INSANEMODE/yuzu-android
1
0
Fork 0
mirror of https://github.com/yuzu-emu/yuzu-android.git synced 2025-06-16 21:07:57 -05:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #6585 from ameerj/hades

Browse Source 
Shader Decompiler Rewrite
This commit is contained in:
bunnei
2021-07-25 11:39:04 -07:00
committed by GitHub
parent c2aaf51370 7f13104c17
commit 98b26b6e12
428 changed files with 49376 additions and 27255 deletions
Download Patch File Download Diff File Expand all files Collapse all files

80
src/video_core/CMakeLists.txt
View File

@ -29,7 +29,6 @@ add_library(video_core STATIC
dirty_flags.h
dma_pusher.cpp
dma_pusher.h
engines/const_buffer_engine_interface.h
engines/const_buffer_info.h
engines/engine_interface.h
engines/engine_upload.cpp
@ -44,9 +43,6 @@ add_library(video_core STATIC
engines/maxwell_3d.h
engines/maxwell_dma.cpp
engines/maxwell_dma.h
engines/shader_bytecode.h
engines/shader_header.h
engines/shader_type.h
framebuffer_config.h
macro/macro.cpp
macro/macro.h
@ -61,8 +57,6 @@ add_library(video_core STATIC
gpu.h
gpu_thread.cpp
gpu_thread.h
guest_driver.cpp
guest_driver.h
memory_manager.cpp
memory_manager.h
query_cache.h
@ -71,26 +65,25 @@ add_library(video_core STATIC
rasterizer_interface.h
renderer_base.cpp
renderer_base.h
renderer_opengl/gl_arb_decompiler.cpp
renderer_opengl/gl_arb_decompiler.h
renderer_opengl/gl_buffer_cache.cpp
renderer_opengl/gl_buffer_cache.h
renderer_opengl/gl_compute_pipeline.cpp
renderer_opengl/gl_compute_pipeline.h
renderer_opengl/gl_device.cpp
renderer_opengl/gl_device.h
renderer_opengl/gl_fence_manager.cpp
renderer_opengl/gl_fence_manager.h
renderer_opengl/gl_graphics_pipeline.cpp
renderer_opengl/gl_graphics_pipeline.h
renderer_opengl/gl_rasterizer.cpp
renderer_opengl/gl_rasterizer.h
renderer_opengl/gl_resource_manager.cpp
renderer_opengl/gl_resource_manager.h
renderer_opengl/gl_shader_cache.cpp
renderer_opengl/gl_shader_cache.h
renderer_opengl/gl_shader_decompiler.cpp
renderer_opengl/gl_shader_decompiler.h
renderer_opengl/gl_shader_disk_cache.cpp
renderer_opengl/gl_shader_disk_cache.h
renderer_opengl/gl_shader_manager.cpp
renderer_opengl/gl_shader_manager.h
renderer_opengl/gl_shader_context.h
renderer_opengl/gl_shader_util.cpp
renderer_opengl/gl_shader_util.h
renderer_opengl/gl_state_tracker.cpp
@ -112,6 +105,7 @@ add_library(video_core STATIC
renderer_vulkan/fixed_pipeline_state.h
renderer_vulkan/maxwell_to_vk.cpp
renderer_vulkan/maxwell_to_vk.h
renderer_vulkan/pipeline_helper.h
renderer_vulkan/renderer_vulkan.h
renderer_vulkan/renderer_vulkan.cpp
renderer_vulkan/vk_blit_screen.cpp
@ -138,12 +132,12 @@ add_library(video_core STATIC
renderer_vulkan/vk_query_cache.h
renderer_vulkan/vk_rasterizer.cpp
renderer_vulkan/vk_rasterizer.h
renderer_vulkan/vk_render_pass_cache.cpp
renderer_vulkan/vk_render_pass_cache.h
renderer_vulkan/vk_resource_pool.cpp
renderer_vulkan/vk_resource_pool.h
renderer_vulkan/vk_scheduler.cpp
renderer_vulkan/vk_scheduler.h
renderer_vulkan/vk_shader_decompiler.cpp
renderer_vulkan/vk_shader_decompiler.h
renderer_vulkan/vk_shader_util.cpp
renderer_vulkan/vk_shader_util.h
renderer_vulkan/vk_staging_buffer_pool.cpp
@ -156,60 +150,12 @@ add_library(video_core STATIC
renderer_vulkan/vk_texture_cache.h
renderer_vulkan/vk_update_descriptor.cpp
renderer_vulkan/vk_update_descriptor.h
shader_cache.cpp
shader_cache.h
shader_environment.cpp
shader_environment.h
shader_notify.cpp
shader_notify.h
shader/decode/arithmetic.cpp
shader/decode/arithmetic_immediate.cpp
shader/decode/bfe.cpp
shader/decode/bfi.cpp
shader/decode/shift.cpp
shader/decode/arithmetic_integer.cpp
shader/decode/arithmetic_integer_immediate.cpp
shader/decode/arithmetic_half.cpp
shader/decode/arithmetic_half_immediate.cpp
shader/decode/ffma.cpp
shader/decode/hfma2.cpp
shader/decode/conversion.cpp
shader/decode/memory.cpp
shader/decode/texture.cpp
shader/decode/image.cpp
shader/decode/float_set_predicate.cpp
shader/decode/integer_set_predicate.cpp
shader/decode/half_set_predicate.cpp
shader/decode/predicate_set_register.cpp
shader/decode/predicate_set_predicate.cpp
shader/decode/register_set_predicate.cpp
shader/decode/float_set.cpp
shader/decode/integer_set.cpp
shader/decode/half_set.cpp
shader/decode/video.cpp
shader/decode/warp.cpp
shader/decode/xmad.cpp
shader/decode/other.cpp
shader/ast.cpp
shader/ast.h
shader/async_shaders.cpp
shader/async_shaders.h
shader/compiler_settings.cpp
shader/compiler_settings.h
shader/control_flow.cpp
shader/control_flow.h
shader/decode.cpp
shader/expr.cpp
shader/expr.h
shader/memory_util.cpp
shader/memory_util.h
shader/node_helper.cpp
shader/node_helper.h
shader/node.h
shader/registry.cpp
shader/registry.h
shader/shader_ir.cpp
shader/shader_ir.h
shader/track.cpp
shader/transform_feedback.cpp
shader/transform_feedback.h
surface.cpp
surface.h
texture_cache/accelerated_swizzle.cpp
@ -242,6 +188,8 @@ add_library(video_core STATIC
textures/decoders.h
textures/texture.cpp
textures/texture.h
transform_feedback.cpp
transform_feedback.h
video_core.cpp
video_core.h
vulkan_common/vulkan_debug_callback.cpp
@ -265,7 +213,7 @@ add_library(video_core STATIC
create_target_directory_groups(video_core)
target_link_libraries(video_core PUBLIC common core)
target_link_libraries(video_core PRIVATE glad xbyak)
target_link_libraries(video_core PUBLIC glad shader_recompiler xbyak)
if (YUZU_USE_BUNDLED_FFMPEG AND NOT WIN32)
add_dependencies(video_core ffmpeg-build)

269
src/video_core/buffer_cache/buffer_cache.h
View File

@ -31,6 +31,7 @@
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/surface.h"
#include "video_core/texture_cache/slot_vector.h"
#include "video_core/texture_cache/types.h"
@ -42,14 +43,19 @@ MICROPROFILE_DECLARE(GPU_DownloadMemory);
using BufferId = SlotId;
using VideoCore::Surface::PixelFormat;
using namespace Common::Literals;
constexpr u32 NUM_VERTEX_BUFFERS = 32;
constexpr u32 NUM_TRANSFORM_FEEDBACK_BUFFERS = 4;
constexpr u32 NUM_GRAPHICS_UNIFORM_BUFFERS = 18;
constexpr u32 NUM_COMPUTE_UNIFORM_BUFFERS = 8;
constexpr u32 NUM_STORAGE_BUFFERS = 16;
constexpr u32 NUM_TEXTURE_BUFFERS = 16;
constexpr u32 NUM_STAGES = 5;
using namespace Common::Literals;
using UniformBufferSizes = std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES>;
using ComputeUniformBufferSizes = std::array<u32, NUM_COMPUTE_UNIFORM_BUFFERS>;
template <typename P>
class BufferCache {
@ -67,6 +73,7 @@ class BufferCache {
static constexpr bool NEEDS_BIND_UNIFORM_INDEX = P::NEEDS_BIND_UNIFORM_INDEX;
static constexpr bool NEEDS_BIND_STORAGE_INDEX = P::NEEDS_BIND_STORAGE_INDEX;
static constexpr bool USE_MEMORY_MAPS = P::USE_MEMORY_MAPS;
static constexpr bool SEPARATE_IMAGE_BUFFERS_BINDINGS = P::SEPARATE_IMAGE_BUFFER_BINDINGS;
static constexpr BufferId NULL_BUFFER_ID{0};
@ -96,6 +103,10 @@ class BufferCache {
BufferId buffer_id;
};
struct TextureBufferBinding : Binding {
PixelFormat format;
};
static constexpr Binding NULL_BINDING{
.cpu_addr = 0,
.size = 0,
@ -133,20 +144,31 @@ public:
void BindHostComputeBuffers();
void SetEnabledUniformBuffers(size_t stage, u32 enabled);
void SetUniformBuffersState(const std::array<u32, NUM_STAGES>& mask,
const UniformBufferSizes* sizes);
void SetEnabledComputeUniformBuffers(u32 enabled);
void SetComputeUniformBufferState(u32 mask, const ComputeUniformBufferSizes* sizes);
void UnbindGraphicsStorageBuffers(size_t stage);
void BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
bool is_written);
void UnbindGraphicsTextureBuffers(size_t stage);
void BindGraphicsTextureBuffer(size_t stage, size_t tbo_index, GPUVAddr gpu_addr, u32 size,
PixelFormat format, bool is_written, bool is_image);
void UnbindComputeStorageBuffers();
void BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
bool is_written);
void UnbindComputeTextureBuffers();
void BindComputeTextureBuffer(size_t tbo_index, GPUVAddr gpu_addr, u32 size, PixelFormat format,
bool is_written, bool is_image);
void FlushCachedWrites();
/// Return true when there are uncommitted buffers to be downloaded
@ -178,6 +200,7 @@ public:
[[nodiscard]] bool IsRegionCpuModified(VAddr addr, size_t size);
std::mutex mutex;
Runtime& runtime;
private:
template <typename Func>
@ -254,12 +277,16 @@ private:
void BindHostGraphicsStorageBuffers(size_t stage);
void BindHostGraphicsTextureBuffers(size_t stage);
void BindHostTransformFeedbackBuffers();
void BindHostComputeUniformBuffers();
void BindHostComputeStorageBuffers();
void BindHostComputeTextureBuffers();
void DoUpdateGraphicsBuffers(bool is_indexed);
void DoUpdateComputeBuffers();
@ -274,6 +301,8 @@ private:
void UpdateStorageBuffers(size_t stage);
void UpdateTextureBuffers(size_t stage);
void UpdateTransformFeedbackBuffers();
void UpdateTransformFeedbackBuffer(u32 index);
@ -282,6 +311,8 @@ private:
void UpdateComputeStorageBuffers();
void UpdateComputeTextureBuffers();
void MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 size);
[[nodiscard]] BufferId FindBuffer(VAddr cpu_addr, u32 size);
@ -323,6 +354,9 @@ private:
[[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr) const;
[[nodiscard]] TextureBufferBinding GetTextureBufferBinding(GPUVAddr gpu_addr, u32 size,
PixelFormat format);
[[nodiscard]] std::span<const u8> ImmediateBufferWithData(VAddr cpu_addr, size_t size);
[[nodiscard]] std::span<u8> ImmediateBuffer(size_t wanted_capacity);
@ -336,7 +370,6 @@ private:
Tegra::Engines::KeplerCompute& kepler_compute;
Tegra::MemoryManager& gpu_memory;
Core::Memory::Memory& cpu_memory;
Runtime& runtime;
SlotVector<Buffer> slot_buffers;
DelayedDestructionRing<Buffer, 8> delayed_destruction_ring;
@ -347,20 +380,30 @@ private:
std::array<Binding, NUM_VERTEX_BUFFERS> vertex_buffers;
std::array<std::array<Binding, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES> uniform_buffers;
std::array<std::array<Binding, NUM_STORAGE_BUFFERS>, NUM_STAGES> storage_buffers;
std::array<std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS>, NUM_STAGES> texture_buffers;
std::array<Binding, NUM_TRANSFORM_FEEDBACK_BUFFERS> transform_feedback_buffers;
std::array<Binding, NUM_COMPUTE_UNIFORM_BUFFERS> compute_uniform_buffers;
std::array<Binding, NUM_STORAGE_BUFFERS> compute_storage_buffers;
std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS> compute_texture_buffers;
std::array<u32, NUM_STAGES> enabled_uniform_buffers{};
u32 enabled_compute_uniform_buffers = 0;
std::array<u32, NUM_STAGES> enabled_uniform_buffer_masks{};
u32 enabled_compute_uniform_buffer_mask = 0;
const UniformBufferSizes* uniform_buffer_sizes{};
const ComputeUniformBufferSizes* compute_uniform_buffer_sizes{};
std::array<u32, NUM_STAGES> enabled_storage_buffers{};
std::array<u32, NUM_STAGES> written_storage_buffers{};
u32 enabled_compute_storage_buffers = 0;
u32 written_compute_storage_buffers = 0;
std::array<u32, NUM_STAGES> fast_bound_uniform_buffers{};
std::array<u32, NUM_STAGES> enabled_texture_buffers{};
std::array<u32, NUM_STAGES> written_texture_buffers{};
std::array<u32, NUM_STAGES> image_texture_buffers{};
u32 enabled_compute_texture_buffers = 0;
u32 written_compute_texture_buffers = 0;
u32 image_compute_texture_buffers = 0;
std::array<u32, 16> uniform_cache_hits{};
std::array<u32, 16> uniform_cache_shots{};
@ -371,6 +414,10 @@ private:
std::conditional_t<HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS, std::array<u32, NUM_STAGES>, Empty>
dirty_uniform_buffers{};
std::conditional_t<IS_OPENGL, std::array<u32, NUM_STAGES>, Empty> fast_bound_uniform_buffers{};
std::conditional_t<HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS,
std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES>, Empty>
uniform_buffer_binding_sizes{};
std::vector<BufferId> cached_write_buffer_ids;
@ -394,8 +441,8 @@ BufferCache<P>::BufferCache(VideoCore::RasterizerInterface& rasterizer_,
Tegra::Engines::KeplerCompute& kepler_compute_,
Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
Runtime& runtime_)
: rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_},
gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_}, runtime{runtime_} {
: runtime{runtime_}, rasterizer{rasterizer_}, maxwell3d{maxwell3d_},
kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_} {
// Ensure the first slot is used for the null buffer
void(slot_buffers.insert(runtime, NullBufferParams{}));
deletion_iterator = slot_buffers.end();
@ -615,6 +662,7 @@ void BufferCache<P>::BindHostStageBuffers(size_t stage) {
MICROPROFILE_SCOPE(GPU_BindUploadBuffers);
BindHostGraphicsUniformBuffers(stage);
BindHostGraphicsStorageBuffers(stage);
BindHostGraphicsTextureBuffers(stage);
}
template <class P>
@ -622,21 +670,30 @@ void BufferCache<P>::BindHostComputeBuffers() {
MICROPROFILE_SCOPE(GPU_BindUploadBuffers);
BindHostComputeUniformBuffers();
BindHostComputeStorageBuffers();
BindHostComputeTextureBuffers();
}
template <class P>
void BufferCache<P>::SetEnabledUniformBuffers(size_t stage, u32 enabled) {
void BufferCache<P>::SetUniformBuffersState(const std::array<u32, NUM_STAGES>& mask,
const UniformBufferSizes* sizes) {
if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
if (enabled_uniform_buffers[stage] != enabled) {
dirty_uniform_buffers[stage] = ~u32{0};
if (enabled_uniform_buffer_masks != mask) {
if constexpr (IS_OPENGL) {
fast_bound_uniform_buffers.fill(0);
}
dirty_uniform_buffers.fill(~u32{0});
uniform_buffer_binding_sizes.fill({});
}
}
enabled_uniform_buffers[stage] = enabled;
enabled_uniform_buffer_masks = mask;
uniform_buffer_sizes = sizes;
}
template <class P>
void BufferCache<P>::SetEnabledComputeUniformBuffers(u32 enabled) {
enabled_compute_uniform_buffers = enabled;
void BufferCache<P>::SetComputeUniformBufferState(u32 mask,
const ComputeUniformBufferSizes* sizes) {
enabled_compute_uniform_buffer_mask = mask;
compute_uniform_buffer_sizes = sizes;
}
template <class P>
@ -656,10 +713,30 @@ void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index,
storage_buffers[stage][ssbo_index] = StorageBufferBinding(ssbo_addr);
}
template <class P>
void BufferCache<P>::UnbindGraphicsTextureBuffers(size_t stage) {
enabled_texture_buffers[stage] = 0;
written_texture_buffers[stage] = 0;
image_texture_buffers[stage] = 0;
}
template <class P>
void BufferCache<P>::BindGraphicsTextureBuffer(size_t stage, size_t tbo_index, GPUVAddr gpu_addr,
u32 size, PixelFormat format, bool is_written,
bool is_image) {
enabled_texture_buffers[stage] |= 1U << tbo_index;
written_texture_buffers[stage] |= (is_written ? 1U : 0U) << tbo_index;
if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
image_texture_buffers[stage] |= (is_image ? 1U : 0U) << tbo_index;
}
texture_buffers[stage][tbo_index] = GetTextureBufferBinding(gpu_addr, size, format);
}
template <class P>
void BufferCache<P>::UnbindComputeStorageBuffers() {
enabled_compute_storage_buffers = 0;
written_compute_storage_buffers = 0;
image_compute_texture_buffers = 0;
}
template <class P>
@ -676,6 +753,24 @@ void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index,
compute_storage_buffers[ssbo_index] = StorageBufferBinding(ssbo_addr);
}
template <class P>
void BufferCache<P>::UnbindComputeTextureBuffers() {
enabled_compute_texture_buffers = 0;
written_compute_texture_buffers = 0;
image_compute_texture_buffers = 0;
}
template <class P>
void BufferCache<P>::BindComputeTextureBuffer(size_t tbo_index, GPUVAddr gpu_addr, u32 size,
PixelFormat format, bool is_written, bool is_image) {
enabled_compute_texture_buffers |= 1U << tbo_index;
written_compute_texture_buffers |= (is_written ? 1U : 0U) << tbo_index;
if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
image_compute_texture_buffers |= (is_image ? 1U : 0U) << tbo_index;
}
compute_texture_buffers[tbo_index] = GetTextureBufferBinding(gpu_addr, size, format);
}
template <class P>
void BufferCache<P>::FlushCachedWrites() {
for (const BufferId buffer_id : cached_write_buffer_ids) {
@ -901,7 +996,7 @@ void BufferCache<P>::BindHostGraphicsUniformBuffers(size_t stage) {
dirty = std::exchange(dirty_uniform_buffers[stage], 0);
}
u32 binding_index = 0;
ForEachEnabledBit(enabled_uniform_buffers[stage], [&](u32 index) {
ForEachEnabledBit(enabled_uniform_buffer_masks[stage], [&](u32 index) {
const bool needs_bind = ((dirty >> index) & 1) != 0;
BindHostGraphicsUniformBuffer(stage, index, binding_index, needs_bind);
if constexpr (NEEDS_BIND_UNIFORM_INDEX) {
@ -915,7 +1010,7 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
bool needs_bind) {
const Binding& binding = uniform_buffers[stage][index];
const VAddr cpu_addr = binding.cpu_addr;
const u32 size = binding.size;
const u32 size = std::min(binding.size, (*uniform_buffer_sizes)[stage][index]);
Buffer& buffer = slot_buffers[binding.buffer_id];
TouchBuffer(buffer);
const bool use_fast_buffer = binding.buffer_id != NULL_BUFFER_ID &&
@ -925,8 +1020,13 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
if constexpr (IS_OPENGL) {
if (runtime.HasFastBufferSubData()) {
// Fast path for Nvidia
if (!HasFastUniformBufferBound(stage, binding_index)) {
const bool should_fast_bind =
!HasFastUniformBufferBound(stage, binding_index) ||
uniform_buffer_binding_sizes[stage][binding_index] != size;
if (should_fast_bind) {
// We only have to bind when the currently bound buffer is not the fast version
fast_bound_uniform_buffers[stage] |= 1U << binding_index;
uniform_buffer_binding_sizes[stage][binding_index] = size;
runtime.BindFastUniformBuffer(stage, binding_index, size);
}
const auto span = ImmediateBufferWithData(cpu_addr, size);
@ -934,8 +1034,10 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
return;
}
}
fast_bound_uniform_buffers[stage] |= 1U << binding_index;
if constexpr (IS_OPENGL) {
fast_bound_uniform_buffers[stage] |= 1U << binding_index;
uniform_buffer_binding_sizes[stage][binding_index] = size;
}
// Stream buffer path to avoid stalling on non-Nvidia drivers or Vulkan
const std::span<u8> span = runtime.BindMappedUniformBuffer(stage, binding_index, size);
cpu_memory.ReadBlockUnsafe(cpu_addr, span.data(), size);
@ -948,14 +1050,27 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
}
++uniform_cache_shots[0];
if (!needs_bind && !HasFastUniformBufferBound(stage, binding_index)) {
// Skip binding if it's not needed and if the bound buffer is not the fast version
// This exists to avoid instances where the fast buffer is bound and a GPU write happens
// Skip binding if it's not needed and if the bound buffer is not the fast version
// This exists to avoid instances where the fast buffer is bound and a GPU write happens
needs_bind |= HasFastUniformBufferBound(stage, binding_index);
if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
needs_bind |= uniform_buffer_binding_sizes[stage][binding_index] != size;
}
if (!needs_bind) {
return;
}
fast_bound_uniform_buffers[stage] &= ~(1U << binding_index);
const u32 offset = buffer.Offset(cpu_addr);
if constexpr (IS_OPENGL) {
// Fast buffer will be unbound
fast_bound_uniform_buffers[stage] &= ~(1U << binding_index);
// Mark the index as dirty if offset doesn't match
const bool is_copy_bind = offset != 0 && !runtime.SupportsNonZeroUniformOffset();
dirty_uniform_buffers[stage] |= (is_copy_bind ? 1U : 0U) << index;
}
if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
uniform_buffer_binding_sizes[stage][binding_index] = size;
}
if constexpr (NEEDS_BIND_UNIFORM_INDEX) {
runtime.BindUniformBuffer(stage, binding_index, buffer, offset, size);
} else {
@ -984,6 +1099,28 @@ void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
});
}
template <class P>
void BufferCache<P>::BindHostGraphicsTextureBuffers(size_t stage) {
ForEachEnabledBit(enabled_texture_buffers[stage], [&](u32 index) {
const TextureBufferBinding& binding = texture_buffers[stage][index];
Buffer& buffer = slot_buffers[binding.buffer_id];
const u32 size = binding.size;
SynchronizeBuffer(buffer, binding.cpu_addr, size);
const u32 offset = buffer.Offset(binding.cpu_addr);
const PixelFormat format = binding.format;
if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
if (((image_texture_buffers[stage] >> index) & 1) != 0) {
runtime.BindImageBuffer(buffer, offset, size, format);
} else {
runtime.BindTextureBuffer(buffer, offset, size, format);
}
} else {
runtime.BindTextureBuffer(buffer, offset, size, format);
}
});
}
template <class P>
void BufferCache<P>::BindHostTransformFeedbackBuffers() {
if (maxwell3d.regs.tfb_enabled == 0) {
@ -1006,13 +1143,14 @@ void BufferCache<P>::BindHostComputeUniformBuffers() {
if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
// Mark all uniform buffers as dirty
dirty_uniform_buffers.fill(~u32{0});
fast_bound_uniform_buffers.fill(0);
}
u32 binding_index = 0;
ForEachEnabledBit(enabled_compute_uniform_buffers, [&](u32 index) {
ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) {
const Binding& binding = compute_uniform_buffers[index];
Buffer& buffer = slot_buffers[binding.buffer_id];
TouchBuffer(buffer);
const u32 size = binding.size;
const u32 size = std::min(binding.size, (*compute_uniform_buffer_sizes)[index]);
SynchronizeBuffer(buffer, binding.cpu_addr, size);
const u32 offset = buffer.Offset(binding.cpu_addr);
@ -1046,6 +1184,28 @@ void BufferCache<P>::BindHostComputeStorageBuffers() {
});
}
template <class P>
void BufferCache<P>::BindHostComputeTextureBuffers() {
ForEachEnabledBit(enabled_compute_texture_buffers, [&](u32 index) {
const TextureBufferBinding& binding = compute_texture_buffers[index];
Buffer& buffer = slot_buffers[binding.buffer_id];
const u32 size = binding.size;
SynchronizeBuffer(buffer, binding.cpu_addr, size);
const u32 offset = buffer.Offset(binding.cpu_addr);
const PixelFormat format = binding.format;
if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) {
if (((image_compute_texture_buffers >> index) & 1) != 0) {
runtime.BindImageBuffer(buffer, offset, size, format);
} else {
runtime.BindTextureBuffer(buffer, offset, size, format);
}
} else {
runtime.BindTextureBuffer(buffer, offset, size, format);
}
});
}
template <class P>
void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) {
if (is_indexed) {
@ -1056,6 +1216,7 @@ void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) {
for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
UpdateUniformBuffers(stage);
UpdateStorageBuffers(stage);
UpdateTextureBuffers(stage);
}
}
@ -1063,6 +1224,7 @@ template <class P>
void BufferCache<P>::DoUpdateComputeBuffers() {
UpdateComputeUniformBuffers();
UpdateComputeStorageBuffers();
UpdateComputeTextureBuffers();
}
template <class P>
@ -1132,7 +1294,7 @@ void BufferCache<P>::UpdateVertexBuffer(u32 index) {
template <class P>
void BufferCache<P>::UpdateUniformBuffers(size_t stage) {
ForEachEnabledBit(enabled_uniform_buffers[stage], [&](u32 index) {
ForEachEnabledBit(enabled_uniform_buffer_masks[stage], [&](u32 index) {
Binding& binding = uniform_buffers[stage][index];
if (binding.buffer_id) {
// Already updated
@ -1162,6 +1324,18 @@ void BufferCache<P>::UpdateStorageBuffers(size_t stage) {
});
}
template <class P>
void BufferCache<P>::UpdateTextureBuffers(size_t stage) {
ForEachEnabledBit(enabled_texture_buffers[stage], [&](u32 index) {
Binding& binding = texture_buffers[stage][index];
binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
// Mark buffer as written if needed
if (((written_texture_buffers[stage] >> index) & 1) != 0) {
MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
}
});
}
template <class P>
void BufferCache<P>::UpdateTransformFeedbackBuffers() {
if (maxwell3d.regs.tfb_enabled == 0) {
@ -1193,7 +1367,7 @@ void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 index) {
template <class P>
void BufferCache<P>::UpdateComputeUniformBuffers() {
ForEachEnabledBit(enabled_compute_uniform_buffers, [&](u32 index) {
ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) {
Binding& binding = compute_uniform_buffers[index];
binding = NULL_BINDING;
const auto& launch_desc = kepler_compute.launch_description;
@ -1214,11 +1388,22 @@ void BufferCache<P>::UpdateComputeStorageBuffers() {
ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) {
// Resolve buffer
Binding& binding = compute_storage_buffers[index];
const BufferId buffer_id = FindBuffer(binding.cpu_addr, binding.size);
binding.buffer_id = buffer_id;
binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
// Mark as written if needed
if (((written_compute_storage_buffers >> index) & 1) != 0) {
MarkWrittenBuffer(buffer_id, binding.cpu_addr, binding.size);
MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
}
});
}
template <class P>
void BufferCache<P>::UpdateComputeTextureBuffers() {
ForEachEnabledBit(enabled_compute_texture_buffers, [&](u32 index) {
Binding& binding = compute_texture_buffers[index];
binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
// Mark as written if needed
if (((written_compute_texture_buffers >> index) & 1) != 0) {
MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size);
}
});
}
@ -1551,6 +1736,7 @@ template <class P>
void BufferCache<P>::NotifyBufferDeletion() {
if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
dirty_uniform_buffers.fill(~u32{0});
uniform_buffer_binding_sizes.fill({});
}
auto& flags = maxwell3d.dirty.flags;
flags[Dirty::IndexBuffer] = true;
@ -1577,6 +1763,25 @@ typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr s
return binding;
}
template <class P>
typename BufferCache<P>::TextureBufferBinding BufferCache<P>::GetTextureBufferBinding(
GPUVAddr gpu_addr, u32 size, PixelFormat format) {
const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
TextureBufferBinding binding;
if (!cpu_addr || size == 0) {
binding.cpu_addr = 0;
binding.size = 0;
binding.buffer_id = NULL_BUFFER_ID;
binding.format = PixelFormat::Invalid;
} else {
binding.cpu_addr = *cpu_addr;
binding.size = size;
binding.buffer_id = BufferId{};
binding.format = format;
}
return binding;
}
template <class P>
std::span<const u8> BufferCache<P>::ImmediateBufferWithData(VAddr cpu_addr, size_t size) {
u8* const base_pointer = cpu_memory.GetPointer(cpu_addr);

6
src/video_core/dirty_flags.cpp
View File

@ -58,6 +58,11 @@ void SetupDirtyRenderTargets(Maxwell3D::DirtyState::Tables& tables) {
FillBlock(table, OFF(zeta), NUM(zeta), flag);
}
}
void SetupDirtyShaders(Maxwell3D::DirtyState::Tables& tables) {
FillBlock(tables[0], OFF(shader_config[0]),
NUM(shader_config[0]) * Maxwell3D::Regs::MaxShaderProgram, Shaders);
}
} // Anonymous namespace
void SetupDirtyFlags(Maxwell3D::DirtyState::Tables& tables) {
@ -65,6 +70,7 @@ void SetupDirtyFlags(Maxwell3D::DirtyState::Tables& tables) {
SetupIndexBuffer(tables);
SetupDirtyDescriptors(tables);
SetupDirtyRenderTargets(tables);
SetupDirtyShaders(tables);
}
} // namespace VideoCommon::Dirty

2
src/video_core/dirty_flags.h
View File

@ -36,6 +36,8 @@ enum : u8 {
IndexBuffer,
Shaders,
LastCommonEntry,
};

