipelineStageFlags
#define GDK_ARRAY_PREALLOC 16
#define GDK_ARRAY_NO_MEMSET 1
#include "gdk/gdkarrayimpl.c"
struct _GskVulkanSemaphores
{
GskSemaphores wait_semaphores;
GskPipelineStages wait_stages;
GskSemaphores signal_semaphores;
};
struct _GskVulkanFrame
{
GskGpuFrame parent_instance;
VkFence vk_fence;
VkCommandBuffer vk_command_buffer;
VkDescriptorPool vk_descriptor_pool;
GskDescriptors descriptors;
gsize pool_n_sets;
gsize pool_n_images;
gsize pool_n_buffers;
};
struct _GskVulkanFrameClass
{
GskGpuFrameClass parent_class;
};
G_DEFINE_TYPE (GskVulkanFrame, gsk_vulkan_frame, GSK_TYPE_GPU_FRAME)
static gboolean
gsk_vulkan_frame_is_busy (GskGpuFrame *frame)
{
GskVulkanFrame *self = GSK_VULKAN_FRAME (frame);
VkDevice device;
device = gsk_vulkan_device_get_vk_device (GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (frame)));
return vkGetFenceStatus (device, self->vk_fence) == VK_NOT_READY;
}
static void
gsk_vulkan_frame_wait (GskGpuFrame *frame)
{
GskVulkanFrame *self = GSK_VULKAN_FRAME (frame);
VkDevice vk_device;
vk_device = gsk_vulkan_device_get_vk_device (GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (frame)));
GSK_VK_CHECK (vkWaitForFences, vk_device,
1,
&self->vk_fence,
VK_FALSE,
INT64_MAX);
}
static void
gsk_vulkan_frame_setup (GskGpuFrame *frame)
{
GskVulkanFrame *self = GSK_VULKAN_FRAME (frame);
GskVulkanDevice *device;
VkDevice vk_device;
VkCommandPool vk_command_pool;
device = GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (frame));
vk_device = gsk_vulkan_device_get_vk_device (device);
vk_command_pool = gsk_vulkan_device_get_vk_command_pool (device);
GSK_VK_CHECK (vkAllocateCommandBuffers, vk_device,
&(VkCommandBufferAllocateInfo) {
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO,
.commandPool = vk_command_pool,
.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY,
.commandBufferCount = 1,
},
&self->vk_command_buffer);
GSK_VK_CHECK (vkCreateFence, vk_device,
&(VkFenceCreateInfo) {
.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO,
.flags = VK_FENCE_CREATE_SIGNALED_BIT
},
NULL,
&self->vk_fence);
}
static void
gsk_vulkan_frame_cleanup (GskGpuFrame *frame)
{
GskVulkanFrame *self = GSK_VULKAN_FRAME (frame);
GskVulkanDevice *device;
VkDevice vk_device;
device = GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (frame));
vk_device = gsk_vulkan_device_get_vk_device (device);
GSK_VK_CHECK (vkWaitForFences, vk_device,
1,
&self->vk_fence,
VK_TRUE,
INT64_MAX);
GSK_VK_CHECK (vkResetFences, vk_device,
1,
&self->vk_fence);
GSK_VK_CHECK (vkResetCommandBuffer, self->vk_command_buffer,
0);
if (self->vk_descriptor_pool != VK_NULL_HANDLE)
{
GSK_VK_CHECK (vkResetDescriptorPool, vk_device,
self->vk_descriptor_pool,
0);
}
gsk_descriptors_set_size (&self->descriptors, 0);
GSK_GPU_FRAME_CLASS (gsk_vulkan_frame_parent_class)->cleanup (frame);
}
static GskGpuImage *
gsk_vulkan_frame_upload_texture (GskGpuFrame *frame,
gboolean with_mipmap,
GdkTexture *texture)
{
#ifdef HAVE_DMABUF
if (GDK_IS_DMABUF_TEXTURE (texture))
{
GskGpuImage *image = gsk_vulkan_image_new_for_dmabuf (GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (frame)),
texture);
if (image)
return image;
}
#endif
return GSK_GPU_FRAME_CLASS (gsk_vulkan_frame_parent_class)->upload_texture (frame, with_mipmap, texture);
}
static void
gsk_vulkan_frame_prepare_descriptors (GskVulkanFrame *self)
{
GskVulkanDevice *device;
VkDevice vk_device;
gsize i, n_images, n_buffers, n_sets;
device = GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (GSK_GPU_FRAME (self)));
vk_device = gsk_vulkan_device_get_vk_device (device);
n_images = 0;
n_buffers = 0;
n_sets = 2 * gsk_descriptors_get_size (&self->descriptors);
for (i = 0; i < gsk_descriptors_get_size (&self->descriptors); i++)
