Files
test/source/blender/gpu/vulkan/vk_context.cc
Miguel Pozo 74224b25a5 GPU: Add GPU_shader_batch_create_from_infos
This is the first commit of the several required to support
subprocess-based parallel compilation on OpenGL.
This provides the base API and implementation, and exposes the max
subprocesses setting on the UI, but it's not used by any code yet.

More information and the rest of the code can be found in #121925.

This one includes:
- A new `GPU_shader_batch` API that allows requesting the compilation
  of multiple shaders at once, allowing GPU backed to compile them in
  parallel and asynchronously without blocking the Blender UI.
- A virtual `ShaderCompiler` class that backends can use to add their
  own implementation.
- A `ShaderCompilerGeneric` class that implements synchronous/blocking
  compilation of batches for backends that don't have their own
  implementation yet.
- A `GLShaderCompiler` that supports parallel compilation using
  subprocesses.
- A new `BLI_subprocess` API, including IPC (required for the
  `GLShaderCompiler` implementation).
- The implementation of the subprocess program in
  `GPU_compilation_subprocess`.
- A new `Max Shader Compilation Subprocesses` option in
  `Preferences > System > Memory & Limits` to enable parallel shader
  compilation and the max number of subprocesses to allocate (each
  subprocess has a relatively high memory footprint).

Implementation Overview:
There's a single `GLShaderCompiler` shared by all OpenGL contexts.
This class stores a pool of up to `GCaps.max_parallel_compilations`
subprocesses that can be used for compilation.
Each subprocess has a shared memory pool used for sending the shader
source code from the main Blender process and for receiving the already
compiled shader binary from the subprocess. This is synchronized using
a series of shared semaphores.
The subprocesses maintain a shader cache on disk inside a
`BLENDER_SHADER_CACHE` folder at the OS temporary folder.
Shaders that fail to compile are tried to be compiled again locally for
proper error reports.
Hanged subprocesses are currently detected using a timeout of 30s.

Pull Request: https://projects.blender.org/blender/blender/pulls/122232
2024-06-05 18:45:57 +02:00