103
src/video_core/engines/const_buffer_engine_interface.h
View File

@ -1,103 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <type_traits>
#include "common/bit_field.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/engines/shader_type.h"
#include "video_core/guest_driver.h"
#include "video_core/textures/texture.h"
namespace Tegra::Engines {
struct SamplerDescriptor {
union {
u32 raw = 0;
BitField<0, 2, Tegra::Shader::TextureType> texture_type;
BitField<2, 3, Tegra::Texture::ComponentType> r_type;
BitField<5, 1, u32> is_array;
BitField<6, 1, u32> is_buffer;
BitField<7, 1, u32> is_shadow;
BitField<8, 3, Tegra::Texture::ComponentType> g_type;
BitField<11, 3, Tegra::Texture::ComponentType> b_type;
BitField<14, 3, Tegra::Texture::ComponentType> a_type;
BitField<17, 7, Tegra::Texture::TextureFormat> format;
};
bool operator==(const SamplerDescriptor& rhs) const noexcept {
return raw == rhs.raw;
}
bool operator!=(const SamplerDescriptor& rhs) const noexcept {
return !operator==(rhs);
}
static SamplerDescriptor FromTIC(const Tegra::Texture::TICEntry& tic) {
using Tegra::Shader::TextureType;
SamplerDescriptor result;
result.format.Assign(tic.format.Value());
result.r_type.Assign(tic.r_type.Value());
result.g_type.Assign(tic.g_type.Value());
result.b_type.Assign(tic.b_type.Value());
result.a_type.Assign(tic.a_type.Value());
switch (tic.texture_type.Value()) {
case Tegra::Texture::TextureType::Texture1D:
result.texture_type.Assign(TextureType::Texture1D);
return result;
case Tegra::Texture::TextureType::Texture2D:
result.texture_type.Assign(TextureType::Texture2D);
return result;
case Tegra::Texture::TextureType::Texture3D:
result.texture_type.Assign(TextureType::Texture3D);
return result;
case Tegra::Texture::TextureType::TextureCubemap:
result.texture_type.Assign(TextureType::TextureCube);
return result;
case Tegra::Texture::TextureType::Texture1DArray:
result.texture_type.Assign(TextureType::Texture1D);
result.is_array.Assign(1);
return result;
case Tegra::Texture::TextureType::Texture2DArray:
result.texture_type.Assign(TextureType::Texture2D);
result.is_array.Assign(1);
return result;
case Tegra::Texture::TextureType::Texture1DBuffer:
result.texture_type.Assign(TextureType::Texture1D);
result.is_buffer.Assign(1);
return result;
case Tegra::Texture::TextureType::Texture2DNoMipmap:
result.texture_type.Assign(TextureType::Texture2D);
return result;
case Tegra::Texture::TextureType::TextureCubeArray:
result.texture_type.Assign(TextureType::TextureCube);
result.is_array.Assign(1);
return result;
default:
result.texture_type.Assign(TextureType::Texture2D);
return result;
}
}
};
static_assert(std::is_trivially_copyable_v<SamplerDescriptor>);
class ConstBufferEngineInterface {
public:
virtual ~ConstBufferEngineInterface() = default;
virtual u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const = 0;
virtual SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const = 0;
virtual SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const = 0;
virtual SamplerDescriptor AccessSampler(u32 handle) const = 0;
virtual u32 GetBoundBuffer() const = 0;
virtual VideoCore::GuestDriverProfile& AccessGuestDriverProfile() = 0;
virtual const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const = 0;
};
} // namespace Tegra::Engines

45
src/video_core/engines/kepler_compute.cpp
View File

@ -8,7 +8,6 @@
#include "core/core.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_base.h"
@ -57,53 +56,11 @@ void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amoun
}
}
u32 KeplerCompute::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const {
ASSERT(stage == ShaderType::Compute);
const auto& buffer = launch_description.const_buffer_config[const_buffer];
u32 result;
std::memcpy(&result, memory_manager.GetPointer(buffer.Address() + offset), sizeof(u32));
return result;
}
SamplerDescriptor KeplerCompute::AccessBoundSampler(ShaderType stage, u64 offset) const {
return AccessBindlessSampler(stage, regs.tex_cb_index, offset * sizeof(Texture::TextureHandle));
}
SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const {
ASSERT(stage == ShaderType::Compute);
const auto& tex_info_buffer = launch_description.const_buffer_config[const_buffer];
const GPUVAddr tex_info_address = tex_info_buffer.Address() + offset;
return AccessSampler(memory_manager.Read<u32>(tex_info_address));
}
SamplerDescriptor KeplerCompute::AccessSampler(u32 handle) const {
const Texture::TextureHandle tex_handle{handle};
const Texture::TICEntry tic = GetTICEntry(tex_handle.tic_id);
const Texture::TSCEntry tsc = GetTSCEntry(tex_handle.tsc_id);
SamplerDescriptor result = SamplerDescriptor::FromTIC(tic);
result.is_shadow.Assign(tsc.depth_compare_enabled.Value());
return result;
}
VideoCore::GuestDriverProfile& KeplerCompute::AccessGuestDriverProfile() {
return rasterizer->AccessGuestDriverProfile();
}
const VideoCore::GuestDriverProfile& KeplerCompute::AccessGuestDriverProfile() const {
return rasterizer->AccessGuestDriverProfile();
}
void KeplerCompute::ProcessLaunch() {
const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address();
memory_manager.ReadBlockUnsafe(launch_desc_loc, &launch_description,
LaunchParams::NUM_LAUNCH_PARAMETERS * sizeof(u32));
const GPUVAddr code_addr = regs.code_loc.Address() + launch_description.program_start;
LOG_TRACE(HW_GPU, "Compute invocation launched at address 0x{:016x}", code_addr);
rasterizer->DispatchCompute(code_addr);
rasterizer->DispatchCompute();
}
Texture::TICEntry KeplerCompute::GetTICEntry(u32 tic_index) const {

21
src/video_core/engines/kepler_compute.h
View File

@ -10,10 +10,8 @@
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/engines/const_buffer_engine_interface.h"
#include "video_core/engines/engine_interface.h"
#include "video_core/engines/engine_upload.h"
#include "video_core/engines/shader_type.h"
#include "video_core/gpu.h"
#include "video_core/textures/texture.h"
@ -40,7 +38,7 @@ namespace Tegra::Engines {
#define KEPLER_COMPUTE_REG_INDEX(field_name) \
(offsetof(Tegra::Engines::KeplerCompute::Regs, field_name) / sizeof(u32))
class KeplerCompute final : public ConstBufferEngineInterface, public EngineInterface {
class KeplerCompute final : public EngineInterface {
public:
explicit KeplerCompute(Core::System& system, MemoryManager& memory_manager);
~KeplerCompute();
@ -209,23 +207,6 @@ public:
void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
u32 methods_pending) override;
u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override;
SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override;
SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const override;
SamplerDescriptor AccessSampler(u32 handle) const override;
u32 GetBoundBuffer() const override {
return regs.tex_cb_index;
}
VideoCore::GuestDriverProfile& AccessGuestDriverProfile() override;
const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override;
private:
void ProcessLaunch();

39
src/video_core/engines/maxwell_3d.cpp
View File

@ -8,7 +8,6 @@
#include "core/core.h"
#include "core/core_timing.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/gpu.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
@ -670,42 +669,4 @@ void Maxwell3D::ProcessClearBuffers() {
rasterizer->Clear();
}
u32 Maxwell3D::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const {
ASSERT(stage != ShaderType::Compute);
const auto& shader_stage = state.shader_stages[static_cast<std::size_t>(stage)];
const auto& buffer = shader_stage.const_buffers[const_buffer];
return memory_manager.Read<u32>(buffer.address + offset);
}
SamplerDescriptor Maxwell3D::AccessBoundSampler(ShaderType stage, u64 offset) const {
return AccessBindlessSampler(stage, regs.tex_cb_index, offset * sizeof(Texture::TextureHandle));
}
SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const {
ASSERT(stage != ShaderType::Compute);
const auto& shader = state.shader_stages[static_cast<std::size_t>(stage)];
const auto& tex_info_buffer = shader.const_buffers[const_buffer];
const GPUVAddr tex_info_address = tex_info_buffer.address + offset;
return AccessSampler(memory_manager.Read<u32>(tex_info_address));
}
SamplerDescriptor Maxwell3D::AccessSampler(u32 handle) const {
const Texture::TextureHandle tex_handle{handle};
const Texture::TICEntry tic = GetTICEntry(tex_handle.tic_id);
const Texture::TSCEntry tsc = GetTSCEntry(tex_handle.tsc_id);
SamplerDescriptor result = SamplerDescriptor::FromTIC(tic);
result.is_shadow.Assign(tsc.depth_compare_enabled.Value());
return result;
}
VideoCore::GuestDriverProfile& Maxwell3D::AccessGuestDriverProfile() {
return rasterizer->AccessGuestDriverProfile();
}
const VideoCore::GuestDriverProfile& Maxwell3D::AccessGuestDriverProfile() const {
return rasterizer->AccessGuestDriverProfile();
}
} // namespace Tegra::Engines

46
src/video_core/engines/maxwell_3d.h
View File

@ -17,11 +17,9 @@
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/math_util.h"
#include "video_core/engines/const_buffer_engine_interface.h"
#include "video_core/engines/const_buffer_info.h"
#include "video_core/engines/engine_interface.h"
#include "video_core/engines/engine_upload.h"
#include "video_core/engines/shader_type.h"
#include "video_core/gpu.h"
#include "video_core/macro/macro.h"
#include "video_core/textures/texture.h"
@ -49,7 +47,7 @@ namespace Tegra::Engines {
#define MAXWELL3D_REG_INDEX(field_name) \
(offsetof(Tegra::Engines::Maxwell3D::Regs, field_name) / sizeof(u32))
class Maxwell3D final : public ConstBufferEngineInterface, public EngineInterface {
class Maxwell3D final : public EngineInterface {
public:
explicit Maxwell3D(Core::System& system, MemoryManager& memory_manager);
~Maxwell3D();
@ -307,10 +305,6 @@ public:
return (type == Type::SignedNorm) || (type == Type::UnsignedNorm);
}
bool IsConstant() const {
return constant;
}
bool IsValid() const {
return size != Size::Invalid;
}
@ -912,7 +906,11 @@ public:
u32 fill_rectangle;
INSERT_PADDING_WORDS_NOINIT(0x8);
INSERT_PADDING_WORDS_NOINIT(0x2);
u32 conservative_raster_enable;
INSERT_PADDING_WORDS_NOINIT(0x5);
std::array<VertexAttribute, NumVertexAttributes> vertex_attrib_format;
@ -959,7 +957,11 @@ public:
SamplerIndex sampler_index;
INSERT_PADDING_WORDS_NOINIT(0x25);
INSERT_PADDING_WORDS_NOINIT(0x2);
std::array<u32, 8> gp_passthrough_mask;
INSERT_PADDING_WORDS_NOINIT(0x1B);
u32 depth_test_enable;
@ -1152,7 +1154,11 @@ public:
u32 index;
} primitive_restart;
INSERT_PADDING_WORDS_NOINIT(0x5F);
INSERT_PADDING_WORDS_NOINIT(0xE);
u32 provoking_vertex_last;
INSERT_PADDING_WORDS_NOINIT(0x50);
struct {
u32 start_addr_high;
@ -1424,23 +1430,6 @@ public:
void FlushMMEInlineDraw();
u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override;
SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override;
SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const override;
SamplerDescriptor AccessSampler(u32 handle) const override;
u32 GetBoundBuffer() const override {
return regs.tex_cb_index;
}
VideoCore::GuestDriverProfile& AccessGuestDriverProfile() override;
const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override;
bool ShouldExecute() const {
return execute_on;
}
@ -1630,6 +1619,7 @@ ASSERT_REG_POSITION(zeta, 0x3F8);
ASSERT_REG_POSITION(render_area, 0x3FD);
ASSERT_REG_POSITION(clear_flags, 0x43E);
ASSERT_REG_POSITION(fill_rectangle, 0x44F);
ASSERT_REG_POSITION(conservative_raster_enable, 0x452);
ASSERT_REG_POSITION(vertex_attrib_format, 0x458);
ASSERT_REG_POSITION(multisample_sample_locations, 0x478);
ASSERT_REG_POSITION(multisample_coverage_to_color, 0x47E);
@ -1638,6 +1628,7 @@ ASSERT_REG_POSITION(zeta_width, 0x48a);
ASSERT_REG_POSITION(zeta_height, 0x48b);
ASSERT_REG_POSITION(zeta_depth, 0x48c);
ASSERT_REG_POSITION(sampler_index, 0x48D);
ASSERT_REG_POSITION(gp_passthrough_mask, 0x490);
ASSERT_REG_POSITION(depth_test_enable, 0x4B3);
ASSERT_REG_POSITION(independent_blend_enable, 0x4B9);
ASSERT_REG_POSITION(depth_write_enabled, 0x4BA);
@ -1690,6 +1681,7 @@ ASSERT_REG_POSITION(point_coord_replace, 0x581);
ASSERT_REG_POSITION(code_address, 0x582);
ASSERT_REG_POSITION(draw, 0x585);
ASSERT_REG_POSITION(primitive_restart, 0x591);
ASSERT_REG_POSITION(provoking_vertex_last, 0x5A1);
ASSERT_REG_POSITION(index_array, 0x5F2);
ASSERT_REG_POSITION(polygon_offset_clamp, 0x61F);
ASSERT_REG_POSITION(instanced_arrays, 0x620);

3
src/video_core/engines/maxwell_dma.cpp
View File

@ -127,7 +127,8 @@ void MaxwellDMA::CopyBlockLinearToPitch() {
// Optimized path for micro copies.
const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count;
if (dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X) {
if (dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X &&
regs.src_params.height > GOB_SIZE_Y) {
FastCopyBlockLinearToPitch();
return;
}

2298
src/video_core/engines/shader_bytecode.h
View File

File diff suppressed because it is too large Load Diff

158
src/video_core/engines/shader_header.h
View File

@ -1,158 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <optional>
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
namespace Tegra::Shader {
enum class OutputTopology : u32 {
PointList = 1,
LineStrip = 6,
TriangleStrip = 7,
};
enum class PixelImap : u8 {
Unused = 0,
Constant = 1,
Perspective = 2,
ScreenLinear = 3,
};
// Documentation in:
// http://download.nvidia.com/open-gpu-doc/Shader-Program-Header/1/Shader-Program-Header.html
struct Header {
union {
BitField<0, 5, u32> sph_type;
BitField<5, 5, u32> version;
BitField<10, 4, u32> shader_type;
BitField<14, 1, u32> mrt_enable;
BitField<15, 1, u32> kills_pixels;
BitField<16, 1, u32> does_global_store;
BitField<17, 4, u32> sass_version;
BitField<21, 5, u32> reserved;
BitField<26, 1, u32> does_load_or_store;
BitField<27, 1, u32> does_fp64;
BitField<28, 4, u32> stream_out_mask;
} common0;
union {
BitField<0, 24, u32> shader_local_memory_low_size;
BitField<24, 8, u32> per_patch_attribute_count;
} common1;
union {
BitField<0, 24, u32> shader_local_memory_high_size;
BitField<24, 8, u32> threads_per_input_primitive;
} common2;
union {
BitField<0, 24, u32> shader_local_memory_crs_size;
BitField<24, 4, OutputTopology> output_topology;
BitField<28, 4, u32> reserved;
} common3;
union {
BitField<0, 12, u32> max_output_vertices;
BitField<12, 8, u32> store_req_start; // NOTE: not used by geometry shaders.
BitField<20, 4, u32> reserved;
BitField<24, 8, u32> store_req_end; // NOTE: not used by geometry shaders.
} common4;
union {
struct {
INSERT_PADDING_BYTES_NOINIT(3); // ImapSystemValuesA
INSERT_PADDING_BYTES_NOINIT(1); // ImapSystemValuesB
INSERT_PADDING_BYTES_NOINIT(16); // ImapGenericVector[32]
INSERT_PADDING_BYTES_NOINIT(2); // ImapColor
union {
BitField<0, 8, u16> clip_distances;
BitField<8, 1, u16> point_sprite_s;
BitField<9, 1, u16> point_sprite_t;
BitField<10, 1, u16> fog_coordinate;
BitField<12, 1, u16> tessellation_eval_point_u;
BitField<13, 1, u16> tessellation_eval_point_v;
BitField<14, 1, u16> instance_id;
BitField<15, 1, u16> vertex_id;
};
INSERT_PADDING_BYTES_NOINIT(5); // ImapFixedFncTexture[10]
INSERT_PADDING_BYTES_NOINIT(1); // ImapReserved
INSERT_PADDING_BYTES_NOINIT(3); // OmapSystemValuesA
INSERT_PADDING_BYTES_NOINIT(1); // OmapSystemValuesB
INSERT_PADDING_BYTES_NOINIT(16); // OmapGenericVector[32]
INSERT_PADDING_BYTES_NOINIT(2); // OmapColor
INSERT_PADDING_BYTES_NOINIT(2); // OmapSystemValuesC
INSERT_PADDING_BYTES_NOINIT(5); // OmapFixedFncTexture[10]
INSERT_PADDING_BYTES_NOINIT(1); // OmapReserved
} vtg;
struct {
INSERT_PADDING_BYTES_NOINIT(3); // ImapSystemValuesA
INSERT_PADDING_BYTES_NOINIT(1); // ImapSystemValuesB
union {
BitField<0, 2, PixelImap> x;
BitField<2, 2, PixelImap> y;
BitField<4, 2, PixelImap> z;
BitField<6, 2, PixelImap> w;
u8 raw;
} imap_generic_vector[32];
INSERT_PADDING_BYTES_NOINIT(2); // ImapColor
INSERT_PADDING_BYTES_NOINIT(2); // ImapSystemValuesC
INSERT_PADDING_BYTES_NOINIT(10); // ImapFixedFncTexture[10]
INSERT_PADDING_BYTES_NOINIT(2); // ImapReserved
struct {
u32 target;
union {
BitField<0, 1, u32> sample_mask;
BitField<1, 1, u32> depth;
BitField<2, 30, u32> reserved;
};
} omap;
bool IsColorComponentOutputEnabled(u32 render_target, u32 component) const {
const u32 bit = render_target * 4 + component;
return omap.target & (1 << bit);
}
PixelImap GetPixelImap(u32 attribute) const {
const auto get_index = [this, attribute](u32 index) {
return static_cast<PixelImap>(
(imap_generic_vector[attribute].raw >> (index * 2)) & 3);
};
std::optional<PixelImap> result;
for (u32 component = 0; component < 4; ++component) {
const PixelImap index = get_index(component);
if (index == PixelImap::Unused) {
continue;
}
if (result && result != index) {
LOG_CRITICAL(HW_GPU, "Generic attribute conflict in interpolation mode");
}
result = index;
}
return result.value_or(PixelImap::Unused);
}
} ps;
std::array<u32, 0xF> raw;
};
u64 GetLocalMemorySize() const {
return (common1.shader_local_memory_low_size |
(common2.shader_local_memory_high_size << 24));
}
};
static_assert(sizeof(Header) == 0x50, "Incorrect structure size");
} // namespace Tegra::Shader

21
src/video_core/engines/shader_type.h
View File

@ -1,21 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/common_types.h"
namespace Tegra::Engines {
enum class ShaderType : u32 {
Vertex = 0,
TesselationControl = 1,
TesselationEval = 2,
Geometry = 3,
Fragment = 4,
Compute = 5,
};
static constexpr std::size_t MaxShaderTypes = 6;
} // namespace Tegra::Engines

37
src/video_core/guest_driver.cpp
View File

@ -1,37 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <limits>
#include <vector>
#include "common/common_types.h"
#include "video_core/guest_driver.h"
namespace VideoCore {
void GuestDriverProfile::DeduceTextureHandlerSize(std::vector<u32> bound_offsets) {
if (texture_handler_size) {
return;
}
const std::size_t size = bound_offsets.size();
if (size < 2) {
return;
}
std::sort(bound_offsets.begin(), bound_offsets.end(), std::less{});
u32 min_val = std::numeric_limits<u32>::max();
for (std::size_t i = 1; i < size; ++i) {
if (bound_offsets[i] == bound_offsets[i - 1]) {
continue;
}
const u32 new_min = bound_offsets[i] - bound_offsets[i - 1];
min_val = std::min(min_val, new_min);
}
if (min_val > 2) {
return;
}
texture_handler_size = min_texture_handler_size * min_val;
}
} // namespace VideoCore

46
src/video_core/guest_driver.h
View File

@ -1,46 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <optional>
#include <vector>
#include "common/common_types.h"
namespace VideoCore {
/**
* The GuestDriverProfile class is used to learn about the GPU drivers behavior and collect
* information necessary for impossible to avoid HLE methods like shader tracks as they are
* Entscheidungsproblems.
*/
class GuestDriverProfile {
public:
explicit GuestDriverProfile() = default;
explicit GuestDriverProfile(std::optional<u32> texture_handler_size_)
: texture_handler_size{texture_handler_size_} {}
void DeduceTextureHandlerSize(std::vector<u32> bound_offsets);
u32 GetTextureHandlerSize() const {
return texture_handler_size.value_or(default_texture_handler_size);
}
bool IsTextureHandlerSizeKnown() const {
return texture_handler_size.has_value();
}
private:
// Minimum size of texture handler any driver can use.
static constexpr u32 min_texture_handler_size = 4;
// This goes with Vulkan and OpenGL standards but Nvidia GPUs can easily use 4 bytes instead.
// Thus, certain drivers may squish the size.
static constexpr u32 default_texture_handler_size = 8;
std::optional<u32> texture_handler_size = default_texture_handler_size;
};
} // namespace VideoCore

1
src/video_core/memory_manager.cpp
View File

@ -69,7 +69,6 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
} else {
UNREACHABLE_MSG("Unmapping non-existent GPU address=0x{:x}", gpu_addr);
}
const auto submapped_ranges = GetSubmappedRange(gpu_addr, size);
for (const auto& map : submapped_ranges) {

16
src/video_core/rasterizer_interface.h
View File

@ -11,7 +11,6 @@
#include "common/common_types.h"
#include "video_core/engines/fermi_2d.h"
#include "video_core/gpu.h"
#include "video_core/guest_driver.h"
namespace Tegra {
class MemoryManager;
@ -45,7 +44,7 @@ public:
virtual void Clear() = 0;
/// Dispatches a compute shader invocation
virtual void DispatchCompute(GPUVAddr code_addr) = 0;
virtual void DispatchCompute() = 0;
/// Resets the counter of a query
virtual void ResetCounter(QueryType type) = 0;
@ -136,18 +135,5 @@ public:
/// Initialize disk cached resources for the game being emulated
virtual void LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const DiskResourceLoadCallback& callback) {}
/// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver.
[[nodiscard]] GuestDriverProfile& AccessGuestDriverProfile() {
return guest_driver_profile;
}
/// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver.
[[nodiscard]] const GuestDriverProfile& AccessGuestDriverProfile() const {
return guest_driver_profile;
}
private:
GuestDriverProfile guest_driver_profile{};
};
} // namespace VideoCore

2124
src/video_core/renderer_opengl/gl_arb_decompiler.cpp
View File

File diff suppressed because it is too large Load Diff

29
src/video_core/renderer_opengl/gl_arb_decompiler.h
View File

@ -1,29 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <string>
#include <string_view>
#include "common/common_types.h"
namespace Tegra::Engines {
enum class ShaderType : u32;
}
namespace VideoCommon::Shader {
class ShaderIR;
class Registry;
} // namespace VideoCommon::Shader
namespace OpenGL {
class Device;
std::string DecompileAssemblyShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
const VideoCommon::Shader::Registry& registry,
Tegra::Engines::ShaderType stage, std::string_view identifier);
} // namespace OpenGL

84
src/video_core/renderer_opengl/gl_buffer_cache.cpp
View File

@ -2,14 +2,18 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <span>
#include "video_core/buffer_cache/buffer_cache.h"
#include "video_core/renderer_opengl/gl_buffer_cache.h"
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/maxwell_to_gl.h"
namespace OpenGL {
namespace {
using VideoCore::Surface::PixelFormat;
struct BindlessSSBO {
GLuint64EXT address;
GLsizei length;
@ -21,6 +25,25 @@ constexpr std::array PROGRAM_LUT{
GL_VERTEX_PROGRAM_NV, GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV,
GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV,
};
[[nodiscard]] GLenum GetTextureBufferFormat(GLenum gl_format) {
switch (gl_format) {
case GL_RGBA8_SNORM:
return GL_RGBA8;
case GL_R8_SNORM:
return GL_R8;
case GL_RGBA16_SNORM:
return GL_RGBA16;
case GL_R16_SNORM:
return GL_R16;
case GL_RG16_SNORM:
return GL_RG16;
case GL_RG8_SNORM:
return GL_RG8;
default:
return gl_format;
}
}
} // Anonymous namespace
Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
@ -62,6 +85,30 @@ void Buffer::MakeResident(GLenum access) noexcept {
glMakeNamedBufferResidentNV(buffer.handle, access);
}
GLuint Buffer::View(u32 offset, u32 size, PixelFormat format) {
const auto it{std::ranges::find_if(views, [offset, size, format](const BufferView& view) {
return offset == view.offset && size == view.size && format == view.format;
})};
if (it != views.end()) {
return it->texture.handle;
}
OGLTexture texture;
texture.Create(GL_TEXTURE_BUFFER);
const GLenum gl_format{MaxwellToGL::GetFormatTuple(format).internal_format};
const GLenum texture_format{GetTextureBufferFormat(gl_format)};
if (texture_format != gl_format) {
LOG_WARNING(Render_OpenGL, "Emulating SNORM texture buffer with UNORM.");
}
glTextureBufferRange(texture.handle, texture_format, buffer.handle, offset, size);
views.push_back({
.offset = offset,
.size = size,
.format = format,
.texture = std::move(texture),
});
return views.back().texture.handle;
}
BufferCacheRuntime::BufferCacheRuntime(const Device& device_)
: device{device_}, has_fast_buffer_sub_data{device.HasFastBufferSubData()},
use_assembly_shaders{device.UseAssemblyShaders()},
@ -144,7 +191,7 @@ void BufferCacheRuntime::BindUniformBuffer(size_t stage, u32 binding_index, Buff
glBindBufferRangeNV(PABO_LUT[stage], binding_index, handle, 0,
static_cast<GLsizeiptr>(size));
} else {
const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer;
const GLuint base_binding = graphics_base_uniform_bindings[stage];
const GLuint binding = base_binding + binding_index;
glBindBufferRange(GL_UNIFORM_BUFFER, binding, buffer.Handle(),
static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
@ -171,7 +218,12 @@ void BufferCacheRuntime::BindComputeUniformBuffer(u32 binding_index, Buffer& buf
void BufferCacheRuntime::BindStorageBuffer(size_t stage, u32 binding_index, Buffer& buffer,
u32 offset, u32 size, bool is_written) {
if (use_assembly_shaders) {
if (use_storage_buffers) {
const GLuint base_binding = graphics_base_storage_bindings[stage];
const GLuint binding = base_binding + binding_index;
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, buffer.Handle(),
static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
} else {
const BindlessSSBO ssbo{
.address = buffer.HostGpuAddr() + offset,
.length = static_cast<GLsizei>(size),
@ -180,17 +232,19 @@ void BufferCacheRuntime::BindStorageBuffer(size_t stage, u32 binding_index, Buff
buffer.MakeResident(is_written ? GL_READ_WRITE : GL_READ_ONLY);
glProgramLocalParametersI4uivNV(PROGRAM_LUT[stage], binding_index, 1,
reinterpret_cast<const GLuint*>(&ssbo));
} else {
const GLuint base_binding = device.GetBaseBindings(stage).shader_storage_buffer;
const GLuint binding = base_binding + binding_index;
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, buffer.Handle(),
static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
}
}
void BufferCacheRuntime::BindComputeStorageBuffer(u32 binding_index, Buffer& buffer, u32 offset,
u32 size, bool is_written) {
if (use_assembly_shaders) {
if (use_storage_buffers) {
if (size != 0) {
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, buffer.Handle(),
static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
} else {
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, 0, 0, 0);
}
} else {
const BindlessSSBO ssbo{
.address = buffer.HostGpuAddr() + offset,
.length = static_cast<GLsizei>(size),
@ -199,11 +253,6 @@ void BufferCacheRuntime::BindComputeStorageBuffer(u32 binding_index, Buffer& buf
buffer.MakeResident(is_written ? GL_READ_WRITE : GL_READ_ONLY);
glProgramLocalParametersI4uivNV(GL_COMPUTE_PROGRAM_NV, binding_index, 1,
reinterpret_cast<const GLuint*>(&ssbo));
} else if (size == 0) {
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, 0, 0, 0);
} else {
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, buffer.Handle(),
static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
}
}
@ -213,4 +262,13 @@ void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, Buffer& buffer,
static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
}
void BufferCacheRuntime::BindTextureBuffer(Buffer& buffer, u32 offset, u32 size,
PixelFormat format) {
*texture_handles++ = buffer.View(offset, size, format);
}
void BufferCacheRuntime::BindImageBuffer(Buffer& buffer, u32 offset, u32 size, PixelFormat format) {
*image_handles++ = buffer.View(offset, size, format);
}
} // namespace OpenGL

57
src/video_core/renderer_opengl/gl_buffer_cache.h
View File

@ -32,6 +32,8 @@ public:
void MakeResident(GLenum access) noexcept;
[[nodiscard]] GLuint View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format);
[[nodiscard]] GLuint64EXT HostGpuAddr() const noexcept {
return address;
}
@ -41,9 +43,17 @@ public:
}
private:
struct BufferView {
u32 offset;
u32 size;
VideoCore::Surface::PixelFormat format;
OGLTexture texture;
};
GLuint64EXT address = 0;
OGLBuffer buffer;
GLenum current_residency_access = GL_NONE;
std::vector<BufferView> views;
};
class BufferCacheRuntime {
@ -75,17 +85,21 @@ public:
void BindTransformFeedbackBuffer(u32 index, Buffer& buffer, u32 offset, u32 size);
void BindTextureBuffer(Buffer& buffer, u32 offset, u32 size,
VideoCore::Surface::PixelFormat format);
void BindImageBuffer(Buffer& buffer, u32 offset, u32 size,
VideoCore::Surface::PixelFormat format);
void BindFastUniformBuffer(size_t stage, u32 binding_index, u32 size) {
const GLuint handle = fast_uniforms[stage][binding_index].handle;
const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size);
if (use_assembly_shaders) {
const GLuint handle = fast_uniforms[stage][binding_index].handle;
const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size);
glBindBufferRangeNV(PABO_LUT[stage], binding_index, handle, 0, gl_size);
} else {
const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer;
const GLuint base_binding = graphics_base_uniform_bindings[stage];
const GLuint binding = base_binding + binding_index;
glBindBufferRange(GL_UNIFORM_BUFFER, binding,
fast_uniforms[stage][binding_index].handle, 0,
static_cast<GLsizeiptr>(size));
glBindBufferRange(GL_UNIFORM_BUFFER, binding, handle, 0, gl_size);
}
}
@ -103,7 +117,7 @@ public:
std::span<u8> BindMappedUniformBuffer(size_t stage, u32 binding_index, u32 size) noexcept {
const auto [mapped_span, offset] = stream_buffer->Request(static_cast<size_t>(size));
const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer;
const GLuint base_binding = graphics_base_uniform_bindings[stage];
const GLuint binding = base_binding + binding_index;
glBindBufferRange(GL_UNIFORM_BUFFER, binding, stream_buffer->Handle(),
static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
@ -118,6 +132,27 @@ public:
return has_fast_buffer_sub_data;
}
[[nodiscard]] bool SupportsNonZeroUniformOffset() const noexcept {
return !use_assembly_shaders;
}
void SetBaseUniformBindings(const std::array<GLuint, 5>& bindings) {
graphics_base_uniform_bindings = bindings;
}
void SetBaseStorageBindings(const std::array<GLuint, 5>& bindings) {
graphics_base_storage_bindings = bindings;
}
void SetImagePointers(GLuint* texture_handles_, GLuint* image_handles_) {
texture_handles = texture_handles_;
image_handles = image_handles_;
}
void SetEnableStorageBuffers(bool use_storage_buffers_) {
use_storage_buffers = use_storage_buffers_;
}
private:
static constexpr std::array PABO_LUT{
GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV,
@ -131,8 +166,15 @@ private:
bool use_assembly_shaders = false;
bool has_unified_vertex_buffers = false;
bool use_storage_buffers = false;
u32 max_attributes = 0;
std::array<GLuint, 5> graphics_base_uniform_bindings{};
std::array<GLuint, 5> graphics_base_storage_bindings{};
GLuint* texture_handles = nullptr;
GLuint* image_handles = nullptr;
std::optional<StreamBuffer> stream_buffer;
std::array<std::array<OGLBuffer, VideoCommon::NUM_GRAPHICS_UNIFORM_BUFFERS>,
@ -156,6 +198,7 @@ struct BufferCacheParams {
static constexpr bool NEEDS_BIND_UNIFORM_INDEX = true;
static constexpr bool NEEDS_BIND_STORAGE_INDEX = true;
static constexpr bool USE_MEMORY_MAPS = false;
static constexpr bool SEPARATE_IMAGE_BUFFER_BINDINGS = true;
};
using BufferCache = VideoCommon::BufferCache<BufferCacheParams>;