{
gsize n_desc_images, n_desc_buffers;
GskVulkanRealDescriptors *desc = gsk_descriptors_get (&self->descriptors, i);
gsk_vulkan_real_descriptors_prepare (desc, &n_desc_images, &n_desc_buffers);
n_images += n_desc_images;
n_buffers += n_desc_buffers;
}
if (n_sets > self->pool_n_sets ||
n_images > self->pool_n_images ||
n_buffers > self->pool_n_buffers)
{
if (self->vk_descriptor_pool != VK_NULL_HANDLE)
{
vkDestroyDescriptorPool (vk_device,
self->vk_descriptor_pool,
NULL);
self->vk_descriptor_pool = VK_NULL_HANDLE;
}
if (n_sets > self->pool_n_sets)
self->pool_n_sets = 4 << g_bit_nth_msf (n_sets - 1, -1);
if (n_images > self->pool_n_images)
self->pool_n_images = 2 << g_bit_nth_msf (n_images - 1, -1);
if (n_buffers > self->pool_n_buffers)
self->pool_n_buffers = 4 << g_bit_nth_msf (n_buffers - 1, -1);
}
if (self->vk_descriptor_pool == VK_NULL_HANDLE)
{
GSK_VK_CHECK (vkCreateDescriptorPool, vk_device,
&(VkDescriptorPoolCreateInfo) {
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
.flags = 0,
.maxSets = self->pool_n_sets,
.poolSizeCount = 2,
.pPoolSizes = (VkDescriptorPoolSize[2]) {
{
.type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
.descriptorCount = self->pool_n_images,
},
{
.type = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = self->pool_n_buffers,
}
}
},
NULL,
&self->vk_descriptor_pool);
}
for (i = 0; i < gsk_descriptors_get_size (&self->descriptors); i++)
{
GskVulkanRealDescriptors *desc = gsk_descriptors_get (&self->descriptors, i);
gsk_vulkan_real_descriptors_update_sets (desc, self->vk_descriptor_pool);
}
}
static GskGpuDescriptors *
gsk_vulkan_frame_create_descriptors (GskGpuFrame *frame)
{
GskVulkanFrame *self = GSK_VULKAN_FRAME (frame);
if (gsk_vulkan_device_has_feature (GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (frame)), GDK_VULKAN_FEATURE_DESCRIPTOR_INDEXING))
{
GskVulkanRealDescriptors *parent;
if (gsk_descriptors_get_size (&self->descriptors) > 0)
{
parent = gsk_descriptors_get (&self->descriptors, gsk_descriptors_get_size (&self->descriptors) - 1);
if (gsk_vulkan_real_descriptors_is_full (parent))
parent = NULL;
}
else
parent = NULL;
if (parent == NULL)
{
parent = gsk_vulkan_real_descriptors_new (self);
gsk_descriptors_append (&self->descriptors, parent);
}
return GSK_GPU_DESCRIPTORS (gsk_vulkan_sub_descriptors_new (GSK_VULKAN_DESCRIPTORS (parent)));
}
else
{
GskVulkanRealDescriptors *desc;
desc = gsk_vulkan_real_descriptors_new (self);
gsk_descriptors_append (&self->descriptors, desc);
return GSK_GPU_DESCRIPTORS (g_object_ref (desc));
}
}
static GskGpuBuffer *
gsk_vulkan_frame_create_vertex_buffer (GskGpuFrame *frame,
gsize size)
{
return gsk_vulkan_buffer_new_vertex (GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (frame)), size);
}
static GskGpuBuffer *
gsk_vulkan_frame_create_storage_buffer (GskGpuFrame *frame,
gsize size)
{
return gsk_vulkan_buffer_new_storage (GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (frame)), size);
}
static void
gsk_vulkan_frame_submit (GskGpuFrame *frame,
GskGpuBuffer *vertex_buffer,
GskGpuOp *op)
{
GskVulkanFrame *self = GSK_VULKAN_FRAME (frame);
GskVulkanSemaphores semaphores;
GskVulkanCommandState state;
if (gsk_descriptors_get_size (&self->descriptors) == 0)
gsk_descriptors_append (&self->descriptors, gsk_vulkan_real_descriptors_new (self));
gsk_vulkan_frame_prepare_descriptors (self);
GSK_VK_CHECK (vkBeginCommandBuffer, self->vk_command_buffer,
&(VkCommandBufferBeginInfo) {
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
});
if (vertex_buffer)
vkCmdBindVertexBuffers (self->vk_command_buffer,
0,
1,
(VkBuffer[1]) {
gsk_vulkan_buffer_get_vk_buffer (GSK_VULKAN_BUFFER (vertex_buffer))
},
(VkDeviceSize[1]) { 0 });
gsk_semaphores_init (&semaphores.wait_semaphores);