396 lines
12 KiB
C++

/* SPDX-FileCopyrightText: 2022 Blender Authors
*
* SPDX-License-Identifier: GPL-2.0-or-later */
/** \file
* \ingroup gpu
*/
#include "vk_context.hh"
#include "vk_debug.hh"
#include "vk_backend.hh"
#include "vk_framebuffer.hh"
#include "vk_immediate.hh"
#include "vk_memory.hh"
#include "vk_shader.hh"
#include "vk_shader_interface.hh"
#include "vk_state_manager.hh"
#include "vk_texture.hh"
#include "GHOST_C-api.h"
namespace blender::gpu {
VKContext::VKContext(void *ghost_window,
void *ghost_context,
render_graph::VKResourceStateTracker &resources)
: render_graph(std::make_unique<render_graph::VKCommandBufferWrapper>(), resources)
{
ghost_window_ = ghost_window;
ghost_context_ = ghost_context;
state_manager = new VKStateManager();
imm = new VKImmediate();
/* For off-screen contexts. Default frame-buffer is empty. */
VKFrameBuffer *framebuffer = new VKFrameBuffer("back_left");
back_left = framebuffer;
active_fb = framebuffer;
compiler = new ShaderCompilerGeneric();
}
VKContext::~VKContext()
{
if (surface_texture_) {
GPU_texture_free(surface_texture_);
surface_texture_ = nullptr;
}
if (use_render_graph) {
render_graph.free_data();
}
VKBackend::get().device_.context_unregister(*this);
delete imm;
imm = nullptr;
delete compiler;
}
void VKContext::sync_backbuffer()
{
if (ghost_context_) {
VKDevice &device = VKBackend::get().device_;
if (!command_buffers_.is_initialized()) {
command_buffers_.init(device);
descriptor_pools_.init(device);
device.init_dummy_buffer(*this);
device.init_dummy_color_attachment();
}
descriptor_pools_.reset();
}
if (ghost_window_) {
GHOST_VulkanSwapChainData swap_chain_data = {};
GHOST_GetVulkanSwapChainFormat((GHOST_WindowHandle)ghost_window_, &swap_chain_data);
const bool reset_framebuffer = swap_chain_format_ != swap_chain_data.format ||
vk_extent_.width != swap_chain_data.extent.width ||
vk_extent_.height != swap_chain_data.extent.height;
if (reset_framebuffer) {
if (has_active_framebuffer()) {
deactivate_framebuffer();
}
if (surface_texture_) {
GPU_texture_free(surface_texture_);
surface_texture_ = nullptr;
}
surface_texture_ = GPU_texture_create_2d("back-left",
swap_chain_data.extent.width,
swap_chain_data.extent.height,
1,
to_gpu_format(swap_chain_data.format),
GPU_TEXTURE_USAGE_ATTACHMENT,
nullptr);
back_left->attachment_set(GPU_FB_COLOR_ATTACHMENT0,
GPU_ATTACHMENT_TEXTURE(surface_texture_));
back_left->bind(false);
swap_chain_format_ = swap_chain_data.format;
vk_extent_ = swap_chain_data.extent;
}
}
}
void VKContext::activate()
{
/* Make sure no other context is already bound to this thread. */
BLI_assert(is_active_ == false);
is_active_ = true;
sync_backbuffer();
immActivate();
}
void VKContext::deactivate()
{
immDeactivate();
is_active_ = false;
}
void VKContext::begin_frame() {}
void VKContext::end_frame()
{
if (!use_render_graph) {
VKDevice &device = VKBackend::get().device_get();
device.destroy_discarded_resources();
}
}
void VKContext::flush()
{
command_buffers_.submit();
}
void VKContext::finish()
{
command_buffers_.finish();
}
void VKContext::memory_statistics_get(int *r_total_mem_kb, int *r_free_mem_kb)
{
const VKDevice &device = VKBackend::get().device_get();
device.memory_statistics_get(r_total_mem_kb, r_free_mem_kb);
}
/* -------------------------------------------------------------------- */
/** \name State manager
* \{ */
VKStateManager &VKContext::state_manager_get() const
{
return *static_cast<VKStateManager *>(state_manager);
}
void VKContext::debug_unbind_all_ubo()
{
state_manager_get().uniform_buffer_unbind_all();
};
void VKContext::debug_unbind_all_ssbo()
{
state_manager_get().storage_buffer_unbind_all();
};
/** \} */
/* -------------------------------------------------------------------- */
/** \name Frame-buffer
* \{ */
void VKContext::activate_framebuffer(VKFrameBuffer &framebuffer)
{
if (has_active_framebuffer()) {
deactivate_framebuffer();
}
BLI_assert(active_fb == nullptr);
active_fb = &framebuffer;
framebuffer.update_size();
framebuffer.update_srgb();
if (use_render_graph) {
framebuffer.rendering_reset();
}
else {
command_buffers_get().begin_render_pass(framebuffer);
}
}
VKFrameBuffer *VKContext::active_framebuffer_get() const
{
return unwrap(active_fb);
}
bool VKContext::has_active_framebuffer() const
{
return active_framebuffer_get() != nullptr;
}
void VKContext::deactivate_framebuffer()
{
VKFrameBuffer *framebuffer = active_framebuffer_get();
BLI_assert(framebuffer != nullptr);
if (use_render_graph) {
framebuffer->rendering_end(*this);
}
else {
command_buffers_get().end_render_pass(*framebuffer);
}
active_fb = nullptr;
}
void VKContext::rendering_end()
{
VKFrameBuffer *framebuffer = active_framebuffer_get();
if (framebuffer) {
framebuffer->rendering_end(*this);
}
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Compute pipeline
* \{ */
void VKContext::bind_compute_pipeline()
{
VKShader *shader = unwrap(this->shader);
BLI_assert(shader);
VKPipeline &pipeline = shader->pipeline_get();
pipeline.bind(*this, VK_PIPELINE_BIND_POINT_COMPUTE);
shader->push_constants.update(*this);
if (shader->has_descriptor_set()) {
descriptor_set_.bind(*this, shader->vk_pipeline_layout_get(), VK_PIPELINE_BIND_POINT_COMPUTE);
}
}
void VKContext::update_pipeline_data(render_graph::VKPipelineData &pipeline_data)