209
src/video_core/renderer_opengl/gl_compute_pipeline.cpp Normal file
View File

@ -0,0 +1,209 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include "common/cityhash.h"
#include "common/settings.h" // for enum class Settings::ShaderBackend
#include "video_core/renderer_opengl/gl_compute_pipeline.h"
#include "video_core/renderer_opengl/gl_shader_manager.h"
#include "video_core/renderer_opengl/gl_shader_util.h"
namespace OpenGL {
using Shader::ImageBufferDescriptor;
using Tegra::Texture::TexturePair;
using VideoCommon::ImageId;
constexpr u32 MAX_TEXTURES = 64;
constexpr u32 MAX_IMAGES = 16;
template <typename Range>
u32 AccumulateCount(const Range& range) {
u32 num{};
for (const auto& desc : range) {
num += desc.count;
}
return num;
}
size_t ComputePipelineKey::Hash() const noexcept {
return static_cast<size_t>(
Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this));
}
bool ComputePipelineKey::operator==(const ComputePipelineKey& rhs) const noexcept {
return std::memcmp(this, &rhs, sizeof *this) == 0;
}
ComputePipeline::ComputePipeline(const Device& device, TextureCache& texture_cache_,
BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_,
Tegra::Engines::KeplerCompute& kepler_compute_,
ProgramManager& program_manager_, const Shader::Info& info_,
std::string code, std::vector<u32> code_v)
: texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, gpu_memory{gpu_memory_},
kepler_compute{kepler_compute_}, program_manager{program_manager_}, info{info_} {
switch (device.GetShaderBackend()) {
case Settings::ShaderBackend::GLSL:
source_program = CreateProgram(code, GL_COMPUTE_SHADER);
break;
case Settings::ShaderBackend::GLASM:
assembly_program = CompileProgram(code, GL_COMPUTE_PROGRAM_NV);
break;
case Settings::ShaderBackend::SPIRV:
source_program = CreateProgram(code_v, GL_COMPUTE_SHADER);
break;
}
std::copy_n(info.constant_buffer_used_sizes.begin(), uniform_buffer_sizes.size(),
uniform_buffer_sizes.begin());
num_texture_buffers = AccumulateCount(info.texture_buffer_descriptors);
num_image_buffers = AccumulateCount(info.image_buffer_descriptors);
const u32 num_textures{num_texture_buffers + AccumulateCount(info.texture_descriptors)};
ASSERT(num_textures <= MAX_TEXTURES);
const u32 num_images{num_image_buffers + AccumulateCount(info.image_descriptors)};
ASSERT(num_images <= MAX_IMAGES);
const bool is_glasm{assembly_program.handle != 0};
const u32 num_storage_buffers{AccumulateCount(info.storage_buffers_descriptors)};
use_storage_buffers =
!is_glasm || num_storage_buffers < device.GetMaxGLASMStorageBufferBlocks();
writes_global_memory = !use_storage_buffers &&
std::ranges::any_of(info.storage_buffers_descriptors,
[](const auto& desc) { return desc.is_written; });
}
void ComputePipeline::Configure() {
buffer_cache.SetComputeUniformBufferState(info.constant_buffer_mask, &uniform_buffer_sizes);
buffer_cache.UnbindComputeStorageBuffers();
size_t ssbo_index{};
for (const auto& desc : info.storage_buffers_descriptors) {
ASSERT(desc.count == 1);
buffer_cache.BindComputeStorageBuffer(ssbo_index, desc.cbuf_index, desc.cbuf_offset,
desc.is_written);
++ssbo_index;
}
texture_cache.SynchronizeComputeDescriptors();
std::array<ImageViewId, MAX_TEXTURES + MAX_IMAGES> image_view_ids;
boost::container::static_vector<u32, MAX_TEXTURES + MAX_IMAGES> image_view_indices;
std::array<GLuint, MAX_TEXTURES> samplers;
std::array<GLuint, MAX_TEXTURES> textures;
std::array<GLuint, MAX_IMAGES> images;
GLsizei sampler_binding{};
GLsizei texture_binding{};
GLsizei image_binding{};
const auto& qmd{kepler_compute.launch_description};
const auto& cbufs{qmd.const_buffer_config};
const bool via_header_index{qmd.linked_tsc != 0};
const auto read_handle{[&](const auto& desc, u32 index) {
ASSERT(((qmd.const_buffer_enable_mask >> desc.cbuf_index) & 1) != 0);
const u32 index_offset{index << desc.size_shift};
const u32 offset{desc.cbuf_offset + index_offset};
const GPUVAddr addr{cbufs[desc.cbuf_index].Address() + offset};
if constexpr (std::is_same_v<decltype(desc), const Shader::TextureDescriptor&> ||
std::is_same_v<decltype(desc), const Shader::TextureBufferDescriptor&>) {
if (desc.has_secondary) {
ASSERT(((qmd.const_buffer_enable_mask >> desc.secondary_cbuf_index) & 1) != 0);
const u32 secondary_offset{desc.secondary_cbuf_offset + index_offset};
const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].Address() +
secondary_offset};
const u32 lhs_raw{gpu_memory.Read<u32>(addr)};
const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)};
return TexturePair(lhs_raw | rhs_raw, via_header_index);
}
}
return TexturePair(gpu_memory.Read<u32>(addr), via_header_index);
}};
const auto add_image{[&](const auto& desc) {
for (u32 index = 0; index < desc.count; ++index) {
const auto handle{read_handle(desc, index)};
image_view_indices.push_back(handle.first);
}
}};
for (const auto& desc : info.texture_buffer_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
const auto handle{read_handle(desc, index)};
image_view_indices.push_back(handle.first);
samplers[sampler_binding++] = 0;
}
}
std::ranges::for_each(info.image_buffer_descriptors, add_image);
for (const auto& desc : info.texture_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
const auto handle{read_handle(desc, index)};
image_view_indices.push_back(handle.first);
Sampler* const sampler = texture_cache.GetComputeSampler(handle.second);
samplers[sampler_binding++] = sampler->Handle();
}
}
std::ranges::for_each(info.image_descriptors, add_image);
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
texture_cache.FillComputeImageViews(indices_span, image_view_ids);
if (assembly_program.handle != 0) {
program_manager.BindComputeAssemblyProgram(assembly_program.handle);
} else {
program_manager.BindComputeProgram(source_program.handle);
}
buffer_cache.UnbindComputeTextureBuffers();
size_t texbuf_index{};
const auto add_buffer{[&](const auto& desc) {
constexpr bool is_image = std::is_same_v<decltype(desc), const ImageBufferDescriptor&>;
for (u32 i = 0; i < desc.count; ++i) {
bool is_written{false};
if constexpr (is_image) {
is_written = desc.is_written;
}
ImageView& image_view{texture_cache.GetImageView(image_view_ids[texbuf_index])};
buffer_cache.BindComputeTextureBuffer(texbuf_index, image_view.GpuAddr(),
image_view.BufferSize(), image_view.format,
is_written, is_image);
++texbuf_index;
}
}};
std::ranges::for_each(info.texture_buffer_descriptors, add_buffer);
std::ranges::for_each(info.image_buffer_descriptors, add_buffer);
buffer_cache.UpdateComputeBuffers();
buffer_cache.runtime.SetEnableStorageBuffers(use_storage_buffers);
buffer_cache.runtime.SetImagePointers(textures.data(), images.data());
buffer_cache.BindHostComputeBuffers();
const ImageId* views_it{image_view_ids.data() + num_texture_buffers + num_image_buffers};
texture_binding += num_texture_buffers;
image_binding += num_image_buffers;
for (const auto& desc : info.texture_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
ImageView& image_view{texture_cache.GetImageView(*(views_it++))};
textures[texture_binding++] = image_view.Handle(desc.type);
}
}
for (const auto& desc : info.image_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
ImageView& image_view{texture_cache.GetImageView(*(views_it++))};
if (desc.is_written) {
texture_cache.MarkModification(image_view.image_id);
}
images[image_binding++] = image_view.StorageView(desc.type, desc.format);
}
}
if (texture_binding != 0) {
ASSERT(texture_binding == sampler_binding);
glBindTextures(0, texture_binding, textures.data());
glBindSamplers(0, sampler_binding, samplers.data());
}
if (image_binding != 0) {
glBindImageTextures(0, image_binding, images.data());
}
}
} // namespace OpenGL

93
src/video_core/renderer_opengl/gl_compute_pipeline.h Normal file
View File

@ -0,0 +1,93 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <type_traits>
#include <utility>
#include "common/common_types.h"
#include "shader_recompiler/shader_info.h"
#include "video_core/renderer_opengl/gl_buffer_cache.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_texture_cache.h"
namespace Tegra {
class MemoryManager;
}
namespace Tegra::Engines {
class KeplerCompute;
}
namespace Shader {
struct Info;
}
namespace OpenGL {
class Device;
class ProgramManager;
struct ComputePipelineKey {
u64 unique_hash;
u32 shared_memory_size;
std::array<u32, 3> workgroup_size;
size_t Hash() const noexcept;
bool operator==(const ComputePipelineKey&) const noexcept;
bool operator!=(const ComputePipelineKey& rhs) const noexcept {
return !operator==(rhs);
}
};
static_assert(std::has_unique_object_representations_v<ComputePipelineKey>);
static_assert(std::is_trivially_copyable_v<ComputePipelineKey>);
static_assert(std::is_trivially_constructible_v<ComputePipelineKey>);
class ComputePipeline {
public:
explicit ComputePipeline(const Device& device, TextureCache& texture_cache_,
BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_,
Tegra::Engines::KeplerCompute& kepler_compute_,
ProgramManager& program_manager_, const Shader::Info& info_,
std::string code, std::vector<u32> code_v);
void Configure();
[[nodiscard]] bool WritesGlobalMemory() const noexcept {
return writes_global_memory;
}
private:
TextureCache& texture_cache;
BufferCache& buffer_cache;
Tegra::MemoryManager& gpu_memory;
Tegra::Engines::KeplerCompute& kepler_compute;
ProgramManager& program_manager;
Shader::Info info;
OGLProgram source_program;
OGLAssemblyProgram assembly_program;
VideoCommon::ComputeUniformBufferSizes uniform_buffer_sizes{};
u32 num_texture_buffers{};
u32 num_image_buffers{};
bool use_storage_buffers{};
bool writes_global_memory{};
};
} // namespace OpenGL
namespace std {
template <>
struct hash<OpenGL::ComputePipelineKey> {
size_t operator()(const OpenGL::ComputePipelineKey& k) const noexcept {
return k.Hash();
}
};
} // namespace std

146
src/video_core/renderer_opengl/gl_device.cpp
View File

@ -17,39 +17,17 @@
#include "common/logging/log.h"
#include "common/scope_exit.h"
#include "common/settings.h"
#include "shader_recompiler/stage.h"
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
namespace OpenGL {
namespace {
// One uniform block is reserved for emulation purposes
constexpr u32 ReservedUniformBlocks = 1;
constexpr u32 NumStages = 5;
constexpr std::array LIMIT_UBOS = {
GL_MAX_VERTEX_UNIFORM_BLOCKS, GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS,
GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS, GL_MAX_GEOMETRY_UNIFORM_BLOCKS,
GL_MAX_FRAGMENT_UNIFORM_BLOCKS, GL_MAX_COMPUTE_UNIFORM_BLOCKS,
};
constexpr std::array LIMIT_SSBOS = {
GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, GL_MAX_TESS_CONTROL_SHADER_STORAGE_BLOCKS,
GL_MAX_TESS_EVALUATION_SHADER_STORAGE_BLOCKS, GL_MAX_GEOMETRY_SHADER_STORAGE_BLOCKS,
GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS, GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS,
};
constexpr std::array LIMIT_SAMPLERS = {
GL_MAX_VERTEX_TEXTURE_IMAGE_UNITS,
GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS,
GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS,
GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS,
GL_MAX_TEXTURE_IMAGE_UNITS,
GL_MAX_COMPUTE_TEXTURE_IMAGE_UNITS,
};
constexpr std::array LIMIT_IMAGES = {
GL_MAX_VERTEX_IMAGE_UNIFORMS, GL_MAX_TESS_CONTROL_IMAGE_UNIFORMS,
GL_MAX_TESS_EVALUATION_IMAGE_UNIFORMS, GL_MAX_GEOMETRY_IMAGE_UNIFORMS,
GL_MAX_FRAGMENT_IMAGE_UNIFORMS, GL_MAX_COMPUTE_IMAGE_UNIFORMS,
};
template <typename T>
T GetInteger(GLenum pname) {
@ -82,81 +60,18 @@ bool HasExtension(std::span<const std::string_view> extensions, std::string_view
return std::ranges::find(extensions, extension) != extensions.end();
}
u32 Extract(u32& base, u32& num, u32 amount, std::optional<GLenum> limit = {}) {
ASSERT(num >= amount);
if (limit) {
amount = std::min(amount, GetInteger<u32>(*limit));
}
num -= amount;
return std::exchange(base, base + amount);
}
std::array<u32, Tegra::Engines::MaxShaderTypes> BuildMaxUniformBuffers() noexcept {
std::array<u32, Tegra::Engines::MaxShaderTypes> max;
std::ranges::transform(LIMIT_UBOS, max.begin(),
[](GLenum pname) { return GetInteger<u32>(pname); });
std::array<u32, Shader::MaxStageTypes> BuildMaxUniformBuffers() noexcept {
std::array<u32, Shader::MaxStageTypes> max;
std::ranges::transform(LIMIT_UBOS, max.begin(), &GetInteger<u32>);
return max;
}
std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindings() noexcept {
std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> bindings;
static constexpr std::array<std::size_t, 5> stage_swizzle{0, 1, 2, 3, 4};
const u32 total_ubos = GetInteger<u32>(GL_MAX_UNIFORM_BUFFER_BINDINGS);
const u32 total_ssbos = GetInteger<u32>(GL_MAX_SHADER_STORAGE_BUFFER_BINDINGS);
const u32 total_samplers = GetInteger<u32>(GL_MAX_COMBINED_TEXTURE_IMAGE_UNITS);
u32 num_ubos = total_ubos - ReservedUniformBlocks;
u32 num_ssbos = total_ssbos;
u32 num_samplers = total_samplers;
u32 base_ubo = ReservedUniformBlocks;
u32 base_ssbo = 0;
u32 base_samplers = 0;
for (std::size_t i = 0; i < NumStages; ++i) {
const std::size_t stage = stage_swizzle[i];
bindings[stage] = {
Extract(base_ubo, num_ubos, total_ubos / NumStages, LIMIT_UBOS[stage]),
Extract(base_ssbo, num_ssbos, total_ssbos / NumStages, LIMIT_SSBOS[stage]),
Extract(base_samplers, num_samplers, total_samplers / NumStages,
LIMIT_SAMPLERS[stage])};
}
u32 num_images = GetInteger<u32>(GL_MAX_IMAGE_UNITS);
u32 base_images = 0;
// GL_MAX_IMAGE_UNITS is guaranteed by the spec to have a minimum value of 8.
// Due to the limitation of GL_MAX_IMAGE_UNITS, reserve at least 4 image bindings on the
// fragment stage, and at least 1 for the rest of the stages.
// So far games are observed to use 1 image binding on vertex and 4 on fragment stages.
// Reserve at least 4 image bindings on the fragment stage.
bindings[4].image =
Extract(base_images, num_images, std::max(4U, num_images / NumStages), LIMIT_IMAGES[4]);
// This is guaranteed to be at least 1.
const u32 total_extracted_images = num_images / (NumStages - 1);
// Reserve the other image bindings.
for (std::size_t i = 0; i < NumStages; ++i) {
const std::size_t stage = stage_swizzle[i];
if (stage == 4) {
continue;
}
bindings[stage].image =
Extract(base_images, num_images, total_extracted_images, LIMIT_IMAGES[stage]);
}
// Compute doesn't care about any of this.
bindings[5] = {0, 0, 0, 0};
return bindings;
}
bool IsASTCSupported() {
static constexpr std::array targets = {GL_TEXTURE_2D, GL_TEXTURE_2D_ARRAY};
static constexpr std::array formats = {
static constexpr std::array targets{
GL_TEXTURE_2D,
GL_TEXTURE_2D_ARRAY,
};
static constexpr std::array formats{
GL_COMPRESSED_RGBA_ASTC_4x4_KHR, GL_COMPRESSED_RGBA_ASTC_5x4_KHR,
GL_COMPRESSED_RGBA_ASTC_5x5_KHR, GL_COMPRESSED_RGBA_ASTC_6x5_KHR,
GL_COMPRESSED_RGBA_ASTC_6x6_KHR, GL_COMPRESSED_RGBA_ASTC_8x5_KHR,
@ -172,11 +87,10 @@ bool IsASTCSupported() {
GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR, GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR,
GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x10_KHR, GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR,
};
static constexpr std::array required_support = {
static constexpr std::array required_support{
GL_VERTEX_TEXTURE, GL_TESS_CONTROL_TEXTURE, GL_TESS_EVALUATION_TEXTURE,
GL_GEOMETRY_TEXTURE, GL_FRAGMENT_TEXTURE, GL_COMPUTE_TEXTURE,
};
for (const GLenum target : targets) {
for (const GLenum format : formats) {
for (const GLenum support : required_support) {
@ -223,14 +137,13 @@ Device::Device() {
"Beta driver 443.24 is known to have issues. There might be performance issues.");
disable_fast_buffer_sub_data = true;
}
max_uniform_buffers = BuildMaxUniformBuffers();
base_bindings = BuildBaseBindings();
uniform_buffer_alignment = GetInteger<size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT);
shader_storage_alignment = GetInteger<size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT);
max_vertex_attributes = GetInteger<u32>(GL_MAX_VERTEX_ATTRIBS);
max_varyings = GetInteger<u32>(GL_MAX_VARYING_VECTORS);
max_compute_shared_memory_size = GetInteger<u32>(GL_MAX_COMPUTE_SHARED_MEMORY_SIZE);
max_glasm_storage_buffer_blocks = GetInteger<u32>(GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS);
has_warp_intrinsics = GLAD_GL_NV_gpu_shader5 && GLAD_GL_NV_shader_thread_group &&
GLAD_GL_NV_shader_thread_shuffle;
has_shader_ballot = GLAD_GL_ARB_shader_ballot;
@ -243,18 +156,30 @@ Device::Device() {
has_precise_bug = TestPreciseBug();
has_broken_texture_view_formats = is_amd || (!is_linux && is_intel);
has_nv_viewport_array2 = GLAD_GL_NV_viewport_array2;
has_derivative_control = GLAD_GL_ARB_derivative_control;
has_vertex_buffer_unified_memory = GLAD_GL_NV_vertex_buffer_unified_memory;
has_debugging_tool_attached = IsDebugToolAttached(extensions);
has_depth_buffer_float = HasExtension(extensions, "GL_NV_depth_buffer_float");
has_geometry_shader_passthrough = GLAD_GL_NV_geometry_shader_passthrough;
has_nv_gpu_shader_5 = GLAD_GL_NV_gpu_shader5;
has_shader_int64 = HasExtension(extensions, "GL_ARB_gpu_shader_int64");
has_amd_shader_half_float = GLAD_GL_AMD_gpu_shader_half_float;
has_sparse_texture_2 = GLAD_GL_ARB_sparse_texture2;
warp_size_potentially_larger_than_guest = !is_nvidia && !is_intel;
need_fastmath_off = is_nvidia;
// At the moment of writing this, only Nvidia's driver optimizes BufferSubData on exclusive
// uniform buffers as "push constants"
has_fast_buffer_sub_data = is_nvidia && !disable_fast_buffer_sub_data;
use_assembly_shaders = Settings::values.use_assembly_shaders.GetValue() &&
shader_backend = Settings::values.shader_backend.GetValue();
use_assembly_shaders = shader_backend == Settings::ShaderBackend::GLASM &&
GLAD_GL_NV_gpu_program5 && GLAD_GL_NV_compute_program5 &&
GLAD_GL_NV_transform_feedback && GLAD_GL_NV_transform_feedback2;
if (shader_backend == Settings::ShaderBackend::GLASM && !use_assembly_shaders) {
LOG_ERROR(Render_OpenGL, "Assembly shaders enabled but not supported");
shader_backend = Settings::ShaderBackend::GLSL;
}
// Blocks AMD and Intel OpenGL drivers on Windows from using asynchronous shader compilation.
use_asynchronous_shaders = Settings::values.use_asynchronous_shaders.GetValue() &&
!(is_amd || (is_intel && !is_linux));
@ -265,11 +190,6 @@ Device::Device() {
LOG_INFO(Render_OpenGL, "Renderer_PreciseBug: {}", has_precise_bug);
LOG_INFO(Render_OpenGL, "Renderer_BrokenTextureViewFormats: {}",
has_broken_texture_view_formats);
if (Settings::values.use_assembly_shaders.GetValue() && !use_assembly_shaders) {
LOG_ERROR(Render_OpenGL, "Assembly shaders enabled but not supported");
}
if (Settings::values.use_asynchronous_shaders.GetValue() && !use_asynchronous_shaders) {
LOG_WARNING(Render_OpenGL, "Asynchronous shader compilation enabled but not supported");
}
@ -325,22 +245,6 @@ std::string Device::GetVendorName() const {
return vendor_name;
}
Device::Device(std::nullptr_t) {
max_uniform_buffers.fill(std::numeric_limits<u32>::max());
uniform_buffer_alignment = 4;
shader_storage_alignment = 4;
max_vertex_attributes = 16;
max_varyings = 15;
max_compute_shared_memory_size = 0x10000;
has_warp_intrinsics = true;
has_shader_ballot = true;
has_vertex_viewport_layer = true;
has_image_load_formatted = true;
has_texture_shadow_lod = true;
has_variable_aoffi = true;
has_depth_buffer_float = true;
}
bool Device::TestVariableAoffi() {
return TestProgram(R"(#version 430 core
// This is a unit test, please ignore me on apitrace bug reports.

84
src/video_core/renderer_opengl/gl_device.h
View File

@ -6,34 +6,22 @@
#include <cstddef>
#include "common/common_types.h"
#include "video_core/engines/shader_type.h"
#include "shader_recompiler/stage.h"
namespace Settings {
enum class ShaderBackend : u32;
};
namespace OpenGL {
class Device {
public:
struct BaseBindings {
u32 uniform_buffer{};
u32 shader_storage_buffer{};
u32 sampler{};
u32 image{};
};
explicit Device();
explicit Device(std::nullptr_t);
[[nodiscard]] std::string GetVendorName() const;
u32 GetMaxUniformBuffers(Tegra::Engines::ShaderType shader_type) const noexcept {
return max_uniform_buffers[static_cast<std::size_t>(shader_type)];
}
const BaseBindings& GetBaseBindings(std::size_t stage_index) const noexcept {
return base_bindings[stage_index];
}
const BaseBindings& GetBaseBindings(Tegra::Engines::ShaderType shader_type) const noexcept {
return GetBaseBindings(static_cast<std::size_t>(shader_type));
u32 GetMaxUniformBuffers(Shader::Stage stage) const noexcept {
return max_uniform_buffers[static_cast<size_t>(stage)];
}
size_t GetUniformBufferAlignment() const {
@ -56,6 +44,10 @@ public:
return max_compute_shared_memory_size;
}
u32 GetMaxGLASMStorageBufferBlocks() const {
return max_glasm_storage_buffer_blocks;
}
bool HasWarpIntrinsics() const {
return has_warp_intrinsics;
}
@ -108,6 +100,10 @@ public:
return has_nv_viewport_array2;
}
bool HasDerivativeControl() const {
return has_derivative_control;
}
bool HasDebuggingToolAttached() const {
return has_debugging_tool_attached;
}
@ -128,18 +124,52 @@ public:
return has_depth_buffer_float;
}
bool HasGeometryShaderPassthrough() const {
return has_geometry_shader_passthrough;
}
bool HasNvGpuShader5() const {
return has_nv_gpu_shader_5;
}
bool HasShaderInt64() const {
return has_shader_int64;
}
bool HasAmdShaderHalfFloat() const {
return has_amd_shader_half_float;
}
bool HasSparseTexture2() const {
return has_sparse_texture_2;
}
bool IsWarpSizePotentiallyLargerThanGuest() const {
return warp_size_potentially_larger_than_guest;
}
bool NeedsFastmathOff() const {
return need_fastmath_off;
}
Settings::ShaderBackend GetShaderBackend() const {
return shader_backend;
}
private:
static bool TestVariableAoffi();
static bool TestPreciseBug();
std::string vendor_name;
std::array<u32, Tegra::Engines::MaxShaderTypes> max_uniform_buffers{};
std::array<BaseBindings, Tegra::Engines::MaxShaderTypes> base_bindings{};
std::array<u32, Shader::MaxStageTypes> max_uniform_buffers{};
size_t uniform_buffer_alignment{};
size_t shader_storage_alignment{};
u32 max_vertex_attributes{};
u32 max_varyings{};
u32 max_compute_shared_memory_size{};
u32 max_glasm_storage_buffer_blocks{};
Settings::ShaderBackend shader_backend{};
bool has_warp_intrinsics{};
bool has_shader_ballot{};
bool has_vertex_viewport_layer{};
@ -153,11 +183,21 @@ private:
bool has_broken_texture_view_formats{};
bool has_fast_buffer_sub_data{};
bool has_nv_viewport_array2{};
bool has_derivative_control{};
bool has_debugging_tool_attached{};
bool use_assembly_shaders{};
bool use_asynchronous_shaders{};
bool use_driver_cache{};
bool has_depth_buffer_float{};
bool has_geometry_shader_passthrough{};
bool has_nv_gpu_shader_5{};
bool has_shader_int64{};
bool has_amd_shader_half_float{};
bool has_sparse_texture_2{};
bool warp_size_potentially_larger_than_guest{};
bool need_fastmath_off{};
std::string vendor_name;
};
} // namespace OpenGL

572
src/video_core/renderer_opengl/gl_graphics_pipeline.cpp Normal file
View File