gsk_pipeline_stages_init (&semaphores.wait_stages);
gsk_semaphores_init (&semaphores.signal_semaphores);
state.vk_command_buffer = self->vk_command_buffer;
state.vk_render_pass = VK_NULL_HANDLE;
state.vk_format = VK_FORMAT_UNDEFINED;
state.blend = GSK_GPU_BLEND_OVER; /* should we have a BLEND_NONE? */
state.desc = GSK_VULKAN_DESCRIPTORS (gsk_descriptors_get (&self->descriptors, 0));
state.semaphores = &semaphores;
gsk_vulkan_descriptors_bind (GSK_VULKAN_DESCRIPTORS (gsk_descriptors_get (&self->descriptors, 0)),
NULL,
state.vk_command_buffer);
while (op)
{
op = gsk_gpu_op_vk_command (op, frame, &state);
}
GSK_VK_CHECK (vkEndCommandBuffer, self->vk_command_buffer);
GSK_VK_CHECK (vkQueueSubmit, gsk_vulkan_device_get_vk_queue (GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (frame))),
1,
&(VkSubmitInfo) {
.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
.commandBufferCount = 1,
.pCommandBuffers = &self->vk_command_buffer,
.pWaitSemaphores = gsk_semaphores_get_data (&semaphores.wait_semaphores),
.pWaitDstStageMask = gsk_pipeline_stages_get_data (&semaphores.wait_stages),
.waitSemaphoreCount = gsk_semaphores_get_size (&semaphores.wait_semaphores),
.pSignalSemaphores = gsk_semaphores_get_data (&semaphores.signal_semaphores),
.signalSemaphoreCount = gsk_semaphores_get_size (&semaphores.signal_semaphores),
},
self->vk_fence);
gsk_semaphores_clear (&semaphores.wait_semaphores);
gsk_pipeline_stages_clear (&semaphores.wait_stages);
gsk_semaphores_clear (&semaphores.signal_semaphores);
}
static void
gsk_vulkan_frame_finalize (GObject *object)
{
GskVulkanFrame *self = GSK_VULKAN_FRAME (object);
GskVulkanDevice *device;
VkDevice vk_device;
VkCommandPool vk_command_pool;
device = GSK_VULKAN_DEVICE (gsk_gpu_frame_get_device (GSK_GPU_FRAME (self)));
vk_device = gsk_vulkan_device_get_vk_device (device);
vk_command_pool = gsk_vulkan_device_get_vk_command_pool (device);
if (self->vk_descriptor_pool != VK_NULL_HANDLE)
{
vkDestroyDescriptorPool (vk_device,
self->vk_descriptor_pool,
NULL);
}
gsk_descriptors_clear (&self->descriptors);
vkFreeCommandBuffers (vk_device,
vk_command_pool,
1, &self->vk_command_buffer);
vkDestroyFence (vk_device,
self->vk_fence,
NULL);
G_OBJECT_CLASS (gsk_vulkan_frame_parent_class)->finalize (object);
}
static void
gsk_vulkan_frame_class_init (GskVulkanFrameClass *klass)
{
GskGpuFrameClass *gpu_frame_class = GSK_GPU_FRAME_CLASS (klass);
GObjectClass *object_class = G_OBJECT_CLASS (klass);
gpu_frame_class->is_busy = gsk_vulkan_frame_is_busy;
gpu_frame_class->wait = gsk_vulkan_frame_wait;
gpu_frame_class->setup = gsk_vulkan_frame_setup;
gpu_frame_class->cleanup = gsk_vulkan_frame_cleanup;
gpu_frame_class->upload_texture = gsk_vulkan_frame_upload_texture;
gpu_frame_class->create_descriptors = gsk_vulkan_frame_create_descriptors;
gpu_frame_class->create_vertex_buffer = gsk_vulkan_frame_create_vertex_buffer;
gpu_frame_class->create_storage_buffer = gsk_vulkan_frame_create_storage_buffer;
gpu_frame_class->submit = gsk_vulkan_frame_submit;
object_class->finalize = gsk_vulkan_frame_finalize;
}
static void
gsk_vulkan_frame_init (GskVulkanFrame *self)
{
gsk_descriptors_init (&self->descriptors);
self->pool_n_sets = 4;
self->pool_n_images = 8;
self->pool_n_buffers = 8;
}
VkFence
gsk_vulkan_frame_get_vk_fence (GskVulkanFrame *self)
{
return self->vk_fence;
}
void
gsk_vulkan_semaphores_add_wait (GskVulkanSemaphores *self,
VkSemaphore semaphore,
VkPipelineStageFlags stage)
{
gsk_semaphores_append (&self->wait_semaphores, semaphore);
gsk_pipeline_stages_append (&self->wait_stages, stage);
}
void
gsk_vulkan_semaphores_add_signal (GskVulkanSemaphores *self,
VkSemaphore semaphore)
{
gsk_semaphores_append (&self->signal_semaphores, semaphore);
}