{
VKShader &vk_shader = unwrap(*shader);
pipeline_data.vk_pipeline_layout = vk_shader.vk_pipeline_layout_get();
pipeline_data.vk_pipeline = vk_shader.ensure_and_get_compute_pipeline();
/* Update push constants. */
pipeline_data.push_constants_data = nullptr;
pipeline_data.push_constants_size = 0;
const VKPushConstants::Layout &push_constants_layout =
vk_shader.interface_get().push_constants_layout_get();
vk_shader.push_constants.update(*this);
if (push_constants_layout.storage_type_get() == VKPushConstants::StorageType::PUSH_CONSTANTS) {
pipeline_data.push_constants_size = push_constants_layout.size_in_bytes();
pipeline_data.push_constants_data = vk_shader.push_constants.data();
}
/* Update descriptor set. */
pipeline_data.vk_descriptor_set = VK_NULL_HANDLE;
if (vk_shader.has_descriptor_set()) {
descriptor_set_.update(*this);
pipeline_data.vk_descriptor_set = descriptor_set_get().active_descriptor_set()->vk_handle();
}
}
render_graph::VKResourceAccessInfo &VKContext::update_and_get_access_info()
{
access_info_.reset();
state_manager_get().apply_bindings(*this, access_info_);
return access_info_;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Graphics pipeline
* \{ */
void VKContext::bind_graphics_pipeline(const GPUPrimType prim_type,
const VKVertexAttributeObject &vertex_attribute_object)
{
VKShader *shader = unwrap(this->shader);
BLI_assert(shader);
BLI_assert_msg(
prim_type != GPU_PRIM_POINTS || shader->interface_get().is_point_shader(),
"GPU_PRIM_POINTS is used with a shader that doesn't set point size before "
"drawing fragments. Calling code should be adapted to use a shader that sets the "
"gl_PointSize before entering the fragment stage. For example `GPU_SHADER_3D_POINT_*`.");
shader->update_graphics_pipeline(*this, prim_type, vertex_attribute_object);
VKPipeline &pipeline = shader->pipeline_get();
pipeline.bind(*this, VK_PIPELINE_BIND_POINT_GRAPHICS);
shader->push_constants.update(*this);
if (shader->has_descriptor_set()) {
descriptor_set_.bind(*this, shader->vk_pipeline_layout_get(), VK_PIPELINE_BIND_POINT_GRAPHICS);
}
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Graphics pipeline
* \{ */
void VKContext::swap_buffers_pre_callback(const GHOST_VulkanSwapChainData *swap_chain_data)
{
VKContext *context = VKContext::get();
BLI_assert(context);
context->swap_buffers_pre_handler(*swap_chain_data);
}
void VKContext::swap_buffers_post_callback()
{
VKContext *context = VKContext::get();
BLI_assert(context);
context->swap_buffers_post_handler();
}
void VKContext::swap_buffers_pre_handler(const GHOST_VulkanSwapChainData &swap_chain_data)
{
VKFrameBuffer &framebuffer = *unwrap(back_left);
VKTexture *color_attachment = unwrap(unwrap(framebuffer.color_tex(0)));
render_graph::VKBlitImageNode::CreateInfo blit_image = {};
blit_image.src_image = color_attachment->vk_image_handle();
blit_image.dst_image = swap_chain_data.image;
blit_image.filter = VK_FILTER_NEAREST;
VkImageBlit &region = blit_image.region;
region.srcOffsets[0] = {0, color_attachment->height_get() - 1, 0};
region.srcOffsets[1] = {color_attachment->width_get(), 0, 1};
region.srcSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
region.srcSubresource.mipLevel = 0;
region.srcSubresource.baseArrayLayer = 0;
region.srcSubresource.layerCount = 1;
region.dstOffsets[0] = {0, 0, 0};
region.dstOffsets[1] = {
int32_t(swap_chain_data.extent.width), int32_t(swap_chain_data.extent.height), 1};
region.dstSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
region.dstSubresource.mipLevel = 0;
region.dstSubresource.baseArrayLayer = 0;
region.dstSubresource.layerCount = 1;
if (use_render_graph) {
/* Swap chain commands are CPU synchronized at this moment, allowing to temporary add the swap
* chain image as device resources. When we move towards GPU swap chain synchronization we need
* to keep track of the swap chain image between frames. */
VKDevice &device = VKBackend::get().device_get();
device.resources.add_image(swap_chain_data.image,
VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
render_graph::ResourceOwner::SWAP_CHAIN);
framebuffer.rendering_end(*this);
render_graph.add_node(blit_image);
render_graph.submit_for_present(swap_chain_data.image);
device.resources.remove_image(swap_chain_data.image);
device.destroy_discarded_resources();
}
else {
/*
* Ensure no graphics/compute commands are scheduled. They could use the back buffer, which
* layout is altered here.
*/
command_buffers_get().submit();
VKTexture wrapper("display_texture");
wrapper.init(swap_chain_data.image,
VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
to_gpu_format(swap_chain_data.format));
wrapper.layout_ensure(*this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
framebuffer.color_attachment_layout_ensure(*this, 0, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
command_buffers_get().blit(wrapper,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
*color_attachment,
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
Span<VkImageBlit>(&region, 1));
wrapper.layout_ensure(*this, VK_IMAGE_LAYOUT_PRESENT_SRC_KHR);
command_buffers_get().submit();
}
}
void VKContext::swap_buffers_post_handler()
{
sync_backbuffer();
}
/** \} */
} // namespace blender::gpu