@ -0,0 +1,572 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <array>
#include <string>
#include <vector>
#include "common/settings.h" // for enum class Settings::ShaderBackend
#include "common/thread_worker.h"
#include "shader_recompiler/shader_info.h"
#include "video_core/renderer_opengl/gl_graphics_pipeline.h"
#include "video_core/renderer_opengl/gl_shader_manager.h"
#include "video_core/renderer_opengl/gl_shader_util.h"
#include "video_core/renderer_opengl/gl_state_tracker.h"
#include "video_core/shader_notify.h"
#include "video_core/texture_cache/texture_cache.h"
#if defined(_MSC_VER) && defined(NDEBUG)
#define LAMBDA_FORCEINLINE [[msvc::forceinline]]
#else
#define LAMBDA_FORCEINLINE
#endif
namespace OpenGL {
namespace {
using Shader::ImageBufferDescriptor;
using Shader::ImageDescriptor;
using Shader::TextureBufferDescriptor;
using Shader::TextureDescriptor;
using Tegra::Texture::TexturePair;
using VideoCommon::ImageId;
constexpr u32 MAX_TEXTURES = 64;
constexpr u32 MAX_IMAGES = 8;
template <typename Range>
u32 AccumulateCount(const Range& range) {
u32 num{};
for (const auto& desc : range) {
num += desc.count;
}
return num;
}
GLenum Stage(size_t stage_index) {
switch (stage_index) {
case 0:
return GL_VERTEX_SHADER;
case 1:
return GL_TESS_CONTROL_SHADER;
case 2:
return GL_TESS_EVALUATION_SHADER;
case 3:
return GL_GEOMETRY_SHADER;
case 4:
return GL_FRAGMENT_SHADER;
}
UNREACHABLE_MSG("{}", stage_index);
return GL_NONE;
}
GLenum AssemblyStage(size_t stage_index) {
switch (stage_index) {
case 0:
return GL_VERTEX_PROGRAM_NV;
case 1:
return GL_TESS_CONTROL_PROGRAM_NV;
case 2:
return GL_TESS_EVALUATION_PROGRAM_NV;
case 3:
return GL_GEOMETRY_PROGRAM_NV;
case 4:
return GL_FRAGMENT_PROGRAM_NV;
}
UNREACHABLE_MSG("{}", stage_index);
return GL_NONE;
}
/// Translates hardware transform feedback indices
/// @param location Hardware location
/// @return Pair of ARB_transform_feedback3 token stream first and third arguments
/// @note Read https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_transform_feedback3.txt
std::pair<GLint, GLint> TransformFeedbackEnum(u8 location) {
const u8 index = location / 4;
if (index >= 8 && index <= 39) {
return {GL_GENERIC_ATTRIB_NV, index - 8};
}
if (index >= 48 && index <= 55) {
return {GL_TEXTURE_COORD_NV, index - 48};
}
switch (index) {
case 7:
return {GL_POSITION, 0};
case 40:
return {GL_PRIMARY_COLOR_NV, 0};
case 41:
return {GL_SECONDARY_COLOR_NV, 0};
case 42:
return {GL_BACK_PRIMARY_COLOR_NV, 0};
case 43:
return {GL_BACK_SECONDARY_COLOR_NV, 0};
}
UNIMPLEMENTED_MSG("index={}", index);
return {GL_POSITION, 0};
}
template <typename Spec>
bool Passes(const std::array<Shader::Info, 5>& stage_infos, u32 enabled_mask) {
for (size_t stage = 0; stage < stage_infos.size(); ++stage) {
if (!Spec::enabled_stages[stage] && ((enabled_mask >> stage) & 1) != 0) {
return false;
}
const auto& info{stage_infos[stage]};
if constexpr (!Spec::has_storage_buffers) {
if (!info.storage_buffers_descriptors.empty()) {
return false;
}
}
if constexpr (!Spec::has_texture_buffers) {
if (!info.texture_buffer_descriptors.empty()) {
return false;
}
}
if constexpr (!Spec::has_image_buffers) {
if (!info.image_buffer_descriptors.empty()) {
return false;
}
}
if constexpr (!Spec::has_images) {
if (!info.image_descriptors.empty()) {
return false;
}
}
}
return true;
}
using ConfigureFuncPtr = void (*)(GraphicsPipeline*, bool);
template <typename Spec, typename... Specs>
ConfigureFuncPtr FindSpec(const std::array<Shader::Info, 5>& stage_infos, u32 enabled_mask) {
if constexpr (sizeof...(Specs) > 0) {
if (!Passes<Spec>(stage_infos, enabled_mask)) {
return FindSpec<Specs...>(stage_infos, enabled_mask);
}
}
return GraphicsPipeline::MakeConfigureSpecFunc<Spec>();
}
struct SimpleVertexFragmentSpec {
static constexpr std::array<bool, 5> enabled_stages{true, false, false, false, true};
static constexpr bool has_storage_buffers = false;
static constexpr bool has_texture_buffers = false;
static constexpr bool has_image_buffers = false;
static constexpr bool has_images = false;
};
struct SimpleVertexSpec {
static constexpr std::array<bool, 5> enabled_stages{true, false, false, false, false};
static constexpr bool has_storage_buffers = false;
static constexpr bool has_texture_buffers = false;
static constexpr bool has_image_buffers = false;
static constexpr bool has_images = false;
};
struct DefaultSpec {
static constexpr std::array<bool, 5> enabled_stages{true, true, true, true, true};
static constexpr bool has_storage_buffers = true;
static constexpr bool has_texture_buffers = true;
static constexpr bool has_image_buffers = true;
static constexpr bool has_images = true;
};
ConfigureFuncPtr ConfigureFunc(const std::array<Shader::Info, 5>& infos, u32 enabled_mask) {
return FindSpec<SimpleVertexSpec, SimpleVertexFragmentSpec, DefaultSpec>(infos, enabled_mask);
}
} // Anonymous namespace
GraphicsPipeline::GraphicsPipeline(
const Device& device, TextureCache& texture_cache_, BufferCache& buffer_cache_,
Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_,
ProgramManager& program_manager_, StateTracker& state_tracker_, ShaderWorker* thread_worker,
VideoCore::ShaderNotify* shader_notify, std::array<std::string, 5> sources,
std::array<std::vector<u32>, 5> sources_spirv, const std::array<const Shader::Info*, 5>& infos,
const GraphicsPipelineKey& key_)
: texture_cache{texture_cache_}, buffer_cache{buffer_cache_},
gpu_memory{gpu_memory_}, maxwell3d{maxwell3d_}, program_manager{program_manager_},
state_tracker{state_tracker_}, key{key_} {
if (shader_notify) {
shader_notify->MarkShaderBuilding();
}
u32 num_textures{};
u32 num_images{};
u32 num_storage_buffers{};
for (size_t stage = 0; stage < base_uniform_bindings.size(); ++stage) {
auto& info{stage_infos[stage]};
if (infos[stage]) {
info = *infos[stage];
enabled_stages_mask |= 1u << stage;
}
if (stage < 4) {
base_uniform_bindings[stage + 1] = base_uniform_bindings[stage];
base_storage_bindings[stage + 1] = base_storage_bindings[stage];
base_uniform_bindings[stage + 1] += AccumulateCount(info.constant_buffer_descriptors);
base_storage_bindings[stage + 1] += AccumulateCount(info.storage_buffers_descriptors);
}
enabled_uniform_buffer_masks[stage] = info.constant_buffer_mask;
std::ranges::copy(info.constant_buffer_used_sizes, uniform_buffer_sizes[stage].begin());
const u32 num_tex_buffer_bindings{AccumulateCount(info.texture_buffer_descriptors)};
num_texture_buffers[stage] += num_tex_buffer_bindings;
num_textures += num_tex_buffer_bindings;
const u32 num_img_buffers_bindings{AccumulateCount(info.image_buffer_descriptors)};
num_image_buffers[stage] += num_img_buffers_bindings;
num_images += num_img_buffers_bindings;
num_textures += AccumulateCount(info.texture_descriptors);
num_images += AccumulateCount(info.image_descriptors);
num_storage_buffers += AccumulateCount(info.storage_buffers_descriptors);
writes_global_memory |= std::ranges::any_of(
info.storage_buffers_descriptors, [](const auto& desc) { return desc.is_written; });
}
ASSERT(num_textures <= MAX_TEXTURES);
ASSERT(num_images <= MAX_IMAGES);
const bool assembly_shaders{assembly_programs[0].handle != 0};
use_storage_buffers =
!assembly_shaders || num_storage_buffers <= device.GetMaxGLASMStorageBufferBlocks();
writes_global_memory &= !use_storage_buffers;
configure_func = ConfigureFunc(stage_infos, enabled_stages_mask);
if (key.xfb_enabled && device.UseAssemblyShaders()) {
GenerateTransformFeedbackState();
}
const bool in_parallel = thread_worker != nullptr;
const auto backend = device.GetShaderBackend();
auto func{[this, sources = std::move(sources), sources_spirv = std::move(sources_spirv),
shader_notify, backend, in_parallel](ShaderContext::Context*) mutable {
for (size_t stage = 0; stage < 5; ++stage) {
switch (backend) {
case Settings::ShaderBackend::GLSL:
if (!sources[stage].empty()) {
source_programs[stage] = CreateProgram(sources[stage], Stage(stage));
}
break;
case Settings::ShaderBackend::GLASM:
if (!sources[stage].empty()) {
assembly_programs[stage] = CompileProgram(sources[stage], AssemblyStage(stage));
if (in_parallel) {
// Make sure program is built before continuing when building in parallel
glGetString(GL_PROGRAM_ERROR_STRING_NV);
}
}
break;
case Settings::ShaderBackend::SPIRV:
if (!sources_spirv[stage].empty()) {
source_programs[stage] = CreateProgram(sources_spirv[stage], Stage(stage));
}
break;
}
}
if (in_parallel && backend != Settings::ShaderBackend::GLASM) {
// Make sure programs have built if we are building shaders in parallel
for (OGLProgram& program : source_programs) {
if (program.handle != 0) {
GLint status{};
glGetProgramiv(program.handle, GL_LINK_STATUS, &status);
}
}
}
if (shader_notify) {
shader_notify->MarkShaderComplete();
}
is_built = true;
built_condvar.notify_one();
}};
if (thread_worker) {
thread_worker->QueueWork(std::move(func));
} else {
func(nullptr);
}
}
template <typename Spec>
void GraphicsPipeline::ConfigureImpl(bool is_indexed) {
std::array<ImageId, MAX_TEXTURES + MAX_IMAGES> image_view_ids;
std::array<u32, MAX_TEXTURES + MAX_IMAGES> image_view_indices;
std::array<GLuint, MAX_TEXTURES> samplers;
size_t image_view_index{};
GLsizei sampler_binding{};
texture_cache.SynchronizeGraphicsDescriptors();
buffer_cache.SetUniformBuffersState(enabled_uniform_buffer_masks, &uniform_buffer_sizes);
buffer_cache.runtime.SetBaseUniformBindings(base_uniform_bindings);
buffer_cache.runtime.SetBaseStorageBindings(base_storage_bindings);
buffer_cache.runtime.SetEnableStorageBuffers(use_storage_buffers);
const auto& regs{maxwell3d.regs};
const bool via_header_index{regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex};
const auto config_stage{[&](size_t stage) LAMBDA_FORCEINLINE {
const Shader::Info& info{stage_infos[stage]};
buffer_cache.UnbindGraphicsStorageBuffers(stage);
if constexpr (Spec::has_storage_buffers) {
size_t ssbo_index{};
for (const auto& desc : info.storage_buffers_descriptors) {
ASSERT(desc.count == 1);
buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, desc.cbuf_index,
desc.cbuf_offset, desc.is_written);
++ssbo_index;
}
}
const auto& cbufs{maxwell3d.state.shader_stages[stage].const_buffers};
const auto read_handle{[&](const auto& desc, u32 index) {
ASSERT(cbufs[desc.cbuf_index].enabled);
const u32 index_offset{index << desc.size_shift};
const u32 offset{desc.cbuf_offset + index_offset};
const GPUVAddr addr{cbufs[desc.cbuf_index].address + offset};
if constexpr (std::is_same_v<decltype(desc), const TextureDescriptor&> ||
std::is_same_v<decltype(desc), const TextureBufferDescriptor&>) {
if (desc.has_secondary) {
ASSERT(cbufs[desc.secondary_cbuf_index].enabled);
const u32 second_offset{desc.secondary_cbuf_offset + index_offset};
const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].address +
second_offset};
const u32 lhs_raw{gpu_memory.Read<u32>(addr)};
const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)};
const u32 raw{lhs_raw | rhs_raw};
return TexturePair(raw, via_header_index);
}
}
return TexturePair(gpu_memory.Read<u32>(addr), via_header_index);
}};
const auto add_image{[&](const auto& desc) {
for (u32 index = 0; index < desc.count; ++index) {
const auto handle{read_handle(desc, index)};
image_view_indices[image_view_index++] = handle.first;
}
}};
if constexpr (Spec::has_texture_buffers) {
for (const auto& desc : info.texture_buffer_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
const auto handle{read_handle(desc, index)};
image_view_indices[image_view_index++] = handle.first;
samplers[sampler_binding++] = 0;
}
}
}
if constexpr (Spec::has_image_buffers) {
for (const auto& desc : info.image_buffer_descriptors) {
add_image(desc);
}
}
for (const auto& desc : info.texture_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
const auto handle{read_handle(desc, index)};
image_view_indices[image_view_index++] = handle.first;
Sampler* const sampler{texture_cache.GetGraphicsSampler(handle.second)};
samplers[sampler_binding++] = sampler->Handle();
}
}
if constexpr (Spec::has_images) {
for (const auto& desc : info.image_descriptors) {
add_image(desc);
}
}
}};
if constexpr (Spec::enabled_stages[0]) {
config_stage(0);
}
if constexpr (Spec::enabled_stages[1]) {
config_stage(1);
}
if constexpr (Spec::enabled_stages[2]) {
config_stage(2);
}
if constexpr (Spec::enabled_stages[3]) {
config_stage(3);
}
if constexpr (Spec::enabled_stages[4]) {
config_stage(4);
}
const std::span indices_span(image_view_indices.data(), image_view_index);
texture_cache.FillGraphicsImageViews(indices_span, image_view_ids);
texture_cache.UpdateRenderTargets(false);
state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle());
ImageId* texture_buffer_index{image_view_ids.data()};
const auto bind_stage_info{[&](size_t stage) LAMBDA_FORCEINLINE {
size_t index{};
const auto add_buffer{[&](const auto& desc) {
constexpr bool is_image = std::is_same_v<decltype(desc), const ImageBufferDescriptor&>;
for (u32 i = 0; i < desc.count; ++i) {
bool is_written{false};
if constexpr (is_image) {
is_written = desc.is_written;
}
ImageView& image_view{texture_cache.GetImageView(*texture_buffer_index)};
buffer_cache.BindGraphicsTextureBuffer(stage, index, image_view.GpuAddr(),
image_view.BufferSize(), image_view.format,
is_written, is_image);
++index;
++texture_buffer_index;
}
}};
const Shader::Info& info{stage_infos[stage]};
buffer_cache.UnbindGraphicsTextureBuffers(stage);
if constexpr (Spec::has_texture_buffers) {
for (const auto& desc : info.texture_buffer_descriptors) {
add_buffer(desc);
}
}
if constexpr (Spec::has_image_buffers) {
for (const auto& desc : info.image_buffer_descriptors) {
add_buffer(desc);
}
}
for (const auto& desc : info.texture_descriptors) {
texture_buffer_index += desc.count;
}
if constexpr (Spec::has_images) {
for (const auto& desc : info.image_descriptors) {
texture_buffer_index += desc.count;
}
}
}};
if constexpr (Spec::enabled_stages[0]) {
bind_stage_info(0);
}
if constexpr (Spec::enabled_stages[1]) {
bind_stage_info(1);
}
if constexpr (Spec::enabled_stages[2]) {
bind_stage_info(2);
}
if constexpr (Spec::enabled_stages[3]) {
bind_stage_info(3);
}
if constexpr (Spec::enabled_stages[4]) {
bind_stage_info(4);
}
buffer_cache.UpdateGraphicsBuffers(is_indexed);
buffer_cache.BindHostGeometryBuffers(is_indexed);
if (!is_built.load(std::memory_order::relaxed)) {
WaitForBuild();
}
if (assembly_programs[0].handle != 0) {
program_manager.BindAssemblyPrograms(assembly_programs, enabled_stages_mask);
} else {
program_manager.BindSourcePrograms(source_programs);
}
const ImageId* views_it{image_view_ids.data()};
GLsizei texture_binding = 0;
GLsizei image_binding = 0;
std::array<GLuint, MAX_TEXTURES> textures;
std::array<GLuint, MAX_IMAGES> images;
const auto prepare_stage{[&](size_t stage) {
buffer_cache.runtime.SetImagePointers(&textures[texture_binding], &images[image_binding]);
buffer_cache.BindHostStageBuffers(stage);
texture_binding += num_texture_buffers[stage];
image_binding += num_image_buffers[stage];
views_it += num_texture_buffers[stage];
views_it += num_image_buffers[stage];
const auto& info{stage_infos[stage]};
for (const auto& desc : info.texture_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
ImageView& image_view{texture_cache.GetImageView(*(views_it++))};
textures[texture_binding++] = image_view.Handle(desc.type);
}
}
for (const auto& desc : info.image_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
ImageView& image_view{texture_cache.GetImageView(*(views_it++))};
if (desc.is_written) {
texture_cache.MarkModification(image_view.image_id);
}
images[image_binding++] = image_view.StorageView(desc.type, desc.format);
}
}
}};
if constexpr (Spec::enabled_stages[0]) {
prepare_stage(0);
}
if constexpr (Spec::enabled_stages[1]) {
prepare_stage(1);
}
if constexpr (Spec::enabled_stages[2]) {
prepare_stage(2);
}
if constexpr (Spec::enabled_stages[3]) {
prepare_stage(3);
}
if constexpr (Spec::enabled_stages[4]) {
prepare_stage(4);
}
if (texture_binding != 0) {
ASSERT(texture_binding == sampler_binding);
glBindTextures(0, texture_binding, textures.data());
glBindSamplers(0, sampler_binding, samplers.data());
}
if (image_binding != 0) {
glBindImageTextures(0, image_binding, images.data());
}
}
void GraphicsPipeline::ConfigureTransformFeedbackImpl() const {
glTransformFeedbackStreamAttribsNV(num_xfb_attribs, xfb_attribs.data(), num_xfb_strides,
xfb_streams.data(), GL_INTERLEAVED_ATTRIBS);
}
void GraphicsPipeline::GenerateTransformFeedbackState() {
// TODO(Rodrigo): Inject SKIP_COMPONENTS*_NV when required. An unimplemented message will signal
// when this is required.
GLint* cursor{xfb_attribs.data()};
GLint* current_stream{xfb_streams.data()};
for (size_t feedback = 0; feedback < Maxwell::NumTransformFeedbackBuffers; ++feedback) {
const auto& layout = key.xfb_state.layouts[feedback];
UNIMPLEMENTED_IF_MSG(layout.stride != layout.varying_count * 4, "Stride padding");
if (layout.varying_count == 0) {
continue;
}
*current_stream = static_cast<GLint>(feedback);
if (current_stream != xfb_streams.data()) {
// When stepping one stream, push the expected token
cursor[0] = GL_NEXT_BUFFER_NV;
cursor[1] = 0;
cursor[2] = 0;
cursor += XFB_ENTRY_STRIDE;
}
++current_stream;
const auto& locations = key.xfb_state.varyings[feedback];
std::optional<u8> current_index;
for (u32 offset = 0; offset < layout.varying_count; ++offset) {
const u8 location = locations[offset];
const u8 index = location / 4;
if (current_index == index) {
// Increase number of components of the previous attachment
++cursor[-2];
continue;
}
current_index = index;
std::tie(cursor[0], cursor[2]) = TransformFeedbackEnum(location);
cursor[1] = 1;
cursor += XFB_ENTRY_STRIDE;
}
}
num_xfb_attribs = static_cast<GLsizei>((cursor - xfb_attribs.data()) / XFB_ENTRY_STRIDE);
num_xfb_strides = static_cast<GLsizei>(current_stream - xfb_streams.data());
}
void GraphicsPipeline::WaitForBuild() {
std::unique_lock lock{built_mutex};
built_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); });
}
} // namespace OpenGL

169
src/video_core/renderer_opengl/gl_graphics_pipeline.h Normal file
View File

@ -0,0 +1,169 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstring>
#include <type_traits>
#include <utility>
#include "common/bit_field.h"
#include "common/cityhash.h"
#include "common/common_types.h"
#include "shader_recompiler/shader_info.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_opengl/gl_buffer_cache.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_texture_cache.h"
#include "video_core/transform_feedback.h"
namespace OpenGL {
namespace ShaderContext {
struct Context;
}
class Device;
class ProgramManager;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
using ShaderWorker = Common::StatefulThreadWorker<ShaderContext::Context>;
struct GraphicsPipelineKey {
std::array<u64, 6> unique_hashes;
union {
u32 raw;
BitField<0, 1, u32> xfb_enabled;
BitField<1, 1, u32> early_z;
BitField<2, 4, Maxwell::PrimitiveTopology> gs_input_topology;
BitField<6, 2, Maxwell::TessellationPrimitive> tessellation_primitive;
BitField<8, 2, Maxwell::TessellationSpacing> tessellation_spacing;
BitField<10, 1, u32> tessellation_clockwise;
};
std::array<u32, 3> padding;
VideoCommon::TransformFeedbackState xfb_state;
size_t Hash() const noexcept {
return static_cast<size_t>(Common::CityHash64(reinterpret_cast<const char*>(this), Size()));
}
bool operator==(const GraphicsPipelineKey& rhs) const noexcept {
return std::memcmp(this, &rhs, Size()) == 0;
}
bool operator!=(const GraphicsPipelineKey& rhs) const noexcept {
return !operator==(rhs);
}
[[nodiscard]] size_t Size() const noexcept {
if (xfb_enabled != 0) {
return sizeof(GraphicsPipelineKey);
} else {
return offsetof(GraphicsPipelineKey, padding);
}
}
};
static_assert(std::has_unique_object_representations_v<GraphicsPipelineKey>);
static_assert(std::is_trivially_copyable_v<GraphicsPipelineKey>);
static_assert(std::is_trivially_constructible_v<GraphicsPipelineKey>);
class GraphicsPipeline {
public:
explicit GraphicsPipeline(const Device& device, TextureCache& texture_cache_,
BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_,
Tegra::Engines::Maxwell3D& maxwell3d_,
ProgramManager& program_manager_, StateTracker& state_tracker_,
ShaderWorker* thread_worker, VideoCore::ShaderNotify* shader_notify,
std::array<std::string, 5> sources,
std::array<std::vector<u32>, 5> sources_spirv,
const std::array<const Shader::Info*, 5>& infos,
const GraphicsPipelineKey& key_);
void Configure(bool is_indexed) {
configure_func(this, is_indexed);
}
void ConfigureTransformFeedback() const {
if (num_xfb_attribs != 0) {
ConfigureTransformFeedbackImpl();
}
}
[[nodiscard]] const GraphicsPipelineKey& Key() const noexcept {
return key;
}
[[nodiscard]] bool WritesGlobalMemory() const noexcept {
return writes_global_memory;
}
[[nodiscard]] bool IsBuilt() const noexcept {
return is_built.load(std::memory_order::relaxed);
}
template <typename Spec>
static auto MakeConfigureSpecFunc() {
return [](GraphicsPipeline* pipeline, bool is_indexed) {
pipeline->ConfigureImpl<Spec>(is_indexed);
};
}
private:
template <typename Spec>
void ConfigureImpl(bool is_indexed);
void ConfigureTransformFeedbackImpl() const;
void GenerateTransformFeedbackState();
void WaitForBuild();
TextureCache& texture_cache;
BufferCache& buffer_cache;
Tegra::MemoryManager& gpu_memory;
Tegra::Engines::Maxwell3D& maxwell3d;
ProgramManager& program_manager;
StateTracker& state_tracker;
const GraphicsPipelineKey key;
void (*configure_func)(GraphicsPipeline*, bool){};
std::array<OGLProgram, 5> source_programs;
std::array<OGLAssemblyProgram, 5> assembly_programs;
u32 enabled_stages_mask{};
std::array<Shader::Info, 5> stage_infos{};
std::array<u32, 5> enabled_uniform_buffer_masks{};
VideoCommon::UniformBufferSizes uniform_buffer_sizes{};
std::array<u32, 5> base_uniform_bindings{};
std::array<u32, 5> base_storage_bindings{};
std::array<u32, 5> num_texture_buffers{};
std::array<u32, 5> num_image_buffers{};
bool use_storage_buffers{};
bool writes_global_memory{};
static constexpr std::size_t XFB_ENTRY_STRIDE = 3;
GLsizei num_xfb_attribs{};
GLsizei num_xfb_strides{};
std::array<GLint, 128 * XFB_ENTRY_STRIDE * Maxwell::NumTransformFeedbackBuffers> xfb_attribs{};
std::array<GLint, Maxwell::NumTransformFeedbackBuffers> xfb_streams{};
std::mutex built_mutex;
std::condition_variable built_condvar;
std::atomic_bool is_built{false};
};
} // namespace OpenGL
namespace std {
template <>
struct hash<OpenGL::GraphicsPipelineKey> {
size_t operator()(const OpenGL::GraphicsPipelineKey& k) const noexcept {
return k.Hash();
}
};
} // namespace std

446
src/video_core/renderer_opengl/gl_rasterizer.cpp
View File

@ -23,7 +23,6 @@
#include "core/memory.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_query_cache.h"
@ -40,7 +39,6 @@ namespace OpenGL {
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
using GLvec4 = std::array<GLfloat, 4>;
using Tegra::Engines::ShaderType;
using VideoCore::Surface::PixelFormat;
using VideoCore::Surface::SurfaceTarget;
using VideoCore::Surface::SurfaceType;
@ -51,112 +49,11 @@ MICROPROFILE_DEFINE(OpenGL_Blits, "OpenGL", "Blits", MP_RGB(128, 128, 192));
MICROPROFILE_DEFINE(OpenGL_CacheManagement, "OpenGL", "Cache Management", MP_RGB(100, 255, 100));
namespace {
constexpr size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16;
struct TextureHandle {
constexpr TextureHandle(u32 data, bool via_header_index) {
const Tegra::Texture::TextureHandle handle{data};
image = handle.tic_id;
sampler = via_header_index ? image : handle.tsc_id.Value();
}
u32 image;
u32 sampler;
};
template <typename Engine, typename Entry>
TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const Entry& entry,
ShaderType shader_type, size_t index = 0) {
if constexpr (std::is_same_v<Entry, SamplerEntry>) {
if (entry.is_separated) {
const u32 buffer_1 = entry.buffer;
const u32 buffer_2 = entry.secondary_buffer;
const u32 offset_1 = entry.offset;
const u32 offset_2 = entry.secondary_offset;
const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1);
const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2);
return TextureHandle(handle_1 | handle_2, via_header_index);
}
}
if (entry.is_bindless) {
const u32 raw = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
return TextureHandle(raw, via_header_index);
}
const u32 buffer = engine.GetBoundBuffer();
const u64 offset = (entry.offset + index) * sizeof(u32);
return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index);
}
/// Translates hardware transform feedback indices
/// @param location Hardware location
/// @return Pair of ARB_transform_feedback3 token stream first and third arguments
/// @note Read https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_transform_feedback3.txt
std::pair<GLint, GLint> TransformFeedbackEnum(u8 location) {
const u8 index = location / 4;
if (index >= 8 && index <= 39) {
return {GL_GENERIC_ATTRIB_NV, index - 8};
}
if (index >= 48 && index <= 55) {
return {GL_TEXTURE_COORD_NV, index - 48};
}
switch (index) {
case 7:
return {GL_POSITION, 0};
case 40:
return {GL_PRIMARY_COLOR_NV, 0};
case 41:
return {GL_SECONDARY_COLOR_NV, 0};
case 42:
return {GL_BACK_PRIMARY_COLOR_NV, 0};
case 43:
return {GL_BACK_SECONDARY_COLOR_NV, 0};
}
UNIMPLEMENTED_MSG("index={}", index);
return {GL_POSITION, 0};
}
void oglEnable(GLenum cap, bool state) {
(state ? glEnable : glDisable)(cap);
}
ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) {
if (entry.is_buffer) {
return ImageViewType::Buffer;
}
switch (entry.type) {
case Tegra::Shader::TextureType::Texture1D:
return entry.is_array ? ImageViewType::e1DArray : ImageViewType::e1D;
case Tegra::Shader::TextureType::Texture2D:
return entry.is_array ? ImageViewType::e2DArray : ImageViewType::e2D;
case Tegra::Shader::TextureType::Texture3D:
return ImageViewType::e3D;
case Tegra::Shader::TextureType::TextureCube:
return entry.is_array ? ImageViewType::CubeArray : ImageViewType::Cube;
}
UNREACHABLE();
return ImageViewType::e2D;
}
ImageViewType ImageViewTypeFromEntry(const ImageEntry& entry) {
switch (entry.type) {
case Tegra::Shader::ImageType::Texture1D:
return ImageViewType::e1D;
case Tegra::Shader::ImageType::Texture1DArray:
return ImageViewType::e1DArray;
case Tegra::Shader::ImageType::Texture2D:
return ImageViewType::e2D;
case Tegra::Shader::ImageType::Texture2DArray:
return ImageViewType::e2DArray;
case Tegra::Shader::ImageType::Texture3D:
return ImageViewType::e3D;
case Tegra::Shader::ImageType::TextureBuffer:
return ImageViewType::Buffer;
}
UNREACHABLE();
return ImageViewType::e2D;
}
} // Anonymous namespace
RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
@ -170,14 +67,10 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra
texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory),
buffer_cache_runtime(device),
buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime),
shader_cache(*this, emu_window_, gpu, maxwell3d, kepler_compute, gpu_memory, device),
shader_cache(*this, emu_window_, maxwell3d, kepler_compute, gpu_memory, device, texture_cache,
buffer_cache, program_manager, state_tracker, gpu.ShaderNotify()),
query_cache(*this, maxwell3d, gpu_memory), accelerate_dma(buffer_cache),
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache),
async_shaders(emu_window_) {
if (device.UseAsynchronousShaders()) {
async_shaders.AllocateWorkers();
}
}
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache) {}
RasterizerOpenGL::~RasterizerOpenGL() = default;
@ -204,7 +97,7 @@ void RasterizerOpenGL::SyncVertexFormats() {
const auto gl_index = static_cast<GLuint>(index);
// Disable constant attributes.
if (attrib.IsConstant()) {
if (attrib.constant) {
glDisableVertexAttribArray(gl_index);
continue;
}
@ -244,116 +137,9 @@ void RasterizerOpenGL::SyncVertexInstances() {
}
}
void RasterizerOpenGL::SetupShaders(bool is_indexed) {
u32 clip_distances = 0;
std::array<Shader*, Maxwell::MaxShaderStage> shaders{};
image_view_indices.clear();
sampler_handles.clear();
texture_cache.SynchronizeGraphicsDescriptors();
for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
const auto& shader_config = maxwell3d.regs.shader_config[index];
const auto program{static_cast<Maxwell::ShaderProgram>(index)};
// Skip stages that are not enabled
if (!maxwell3d.regs.IsShaderConfigEnabled(index)) {
switch (program) {
case Maxwell::ShaderProgram::Geometry:
program_manager.UseGeometryShader(0);
break;
case Maxwell::ShaderProgram::Fragment:
program_manager.UseFragmentShader(0);
break;
default:
break;
}
continue;
}
// Currently this stages are not supported in the OpenGL backend.
// TODO(Blinkhawk): Port tesselation shaders from Vulkan to OpenGL
if (program == Maxwell::ShaderProgram::TesselationControl ||
program == Maxwell::ShaderProgram::TesselationEval) {
continue;
}
Shader* const shader = shader_cache.GetStageProgram(program, async_shaders);
const GLuint program_handle = shader->IsBuilt() ? shader->GetHandle() : 0;
switch (program) {
case Maxwell::ShaderProgram::VertexA:
case Maxwell::ShaderProgram::VertexB:
program_manager.UseVertexShader(program_handle);
break;
case Maxwell::ShaderProgram::Geometry:
program_manager.UseGeometryShader(program_handle);
break;
case Maxwell::ShaderProgram::Fragment:
program_manager.UseFragmentShader(program_handle);
break;
default:
UNIMPLEMENTED_MSG("Unimplemented shader index={}, enable={}, offset=0x{:08X}", index,
shader_config.enable.Value(), shader_config.offset);
break;
}
// Stage indices are 0 - 5
const size_t stage = index == 0 ? 0 : index - 1;
shaders[stage] = shader;
SetupDrawTextures(shader, stage);
SetupDrawImages(shader, stage);
buffer_cache.SetEnabledUniformBuffers(stage, shader->GetEntries().enabled_uniform_buffers);
buffer_cache.UnbindGraphicsStorageBuffers(stage);
u32 ssbo_index = 0;
for (const auto& buffer : shader->GetEntries().global_memory_entries) {
buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, buffer.cbuf_index,
buffer.cbuf_offset, buffer.is_written);
++ssbo_index;
}
// Workaround for Intel drivers.
// When a clip distance is enabled but not set in the shader it crops parts of the screen
// (sometimes it's half the screen, sometimes three quarters). To avoid this, enable the
// clip distances only when it's written by a shader stage.
clip_distances |= shader->GetEntries().clip_distances;
// When VertexA is enabled, we have dual vertex shaders
if (program == Maxwell::ShaderProgram::VertexA) {
// VertexB was combined with VertexA, so we skip the VertexB iteration
++index;
}
}
SyncClipEnabled(clip_distances);
maxwell3d.dirty.flags[Dirty::Shaders] = false;
buffer_cache.UpdateGraphicsBuffers(is_indexed);
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
texture_cache.FillGraphicsImageViews(indices_span, image_view_ids);
buffer_cache.BindHostGeometryBuffers(is_indexed);
size_t image_view_index = 0;
size_t texture_index = 0;
size_t image_index = 0;
for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
const Shader* const shader = shaders[stage];
if (!shader) {
continue;
}
buffer_cache.BindHostStageBuffers(stage);
const auto& base = device.GetBaseBindings(stage);
BindTextures(shader->GetEntries(), base.sampler, base.image, image_view_index,
texture_index, image_index);
}
}
void RasterizerOpenGL::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) {
shader_cache.LoadDiskCache(title_id, stop_loading, callback);
shader_cache.LoadDiskResources(title_id, stop_loading, callback);
}
void RasterizerOpenGL::Clear() {
@ -432,16 +218,15 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
SyncState();
// Setup shaders and their used resources.
GraphicsPipeline* const pipeline{shader_cache.CurrentGraphicsPipeline()};
if (!pipeline) {
return;
}
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
SetupShaders(is_indexed);
texture_cache.UpdateRenderTargets(false);
state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle());
program_manager.BindGraphicsPipeline();
pipeline->Configure(is_indexed);
const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d.regs.draw.topology);
BeginTransformFeedback(primitive_mode);
BeginTransformFeedback(pipeline, primitive_mode);
const GLuint base_instance = static_cast<GLuint>(maxwell3d.regs.vb_base_instance);
const GLsizei num_instances =
@ -480,35 +265,24 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
num_instances, base_instance);
}
}
EndTransformFeedback();
++num_queued_commands;
has_written_global_memory |= pipeline->WritesGlobalMemory();
gpu.TickWork();
}
void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
Shader* const kernel = shader_cache.GetComputeKernel(code_addr);
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
BindComputeTextures(kernel);
const auto& entries = kernel->GetEntries();
buffer_cache.SetEnabledComputeUniformBuffers(entries.enabled_uniform_buffers);
buffer_cache.UnbindComputeStorageBuffers();
u32 ssbo_index = 0;
for (const auto& buffer : entries.global_memory_entries) {
buffer_cache.BindComputeStorageBuffer(ssbo_index, buffer.cbuf_index, buffer.cbuf_offset,
buffer.is_written);
++ssbo_index;
void RasterizerOpenGL::DispatchCompute() {
ComputePipeline* const pipeline{shader_cache.CurrentComputePipeline()};
if (!pipeline) {
return;
}
buffer_cache.UpdateComputeBuffers();
buffer_cache.BindHostComputeBuffers();
const auto& launch_desc = kepler_compute.launch_description;
glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z);
pipeline->Configure();
const auto& qmd{kepler_compute.launch_description};
glDispatchCompute(qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z);
++num_queued_commands;
has_written_global_memory |= pipeline->WritesGlobalMemory();
}
void RasterizerOpenGL::ResetCounter(VideoCore::QueryType type) {
@ -661,7 +435,7 @@ void RasterizerOpenGL::WaitForIdle() {
}
void RasterizerOpenGL::FragmentBarrier() {
glMemoryBarrier(GL_FRAMEBUFFER_BARRIER_BIT);
glMemoryBarrier(GL_FRAMEBUFFER_BARRIER_BIT | GL_TEXTURE_FETCH_BARRIER_BIT);
}
void RasterizerOpenGL::TiledCacheBarrier() {
@ -674,6 +448,13 @@ void RasterizerOpenGL::FlushCommands() {
return;
}
num_queued_commands = 0;
// Make sure memory stored from the previous GL command stream is visible
// This is only needed on assembly shaders where we write to GPU memory with raw pointers
if (has_written_global_memory) {
has_written_global_memory = false;
glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
}
glFlush();
}
@ -721,111 +502,11 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
// ASSERT_MSG(image_view->size.width == config.width, "Framebuffer width is different");
// ASSERT_MSG(image_view->size.height == config.height, "Framebuffer height is different");
screen_info.display_texture = image_view->Handle(ImageViewType::e2D);
screen_info.display_texture = image_view->Handle(Shader::TextureType::Color2D);
screen_info.display_srgb = VideoCore::Surface::IsPixelFormatSRGB(image_view->format);
return true;
}
void RasterizerOpenGL::BindComputeTextures(Shader* kernel) {
image_view_indices.clear();
sampler_handles.clear();
texture_cache.SynchronizeComputeDescriptors();
SetupComputeTextures(kernel);
SetupComputeImages(kernel);
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
texture_cache.FillComputeImageViews(indices_span, image_view_ids);
program_manager.BindCompute(kernel->GetHandle());
size_t image_view_index = 0;
size_t texture_index = 0;
size_t image_index = 0;
BindTextures(kernel->GetEntries(), 0, 0, image_view_index, texture_index, image_index);
}
void RasterizerOpenGL::BindTextures(const ShaderEntries& entries, GLuint base_texture,
GLuint base_image, size_t& image_view_index,
size_t& texture_index, size_t& image_index) {
const GLuint* const samplers = sampler_handles.data() + texture_index;
const GLuint* const textures = texture_handles.data() + texture_index;
const GLuint* const images = image_handles.data() + image_index;
const size_t num_samplers = entries.samplers.size();
for (const auto& sampler : entries.samplers) {
for (size_t i = 0; i < sampler.size; ++i) {
const ImageViewId image_view_id = image_view_ids[image_view_index++];
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
const GLuint handle = image_view.Handle(ImageViewTypeFromEntry(sampler));
texture_handles[texture_index++] = handle;
}
}
const size_t num_images = entries.images.size();
for (size_t unit = 0; unit < num_images; ++unit) {
// TODO: Mark as modified
const ImageViewId image_view_id = image_view_ids[image_view_index++];
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
const GLuint handle = image_view.Handle(ImageViewTypeFromEntry(entries.images[unit]));
image_handles[image_index] = handle;
++image_index;
}
if (num_samplers > 0) {
glBindSamplers(base_texture, static_cast<GLsizei>(num_samplers), samplers);
glBindTextures(base_texture, static_cast<GLsizei>(num_samplers), textures);
}
if (num_images > 0) {
glBindImageTextures(base_image, static_cast<GLsizei>(num_images), images);
}
}
void RasterizerOpenGL::SetupDrawTextures(const Shader* shader, size_t stage_index) {
const bool via_header_index =
maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : shader->GetEntries().samplers) {
const auto shader_type = static_cast<ShaderType>(stage_index);
for (size_t index = 0; index < entry.size; ++index) {
const auto handle =
GetTextureInfo(maxwell3d, via_header_index, entry, shader_type, index);
const Sampler* const sampler = texture_cache.GetGraphicsSampler(handle.sampler);
sampler_handles.push_back(sampler->Handle());
image_view_indices.push_back(handle.image);
}
}
}
void RasterizerOpenGL::SetupComputeTextures(const Shader* kernel) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : kernel->GetEntries().samplers) {
for (size_t i = 0; i < entry.size; ++i) {
const auto handle =
GetTextureInfo(kepler_compute, via_header_index, entry, ShaderType::Compute, i);
const Sampler* const sampler = texture_cache.GetComputeSampler(handle.sampler);
sampler_handles.push_back(sampler->Handle());
image_view_indices.push_back(handle.image);
}
}
}
void RasterizerOpenGL::SetupDrawImages(const Shader* shader, size_t stage_index) {
const bool via_header_index =
maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : shader->GetEntries().images) {
const auto shader_type = static_cast<ShaderType>(stage_index);
const auto handle = GetTextureInfo(maxwell3d, via_header_index, entry, shader_type);
image_view_indices.push_back(handle.image);
}
}
void RasterizerOpenGL::SetupComputeImages(const Shader* shader) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : shader->GetEntries().images) {
const auto handle =
GetTextureInfo(kepler_compute, via_header_index, entry, ShaderType::Compute);
image_view_indices.push_back(handle.image);
}
}
void RasterizerOpenGL::SyncState() {
SyncViewport();
SyncRasterizeEnable();
@ -941,7 +622,7 @@ void RasterizerOpenGL::SyncDepthClamp() {
void RasterizerOpenGL::SyncClipEnabled(u32 clip_mask) {
auto& flags = maxwell3d.dirty.flags;
if (!flags[Dirty::ClipDistances] && !flags[Dirty::Shaders]) {
if (!flags[Dirty::ClipDistances] && !flags[VideoCommon::Dirty::Shaders]) {
return;
}
flags[Dirty::ClipDistances] = false;
@ -1318,68 +999,13 @@ void RasterizerOpenGL::SyncFramebufferSRGB() {
oglEnable(GL_FRAMEBUFFER_SRGB, maxwell3d.regs.framebuffer_srgb);
}
void RasterizerOpenGL::SyncTransformFeedback() {
// TODO(Rodrigo): Inject SKIP_COMPONENTS*_NV when required. An unimplemented message will signal
// when this is required.
const auto& regs = maxwell3d.regs;
static constexpr std::size_t STRIDE = 3;
std::array<GLint, 128 * STRIDE * Maxwell::NumTransformFeedbackBuffers> attribs;
std::array<GLint, Maxwell::NumTransformFeedbackBuffers> streams;
GLint* cursor = attribs.data();
GLint* current_stream = streams.data();
for (std::size_t feedback = 0; feedback < Maxwell::NumTransformFeedbackBuffers; ++feedback) {
const auto& layout = regs.tfb_layouts[feedback];
UNIMPLEMENTED_IF_MSG(layout.stride != layout.varying_count * 4, "Stride padding");
if (layout.varying_count == 0) {
continue;
}
*current_stream = static_cast<GLint>(feedback);
if (current_stream != streams.data()) {
// When stepping one stream, push the expected token
cursor[0] = GL_NEXT_BUFFER_NV;
cursor[1] = 0;
cursor[2] = 0;
cursor += STRIDE;
}
++current_stream;
const auto& locations = regs.tfb_varying_locs[feedback];
std::optional<u8> current_index;
for (u32 offset = 0; offset < layout.varying_count; ++offset) {
const u8 location = locations[offset];
const u8 index = location / 4;
if (current_index == index) {
// Increase number of components of the previous attachment
++cursor[-2];
continue;
}
current_index = index;
std::tie(cursor[0], cursor[2]) = TransformFeedbackEnum(location);
cursor[1] = 1;
cursor += STRIDE;
}
}
const GLsizei num_attribs = static_cast<GLsizei>((cursor - attribs.data()) / STRIDE);
const GLsizei num_strides = static_cast<GLsizei>(current_stream - streams.data());
glTransformFeedbackStreamAttribsNV(num_attribs, attribs.data(), num_strides, streams.data(),
GL_INTERLEAVED_ATTRIBS);
}
void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) {
void RasterizerOpenGL::BeginTransformFeedback(GraphicsPipeline* program, GLenum primitive_mode) {
const auto& regs = maxwell3d.regs;
if (regs.tfb_enabled == 0) {
return;
}
if (device.UseAssemblyShaders()) {
SyncTransformFeedback();
}
program->ConfigureTransformFeedback();
UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) ||
regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) ||
regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry));
@ -1393,11 +1019,9 @@ void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) {
}
void RasterizerOpenGL::EndTransformFeedback() {
const auto& regs = maxwell3d.regs;
if (regs.tfb_enabled == 0) {
return;
if (maxwell3d.regs.tfb_enabled != 0) {
glEndTransformFeedback();
}
glEndTransformFeedback();
}
AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {}

44
src/video_core/renderer_opengl/gl_rasterizer.h
View File

@ -28,11 +28,9 @@
#include "video_core/renderer_opengl/gl_query_cache.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_opengl/gl_shader_manager.h"
#include "video_core/renderer_opengl/gl_state_tracker.h"
#include "video_core/renderer_opengl/gl_texture_cache.h"
#include "video_core/shader/async_shaders.h"
#include "video_core/textures/texture.h"
namespace Core::Memory {
@ -81,7 +79,7 @@ public:
void Draw(bool is_indexed, bool is_instanced) override;
void Clear() override;
void DispatchCompute(GPUVAddr code_addr) override;
void DispatchCompute() override;
void ResetCounter(VideoCore::QueryType type) override;
void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
@ -118,36 +116,11 @@ public:
return num_queued_commands > 0;
}
VideoCommon::Shader::AsyncShaders& GetAsyncShaders() {
return async_shaders;
}
const VideoCommon::Shader::AsyncShaders& GetAsyncShaders() const {
return async_shaders;
}
private:
static constexpr size_t MAX_TEXTURES = 192;
static constexpr size_t MAX_IMAGES = 48;
static constexpr size_t MAX_IMAGE_VIEWS = MAX_TEXTURES + MAX_IMAGES;
void BindComputeTextures(Shader* kernel);
void BindTextures(const ShaderEntries& entries, GLuint base_texture, GLuint base_image,
size_t& image_view_index, size_t& texture_index, size_t& image_index);
/// Configures the current textures to use for the draw command.
void SetupDrawTextures(const Shader* shader, size_t stage_index);
/// Configures the textures used in a compute shader.
void SetupComputeTextures(const Shader* kernel);
/// Configures images in a graphics shader.
void SetupDrawImages(const Shader* shader, size_t stage_index);
/// Configures images in a compute shader.
void SetupComputeImages(const Shader* shader);
/// Syncs state to match guest's
void SyncState();
@ -220,18 +193,12 @@ private:
/// Syncs vertex instances to match the guest state
void SyncVertexInstances();
/// Syncs transform feedback state to match guest state
/// @note Only valid on assembly shaders
void SyncTransformFeedback();
/// Begin a transform feedback
void BeginTransformFeedback(GLenum primitive_mode);
void BeginTransformFeedback(GraphicsPipeline* pipeline, GLenum primitive_mode);
/// End a transform feedback
void EndTransformFeedback();
void SetupShaders(bool is_indexed);
Tegra::GPU& gpu;
Tegra::Engines::Maxwell3D& maxwell3d;
Tegra::Engines::KeplerCompute& kepler_compute;
@ -246,13 +213,11 @@ private:
TextureCache texture_cache;
BufferCacheRuntime buffer_cache_runtime;
BufferCache buffer_cache;
ShaderCacheOpenGL shader_cache;
ShaderCache shader_cache;
QueryCache query_cache;
AccelerateDMA accelerate_dma;
FenceManagerOpenGL fence_manager;
VideoCommon::Shader::AsyncShaders async_shaders;
boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices;
std::array<ImageViewId, MAX_IMAGE_VIEWS> image_view_ids;
boost::container::static_vector<GLuint, MAX_TEXTURES> sampler_handles;
@ -260,7 +225,8 @@ private:
std::array<GLuint, MAX_IMAGES> image_handles{};
/// Number of commands queued to the OpenGL driver. Resetted on flush.
std::size_t num_queued_commands = 0;
size_t num_queued_commands = 0;
bool has_written_global_memory = false;
u32 last_clip_distance_mask = 0;
};

27
src/video_core/renderer_opengl/gl_resource_manager.cpp
View File

@ -83,18 +83,6 @@ void OGLSampler::Release() {
handle = 0;
}
void OGLShader::Create(std::string_view source, GLenum type) {
if (handle != 0) {
return;
}
if (source.empty()) {
return;
}
MICROPROFILE_SCOPE(OpenGL_ResourceCreation);
handle = GLShader::LoadShader(source, type);
}
void OGLShader::Release() {
if (handle == 0)
return;
@ -104,21 +92,6 @@ void OGLShader::Release() {
handle = 0;
}
void OGLProgram::CreateFromSource(const char* vert_shader, const char* geo_shader,
const char* frag_shader, bool separable_program,
bool hint_retrievable) {
OGLShader vert, geo, frag;
if (vert_shader)
vert.Create(vert_shader, GL_VERTEX_SHADER);
if (geo_shader)
geo.Create(geo_shader, GL_GEOMETRY_SHADER);
if (frag_shader)
frag.Create(frag_shader, GL_FRAGMENT_SHADER);
MICROPROFILE_SCOPE(OpenGL_ResourceCreation);
Create(separable_program, hint_retrievable, vert.handle, geo.handle, frag.handle);
}
void OGLProgram::Release() {
if (handle == 0)
return;

14
src/video_core/renderer_opengl/gl_resource_manager.h
View File

@ -8,7 +8,6 @@
#include <utility>
#include <glad/glad.h>
#include "common/common_types.h"
#include "video_core/renderer_opengl/gl_shader_util.h"
namespace OpenGL {
@ -128,8 +127,6 @@ public:
return *this;
}
void Create(std::string_view source, GLenum type);
void Release();
GLuint handle = 0;
@ -151,17 +148,6 @@ public:
return *this;
}
template <typename... T>
void Create(bool separable_program, bool hint_retrievable, T... shaders) {
if (handle != 0)
return;
handle = GLShader::LoadProgram(separable_program, hint_retrievable, shaders...);
}
/// Creates a new internal OpenGL resource and stores the handle
void CreateFromSource(const char* vert_shader, const char* geo_shader, const char* frag_shader,
bool separable_program = false, bool hint_retrievable = false);
/// Deletes the internal OpenGL resource
void Release();

998
src/video_core/renderer_opengl/gl_shader_cache.cpp
View File

File diff suppressed because it is too large Load Diff

170
src/video_core/renderer_opengl/gl_shader_cache.h
View File

@ -5,157 +5,93 @@
#pragma once
#include <array>
#include <atomic>
#include <bitset>
#include <memory>
#include <string>
#include <tuple>
#include <filesystem>
#include <stop_token>
#include <unordered_map>
#include <unordered_set>
#include <vector>
#include <glad/glad.h>
#include "common/common_types.h"
#include "video_core/engines/shader_type.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
#include "common/thread_worker.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/host_translate_info.h"
#include "shader_recompiler/object_pool.h"
#include "shader_recompiler/profile.h"
#include "video_core/renderer_opengl/gl_compute_pipeline.h"
#include "video_core/renderer_opengl/gl_graphics_pipeline.h"
#include "video_core/renderer_opengl/gl_shader_context.h"
#include "video_core/shader_cache.h"
namespace Tegra {
class MemoryManager;
}
namespace Core::Frontend {
class EmuWindow;
}
namespace VideoCommon::Shader {
class AsyncShaders;
}
namespace OpenGL {
class Device;
class ProgramManager;
class RasterizerOpenGL;
using ShaderWorker = Common::StatefulThreadWorker<ShaderContext::Context>;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
struct ProgramHandle {
OGLProgram source_program;
OGLAssemblyProgram assembly_program;
};
using ProgramSharedPtr = std::shared_ptr<ProgramHandle>;
struct PrecompiledShader {
ProgramSharedPtr program;
std::shared_ptr<VideoCommon::Shader::Registry> registry;
ShaderEntries entries;
};
struct ShaderParameters {
Tegra::GPU& gpu;
Tegra::Engines::ConstBufferEngineInterface& engine;
ShaderDiskCacheOpenGL& disk_cache;
const Device& device;
VAddr cpu_addr;
const u8* host_ptr;
u64 unique_identifier;
};
ProgramSharedPtr BuildShader(const Device& device, Tegra::Engines::ShaderType shader_type,
u64 unique_identifier, const VideoCommon::Shader::ShaderIR& ir,
const VideoCommon::Shader::Registry& registry,
bool hint_retrievable = false);
class Shader final {
class ShaderCache : public VideoCommon::ShaderCache {
public:
~Shader();
explicit ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindow& emu_window_,
Tegra::Engines::Maxwell3D& maxwell3d_,
Tegra::Engines::KeplerCompute& kepler_compute_,
Tegra::MemoryManager& gpu_memory_, const Device& device_,
TextureCache& texture_cache_, BufferCache& buffer_cache_,
ProgramManager& program_manager_, StateTracker& state_tracker_,
VideoCore::ShaderNotify& shader_notify_);
~ShaderCache();
/// Gets the GL program handle for the shader
GLuint GetHandle() const;
void LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback);
bool IsBuilt() const;
[[nodiscard]] GraphicsPipeline* CurrentGraphicsPipeline();
/// Gets the shader entries for the shader
const ShaderEntries& GetEntries() const {
return entries;
}
const VideoCommon::Shader::Registry& GetRegistry() const {
return *registry;
}
/// Mark a OpenGL shader as built
void AsyncOpenGLBuilt(OGLProgram new_program);
/// Mark a GLASM shader as built
void AsyncGLASMBuilt(OGLAssemblyProgram new_program);
static std::unique_ptr<Shader> CreateStageFromMemory(
const ShaderParameters& params, Maxwell::ShaderProgram program_type,
ProgramCode program_code, ProgramCode program_code_b,
VideoCommon::Shader::AsyncShaders& async_shaders, VAddr cpu_addr);
static std::unique_ptr<Shader> CreateKernelFromMemory(const ShaderParameters& params,
ProgramCode code);
static std::unique_ptr<Shader> CreateFromCache(const ShaderParameters& params,
const PrecompiledShader& precompiled_shader);
[[nodiscard]] ComputePipeline* CurrentComputePipeline();
private:
explicit Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry, ShaderEntries entries,
ProgramSharedPtr program, bool is_built_ = true);
GraphicsPipeline* CurrentGraphicsPipelineSlowPath();
std::shared_ptr<VideoCommon::Shader::Registry> registry;
ShaderEntries entries;
ProgramSharedPtr program;
GLuint handle = 0;
bool is_built{};
};
[[nodiscard]] GraphicsPipeline* BuiltPipeline(GraphicsPipeline* pipeline) const noexcept;
class ShaderCacheOpenGL final : public VideoCommon::ShaderCache<Shader> {
public:
explicit ShaderCacheOpenGL(RasterizerOpenGL& rasterizer_,
Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu,
Tegra::Engines::Maxwell3D& maxwell3d_,
Tegra::Engines::KeplerCompute& kepler_compute_,
Tegra::MemoryManager& gpu_memory_, const Device& device_);
~ShaderCacheOpenGL() override;
std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline();
/// Loads disk cache for the current game
void LoadDiskCache(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback);
std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(
ShaderContext::ShaderPools& pools, const GraphicsPipelineKey& key,
std::span<Shader::Environment* const> envs, bool build_in_parallel);
/// Gets the current specified shader stage program
Shader* GetStageProgram(Maxwell::ShaderProgram program,
VideoCommon::Shader::AsyncShaders& async_shaders);
std::unique_ptr<ComputePipeline> CreateComputePipeline(const ComputePipelineKey& key,
const VideoCommon::ShaderInfo* shader);
/// Gets a compute kernel in the passed address
Shader* GetComputeKernel(GPUVAddr code_addr);
std::unique_ptr<ComputePipeline> CreateComputePipeline(ShaderContext::ShaderPools& pools,
const ComputePipelineKey& key,
Shader::Environment& env);
private:
ProgramSharedPtr GeneratePrecompiledProgram(
const ShaderDiskCacheEntry& entry, const ShaderDiskCachePrecompiled& precompiled_entry,
const std::unordered_set<GLenum>& supported_formats);
std::unique_ptr<ShaderWorker> CreateWorkers() const;
Core::Frontend::EmuWindow& emu_window;
Tegra::GPU& gpu;
Tegra::MemoryManager& gpu_memory;
Tegra::Engines::Maxwell3D& maxwell3d;
Tegra::Engines::KeplerCompute& kepler_compute;
const Device& device;
TextureCache& texture_cache;
BufferCache& buffer_cache;
ProgramManager& program_manager;
StateTracker& state_tracker;
VideoCore::ShaderNotify& shader_notify;
const bool use_asynchronous_shaders;
ShaderDiskCacheOpenGL disk_cache;
std::unordered_map<u64, PrecompiledShader> runtime_cache;
GraphicsPipelineKey graphics_key{};
GraphicsPipeline* current_pipeline{};
std::unique_ptr<Shader> null_shader;
std::unique_ptr<Shader> null_kernel;
ShaderContext::ShaderPools main_pools;
std::unordered_map<GraphicsPipelineKey, std::unique_ptr<GraphicsPipeline>> graphics_cache;
std::unordered_map<ComputePipelineKey, std::unique_ptr<ComputePipeline>> compute_cache;
std::array<Shader*, Maxwell::MaxShaderProgram> last_shaders{};
Shader::Profile profile;
Shader::HostTranslateInfo host_info;
std::filesystem::path shader_cache_filename;
std::unique_ptr<ShaderWorker> workers;
};
} // namespace OpenGL

33
src/video_core/renderer_opengl/gl_shader_context.h Normal file
View File

@ -0,0 +1,33 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "core/frontend/emu_window.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
namespace OpenGL::ShaderContext {
struct ShaderPools {
void ReleaseContents() {
flow_block.ReleaseContents();
block.ReleaseContents();
inst.ReleaseContents();
}
Shader::ObjectPool<Shader::IR::Inst> inst;
Shader::ObjectPool<Shader::IR::Block> block;
Shader::ObjectPool<Shader::Maxwell::Flow::Block> flow_block;
};
struct Context {
explicit Context(Core::Frontend::EmuWindow& emu_window)
: gl_context{emu_window.CreateSharedContext()}, scoped{*gl_context} {}
std::unique_ptr<Core::Frontend::GraphicsContext> gl_context;
Core::Frontend::GraphicsContext::Scoped scoped;
ShaderPools pools;
};
} // namespace OpenGL::ShaderContext

2986
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View File

File diff suppressed because it is too large Load Diff

69
src/video_core/renderer_opengl/gl_shader_decompiler.h
View File

@ -1,69 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <string>
#include <string_view>
#include <utility>
#include <vector>
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace OpenGL {
class Device;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
using SamplerEntry = VideoCommon::Shader::SamplerEntry;
using ImageEntry = VideoCommon::Shader::ImageEntry;
class ConstBufferEntry : public VideoCommon::Shader::ConstBuffer {
public:
explicit ConstBufferEntry(u32 max_offset_, bool is_indirect_, u32 index_)
: ConstBuffer{max_offset_, is_indirect_}, index{index_} {}
u32 GetIndex() const {
return index;
}
private:
u32 index = 0;
};
struct GlobalMemoryEntry {
constexpr explicit GlobalMemoryEntry(u32 cbuf_index_, u32 cbuf_offset_, bool is_read_,
bool is_written_)
: cbuf_index{cbuf_index_}, cbuf_offset{cbuf_offset_}, is_read{is_read_}, is_written{
is_written_} {}
u32 cbuf_index = 0;
u32 cbuf_offset = 0;
bool is_read = false;
bool is_written = false;
};
struct ShaderEntries {
std::vector<ConstBufferEntry> const_buffers;
std::vector<GlobalMemoryEntry> global_memory_entries;
std::vector<SamplerEntry> samplers;
std::vector<ImageEntry> images;
std::size_t shader_length{};
u32 clip_distances{};
u32 enabled_uniform_buffers{};
};
ShaderEntries MakeEntries(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
Tegra::Engines::ShaderType stage);
std::string DecompileShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
const VideoCommon::Shader::Registry& registry,
Tegra::Engines::ShaderType stage, std::string_view identifier,
std::string_view suffix = {});
} // namespace OpenGL

482
src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
View File

@ -1,482 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <fmt/format.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "common/fs/file.h"
#include "common/fs/fs.h"
#include "common/fs/path_util.h"
#include "common/logging/log.h"
#include "common/scm_rev.h"
#include "common/settings.h"
#include "common/zstd_compression.h"
#include "core/core.h"
#include "core/hle/kernel/k_process.h"
#include "video_core/engines/shader_type.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
namespace OpenGL {
using Tegra::Engines::ShaderType;
using VideoCommon::Shader::BindlessSamplerMap;
using VideoCommon::Shader::BoundSamplerMap;
using VideoCommon::Shader::KeyMap;
using VideoCommon::Shader::SeparateSamplerKey;
using ShaderCacheVersionHash = std::array<u8, 64>;
struct ConstBufferKey {
u32 cbuf = 0;
u32 offset = 0;
u32 value = 0;
};
struct BoundSamplerEntry {
u32 offset = 0;
Tegra::Engines::SamplerDescriptor sampler;
};
struct SeparateSamplerEntry {
u32 cbuf1 = 0;
u32 cbuf2 = 0;
u32 offset1 = 0;
u32 offset2 = 0;
Tegra::Engines::SamplerDescriptor sampler;
};
struct BindlessSamplerEntry {
u32 cbuf = 0;
u32 offset = 0;
Tegra::Engines::SamplerDescriptor sampler;
};
namespace {
constexpr u32 NativeVersion = 21;
ShaderCacheVersionHash GetShaderCacheVersionHash() {
ShaderCacheVersionHash hash{};
const std::size_t length = std::min(std::strlen(Common::g_shader_cache_version), hash.size());
std::memcpy(hash.data(), Common::g_shader_cache_version, length);
return hash;
}
} // Anonymous namespace
ShaderDiskCacheEntry::ShaderDiskCacheEntry() = default;
ShaderDiskCacheEntry::~ShaderDiskCacheEntry() = default;
bool ShaderDiskCacheEntry::Load(Common::FS::IOFile& file) {
if (!file.ReadObject(type)) {
return false;
}
u32 code_size;
u32 code_size_b;
if (!file.ReadObject(code_size) || !file.ReadObject(code_size_b)) {
return false;
}
code.resize(code_size);
code_b.resize(code_size_b);
if (file.Read(code) != code_size) {
return false;
}
if (HasProgramA() && file.Read(code_b) != code_size_b) {
return false;
}
u8 is_texture_handler_size_known;
u32 texture_handler_size_value;
u32 num_keys;
u32 num_bound_samplers;
u32 num_separate_samplers;
u32 num_bindless_samplers;
if (!file.ReadObject(unique_identifier) || !file.ReadObject(bound_buffer) ||
!file.ReadObject(is_texture_handler_size_known) ||
!file.ReadObject(texture_handler_size_value) || !file.ReadObject(graphics_info) ||
!file.ReadObject(compute_info) || !file.ReadObject(num_keys) ||
!file.ReadObject(num_bound_samplers) || !file.ReadObject(num_separate_samplers) ||
!file.ReadObject(num_bindless_samplers)) {
return false;
}
if (is_texture_handler_size_known) {
texture_handler_size = texture_handler_size_value;
}
std::vector<ConstBufferKey> flat_keys(num_keys);
std::vector<BoundSamplerEntry> flat_bound_samplers(num_bound_samplers);
std::vector<SeparateSamplerEntry> flat_separate_samplers(num_separate_samplers);
std::vector<BindlessSamplerEntry> flat_bindless_samplers(num_bindless_samplers);
if (file.Read(flat_keys) != flat_keys.size() ||
file.Read(flat_bound_samplers) != flat_bound_samplers.size() ||
file.Read(flat_separate_samplers) != flat_separate_samplers.size() ||
file.Read(flat_bindless_samplers) != flat_bindless_samplers.size()) {
return false;
}
for (const auto& entry : flat_keys) {
keys.insert({{entry.cbuf, entry.offset}, entry.value});
}
for (const auto& entry : flat_bound_samplers) {
bound_samplers.emplace(entry.offset, entry.sampler);
}
for (const auto& entry : flat_separate_samplers) {
SeparateSamplerKey key;
key.buffers = {entry.cbuf1, entry.cbuf2};
key.offsets = {entry.offset1, entry.offset2};
separate_samplers.emplace(key, entry.sampler);
}
for (const auto& entry : flat_bindless_samplers) {
bindless_samplers.insert({{entry.cbuf, entry.offset}, entry.sampler});
}
return true;
}
bool ShaderDiskCacheEntry::Save(Common::FS::IOFile& file) const {
if (!file.WriteObject(static_cast<u32>(type)) ||
!file.WriteObject(static_cast<u32>(code.size())) ||
!file.WriteObject(static_cast<u32>(code_b.size()))) {
return false;
}
if (file.Write(code) != code.size()) {
return false;
}
if (HasProgramA() && file.Write(code_b) != code_b.size()) {
return false;
}
if (!file.WriteObject(unique_identifier) || !file.WriteObject(bound_buffer) ||
!file.WriteObject(static_cast<u8>(texture_handler_size.has_value())) ||
!file.WriteObject(texture_handler_size.value_or(0)) || !file.WriteObject(graphics_info) ||
!file.WriteObject(compute_info) || !file.WriteObject(static_cast<u32>(keys.size())) ||
!file.WriteObject(static_cast<u32>(bound_samplers.size())) ||
!file.WriteObject(static_cast<u32>(separate_samplers.size())) ||
!file.WriteObject(static_cast<u32>(bindless_samplers.size()))) {
return false;
}
std::vector<ConstBufferKey> flat_keys;
flat_keys.reserve(keys.size());
for (const auto& [address, value] : keys) {
flat_keys.push_back(ConstBufferKey{address.first, address.second, value});
}
std::vector<BoundSamplerEntry> flat_bound_samplers;
flat_bound_samplers.reserve(bound_samplers.size());
for (const auto& [address, sampler] : bound_samplers) {
flat_bound_samplers.push_back(BoundSamplerEntry{address, sampler});
}
std::vector<SeparateSamplerEntry> flat_separate_samplers;
flat_separate_samplers.reserve(separate_samplers.size());
for (const auto& [key, sampler] : separate_samplers) {
SeparateSamplerEntry entry;
std::tie(entry.cbuf1, entry.cbuf2) = key.buffers;
std::tie(entry.offset1, entry.offset2) = key.offsets;
entry.sampler = sampler;
flat_separate_samplers.push_back(entry);
}
std::vector<BindlessSamplerEntry> flat_bindless_samplers;
flat_bindless_samplers.reserve(bindless_samplers.size());
for (const auto& [address, sampler] : bindless_samplers) {
flat_bindless_samplers.push_back(
BindlessSamplerEntry{address.first, address.second, sampler});
}
return file.Write(flat_keys) == flat_keys.size() &&
file.Write(flat_bound_samplers) == flat_bound_samplers.size() &&
file.Write(flat_separate_samplers) == flat_separate_samplers.size() &&
file.Write(flat_bindless_samplers) == flat_bindless_samplers.size();
}
ShaderDiskCacheOpenGL::ShaderDiskCacheOpenGL() = default;
ShaderDiskCacheOpenGL::~ShaderDiskCacheOpenGL() = default;
void ShaderDiskCacheOpenGL::BindTitleID(u64 title_id_) {
title_id = title_id_;
}
std::optional<std::vector<ShaderDiskCacheEntry>> ShaderDiskCacheOpenGL::LoadTransferable() {
// Skip games without title id
const bool has_title_id = title_id != 0;
if (!Settings::values.use_disk_shader_cache.GetValue() || !has_title_id) {
return std::nullopt;
}
Common::FS::IOFile file{GetTransferablePath(), Common::FS::FileAccessMode::Read,
Common::FS::FileType::BinaryFile};
if (!file.IsOpen()) {
LOG_INFO(Render_OpenGL, "No transferable shader cache found");
is_usable = true;
return std::nullopt;
}
u32 version{};
if (!file.ReadObject(version)) {
LOG_ERROR(Render_OpenGL, "Failed to get transferable cache version, skipping it");
return std::nullopt;
}
if (version < NativeVersion) {
LOG_INFO(Render_OpenGL, "Transferable shader cache is old, removing");
file.Close();
InvalidateTransferable();
is_usable = true;
return std::nullopt;
}
if (version > NativeVersion) {
LOG_WARNING(Render_OpenGL, "Transferable shader cache was generated with a newer version "
"of the emulator, skipping");
return std::nullopt;
}
// Version is valid, load the shaders
std::vector<ShaderDiskCacheEntry> entries;
while (static_cast<u64>(file.Tell()) < file.GetSize()) {
ShaderDiskCacheEntry& entry = entries.emplace_back();
if (!entry.Load(file)) {
LOG_ERROR(Render_OpenGL, "Failed to load transferable raw entry, skipping");
return std::nullopt;
}
}
is_usable = true;
return {std::move(entries)};
}
std::vector<ShaderDiskCachePrecompiled> ShaderDiskCacheOpenGL::LoadPrecompiled() {
if (!is_usable) {
return {};
}
Common::FS::IOFile file{GetPrecompiledPath(), Common::FS::FileAccessMode::Read,
Common::FS::FileType::BinaryFile};
if (!file.IsOpen()) {
LOG_INFO(Render_OpenGL, "No precompiled shader cache found");
return {};
}
if (const auto result = LoadPrecompiledFile(file)) {
return *result;
}
LOG_INFO(Render_OpenGL, "Failed to load precompiled cache");
file.Close();
InvalidatePrecompiled();
return {};
}
std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::LoadPrecompiledFile(
Common::FS::IOFile& file) {
// Read compressed file from disk and decompress to virtual precompiled cache file
std::vector<u8> compressed(file.GetSize());
if (file.Read(compressed) != file.GetSize()) {
return std::nullopt;
}
const std::vector<u8> decompressed = Common::Compression::DecompressDataZSTD(compressed);
SaveArrayToPrecompiled(decompressed.data(), decompressed.size());
precompiled_cache_virtual_file_offset = 0;
ShaderCacheVersionHash file_hash{};
if (!LoadArrayFromPrecompiled(file_hash.data(), file_hash.size())) {
precompiled_cache_virtual_file_offset = 0;
return std::nullopt;
}
if (GetShaderCacheVersionHash() != file_hash) {
LOG_INFO(Render_OpenGL, "Precompiled cache is from another version of the emulator");
precompiled_cache_virtual_file_offset = 0;
return std::nullopt;
}
std::vector<ShaderDiskCachePrecompiled> entries;
while (precompiled_cache_virtual_file_offset < precompiled_cache_virtual_file.GetSize()) {
u32 binary_size;
auto& entry = entries.emplace_back();
if (!LoadObjectFromPrecompiled(entry.unique_identifier) ||
!LoadObjectFromPrecompiled(entry.binary_format) ||
!LoadObjectFromPrecompiled(binary_size)) {
return std::nullopt;
}
entry.binary.resize(binary_size);
if (!LoadArrayFromPrecompiled(entry.binary.data(), entry.binary.size())) {
return std::nullopt;
}
}
return entries;
}
void ShaderDiskCacheOpenGL::InvalidateTransferable() {
if (!Common::FS::RemoveFile(GetTransferablePath())) {
LOG_ERROR(Render_OpenGL, "Failed to invalidate transferable file={}",
Common::FS::PathToUTF8String(GetTransferablePath()));
}
InvalidatePrecompiled();
}
void ShaderDiskCacheOpenGL::InvalidatePrecompiled() {
// Clear virtaul precompiled cache file
precompiled_cache_virtual_file.Resize(0);
if (!Common::FS::RemoveFile(GetPrecompiledPath())) {
LOG_ERROR(Render_OpenGL, "Failed to invalidate precompiled file={}",
Common::FS::PathToUTF8String(GetPrecompiledPath()));
}
}
void ShaderDiskCacheOpenGL::SaveEntry(const ShaderDiskCacheEntry& entry) {
if (!is_usable) {
return;
}
const u64 id = entry.unique_identifier;
if (stored_transferable.contains(id)) {
// The shader already exists
return;
}
Common::FS::IOFile file = AppendTransferableFile();
if (!file.IsOpen()) {
return;
}
if (!entry.Save(file)) {
LOG_ERROR(Render_OpenGL, "Failed to save raw transferable cache entry, removing");
file.Close();
InvalidateTransferable();
return;
}
stored_transferable.insert(id);
}
void ShaderDiskCacheOpenGL::SavePrecompiled(u64 unique_identifier, GLuint program) {
if (!is_usable) {
return;
}
// TODO(Rodrigo): This is a design smell. I shouldn't be having to manually write the header
// when writing the dump. This should be done the moment I get access to write to the virtual
// file.
if (precompiled_cache_virtual_file.GetSize() == 0) {
SavePrecompiledHeaderToVirtualPrecompiledCache();
}
GLint binary_length;
glGetProgramiv(program, GL_PROGRAM_BINARY_LENGTH, &binary_length);
GLenum binary_format;
std::vector<u8> binary(binary_length);
glGetProgramBinary(program, binary_length, nullptr, &binary_format, binary.data());
if (!SaveObjectToPrecompiled(unique_identifier) || !SaveObjectToPrecompiled(binary_format) ||
!SaveObjectToPrecompiled(static_cast<u32>(binary.size())) ||
!SaveArrayToPrecompiled(binary.data(), binary.size())) {
LOG_ERROR(Render_OpenGL, "Failed to save binary program file in shader={:016X}, removing",
unique_identifier);
InvalidatePrecompiled();
}
}
Common::FS::IOFile ShaderDiskCacheOpenGL::AppendTransferableFile() const {
if (!EnsureDirectories()) {
return {};
}
const auto transferable_path{GetTransferablePath()};
const bool existed = Common::FS::Exists(transferable_path);
Common::FS::IOFile file{transferable_path, Common::FS::FileAccessMode::Append,
Common::FS::FileType::BinaryFile};
if (!file.IsOpen()) {
LOG_ERROR(Render_OpenGL, "Failed to open transferable cache in path={}",
Common::FS::PathToUTF8String(transferable_path));
return {};
}
if (!existed || file.GetSize() == 0) {
// If the file didn't exist, write its version
if (!file.WriteObject(NativeVersion)) {
LOG_ERROR(Render_OpenGL, "Failed to write transferable cache version in path={}",
Common::FS::PathToUTF8String(transferable_path));
return {};
}
}
return file;
}
void ShaderDiskCacheOpenGL::SavePrecompiledHeaderToVirtualPrecompiledCache() {
const auto hash{GetShaderCacheVersionHash()};
if (!SaveArrayToPrecompiled(hash.data(), hash.size())) {
LOG_ERROR(
Render_OpenGL,
"Failed to write precompiled cache version hash to virtual precompiled cache file");
}
}
void ShaderDiskCacheOpenGL::SaveVirtualPrecompiledFile() {
precompiled_cache_virtual_file_offset = 0;
const std::vector<u8> uncompressed = precompiled_cache_virtual_file.ReadAllBytes();
const std::vector<u8> compressed =
Common::Compression::CompressDataZSTDDefault(uncompressed.data(), uncompressed.size());
const auto precompiled_path = GetPrecompiledPath();
Common::FS::IOFile file{precompiled_path, Common::FS::FileAccessMode::Write,
Common::FS::FileType::BinaryFile};
if (!file.IsOpen()) {
LOG_ERROR(Render_OpenGL, "Failed to open precompiled cache in path={}",
Common::FS::PathToUTF8String(precompiled_path));
return;
}
if (file.Write(compressed) != compressed.size()) {
LOG_ERROR(Render_OpenGL, "Failed to write precompiled cache version in path={}",
Common::FS::PathToUTF8String(precompiled_path));
}
}
bool ShaderDiskCacheOpenGL::EnsureDirectories() const {
const auto CreateDir = [](const std::filesystem::path& dir) {
if (!Common::FS::CreateDir(dir)) {
LOG_ERROR(Render_OpenGL, "Failed to create directory={}",
Common::FS::PathToUTF8String(dir));
return false;
}
return true;
};
return CreateDir(Common::FS::GetYuzuPath(Common::FS::YuzuPath::ShaderDir)) &&
CreateDir(GetBaseDir()) && CreateDir(GetTransferableDir()) &&
CreateDir(GetPrecompiledDir());
}
std::filesystem::path ShaderDiskCacheOpenGL::GetTransferablePath() const {
return GetTransferableDir() / fmt::format("{}.bin", GetTitleID());
}
std::filesystem::path ShaderDiskCacheOpenGL::GetPrecompiledPath() const {
return GetPrecompiledDir() / fmt::format("{}.bin", GetTitleID());
}
std::filesystem::path ShaderDiskCacheOpenGL::GetTransferableDir() const {
return GetBaseDir() / "transferable";
}
std::filesystem::path ShaderDiskCacheOpenGL::GetPrecompiledDir() const {
return GetBaseDir() / "precompiled";
}
std::filesystem::path ShaderDiskCacheOpenGL::GetBaseDir() const {
return Common::FS::GetYuzuPath(Common::FS::YuzuPath::ShaderDir) / "opengl";
}
std::string ShaderDiskCacheOpenGL::GetTitleID() const {
return fmt::format("{:016X}", title_id);
}
} // namespace OpenGL

176
src/video_core/renderer_opengl/gl_shader_disk_cache.h
View File

@ -1,176 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <filesystem>
#include <optional>
#include <string>
#include <tuple>
#include <type_traits>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
#include <glad/glad.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "core/file_sys/vfs_vector.h"
#include "video_core/engines/shader_type.h"
#include "video_core/shader/registry.h"
namespace Common::FS {
class IOFile;
}
namespace OpenGL {
using ProgramCode = std::vector<u64>;
/// Describes a shader and how it's used by the guest GPU
struct ShaderDiskCacheEntry {
ShaderDiskCacheEntry();
~ShaderDiskCacheEntry();
bool Load(Common::FS::IOFile& file);
bool Save(Common::FS::IOFile& file) const;
bool HasProgramA() const {
return !code.empty() && !code_b.empty();
}
Tegra::Engines::ShaderType type{};
ProgramCode code;
ProgramCode code_b;
u64 unique_identifier = 0;
std::optional<u32> texture_handler_size;
u32 bound_buffer = 0;
VideoCommon::Shader::GraphicsInfo graphics_info;
VideoCommon::Shader::ComputeInfo compute_info;
VideoCommon::Shader::KeyMap keys;
VideoCommon::Shader::BoundSamplerMap bound_samplers;
VideoCommon::Shader::SeparateSamplerMap separate_samplers;
VideoCommon::Shader::BindlessSamplerMap bindless_samplers;
};
/// Contains an OpenGL dumped binary program
struct ShaderDiskCachePrecompiled {
u64 unique_identifier = 0;
GLenum binary_format = 0;
std::vector<u8> binary;
};
class ShaderDiskCacheOpenGL {
public:
explicit ShaderDiskCacheOpenGL();
~ShaderDiskCacheOpenGL();
/// Binds a title ID for all future operations.
void BindTitleID(u64 title_id);
/// Loads transferable cache. If file has a old version or on failure, it deletes the file.
std::optional<std::vector<ShaderDiskCacheEntry>> LoadTransferable();
/// Loads current game's precompiled cache. Invalidates on failure.
std::vector<ShaderDiskCachePrecompiled> LoadPrecompiled();
/// Removes the transferable (and precompiled) cache file.
void InvalidateTransferable();
/// Removes the precompiled cache file and clears virtual precompiled cache file.
void InvalidatePrecompiled();
/// Saves a raw dump to the transferable file. Checks for collisions.
void SaveEntry(const ShaderDiskCacheEntry& entry);
/// Saves a dump entry to the precompiled file. Does not check for collisions.
void SavePrecompiled(u64 unique_identifier, GLuint program);
/// Serializes virtual precompiled shader cache file to real file
void SaveVirtualPrecompiledFile();
private:
/// Loads the transferable cache. Returns empty on failure.
std::optional<std::vector<ShaderDiskCachePrecompiled>> LoadPrecompiledFile(
Common::FS::IOFile& file);
/// Opens current game's transferable file and write it's header if it doesn't exist
Common::FS::IOFile AppendTransferableFile() const;
/// Save precompiled header to precompiled_cache_in_memory
void SavePrecompiledHeaderToVirtualPrecompiledCache();
/// Create shader disk cache directories. Returns true on success.
bool EnsureDirectories() const;
/// Gets current game's transferable file path
std::filesystem::path GetTransferablePath() const;
/// Gets current game's precompiled file path
std::filesystem::path GetPrecompiledPath() const;
/// Get user's transferable directory path
std::filesystem::path GetTransferableDir() const;
/// Get user's precompiled directory path
std::filesystem::path GetPrecompiledDir() const;
/// Get user's shader directory path
std::filesystem::path GetBaseDir() const;
/// Get current game's title id
std::string GetTitleID() const;
template <typename T>
bool SaveArrayToPrecompiled(const T* data, std::size_t length) {
const std::size_t write_length = precompiled_cache_virtual_file.WriteArray(
data, length, precompiled_cache_virtual_file_offset);
precompiled_cache_virtual_file_offset += write_length;
return write_length == sizeof(T) * length;
}
template <typename T>
bool LoadArrayFromPrecompiled(T* data, std::size_t length) {
const std::size_t read_length = precompiled_cache_virtual_file.ReadArray(
data, length, precompiled_cache_virtual_file_offset);
precompiled_cache_virtual_file_offset += read_length;
return read_length == sizeof(T) * length;
}
template <typename T>
bool SaveObjectToPrecompiled(const T& object) {
return SaveArrayToPrecompiled(&object, 1);
}
bool SaveObjectToPrecompiled(bool object) {
const auto value = static_cast<u8>(object);
return SaveArrayToPrecompiled(&value, 1);
}
template <typename T>
bool LoadObjectFromPrecompiled(T& object) {
return LoadArrayFromPrecompiled(&object, 1);
}
// Stores whole precompiled cache which will be read from or saved to the precompiled chache
// file
FileSys::VectorVfsFile precompiled_cache_virtual_file;
// Stores the current offset of the precompiled cache file for IO purposes
std::size_t precompiled_cache_virtual_file_offset = 0;
// Stored transferable shaders
std::unordered_set<u64> stored_transferable;
/// Title ID to operate on
u64 title_id = 0;
// The cache has been loaded at boot
bool is_usable = false;
};
} // namespace OpenGL

146
src/video_core/renderer_opengl/gl_shader_manager.cpp
View File

@ -1,149 +1,3 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <glad/glad.h>
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_shader_manager.h"
namespace OpenGL {
namespace {
void BindProgram(GLenum stage, GLuint current, GLuint old, bool& enabled) {
if (current == old) {
return;
}
if (current == 0) {
if (enabled) {
enabled = false;
glDisable(stage);
}
return;
}
if (!enabled) {
enabled = true;
glEnable(stage);
}
glBindProgramARB(stage, current);
}
} // Anonymous namespace
ProgramManager::ProgramManager(const Device& device)
: use_assembly_programs{device.UseAssemblyShaders()} {
if (use_assembly_programs) {
glEnable(GL_COMPUTE_PROGRAM_NV);
} else {
graphics_pipeline.Create();
glBindProgramPipeline(graphics_pipeline.handle);
}
}
ProgramManager::~ProgramManager() = default;
void ProgramManager::BindCompute(GLuint program) {
if (use_assembly_programs) {
glBindProgramARB(GL_COMPUTE_PROGRAM_NV, program);
} else {
is_graphics_bound = false;
glUseProgram(program);
}
}
void ProgramManager::BindGraphicsPipeline() {
if (!use_assembly_programs) {
UpdateSourcePrograms();
}
}
void ProgramManager::BindHostPipeline(GLuint pipeline) {
if (use_assembly_programs) {
if (geometry_enabled) {
geometry_enabled = false;
old_state.geometry = 0;
glDisable(GL_GEOMETRY_PROGRAM_NV);
}
} else {
if (!is_graphics_bound) {
glUseProgram(0);
}
}
glBindProgramPipeline(pipeline);
}
void ProgramManager::RestoreGuestPipeline() {
if (use_assembly_programs) {
glBindProgramPipeline(0);
} else {
glBindProgramPipeline(graphics_pipeline.handle);
}
}
void ProgramManager::BindHostCompute(GLuint program) {
if (use_assembly_programs) {
glDisable(GL_COMPUTE_PROGRAM_NV);
}
glUseProgram(program);
is_graphics_bound = false;
}
void ProgramManager::RestoreGuestCompute() {
if (use_assembly_programs) {
glEnable(GL_COMPUTE_PROGRAM_NV);
glUseProgram(0);
}
}
void ProgramManager::UseVertexShader(GLuint program) {
if (use_assembly_programs) {
BindProgram(GL_VERTEX_PROGRAM_NV, program, current_state.vertex, vertex_enabled);
}
current_state.vertex = program;
}
void ProgramManager::UseGeometryShader(GLuint program) {
if (use_assembly_programs) {
BindProgram(GL_GEOMETRY_PROGRAM_NV, program, current_state.vertex, geometry_enabled);
}
current_state.geometry = program;
}
void ProgramManager::UseFragmentShader(GLuint program) {
if (use_assembly_programs) {
BindProgram(GL_FRAGMENT_PROGRAM_NV, program, current_state.vertex, fragment_enabled);
}
current_state.fragment = program;
}
void ProgramManager::UpdateSourcePrograms() {
if (!is_graphics_bound) {
is_graphics_bound = true;
glUseProgram(0);
}
const GLuint handle = graphics_pipeline.handle;
const auto update_state = [handle](GLenum stage, GLuint current, GLuint old) {
if (current == old) {
return;
}
glUseProgramStages(handle, stage, current);
};
update_state(GL_VERTEX_SHADER_BIT, current_state.vertex, old_state.vertex);
update_state(GL_GEOMETRY_SHADER_BIT, current_state.geometry, old_state.geometry);
update_state(GL_FRAGMENT_SHADER_BIT, current_state.fragment, old_state.fragment);
old_state = current_state;
}
void MaxwellUniformData::SetFromRegs(const Tegra::Engines::Maxwell3D& maxwell) {
const auto& regs = maxwell.regs;
// Y_NEGATE controls what value S2R returns for the Y_DIRECTION system value.
y_direction = regs.screen_y_control.y_negate == 0 ? 1.0f : -1.0f;
}
} // namespace OpenGL

177
src/video_core/renderer_opengl/gl_shader_manager.h
View File

@ -4,79 +4,142 @@
#pragma once
#include <cstddef>
#include <array>
#include <span>
#include <glad/glad.h>
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/maxwell_to_gl.h"
namespace OpenGL {
class Device;
/// Uniform structure for the Uniform Buffer Object, all vectors must be 16-byte aligned
/// @note Always keep a vec4 at the end. The GL spec is not clear whether the alignment at
/// the end of a uniform block is included in UNIFORM_BLOCK_DATA_SIZE or not.
/// Not following that rule will cause problems on some AMD drivers.
struct alignas(16) MaxwellUniformData {
void SetFromRegs(const Tegra::Engines::Maxwell3D& maxwell);
GLfloat y_direction;
};
static_assert(sizeof(MaxwellUniformData) == 16, "MaxwellUniformData structure size is incorrect");
static_assert(sizeof(MaxwellUniformData) < 16384,
"MaxwellUniformData structure must be less than 16kb as per the OpenGL spec");
class ProgramManager {
public:
explicit ProgramManager(const Device& device);
~ProgramManager();
static constexpr size_t NUM_STAGES = 5;
/// Binds a compute program
void BindCompute(GLuint program);
/// Updates bound programs.
void BindGraphicsPipeline();
/// Binds an OpenGL pipeline object unsynchronized with the guest state.
void BindHostPipeline(GLuint pipeline);
/// Rewinds BindHostPipeline state changes.
void RestoreGuestPipeline();
/// Binds an OpenGL GLSL program object unsynchronized with the guest state.
void BindHostCompute(GLuint program);
/// Rewinds BindHostCompute state changes.
void RestoreGuestCompute();
void UseVertexShader(GLuint program);
void UseGeometryShader(GLuint program);
void UseFragmentShader(GLuint program);
private:
struct PipelineState {
GLuint vertex = 0;
GLuint geometry = 0;
GLuint fragment = 0;
static constexpr std::array ASSEMBLY_PROGRAM_ENUMS{
GL_VERTEX_PROGRAM_NV, GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV,
GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV,
};
/// Update GLSL programs.
void UpdateSourcePrograms();
public:
explicit ProgramManager(const Device& device) {
glCreateProgramPipelines(1, &pipeline.handle);
if (device.UseAssemblyShaders()) {
glEnable(GL_COMPUTE_PROGRAM_NV);
}
}
OGLPipeline graphics_pipeline;
void BindComputeProgram(GLuint program) {
glUseProgram(program);
is_compute_bound = true;
}
PipelineState current_state;
PipelineState old_state;
void BindComputeAssemblyProgram(GLuint program) {
if (current_assembly_compute_program != program) {
current_assembly_compute_program = program;
glBindProgramARB(GL_COMPUTE_PROGRAM_NV, program);
}
UnbindPipeline();
}
bool use_assembly_programs = false;
void BindSourcePrograms(std::span<const OGLProgram, NUM_STAGES> programs) {
static constexpr std::array<GLenum, 5> stage_enums{
GL_VERTEX_SHADER_BIT, GL_TESS_CONTROL_SHADER_BIT, GL_TESS_EVALUATION_SHADER_BIT,
GL_GEOMETRY_SHADER_BIT, GL_FRAGMENT_SHADER_BIT,
};
for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
if (current_programs[stage] != programs[stage].handle) {
current_programs[stage] = programs[stage].handle;
glUseProgramStages(pipeline.handle, stage_enums[stage], programs[stage].handle);
}
}
BindPipeline();
}
bool is_graphics_bound = true;
void BindPresentPrograms(GLuint vertex, GLuint fragment) {
if (current_programs[0] != vertex) {
current_programs[0] = vertex;
glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, vertex);
}
if (current_programs[4] != fragment) {
current_programs[4] = fragment;
glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, fragment);
}
glUseProgramStages(
pipeline.handle,
GL_TESS_CONTROL_SHADER_BIT | GL_TESS_EVALUATION_SHADER_BIT | GL_GEOMETRY_SHADER_BIT, 0);
current_programs[1] = 0;
current_programs[2] = 0;
current_programs[3] = 0;
bool vertex_enabled = false;
bool geometry_enabled = false;
bool fragment_enabled = false;
if (current_stage_mask != 0) {
current_stage_mask = 0;
for (const GLenum program_type : ASSEMBLY_PROGRAM_ENUMS) {
glDisable(program_type);
}
}
BindPipeline();
}
void BindAssemblyPrograms(std::span<const OGLAssemblyProgram, NUM_STAGES> programs,
u32 stage_mask) {
const u32 changed_mask = current_stage_mask ^ stage_mask;
current_stage_mask = stage_mask;
if (changed_mask != 0) {
for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
if (((changed_mask >> stage) & 1) != 0) {
if (((stage_mask >> stage) & 1) != 0) {
glEnable(ASSEMBLY_PROGRAM_ENUMS[stage]);
} else {
glDisable(ASSEMBLY_PROGRAM_ENUMS[stage]);
}
}
}
}
for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
if (current_programs[stage] != programs[stage].handle) {
current_programs[stage] = programs[stage].handle;
glBindProgramARB(ASSEMBLY_PROGRAM_ENUMS[stage], programs[stage].handle);
}
}
UnbindPipeline();
}
void RestoreGuestCompute() {}
private:
void BindPipeline() {
if (!is_pipeline_bound) {
is_pipeline_bound = true;
glBindProgramPipeline(pipeline.handle);
}
UnbindCompute();
}
void UnbindPipeline() {
if (is_pipeline_bound) {
is_pipeline_bound = false;
glBindProgramPipeline(0);
}
UnbindCompute();
}
void UnbindCompute() {
if (is_compute_bound) {
is_compute_bound = false;
glUseProgram(0);
}
}
OGLPipeline pipeline;
bool is_pipeline_bound{};
bool is_compute_bound{};
u32 current_stage_mask = 0;
std::array<GLuint, NUM_STAGES> current_programs{};
GLuint current_assembly_compute_program = 0;
};
} // namespace OpenGL

125
src/video_core/renderer_opengl/gl_shader_util.cpp
View File

@ -5,57 +5,108 @@
#include <string_view>
#include <vector>
#include <glad/glad.h>
#include "common/assert.h"
#include "common/logging/log.h"
#include "common/settings.h"
#include "video_core/renderer_opengl/gl_shader_util.h"
namespace OpenGL::GLShader {
namespace OpenGL {
namespace {
std::string_view StageDebugName(GLenum type) {
switch (type) {
case GL_VERTEX_SHADER:
return "vertex";
case GL_GEOMETRY_SHADER:
return "geometry";
case GL_FRAGMENT_SHADER:
return "fragment";
case GL_COMPUTE_SHADER:
return "compute";
static OGLProgram LinkSeparableProgram(GLuint shader) {
OGLProgram program;
program.handle = glCreateProgram();
glProgramParameteri(program.handle, GL_PROGRAM_SEPARABLE, GL_TRUE);
glAttachShader(program.handle, shader);
glLinkProgram(program.handle);
if (!Settings::values.renderer_debug) {
return program;
}
UNIMPLEMENTED();
return "unknown";
GLint link_status{};
glGetProgramiv(program.handle, GL_LINK_STATUS, &link_status);
GLint log_length{};
glGetProgramiv(program.handle, GL_INFO_LOG_LENGTH, &log_length);
if (log_length == 0) {
return program;
}
std::string log(log_length, 0);
glGetProgramInfoLog(program.handle, log_length, nullptr, log.data());
if (link_status == GL_FALSE) {
LOG_ERROR(Render_OpenGL, "{}", log);
} else {
LOG_WARNING(Render_OpenGL, "{}", log);
}
return program;
}
} // Anonymous namespace
static void LogShader(GLuint shader, std::string_view code = {}) {
GLint shader_status{};
glGetShaderiv(shader, GL_COMPILE_STATUS, &shader_status);
if (shader_status == GL_FALSE) {
LOG_ERROR(Render_OpenGL, "Failed to build shader");
}
GLint log_length{};
glGetShaderiv(shader, GL_INFO_LOG_LENGTH, &log_length);
if (log_length == 0) {
return;
}
std::string log(log_length, 0);
glGetShaderInfoLog(shader, log_length, nullptr, log.data());
if (shader_status == GL_FALSE) {
LOG_ERROR(Render_OpenGL, "{}", log);
if (!code.empty()) {
LOG_INFO(Render_OpenGL, "\n{}", code);
}
} else {
LOG_WARNING(Render_OpenGL, "{}", log);
}
}
GLuint LoadShader(std::string_view source, GLenum type) {
const std::string_view debug_type = StageDebugName(type);
const GLuint shader_id = glCreateShader(type);
OGLProgram CreateProgram(std::string_view code, GLenum stage) {
OGLShader shader;
shader.handle = glCreateShader(stage);
const GLchar* source_string = source.data();
const GLint source_length = static_cast<GLint>(source.size());
const GLint length = static_cast<GLint>(code.size());
const GLchar* const code_ptr = code.data();
glShaderSource(shader.handle, 1, &code_ptr, &length);
glCompileShader(shader.handle);
if (Settings::values.renderer_debug) {
LogShader(shader.handle, code);
}
return LinkSeparableProgram(shader.handle);
}
glShaderSource(shader_id, 1, &source_string, &source_length);
LOG_DEBUG(Render_OpenGL, "Compiling {} shader...", debug_type);
glCompileShader(shader_id);
OGLProgram CreateProgram(std::span<const u32> code, GLenum stage) {
OGLShader shader;
shader.handle = glCreateShader(stage);
GLint result = GL_FALSE;
GLint info_log_length;
glGetShaderiv(shader_id, GL_COMPILE_STATUS, &result);
glGetShaderiv(shader_id, GL_INFO_LOG_LENGTH, &info_log_length);
glShaderBinary(1, &shader.handle, GL_SHADER_BINARY_FORMAT_SPIR_V_ARB, code.data(),
static_cast<GLsizei>(code.size_bytes()));
glSpecializeShader(shader.handle, "main", 0, nullptr, nullptr);
if (Settings::values.renderer_debug) {
LogShader(shader.handle);
}
return LinkSeparableProgram(shader.handle);
}
if (info_log_length > 1) {
std::string shader_error(info_log_length, ' ');
glGetShaderInfoLog(shader_id, info_log_length, nullptr, &shader_error[0]);
if (result == GL_TRUE) {
LOG_DEBUG(Render_OpenGL, "{}", shader_error);
} else {
LOG_ERROR(Render_OpenGL, "Error compiling {} shader:\n{}", debug_type, shader_error);
OGLAssemblyProgram CompileProgram(std::string_view code, GLenum target) {
OGLAssemblyProgram program;
glGenProgramsARB(1, &program.handle);
glNamedProgramStringEXT(program.handle, target, GL_PROGRAM_FORMAT_ASCII_ARB,
static_cast<GLsizei>(code.size()), code.data());
if (Settings::values.renderer_debug) {
const auto err = reinterpret_cast<const char*>(glGetString(GL_PROGRAM_ERROR_STRING_NV));
if (err && *err) {
if (std::strstr(err, "error")) {
LOG_CRITICAL(Render_OpenGL, "\n{}", err);
LOG_INFO(Render_OpenGL, "\n{}", code);
} else {
LOG_WARNING(Render_OpenGL, "\n{}", err);
}
}
}
return shader_id;
return program;
}
} // namespace OpenGL::GLShader
} // namespace OpenGL

89
src/video_core/renderer_opengl/gl_shader_util.h
View File

@ -4,92 +4,23 @@
#pragma once
#include <span>
#include <string>
#include <string_view>
#include <vector>
#include <glad/glad.h>
#include "common/assert.h"
#include "common/logging/log.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
namespace OpenGL::GLShader {
namespace OpenGL {
/**
* Utility function to log the source code of a list of shaders.
* @param shaders The OpenGL shaders whose source we will print.
*/
template <typename... T>
void LogShaderSource(T... shaders) {
auto shader_list = {shaders...};
OGLProgram CreateProgram(std::string_view code, GLenum stage);
for (const auto& shader : shader_list) {
if (shader == 0)
continue;
OGLProgram CreateProgram(std::span<const u32> code, GLenum stage);
GLint source_length;
glGetShaderiv(shader, GL_SHADER_SOURCE_LENGTH, &source_length);
OGLAssemblyProgram CompileProgram(std::string_view code, GLenum target);
std::string source(source_length, ' ');
glGetShaderSource(shader, source_length, nullptr, &source[0]);
LOG_INFO(Render_OpenGL, "Shader source {}", source);
}
}
/**
* Utility function to create and compile an OpenGL GLSL shader
* @param source String of the GLSL shader program
* @param type Type of the shader (GL_VERTEX_SHADER, GL_GEOMETRY_SHADER or GL_FRAGMENT_SHADER)
*/
GLuint LoadShader(std::string_view source, GLenum type);
/**
* Utility function to create and compile an OpenGL GLSL shader program (vertex + fragment shader)
* @param separable_program whether to create a separable program
* @param shaders ID of shaders to attach to the program
* @returns Handle of the newly created OpenGL program object
*/
template <typename... T>
GLuint LoadProgram(bool separable_program, bool hint_retrievable, T... shaders) {
// Link the program
LOG_DEBUG(Render_OpenGL, "Linking program...");
GLuint program_id = glCreateProgram();
((shaders == 0 ? (void)0 : glAttachShader(program_id, shaders)), ...);
if (separable_program) {
glProgramParameteri(program_id, GL_PROGRAM_SEPARABLE, GL_TRUE);
}
if (hint_retrievable) {
glProgramParameteri(program_id, GL_PROGRAM_BINARY_RETRIEVABLE_HINT, GL_TRUE);
}
glLinkProgram(program_id);
// Check the program
GLint result = GL_FALSE;
GLint info_log_length;
glGetProgramiv(program_id, GL_LINK_STATUS, &result);
glGetProgramiv(program_id, GL_INFO_LOG_LENGTH, &info_log_length);
if (info_log_length > 1) {
std::string program_error(info_log_length, ' ');
glGetProgramInfoLog(program_id, info_log_length, nullptr, &program_error[0]);
if (result == GL_TRUE) {
LOG_DEBUG(Render_OpenGL, "{}", program_error);
} else {
LOG_ERROR(Render_OpenGL, "Error linking shader:\n{}", program_error);
}
}
if (result == GL_FALSE) {
// There was a problem linking the shader, print the source for debugging purposes.
LogShaderSource(shaders...);
}
ASSERT_MSG(result == GL_TRUE, "Shader not linked");
((shaders == 0 ? (void)0 : glDetachShader(program_id, shaders)), ...);
return program_id;
}
} // namespace OpenGL::GLShader
} // namespace OpenGL

6
src/video_core/renderer_opengl/gl_state_tracker.cpp
View File

@ -83,11 +83,6 @@ void SetupDirtyScissors(Tables& tables) {
FillBlock(tables[1], OFF(scissor_test), NUM(scissor_test), Scissors);
}
void SetupDirtyShaders(Tables& tables) {
FillBlock(tables[0], OFF(shader_config[0]), NUM(shader_config[0]) * Regs::MaxShaderProgram,
Shaders);
}
void SetupDirtyPolygonModes(Tables& tables) {
tables[0][OFF(polygon_mode_front)] = PolygonModeFront;
tables[0][OFF(polygon_mode_back)] = PolygonModeBack;
@ -217,7 +212,6 @@ StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags}
SetupDirtyScissors(tables);
SetupDirtyVertexInstances(tables);
SetupDirtyVertexFormat(tables);
SetupDirtyShaders(tables);
SetupDirtyPolygonModes(tables);
SetupDirtyDepthTest(tables);
SetupDirtyStencilTest(tables);

1
src/video_core/renderer_opengl/gl_state_tracker.h
View File

@ -52,7 +52,6 @@ enum : u8 {
BlendState0,
BlendState7 = BlendState0 + 7,
Shaders,
ClipDistances,
PolygonModes,

361
src/video_core/renderer_opengl/gl_texture_cache.cpp
View File

@ -24,9 +24,7 @@
#include "video_core/textures/decoders.h"
namespace OpenGL {
namespace {
using Tegra::Texture::SwizzleSource;
using Tegra::Texture::TextureMipmapFilter;
using Tegra::Texture::TextureType;
@ -59,107 +57,6 @@ struct CopyRegion {
GLsizei depth;
};
struct FormatTuple {
GLenum internal_format;
GLenum format = GL_NONE;
GLenum type = GL_NONE;
};
constexpr std::array<FormatTuple, MaxPixelFormat> FORMAT_TABLE = {{
{GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // A8B8G8R8_UNORM
{GL_RGBA8_SNORM, GL_RGBA, GL_BYTE}, // A8B8G8R8_SNORM
{GL_RGBA8I, GL_RGBA_INTEGER, GL_BYTE}, // A8B8G8R8_SINT
{GL_RGBA8UI, GL_RGBA_INTEGER, GL_UNSIGNED_BYTE}, // A8B8G8R8_UINT
{GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5}, // R5G6B5_UNORM
{GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV}, // B5G6R5_UNORM
{GL_RGB5_A1, GL_BGRA, GL_UNSIGNED_SHORT_1_5_5_5_REV}, // A1R5G5B5_UNORM
{GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10_UNORM
{GL_RGB10_A2UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10_UINT
{GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV}, // A1B5G5R5_UNORM
{GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // R8_UNORM
{GL_R8_SNORM, GL_RED, GL_BYTE}, // R8_SNORM
{GL_R8I, GL_RED_INTEGER, GL_BYTE}, // R8_SINT
{GL_R8UI, GL_RED_INTEGER, GL_UNSIGNED_BYTE}, // R8_UINT
{GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT}, // R16G16B16A16_FLOAT
{GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT}, // R16G16B16A16_UNORM
{GL_RGBA16_SNORM, GL_RGBA, GL_SHORT}, // R16G16B16A16_SNORM
{GL_RGBA16I, GL_RGBA_INTEGER, GL_SHORT}, // R16G16B16A16_SINT
{GL_RGBA16UI, GL_RGBA_INTEGER, GL_UNSIGNED_SHORT}, // R16G16B16A16_UINT
{GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV}, // B10G11R11_FLOAT
{GL_RGBA32UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT}, // R32G32B32A32_UINT
{GL_COMPRESSED_RGBA_S3TC_DXT1_EXT}, // BC1_RGBA_UNORM
{GL_COMPRESSED_RGBA_S3TC_DXT3_EXT}, // BC2_UNORM
{GL_COMPRESSED_RGBA_S3TC_DXT5_EXT}, // BC3_UNORM
{GL_COMPRESSED_RED_RGTC1}, // BC4_UNORM
{GL_COMPRESSED_SIGNED_RED_RGTC1}, // BC4_SNORM
{GL_COMPRESSED_RG_RGTC2}, // BC5_UNORM
{GL_COMPRESSED_SIGNED_RG_RGTC2}, // BC5_SNORM
{GL_COMPRESSED_RGBA_BPTC_UNORM}, // BC7_UNORM
{GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT}, // BC6H_UFLOAT
{GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT}, // BC6H_SFLOAT
{GL_COMPRESSED_RGBA_ASTC_4x4_KHR}, // ASTC_2D_4X4_UNORM
{GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE}, // B8G8R8A8_UNORM
{GL_RGBA32F, GL_RGBA, GL_FLOAT}, // R32G32B32A32_FLOAT
{GL_RGBA32I, GL_RGBA_INTEGER, GL_INT}, // R32G32B32A32_SINT
{GL_RG32F, GL_RG, GL_FLOAT}, // R32G32_FLOAT
{GL_RG32I, GL_RG_INTEGER, GL_INT}, // R32G32_SINT
{GL_R32F, GL_RED, GL_FLOAT}, // R32_FLOAT
{GL_R16F, GL_RED, GL_HALF_FLOAT}, // R16_FLOAT
{GL_R16, GL_RED, GL_UNSIGNED_SHORT}, // R16_UNORM
{GL_R16_SNORM, GL_RED, GL_SHORT}, // R16_SNORM
{GL_R16UI, GL_RED_INTEGER, GL_UNSIGNED_SHORT}, // R16_UINT
{GL_R16I, GL_RED_INTEGER, GL_SHORT}, // R16_SINT
{GL_RG16, GL_RG, GL_UNSIGNED_SHORT}, // R16G16_UNORM
{GL_RG16F, GL_RG, GL_HALF_FLOAT}, // R16G16_FLOAT
{GL_RG16UI, GL_RG_INTEGER, GL_UNSIGNED_SHORT}, // R16G16_UINT
{GL_RG16I, GL_RG_INTEGER, GL_SHORT}, // R16G16_SINT
{GL_RG16_SNORM, GL_RG, GL_SHORT}, // R16G16_SNORM
{GL_RGB32F, GL_RGB, GL_FLOAT}, // R32G32B32_FLOAT
{GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // A8B8G8R8_SRGB
{GL_RG8, GL_RG, GL_UNSIGNED_BYTE}, // R8G8_UNORM
{GL_RG8_SNORM, GL_RG, GL_BYTE}, // R8G8_SNORM
{GL_RG8I, GL_RG_INTEGER, GL_BYTE}, // R8G8_SINT
{GL_RG8UI, GL_RG_INTEGER, GL_UNSIGNED_BYTE}, // R8G8_UINT
{GL_RG32UI, GL_RG_INTEGER, GL_UNSIGNED_INT}, // R32G32_UINT
{GL_RGB16F, GL_RGBA, GL_HALF_FLOAT}, // R16G16B16X16_FLOAT
{GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT}, // R32_UINT
{GL_R32I, GL_RED_INTEGER, GL_INT}, // R32_SINT
{GL_COMPRESSED_RGBA_ASTC_8x8_KHR}, // ASTC_2D_8X8_UNORM
{GL_COMPRESSED_RGBA_ASTC_8x5_KHR}, // ASTC_2D_8X5_UNORM
{GL_COMPRESSED_RGBA_ASTC_5x4_KHR}, // ASTC_2D_5X4_UNORM
{GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE}, // B8G8R8A8_SRGB
{GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT}, // BC1_RGBA_SRGB
{GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT}, // BC2_SRGB
{GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT}, // BC3_SRGB
{GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM}, // BC7_SRGB
{GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV}, // A4B4G4R4_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_4x4_KHR}, // ASTC_2D_4X4_SRGB
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x8_KHR}, // ASTC_2D_8X8_SRGB
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x5_KHR}, // ASTC_2D_8X5_SRGB
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x4_KHR}, // ASTC_2D_5X4_SRGB
{GL_COMPRESSED_RGBA_ASTC_5x5_KHR}, // ASTC_2D_5X5_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x5_KHR}, // ASTC_2D_5X5_SRGB
{GL_COMPRESSED_RGBA_ASTC_10x8_KHR}, // ASTC_2D_10X8_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR}, // ASTC_2D_10X8_SRGB
{GL_COMPRESSED_RGBA_ASTC_6x6_KHR}, // ASTC_2D_6X6_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x6_KHR}, // ASTC_2D_6X6_SRGB
{GL_COMPRESSED_RGBA_ASTC_10x10_KHR}, // ASTC_2D_10X10_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR}, // ASTC_2D_10X10_SRGB
{GL_COMPRESSED_RGBA_ASTC_12x12_KHR}, // ASTC_2D_12X12_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR}, // ASTC_2D_12X12_SRGB
{GL_COMPRESSED_RGBA_ASTC_8x6_KHR}, // ASTC_2D_8X6_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x6_KHR}, // ASTC_2D_8X6_SRGB
{GL_COMPRESSED_RGBA_ASTC_6x5_KHR}, // ASTC_2D_6X5_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x5_KHR}, // ASTC_2D_6X5_SRGB
{GL_RGB9_E5, GL_RGB, GL_UNSIGNED_INT_5_9_9_9_REV}, // E5B9G9R9_FLOAT
{GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // D32_FLOAT
{GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16_UNORM
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24_UNORM_S8_UINT
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // S8_UINT_D24_UNORM
{GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL,
GL_FLOAT_32_UNSIGNED_INT_24_8_REV}, // D32_FLOAT_S8_UINT
}};
constexpr std::array ACCELERATED_FORMATS{
GL_RGBA32F, GL_RGBA16F, GL_RG32F, GL_RG16F, GL_R11F_G11F_B10F, GL_R32F,
GL_R16F, GL_RGBA32UI, GL_RGBA16UI, GL_RGB10_A2UI, GL_RGBA8UI, GL_RG32UI,
@ -170,11 +67,6 @@ constexpr std::array ACCELERATED_FORMATS{
GL_RG8_SNORM, GL_R16_SNORM, GL_R8_SNORM,
};
const FormatTuple& GetFormatTuple(PixelFormat pixel_format) {
ASSERT(static_cast<size_t>(pixel_format) < FORMAT_TABLE.size());
return FORMAT_TABLE[static_cast<size_t>(pixel_format)];
}
GLenum ImageTarget(const VideoCommon::ImageInfo& info) {
switch (info.type) {
case ImageType::e1D:
@ -195,26 +87,24 @@ GLenum ImageTarget(const VideoCommon::ImageInfo& info) {
return GL_NONE;
}
GLenum ImageTarget(ImageViewType type, int num_samples = 1) {
GLenum ImageTarget(Shader::TextureType type, int num_samples = 1) {
const bool is_multisampled = num_samples > 1;
switch (type) {
case ImageViewType::e1D:
case Shader::TextureType::Color1D:
return GL_TEXTURE_1D;
case ImageViewType::e2D:
case Shader::TextureType::Color2D:
return is_multisampled ? GL_TEXTURE_2D_MULTISAMPLE : GL_TEXTURE_2D;
case ImageViewType::Cube:
case Shader::TextureType::ColorCube:
return GL_TEXTURE_CUBE_MAP;
case ImageViewType::e3D:
case Shader::TextureType::Color3D:
return GL_TEXTURE_3D;
case ImageViewType::e1DArray:
case Shader::TextureType::ColorArray1D:
return GL_TEXTURE_1D_ARRAY;
case ImageViewType::e2DArray:
case Shader::TextureType::ColorArray2D:
return is_multisampled ? GL_TEXTURE_2D_MULTISAMPLE_ARRAY : GL_TEXTURE_2D_ARRAY;
case ImageViewType::CubeArray:
case Shader::TextureType::ColorArrayCube:
return GL_TEXTURE_CUBE_MAP_ARRAY;
case ImageViewType::Rect:
return GL_TEXTURE_RECTANGLE;
case ImageViewType::Buffer:
case Shader::TextureType::Buffer:
return GL_TEXTURE_BUFFER;
}
UNREACHABLE_MSG("Invalid image view type={}", type);
@ -322,7 +212,7 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4
default:
return false;
}
const GLenum internal_format = GetFormatTuple(info.format).internal_format;
const GLenum internal_format = MaxwellToGL::GetFormatTuple(info.format).internal_format;
const auto& format_info = runtime.FormatInfo(info.type, internal_format);
if (format_info.is_compressed) {
return false;
@ -414,11 +304,10 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4
void AttachTexture(GLuint fbo, GLenum attachment, const ImageView* image_view) {
if (False(image_view->flags & VideoCommon::ImageViewFlagBits::Slice)) {
const GLuint texture = image_view->DefaultHandle();
glNamedFramebufferTexture(fbo, attachment, texture, 0);
glNamedFramebufferTexture(fbo, attachment, image_view->DefaultHandle(), 0);
return;
}
const GLuint texture = image_view->Handle(ImageViewType::e3D);
const GLuint texture = image_view->Handle(Shader::TextureType::Color3D);
if (image_view->range.extent.layers > 1) {
// TODO: OpenGL doesn't support rendering to a fixed number of slices
glNamedFramebufferTexture(fbo, attachment, texture, 0);
@ -439,6 +328,28 @@ void AttachTexture(GLuint fbo, GLenum attachment, const ImageView* image_view) {
}
}
[[nodiscard]] GLenum ShaderFormat(Shader::ImageFormat format) {
switch (format) {
case Shader::ImageFormat::Typeless:
break;
case Shader::ImageFormat::R8_SINT:
return GL_R8I;
case Shader::ImageFormat::R8_UINT:
return GL_R8UI;
case Shader::ImageFormat::R16_UINT:
return GL_R16UI;
case Shader::ImageFormat::R16_SINT:
return GL_R16I;
case Shader::ImageFormat::R32_UINT:
return GL_R32UI;
case Shader::ImageFormat::R32G32_UINT:
return GL_RG32UI;
case Shader::ImageFormat::R32G32B32A32_UINT:
return GL_RGBA32UI;
}
UNREACHABLE_MSG("Invalid image format={}", format);
return GL_R32UI;
}
} // Anonymous namespace
ImageBufferMap::~ImageBufferMap() {
@ -453,7 +364,7 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager&
static constexpr std::array TARGETS{GL_TEXTURE_1D_ARRAY, GL_TEXTURE_2D_ARRAY, GL_TEXTURE_3D};
for (size_t i = 0; i < TARGETS.size(); ++i) {
const GLenum target = TARGETS[i];
for (const FormatTuple& tuple : FORMAT_TABLE) {
for (const MaxwellToGL::FormatTuple& tuple : MaxwellToGL::FORMAT_TABLE) {
const GLenum format = tuple.internal_format;
GLint compat_class;
GLint compat_type;
@ -475,11 +386,9 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager&
null_image_1d_array.Create(GL_TEXTURE_1D_ARRAY);
null_image_cube_array.Create(GL_TEXTURE_CUBE_MAP_ARRAY);
null_image_3d.Create(GL_TEXTURE_3D);
null_image_rect.Create(GL_TEXTURE_RECTANGLE);
glTextureStorage2D(null_image_1d_array.handle, 1, GL_R8, 1, 1);
glTextureStorage3D(null_image_cube_array.handle, 1, GL_R8, 1, 1, 6);
glTextureStorage3D(null_image_3d.handle, 1, GL_R8, 1, 1, 1);
glTextureStorage2D(null_image_rect.handle, 1, GL_R8, 1, 1);
std::array<GLuint, 4> new_handles;
glGenTextures(static_cast<GLsizei>(new_handles.size()), new_handles.data());
@ -496,29 +405,28 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager&
glTextureView(null_image_view_cube.handle, GL_TEXTURE_CUBE_MAP, null_image_cube_array.handle,
GL_R8, 0, 1, 0, 6);
const std::array texture_handles{
null_image_1d_array.handle, null_image_cube_array.handle, null_image_3d.handle,
null_image_rect.handle, null_image_view_1d.handle, null_image_view_2d.handle,
null_image_view_2d_array.handle, null_image_view_cube.handle,
null_image_1d_array.handle, null_image_cube_array.handle, null_image_3d.handle,
null_image_view_1d.handle, null_image_view_2d.handle, null_image_view_2d_array.handle,
null_image_view_cube.handle,
};
for (const GLuint handle : texture_handles) {
static constexpr std::array NULL_SWIZZLE{GL_ZERO, GL_ZERO, GL_ZERO, GL_ZERO};
glTextureParameteriv(handle, GL_TEXTURE_SWIZZLE_RGBA, NULL_SWIZZLE.data());
}
const auto set_view = [this](ImageViewType type, GLuint handle) {
const auto set_view = [this](Shader::TextureType type, GLuint handle) {
if (device.HasDebuggingToolAttached()) {
const std::string name = fmt::format("NullImage {}", type);
glObjectLabel(GL_TEXTURE, handle, static_cast<GLsizei>(name.size()), name.data());
}
null_image_views[static_cast<size_t>(type)] = handle;
};
set_view(ImageViewType::e1D, null_image_view_1d.handle);
set_view(ImageViewType::e2D, null_image_view_2d.handle);
set_view(ImageViewType::Cube, null_image_view_cube.handle);
set_view(ImageViewType::e3D, null_image_3d.handle);
set_view(ImageViewType::e1DArray, null_image_1d_array.handle);
set_view(ImageViewType::e2DArray, null_image_view_2d_array.handle);
set_view(ImageViewType::CubeArray, null_image_cube_array.handle);
set_view(ImageViewType::Rect, null_image_rect.handle);
set_view(Shader::TextureType::Color1D, null_image_view_1d.handle);
set_view(Shader::TextureType::Color2D, null_image_view_2d.handle);
set_view(Shader::TextureType::ColorCube, null_image_view_cube.handle);
set_view(Shader::TextureType::Color3D, null_image_3d.handle);
set_view(Shader::TextureType::ColorArray1D, null_image_1d_array.handle);
set_view(Shader::TextureType::ColorArray2D, null_image_view_2d_array.handle);
set_view(Shader::TextureType::ColorArrayCube, null_image_cube_array.handle);
}
TextureCacheRuntime::~TextureCacheRuntime() = default;
@ -710,7 +618,7 @@ Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_,
gl_format = GL_RGBA;
gl_type = GL_UNSIGNED_INT_8_8_8_8_REV;
} else {
const auto& tuple = GetFormatTuple(info.format);
const auto& tuple = MaxwellToGL::GetFormatTuple(info.format);
gl_internal_format = tuple.internal_format;
gl_format = tuple.format;
gl_type = tuple.type;
@ -750,8 +658,7 @@ Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_,
glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, depth);
break;
case GL_TEXTURE_BUFFER:
buffer.Create();
glNamedBufferStorage(buffer.handle, guest_size_bytes, nullptr, 0);
UNREACHABLE();
break;
default:
UNREACHABLE_MSG("Invalid target=0x{:x}", target);
@ -789,14 +696,6 @@ void Image::UploadMemory(const ImageBufferMap& map,
}
}
void Image::UploadMemory(const ImageBufferMap& map,
std::span<const VideoCommon::BufferCopy> copies) {
for (const VideoCommon::BufferCopy& copy : copies) {
glCopyNamedBufferSubData(map.buffer, buffer.handle, copy.src_offset + map.offset,
copy.dst_offset, copy.size);
}
}
void Image::DownloadMemory(ImageBufferMap& map,
std::span<const VideoCommon::BufferImageCopy> copies) {
glMemoryBarrier(GL_PIXEL_BUFFER_BARRIER_BIT); // TODO: Move this to its own API
@ -958,23 +857,30 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
if (True(image.flags & ImageFlagBits::Converted)) {
internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8;
} else {
internal_format = GetFormatTuple(format).internal_format;
internal_format = MaxwellToGL::GetFormatTuple(format).internal_format;
}
full_range = info.range;
flat_range = info.range;
set_object_label = device.HasDebuggingToolAttached();
is_render_target = info.IsRenderTarget();
original_texture = image.texture.handle;
num_samples = image.info.num_samples;
if (!is_render_target) {
swizzle[0] = info.x_source;
swizzle[1] = info.y_source;
swizzle[2] = info.z_source;
swizzle[3] = info.w_source;
}
VideoCommon::SubresourceRange flatten_range = info.range;
std::array<GLuint, 2> handles;
stored_views.reserve(2);
switch (info.type) {
case ImageViewType::e1DArray:
flatten_range.extent.layers = 1;
flat_range.extent.layers = 1;
[[fallthrough]];
case ImageViewType::e1D:
glGenTextures(2, handles.data());
SetupView(device, image, ImageViewType::e1D, handles[0], info, flatten_range);
SetupView(device, image, ImageViewType::e1DArray, handles[1], info, info.range);
SetupView(Shader::TextureType::Color1D);
SetupView(Shader::TextureType::ColorArray1D);
break;
case ImageViewType::e2DArray:
flatten_range.extent.layers = 1;
flat_range.extent.layers = 1;
[[fallthrough]];
case ImageViewType::e2D:
if (True(flags & VideoCommon::ImageViewFlagBits::Slice)) {
@ -984,63 +890,126 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
.base = {.level = info.range.base.level, .layer = 0},
.extent = {.levels = 1, .layers = 1},
};
glGenTextures(1, handles.data());
SetupView(device, image, ImageViewType::e3D, handles[0], info, slice_range);
break;
full_range = slice_range;
SetupView(Shader::TextureType::Color3D);
} else {
SetupView(Shader::TextureType::Color2D);
SetupView(Shader::TextureType::ColorArray2D);
}
glGenTextures(2, handles.data());
SetupView(device, image, ImageViewType::e2D, handles[0], info, flatten_range);
SetupView(device, image, ImageViewType::e2DArray, handles[1], info, info.range);
break;
case ImageViewType::e3D:
glGenTextures(1, handles.data());
SetupView(device, image, ImageViewType::e3D, handles[0], info, info.range);
SetupView(Shader::TextureType::Color3D);
break;
case ImageViewType::CubeArray:
flatten_range.extent.layers = 6;
flat_range.extent.layers = 6;
[[fallthrough]];
case ImageViewType::Cube:
glGenTextures(2, handles.data());
SetupView(device, image, ImageViewType::Cube, handles[0], info, flatten_range);
SetupView(device, image, ImageViewType::CubeArray, handles[1], info, info.range);
SetupView(Shader::TextureType::ColorCube);
SetupView(Shader::TextureType::ColorArrayCube);
break;
case ImageViewType::Rect:
glGenTextures(1, handles.data());
SetupView(device, image, ImageViewType::Rect, handles[0], info, info.range);
UNIMPLEMENTED();
break;
case ImageViewType::Buffer:
glCreateTextures(GL_TEXTURE_BUFFER, 1, handles.data());
SetupView(device, image, ImageViewType::Buffer, handles[0], info, info.range);
UNREACHABLE();
break;
}
switch (info.type) {
case ImageViewType::e1D:
default_handle = Handle(Shader::TextureType::Color1D);
break;
case ImageViewType::e1DArray:
default_handle = Handle(Shader::TextureType::ColorArray1D);
break;
case ImageViewType::e2D:
default_handle = Handle(Shader::TextureType::Color2D);
break;
case ImageViewType::e2DArray:
default_handle = Handle(Shader::TextureType::ColorArray2D);
break;
case ImageViewType::e3D:
default_handle = Handle(Shader::TextureType::Color3D);
break;
case ImageViewType::Cube:
default_handle = Handle(Shader::TextureType::ColorCube);
break;
case ImageViewType::CubeArray:
default_handle = Handle(Shader::TextureType::ColorArrayCube);
break;
default:
break;
}
default_handle = Handle(info.type);
}
ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info,
const VideoCommon::ImageViewInfo& view_info, GPUVAddr gpu_addr_)
: VideoCommon::ImageViewBase{info, view_info}, gpu_addr{gpu_addr_},
buffer_size{VideoCommon::CalculateGuestSizeInBytes(info)} {}
ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info,
const VideoCommon::ImageViewInfo& view_info)
: VideoCommon::ImageViewBase{info, view_info} {}
ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::NullImageParams& params)
: VideoCommon::ImageViewBase{params}, views{runtime.null_image_views} {}
void ImageView::SetupView(const Device& device, Image& image, ImageViewType view_type,
GLuint handle, const VideoCommon::ImageViewInfo& info,
VideoCommon::SubresourceRange view_range) {
if (info.type == ImageViewType::Buffer) {
// TODO: Take offset from buffer cache
glTextureBufferRange(handle, internal_format, image.buffer.handle, 0,
image.guest_size_bytes);
} else {
const GLuint parent = image.texture.handle;
const GLenum target = ImageTarget(view_type, image.info.num_samples);
glTextureView(handle, target, parent, internal_format, view_range.base.level,
view_range.extent.levels, view_range.base.layer, view_range.extent.layers);
if (!info.IsRenderTarget()) {
ApplySwizzle(handle, format, info.Swizzle());
}
GLuint ImageView::StorageView(Shader::TextureType texture_type, Shader::ImageFormat image_format) {
if (image_format == Shader::ImageFormat::Typeless) {
return Handle(texture_type);
}
if (device.HasDebuggingToolAttached()) {
const std::string name = VideoCommon::Name(*this, view_type);
glObjectLabel(GL_TEXTURE, handle, static_cast<GLsizei>(name.size()), name.data());
const bool is_signed{image_format == Shader::ImageFormat::R8_SINT ||
image_format == Shader::ImageFormat::R16_SINT};
if (!storage_views) {
storage_views = std::make_unique<StorageViews>();
}
stored_views.emplace_back().handle = handle;
views[static_cast<size_t>(view_type)] = handle;
auto& type_views{is_signed ? storage_views->signeds : storage_views->unsigneds};
GLuint& view{type_views[static_cast<size_t>(texture_type)]};
if (view == 0) {
view = MakeView(texture_type, ShaderFormat(image_format));
}
return view;
}
void ImageView::SetupView(Shader::TextureType view_type) {
views[static_cast<size_t>(view_type)] = MakeView(view_type, internal_format);
}
GLuint ImageView::MakeView(Shader::TextureType view_type, GLenum view_format) {
VideoCommon::SubresourceRange view_range;
switch (view_type) {
case Shader::TextureType::Color1D:
case Shader::TextureType::Color2D:
case Shader::TextureType::ColorCube:
view_range = flat_range;
break;
case Shader::TextureType::ColorArray1D:
case Shader::TextureType::ColorArray2D:
case Shader::TextureType::Color3D:
case Shader::TextureType::ColorArrayCube:
view_range = full_range;
break;
default:
UNREACHABLE();
}
OGLTextureView& view = stored_views.emplace_back();
view.Create();
const GLenum target = ImageTarget(view_type, num_samples);
glTextureView(view.handle, target, original_texture, view_format, view_range.base.level,
view_range.extent.levels, view_range.base.layer, view_range.extent.layers);
if (!is_render_target) {
std::array<SwizzleSource, 4> casted_swizzle;
std::ranges::transform(swizzle, casted_swizzle.begin(), [](u8 component_swizzle) {
return static_cast<SwizzleSource>(component_swizzle);
});
ApplySwizzle(view.handle, format, casted_swizzle);
}
if (set_object_label) {
const std::string name = VideoCommon::Name(*this);
glObjectLabel(GL_TEXTURE, view.handle, static_cast<GLsizei>(name.size()), name.data());
}
return view.handle;
}
Sampler::Sampler(TextureCacheRuntime& runtime, const TSCEntry& config) {

53
src/video_core/renderer_opengl/gl_texture_cache.h
View File

@ -9,6 +9,7 @@
#include <glad/glad.h>
#include "shader_recompiler/shader_info.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/util_shaders.h"
#include "video_core/texture_cache/texture_cache.h"
@ -127,13 +128,12 @@ private:
OGLTexture null_image_1d_array;
OGLTexture null_image_cube_array;
OGLTexture null_image_3d;
OGLTexture null_image_rect;
OGLTextureView null_image_view_1d;
OGLTextureView null_image_view_2d;
OGLTextureView null_image_view_2d_array;
OGLTextureView null_image_view_cube;
std::array<GLuint, VideoCommon::NUM_IMAGE_VIEW_TYPES> null_image_views;
std::array<GLuint, Shader::NUM_TEXTURE_TYPES> null_image_views{};
};
class Image : public VideoCommon::ImageBase {
@ -154,8 +154,6 @@ public:
void UploadMemory(const ImageBufferMap& map,
std::span<const VideoCommon::BufferImageCopy> copies);
void UploadMemory(const ImageBufferMap& map, std::span<const VideoCommon::BufferCopy> copies);
void DownloadMemory(ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies);
GLuint StorageHandle() noexcept;
@ -170,7 +168,6 @@ private:
void CopyImageToBuffer(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset);
OGLTexture texture;
OGLBuffer buffer;
OGLTextureView store_view;
GLenum gl_internal_format = GL_NONE;
GLenum gl_format = GL_NONE;
@ -182,10 +179,17 @@ class ImageView : public VideoCommon::ImageViewBase {
public:
explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageViewInfo&, ImageId, Image&);
explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo&,
const VideoCommon::ImageViewInfo&, GPUVAddr);
explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info,
const VideoCommon::ImageViewInfo& view_info);
explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams&);
[[nodiscard]] GLuint Handle(ImageViewType query_type) const noexcept {
return views[static_cast<size_t>(query_type)];
[[nodiscard]] GLuint StorageView(Shader::TextureType texture_type,
Shader::ImageFormat image_format);
[[nodiscard]] GLuint Handle(Shader::TextureType handle_type) const noexcept {
return views[static_cast<size_t>(handle_type)];
}
[[nodiscard]] GLuint DefaultHandle() const noexcept {
@ -196,15 +200,38 @@ public:
return internal_format;
}
private:
void SetupView(const Device& device, Image& image, ImageViewType view_type, GLuint handle,
const VideoCommon::ImageViewInfo& info,
VideoCommon::SubresourceRange view_range);
[[nodiscard]] GPUVAddr GpuAddr() const noexcept {
return gpu_addr;
}
std::array<GLuint, VideoCommon::NUM_IMAGE_VIEW_TYPES> views{};
[[nodiscard]] u32 BufferSize() const noexcept {
return buffer_size;
}
private:
struct StorageViews {
std::array<GLuint, Shader::NUM_TEXTURE_TYPES> signeds{};
std::array<GLuint, Shader::NUM_TEXTURE_TYPES> unsigneds{};
};
void SetupView(Shader::TextureType view_type);
GLuint MakeView(Shader::TextureType view_type, GLenum view_format);
std::array<GLuint, Shader::NUM_TEXTURE_TYPES> views{};
std::vector<OGLTextureView> stored_views;
GLuint default_handle = 0;
std::unique_ptr<StorageViews> storage_views;
GLenum internal_format = GL_NONE;
GLuint default_handle = 0;
GPUVAddr gpu_addr = 0;
u32 buffer_size = 0;
GLuint original_texture = 0;
int num_samples = 0;
VideoCommon::SubresourceRange flat_range;
VideoCommon::SubresourceRange full_range;
std::array<u8, 4> swizzle{};
bool set_object_label = false;
bool is_render_target = false;
};
class ImageAlloc : public VideoCommon::ImageAllocBase {};

108
src/video_core/renderer_opengl/maxwell_to_gl.h
View File

@ -5,12 +5,120 @@
#pragma once
#include <glad/glad.h>
#include "video_core/engines/maxwell_3d.h"
#include "video_core/surface.h"
namespace OpenGL::MaxwellToGL {
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
struct FormatTuple {
GLenum internal_format;
GLenum format = GL_NONE;
GLenum type = GL_NONE;
};
constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> FORMAT_TABLE = {{
{GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // A8B8G8R8_UNORM
{GL_RGBA8_SNORM, GL_RGBA, GL_BYTE}, // A8B8G8R8_SNORM
{GL_RGBA8I, GL_RGBA_INTEGER, GL_BYTE}, // A8B8G8R8_SINT
{GL_RGBA8UI, GL_RGBA_INTEGER, GL_UNSIGNED_BYTE}, // A8B8G8R8_UINT
{GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5}, // R5G6B5_UNORM
{GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV}, // B5G6R5_UNORM
{GL_RGB5_A1, GL_BGRA, GL_UNSIGNED_SHORT_1_5_5_5_REV}, // A1R5G5B5_UNORM
{GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10_UNORM
{GL_RGB10_A2UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10_UINT
{GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV}, // A1B5G5R5_UNORM
{GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // R8_UNORM
{GL_R8_SNORM, GL_RED, GL_BYTE}, // R8_SNORM
{GL_R8I, GL_RED_INTEGER, GL_BYTE}, // R8_SINT
{GL_R8UI, GL_RED_INTEGER, GL_UNSIGNED_BYTE}, // R8_UINT
{GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT}, // R16G16B16A16_FLOAT
{GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT}, // R16G16B16A16_UNORM
{GL_RGBA16_SNORM, GL_RGBA, GL_SHORT}, // R16G16B16A16_SNORM
{GL_RGBA16I, GL_RGBA_INTEGER, GL_SHORT}, // R16G16B16A16_SINT
{GL_RGBA16UI, GL_RGBA_INTEGER, GL_UNSIGNED_SHORT}, // R16G16B16A16_UINT
{GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV}, // B10G11R11_FLOAT
{GL_RGBA32UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT}, // R32G32B32A32_UINT
{GL_COMPRESSED_RGBA_S3TC_DXT1_EXT}, // BC1_RGBA_UNORM
{GL_COMPRESSED_RGBA_S3TC_DXT3_EXT}, // BC2_UNORM
{GL_COMPRESSED_RGBA_S3TC_DXT5_EXT}, // BC3_UNORM
{GL_COMPRESSED_RED_RGTC1}, // BC4_UNORM
{GL_COMPRESSED_SIGNED_RED_RGTC1}, // BC4_SNORM
{GL_COMPRESSED_RG_RGTC2}, // BC5_UNORM
{GL_COMPRESSED_SIGNED_RG_RGTC2}, // BC5_SNORM
{GL_COMPRESSED_RGBA_BPTC_UNORM}, // BC7_UNORM
{GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT}, // BC6H_UFLOAT
{GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT}, // BC6H_SFLOAT
{GL_COMPRESSED_RGBA_ASTC_4x4_KHR}, // ASTC_2D_4X4_UNORM
{GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE}, // B8G8R8A8_UNORM
{GL_RGBA32F, GL_RGBA, GL_FLOAT}, // R32G32B32A32_FLOAT
{GL_RGBA32I, GL_RGBA_INTEGER, GL_INT}, // R32G32B32A32_SINT
{GL_RG32F, GL_RG, GL_FLOAT}, // R32G32_FLOAT
{GL_RG32I, GL_RG_INTEGER, GL_INT}, // R32G32_SINT
{GL_R32F, GL_RED, GL_FLOAT}, // R32_FLOAT
{GL_R16F, GL_RED, GL_HALF_FLOAT}, // R16_FLOAT
{GL_R16, GL_RED, GL_UNSIGNED_SHORT}, // R16_UNORM
{GL_R16_SNORM, GL_RED, GL_SHORT}, // R16_SNORM
{GL_R16UI, GL_RED_INTEGER, GL_UNSIGNED_SHORT}, // R16_UINT
{GL_R16I, GL_RED_INTEGER, GL_SHORT}, // R16_SINT
{GL_RG16, GL_RG, GL_UNSIGNED_SHORT}, // R16G16_UNORM
{GL_RG16F, GL_RG, GL_HALF_FLOAT}, // R16G16_FLOAT
{GL_RG16UI, GL_RG_INTEGER, GL_UNSIGNED_SHORT}, // R16G16_UINT
{GL_RG16I, GL_RG_INTEGER, GL_SHORT}, // R16G16_SINT
{GL_RG16_SNORM, GL_RG, GL_SHORT}, // R16G16_SNORM
{GL_RGB32F, GL_RGB, GL_FLOAT}, // R32G32B32_FLOAT
{GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // A8B8G8R8_SRGB
{GL_RG8, GL_RG, GL_UNSIGNED_BYTE}, // R8G8_UNORM
{GL_RG8_SNORM, GL_RG, GL_BYTE}, // R8G8_SNORM
{GL_RG8I, GL_RG_INTEGER, GL_BYTE}, // R8G8_SINT
{GL_RG8UI, GL_RG_INTEGER, GL_UNSIGNED_BYTE}, // R8G8_UINT
{GL_RG32UI, GL_RG_INTEGER, GL_UNSIGNED_INT}, // R32G32_UINT
{GL_RGB16F, GL_RGBA, GL_HALF_FLOAT}, // R16G16B16X16_FLOAT
{GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT}, // R32_UINT
{GL_R32I, GL_RED_INTEGER, GL_INT}, // R32_SINT
{GL_COMPRESSED_RGBA_ASTC_8x8_KHR}, // ASTC_2D_8X8_UNORM
{GL_COMPRESSED_RGBA_ASTC_8x5_KHR}, // ASTC_2D_8X5_UNORM
{GL_COMPRESSED_RGBA_ASTC_5x4_KHR}, // ASTC_2D_5X4_UNORM
{GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE}, // B8G8R8A8_SRGB
{GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT}, // BC1_RGBA_SRGB
{GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT}, // BC2_SRGB
{GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT}, // BC3_SRGB
{GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM}, // BC7_SRGB
{GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV}, // A4B4G4R4_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_4x4_KHR}, // ASTC_2D_4X4_SRGB
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x8_KHR}, // ASTC_2D_8X8_SRGB
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x5_KHR}, // ASTC_2D_8X5_SRGB
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x4_KHR}, // ASTC_2D_5X4_SRGB
{GL_COMPRESSED_RGBA_ASTC_5x5_KHR}, // ASTC_2D_5X5_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x5_KHR}, // ASTC_2D_5X5_SRGB
{GL_COMPRESSED_RGBA_ASTC_10x8_KHR}, // ASTC_2D_10X8_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR}, // ASTC_2D_10X8_SRGB
{GL_COMPRESSED_RGBA_ASTC_6x6_KHR}, // ASTC_2D_6X6_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x6_KHR}, // ASTC_2D_6X6_SRGB
{GL_COMPRESSED_RGBA_ASTC_10x10_KHR}, // ASTC_2D_10X10_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR}, // ASTC_2D_10X10_SRGB
{GL_COMPRESSED_RGBA_ASTC_12x12_KHR}, // ASTC_2D_12X12_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR}, // ASTC_2D_12X12_SRGB
{GL_COMPRESSED_RGBA_ASTC_8x6_KHR}, // ASTC_2D_8X6_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x6_KHR}, // ASTC_2D_8X6_SRGB
{GL_COMPRESSED_RGBA_ASTC_6x5_KHR}, // ASTC_2D_6X5_UNORM
{GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x5_KHR}, // ASTC_2D_6X5_SRGB
{GL_RGB9_E5, GL_RGB, GL_UNSIGNED_INT_5_9_9_9_REV}, // E5B9G9R9_FLOAT
{GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // D32_FLOAT
{GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16_UNORM
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24_UNORM_S8_UINT
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // S8_UINT_D24_UNORM
{GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL,
GL_FLOAT_32_UNSIGNED_INT_24_8_REV}, // D32_FLOAT_S8_UINT
}};
inline const FormatTuple& GetFormatTuple(VideoCore::Surface::PixelFormat pixel_format) {
ASSERT(static_cast<size_t>(pixel_format) < FORMAT_TABLE.size());
return FORMAT_TABLE[static_cast<size_t>(pixel_format)];
}
inline GLenum VertexFormat(Maxwell::VertexAttribute attrib) {
switch (attrib.type) {
case Maxwell::VertexAttribute::Type::UnsignedNorm:

60
src/video_core/renderer_opengl/renderer_opengl.cpp
View File

@ -25,6 +25,7 @@
#include "video_core/host_shaders/opengl_present_vert.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/renderer_opengl/gl_shader_manager.h"
#include "video_core/renderer_opengl/gl_shader_util.h"
#include "video_core/renderer_opengl/renderer_opengl.h"
#include "video_core/textures/decoders.h"
@ -139,6 +140,26 @@ RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_,
}
AddTelemetryFields();
InitOpenGLObjects();
// Initialize default attributes to match hardware's disabled attributes
GLint max_attribs{};
glGetIntegerv(GL_MAX_VERTEX_ATTRIBS, &max_attribs);
for (GLint attrib = 0; attrib < max_attribs; ++attrib) {
glVertexAttrib4f(attrib, 0.0f, 0.0f, 0.0f, 1.0f);
}
// Enable seamless cubemaps when per texture parameters are not available
if (!GLAD_GL_ARB_seamless_cubemap_per_texture && !GLAD_GL_AMD_seamless_cubemap_per_texture) {
glEnable(GL_TEXTURE_CUBE_MAP_SEAMLESS);
}
// Enable unified vertex attributes and query vertex buffer address when the driver supports it
if (device.HasVertexBufferUnifiedMemory()) {
glEnableClientState(GL_VERTEX_ATTRIB_ARRAY_UNIFIED_NV);
glEnableClientState(GL_ELEMENT_ARRAY_UNIFIED_NV);
glMakeNamedBufferResidentNV(vertex_buffer.handle, GL_READ_ONLY);
glGetNamedBufferParameterui64vNV(vertex_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV,
&vertex_buffer_address);
}
}
RendererOpenGL::~RendererOpenGL() = default;
@ -230,18 +251,8 @@ void RendererOpenGL::LoadColorToActiveGLTexture(u8 color_r, u8 color_g, u8 color
void RendererOpenGL::InitOpenGLObjects() {
// Create shader programs
OGLShader vertex_shader;
vertex_shader.Create(HostShaders::OPENGL_PRESENT_VERT, GL_VERTEX_SHADER);
OGLShader fragment_shader;
fragment_shader.Create(HostShaders::OPENGL_PRESENT_FRAG, GL_FRAGMENT_SHADER);
vertex_program.Create(true, false, vertex_shader.handle);
fragment_program.Create(true, false, fragment_shader.handle);
pipeline.Create();
glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, vertex_program.handle);
glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, fragment_program.handle);
present_vertex = CreateProgram(HostShaders::OPENGL_PRESENT_VERT, GL_VERTEX_SHADER);
present_fragment = CreateProgram(HostShaders::OPENGL_PRESENT_FRAG, GL_FRAGMENT_SHADER);
// Generate presentation sampler
present_sampler.Create();
@ -263,21 +274,6 @@ void RendererOpenGL::InitOpenGLObjects() {
// Clear screen to black
LoadColorToActiveGLTexture(0, 0, 0, 0, screen_info.texture);
// Enable seamless cubemaps when per texture parameters are not available
if (!GLAD_GL_ARB_seamless_cubemap_per_texture && !GLAD_GL_AMD_seamless_cubemap_per_texture) {
glEnable(GL_TEXTURE_CUBE_MAP_SEAMLESS);
}
// Enable unified vertex attributes and query vertex buffer address when the driver supports it
if (device.HasVertexBufferUnifiedMemory()) {
glEnableClientState(GL_VERTEX_ATTRIB_ARRAY_UNIFIED_NV);
glEnableClientState(GL_ELEMENT_ARRAY_UNIFIED_NV);
glMakeNamedBufferResidentNV(vertex_buffer.handle, GL_READ_ONLY);
glGetNamedBufferParameterui64vNV(vertex_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV,
&vertex_buffer_address);
}
}
void RendererOpenGL::AddTelemetryFields() {
@ -342,8 +338,9 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
// Set projection matrix
const std::array ortho_matrix =
MakeOrthographicMatrix(static_cast<float>(layout.width), static_cast<float>(layout.height));
glProgramUniformMatrix3x2fv(vertex_program.handle, ModelViewMatrixLocation, 1, GL_FALSE,
std::data(ortho_matrix));
program_manager.BindPresentPrograms(present_vertex.handle, present_fragment.handle);
glProgramUniformMatrix3x2fv(present_vertex.handle, ModelViewMatrixLocation, 1, GL_FALSE,
ortho_matrix.data());
const auto& texcoords = screen_info.display_texcoords;
auto left = texcoords.left;
@ -404,8 +401,6 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
state_tracker.NotifyClipControl();
state_tracker.NotifyAlphaTest();
program_manager.BindHostPipeline(pipeline.handle);
state_tracker.ClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE);
glEnable(GL_CULL_FACE);
if (screen_info.display_srgb) {
@ -453,7 +448,8 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
glClear(GL_COLOR_BUFFER_BIT);
glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
program_manager.RestoreGuestPipeline();
// TODO
// program_manager.RestoreGuestPipeline();
}
void RendererOpenGL::RenderScreenshot() {

6
src/video_core/renderer_opengl/renderer_opengl.h
View File

@ -12,7 +12,6 @@
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_manager.h"
#include "video_core/renderer_opengl/gl_state_tracker.h"
namespace Core {
@ -111,9 +110,8 @@ private:
// OpenGL object IDs
OGLSampler present_sampler;
OGLBuffer vertex_buffer;
OGLProgram vertex_program;
OGLProgram fragment_program;
OGLPipeline pipeline;
OGLProgram present_vertex;
OGLProgram present_fragment;
OGLFramebuffer screenshot_framebuffer;
// GPU address of the vertex buffer

23
src/video_core/renderer_opengl/util_shaders.cpp
View File

@ -16,8 +16,8 @@
#include "video_core/host_shaders/opengl_copy_bc4_comp.h"
#include "video_core/host_shaders/opengl_copy_bgra_comp.h"
#include "video_core/host_shaders/pitch_unswizzle_comp.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_manager.h"
#include "video_core/renderer_opengl/gl_shader_util.h"
#include "video_core/renderer_opengl/gl_texture_cache.h"
#include "video_core/renderer_opengl/util_shaders.h"
#include "video_core/texture_cache/accelerated_swizzle.h"
@ -41,21 +41,14 @@ using VideoCommon::Accelerated::MakeBlockLinearSwizzle3DParams;
using VideoCore::Surface::BytesPerBlock;
namespace {
OGLProgram MakeProgram(std::string_view source) {
OGLShader shader;
shader.Create(source, GL_COMPUTE_SHADER);
OGLProgram program;
program.Create(true, false, shader.handle);
return program;
return CreateProgram(source, GL_COMPUTE_SHADER);
}
size_t NumPixelsInCopy(const VideoCommon::ImageCopy& copy) {
return static_cast<size_t>(copy.extent.width * copy.extent.height *
copy.src_subresource.num_layers);
}
} // Anonymous namespace
UtilShaders::UtilShaders(ProgramManager& program_manager_)
@ -86,7 +79,7 @@ void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
.width = VideoCore::Surface::DefaultBlockWidth(image.info.format),
.height = VideoCore::Surface::DefaultBlockHeight(image.info.format),
};
program_manager.BindHostCompute(astc_decoder_program.handle);
program_manager.BindComputeProgram(astc_decoder_program.handle);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_ENC_BUFFER, astc_buffer.handle);
@ -134,7 +127,7 @@ void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map,
static constexpr GLuint BINDING_INPUT_BUFFER = 1;
static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
program_manager.BindHostCompute(block_linear_unswizzle_2d_program.handle);
program_manager.BindComputeProgram(block_linear_unswizzle_2d_program.handle);
glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
@ -173,7 +166,7 @@ void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map,
static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes);
program_manager.BindHostCompute(block_linear_unswizzle_3d_program.handle);
program_manager.BindComputeProgram(block_linear_unswizzle_3d_program.handle);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
const GLenum store_format = StoreFormat(BytesPerBlock(image.info.format));
@ -222,7 +215,7 @@ void UtilShaders::PitchUpload(Image& image, const ImageBufferMap& map,
UNIMPLEMENTED_IF_MSG(!std::has_single_bit(bytes_per_block),
"Non-power of two images are not implemented");
program_manager.BindHostCompute(pitch_unswizzle_program.handle);
program_manager.BindComputeProgram(pitch_unswizzle_program.handle);
glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes);
glUniform2ui(LOC_ORIGIN, 0, 0);
glUniform2i(LOC_DESTINATION, 0, 0);
@ -250,7 +243,7 @@ void UtilShaders::CopyBC4(Image& dst_image, Image& src_image, std::span<const Im
static constexpr GLuint LOC_SRC_OFFSET = 0;
static constexpr GLuint LOC_DST_OFFSET = 1;
program_manager.BindHostCompute(copy_bc4_program.handle);
program_manager.BindComputeProgram(copy_bc4_program.handle);
for (const ImageCopy& copy : copies) {
ASSERT(copy.src_subresource.base_layer == 0);
@ -286,7 +279,7 @@ void UtilShaders::CopyBGR(Image& dst_image, Image& src_image,
break;
case 4: {
// BGRA8 copy
program_manager.BindHostCompute(copy_bgra_program.handle);
program_manager.BindComputeProgram(copy_bgra_program.handle);
constexpr GLenum FORMAT = GL_RGBA8;
for (const ImageCopy& copy : copies) {
ASSERT(copy.src_offset == zero_offset);

40
src/video_core/renderer_vulkan/blit_image.cpp
View File

@ -49,6 +49,16 @@ constexpr VkDescriptorSetLayoutCreateInfo ONE_TEXTURE_DESCRIPTOR_SET_LAYOUT_CREA
.bindingCount = 1,
.pBindings = &TEXTURE_DESCRIPTOR_SET_LAYOUT_BINDING<0>,
};
template <u32 num_textures>
inline constexpr DescriptorBankInfo TEXTURE_DESCRIPTOR_BANK_INFO{
.uniform_buffers = 0,
.storage_buffers = 0,
.texture_buffers = 0,
.image_buffers = 0,
.textures = num_textures,
.images = 0,
.score = 2,
};
constexpr VkDescriptorSetLayoutCreateInfo TWO_TEXTURES_DESCRIPTOR_SET_LAYOUT_CREATE_INFO{
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
.pNext = nullptr,
@ -323,18 +333,19 @@ void BindBlitState(vk::CommandBuffer cmdbuf, VkPipelineLayout layout, const Regi
cmdbuf.SetScissor(0, scissor);
cmdbuf.PushConstants(layout, VK_SHADER_STAGE_VERTEX_BIT, push_constants);
}
} // Anonymous namespace
BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_,
StateTracker& state_tracker_, VKDescriptorPool& descriptor_pool)
StateTracker& state_tracker_, DescriptorPool& descriptor_pool)
: device{device_}, scheduler{scheduler_}, state_tracker{state_tracker_},
one_texture_set_layout(device.GetLogical().CreateDescriptorSetLayout(
ONE_TEXTURE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO)),
two_textures_set_layout(device.GetLogical().CreateDescriptorSetLayout(
TWO_TEXTURES_DESCRIPTOR_SET_LAYOUT_CREATE_INFO)),
one_texture_descriptor_allocator(descriptor_pool, *one_texture_set_layout),
two_textures_descriptor_allocator(descriptor_pool, *two_textures_set_layout),
one_texture_descriptor_allocator{
descriptor_pool.Allocator(*one_texture_set_layout, TEXTURE_DESCRIPTOR_BANK_INFO<1>)},
two_textures_descriptor_allocator{
descriptor_pool.Allocator(*two_textures_set_layout, TEXTURE_DESCRIPTOR_BANK_INFO<2>)},
one_texture_pipeline_layout(device.GetLogical().CreatePipelineLayout(
PipelineLayoutCreateInfo(one_texture_set_layout.address()))),
two_textures_pipeline_layout(device.GetLogical().CreatePipelineLayout(
@ -362,14 +373,14 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageV
.operation = operation,
};
const VkPipelineLayout layout = *one_texture_pipeline_layout;
const VkImageView src_view = src_image_view.Handle(ImageViewType::e2D);
const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D);
const VkSampler sampler = is_linear ? *linear_sampler : *nearest_sampler;
const VkPipeline pipeline = FindOrEmplacePipeline(key);
const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
scheduler.RequestRenderpass(dst_framebuffer);
scheduler.Record([dst_region, src_region, pipeline, layout, sampler, src_view, descriptor_set,
&device = device](vk::CommandBuffer cmdbuf) {
scheduler.Record([this, dst_region, src_region, pipeline, layout, sampler,
src_view](vk::CommandBuffer cmdbuf) {
// TODO: Barriers
const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
@ -391,12 +402,11 @@ void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer,
const VkPipelineLayout layout = *two_textures_pipeline_layout;
const VkSampler sampler = *nearest_sampler;
const VkPipeline pipeline = BlitDepthStencilPipeline(dst_framebuffer->RenderPass());
const VkDescriptorSet descriptor_set = two_textures_descriptor_allocator.Commit();
scheduler.RequestRenderpass(dst_framebuffer);
scheduler.Record([dst_region, src_region, pipeline, layout, sampler, src_depth_view,
src_stencil_view, descriptor_set,
&device = device](vk::CommandBuffer cmdbuf) {
src_stencil_view, this](vk::CommandBuffer cmdbuf) {
// TODO: Barriers
const VkDescriptorSet descriptor_set = two_textures_descriptor_allocator.Commit();
UpdateTwoTexturesDescriptorSet(device, descriptor_set, sampler, src_depth_view,
src_stencil_view);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
@ -416,7 +426,6 @@ void BlitImageHelper::ConvertD32ToR32(const Framebuffer* dst_framebuffer,
void BlitImageHelper::ConvertR32ToD32(const Framebuffer* dst_framebuffer,
const ImageView& src_image_view) {
ConvertColorToDepthPipeline(convert_r32_to_d32_pipeline, dst_framebuffer->RenderPass());
Convert(*convert_r32_to_d32_pipeline, dst_framebuffer, src_image_view);
}
@ -436,16 +445,14 @@ void BlitImageHelper::ConvertR16ToD16(const Framebuffer* dst_framebuffer,
void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
const ImageView& src_image_view) {
const VkPipelineLayout layout = *one_texture_pipeline_layout;
const VkImageView src_view = src_image_view.Handle(ImageViewType::e2D);
const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D);
const VkSampler sampler = *nearest_sampler;
const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
const VkExtent2D extent{
.width = src_image_view.size.width,
.height = src_image_view.size.height,
};
scheduler.RequestRenderpass(dst_framebuffer);
scheduler.Record([pipeline, layout, sampler, src_view, descriptor_set, extent,
&device = device](vk::CommandBuffer cmdbuf) {
scheduler.Record([pipeline, layout, sampler, src_view, extent, this](vk::CommandBuffer cmdbuf) {
const VkOffset2D offset{
.x = 0,
.y = 0,
@ -466,6 +473,7 @@ void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_frameb
.tex_scale = {viewport.width, viewport.height},
.tex_offset = {0.0f, 0.0f},
};
const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view);
// TODO: Barriers

2
src/video_core/renderer_vulkan/blit_image.h
View File

@ -31,7 +31,7 @@ struct BlitImagePipelineKey {
class BlitImageHelper {
public:
explicit BlitImageHelper(const Device& device, VKScheduler& scheduler,
StateTracker& state_tracker, VKDescriptorPool& descriptor_pool);
StateTracker& state_tracker, DescriptorPool& descriptor_pool);
~BlitImageHelper();
void BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,

90
src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
View File

@ -15,9 +15,7 @@
#include "video_core/renderer_vulkan/vk_state_tracker.h"
namespace Vulkan {
namespace {
constexpr size_t POINT = 0;
constexpr size_t LINE = 1;
constexpr size_t POLYGON = 2;
@ -39,10 +37,20 @@ constexpr std::array POLYGON_OFFSET_ENABLE_LUT = {
POLYGON, // Patches
};
void RefreshXfbState(VideoCommon::TransformFeedbackState& state, const Maxwell& regs) {
std::ranges::transform(regs.tfb_layouts, state.layouts.begin(), [](const auto& layout) {
return VideoCommon::TransformFeedbackState::Layout{
.stream = layout.stream,
.varying_count = layout.varying_count,
.stride = layout.stride,
};
});
state.varyings = regs.tfb_varying_locs;
}
} // Anonymous namespace
void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d,
bool has_extended_dynamic_state) {
bool has_extended_dynamic_state, bool has_dynamic_vertex_input) {
const Maxwell& regs = maxwell3d.regs;
const std::array enabled_lut{
regs.polygon_offset_point_enable,
@ -52,6 +60,9 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d,
const u32 topology_index = static_cast<u32>(regs.draw.topology.Value());
raw1 = 0;
extended_dynamic_state.Assign(has_extended_dynamic_state ? 1 : 0);
dynamic_vertex_input.Assign(has_dynamic_vertex_input ? 1 : 0);
xfb_enabled.Assign(regs.tfb_enabled != 0);
primitive_restart_enable.Assign(regs.primitive_restart.enabled != 0 ? 1 : 0);
depth_bias_enable.Assign(enabled_lut[POLYGON_OFFSET_ENABLE_LUT[topology_index]] != 0 ? 1 : 0);
depth_clamp_disabled.Assign(regs.view_volume_clip_control.depth_clamp_disabled.Value());
@ -63,37 +74,66 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d,
tessellation_clockwise.Assign(regs.tess_mode.cw.Value());
logic_op_enable.Assign(regs.logic_op.enable != 0 ? 1 : 0);
logic_op.Assign(PackLogicOp(regs.logic_op.operation));
rasterize_enable.Assign(regs.rasterize_enable != 0 ? 1 : 0);
topology.Assign(regs.draw.topology);
msaa_mode.Assign(regs.multisample_mode);
raw2 = 0;
rasterize_enable.Assign(regs.rasterize_enable != 0 ? 1 : 0);
const auto test_func =
regs.alpha_test_enabled != 0 ? regs.alpha_test_func : Maxwell::ComparisonOp::Always;
alpha_test_func.Assign(PackComparisonOp(test_func));
early_z.Assign(regs.force_early_fragment_tests != 0 ? 1 : 0);
depth_enabled.Assign(regs.zeta_enable != 0 ? 1 : 0);
depth_format.Assign(static_cast<u32>(regs.zeta.format));
y_negate.Assign(regs.screen_y_control.y_negate != 0 ? 1 : 0);
provoking_vertex_last.Assign(regs.provoking_vertex_last != 0 ? 1 : 0);
conservative_raster_enable.Assign(regs.conservative_raster_enable != 0 ? 1 : 0);
smooth_lines.Assign(regs.line_smooth_enable != 0 ? 1 : 0);
for (size_t i = 0; i < regs.rt.size(); ++i) {
color_formats[i] = static_cast<u8>(regs.rt[i].format);
}
alpha_test_ref = Common::BitCast<u32>(regs.alpha_test_ref);
point_size = Common::BitCast<u32>(regs.point_size);
if (maxwell3d.dirty.flags[Dirty::InstanceDivisors]) {
maxwell3d.dirty.flags[Dirty::InstanceDivisors] = false;
for (size_t index = 0; index < Maxwell::NumVertexArrays; ++index) {
const bool is_enabled = regs.instanced_arrays.IsInstancingEnabled(index);
binding_divisors[index] = is_enabled ? regs.vertex_array[index].divisor : 0;
}
}
if (maxwell3d.dirty.flags[Dirty::VertexAttributes]) {
maxwell3d.dirty.flags[Dirty::VertexAttributes] = false;
for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) {
const auto& input = regs.vertex_attrib_format[index];
auto& attribute = attributes[index];
attribute.raw = 0;
attribute.enabled.Assign(input.IsConstant() ? 0 : 1);
attribute.buffer.Assign(input.buffer);
attribute.offset.Assign(input.offset);
attribute.type.Assign(static_cast<u32>(input.type.Value()));
attribute.size.Assign(static_cast<u32>(input.size.Value()));
if (maxwell3d.dirty.flags[Dirty::VertexInput]) {
if (has_dynamic_vertex_input) {
// Dirty flag will be reset by the command buffer update
static constexpr std::array LUT{
0u, // Invalid
1u, // SignedNorm
1u, // UnsignedNorm
2u, // SignedInt
3u, // UnsignedInt
1u, // UnsignedScaled
1u, // SignedScaled
1u, // Float
};
const auto& attrs = regs.vertex_attrib_format;
attribute_types = 0;
for (size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) {
const u32 mask = attrs[i].constant != 0 ? 0 : 3;
const u32 type = LUT[static_cast<size_t>(attrs[i].type.Value())];
attribute_types |= static_cast<u64>(type & mask) << (i * 2);
}
} else {
maxwell3d.dirty.flags[Dirty::VertexInput] = false;
enabled_divisors = 0;
for (size_t index = 0; index < Maxwell::NumVertexArrays; ++index) {
const bool is_enabled = regs.instanced_arrays.IsInstancingEnabled(index);
binding_divisors[index] = is_enabled ? regs.vertex_array[index].divisor : 0;
enabled_divisors |= (is_enabled ? u64{1} : 0) << index;
}
for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) {
const auto& input = regs.vertex_attrib_format[index];
auto& attribute = attributes[index];
attribute.raw = 0;
attribute.enabled.Assign(input.constant ? 0 : 1);
attribute.buffer.Assign(input.buffer);
attribute.offset.Assign(input.offset);
attribute.type.Assign(static_cast<u32>(input.type.Value()));
attribute.size.Assign(static_cast<u32>(input.size.Value()));
}
}
}
if (maxwell3d.dirty.flags[Dirty::Blending]) {
@ -109,10 +149,12 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d,
return static_cast<u16>(viewport.swizzle.raw);
});
}
if (!has_extended_dynamic_state) {
no_extended_dynamic_state.Assign(1);
if (!extended_dynamic_state) {
dynamic_state.Refresh(regs);
}
if (xfb_enabled) {
RefreshXfbState(xfb_state, regs);
}
}
void FixedPipelineState::BlendingAttachment::Refresh(const Maxwell& regs, size_t index) {

81
src/video_core/renderer_vulkan/fixed_pipeline_state.h
View File

@ -12,6 +12,7 @@
#include "video_core/engines/maxwell_3d.h"
#include "video_core/surface.h"
#include "video_core/transform_feedback.h"
namespace Vulkan {
@ -60,7 +61,7 @@ struct FixedPipelineState {
void Refresh(const Maxwell& regs, size_t index);
constexpr std::array<bool, 4> Mask() const noexcept {
std::array<bool, 4> Mask() const noexcept {
return {mask_r != 0, mask_g != 0, mask_b != 0, mask_a != 0};
}
@ -97,11 +98,11 @@ struct FixedPipelineState {
BitField<20, 3, u32> type;
BitField<23, 6, u32> size;
constexpr Maxwell::VertexAttribute::Type Type() const noexcept {
Maxwell::VertexAttribute::Type Type() const noexcept {
return static_cast<Maxwell::VertexAttribute::Type>(type.Value());
}
constexpr Maxwell::VertexAttribute::Size Size() const noexcept {
Maxwell::VertexAttribute::Size Size() const noexcept {
return static_cast<Maxwell::VertexAttribute::Size>(size.Value());
}
};
@ -167,37 +168,53 @@ struct FixedPipelineState {
union {
u32 raw1;
BitField<0, 1, u32> no_extended_dynamic_state;
BitField<2, 1, u32> primitive_restart_enable;
BitField<3, 1, u32> depth_bias_enable;
BitField<4, 1, u32> depth_clamp_disabled;
BitField<5, 1, u32> ndc_minus_one_to_one;
BitField<6, 2, u32> polygon_mode;
BitField<8, 5, u32> patch_control_points_minus_one;
BitField<13, 2, u32> tessellation_primitive;
BitField<15, 2, u32> tessellation_spacing;
BitField<17, 1, u32> tessellation_clockwise;
BitField<18, 1, u32> logic_op_enable;
BitField<19, 4, u32> logic_op;
BitField<23, 1, u32> rasterize_enable;
BitField<0, 1, u32> extended_dynamic_state;
BitField<1, 1, u32> dynamic_vertex_input;
BitField<2, 1, u32> xfb_enabled;
BitField<3, 1, u32> primitive_restart_enable;
BitField<4, 1, u32> depth_bias_enable;
BitField<5, 1, u32> depth_clamp_disabled;
BitField<6, 1, u32> ndc_minus_one_to_one;
BitField<7, 2, u32> polygon_mode;
BitField<9, 5, u32> patch_control_points_minus_one;
BitField<14, 2, u32> tessellation_primitive;
BitField<16, 2, u32> tessellation_spacing;
BitField<18, 1, u32> tessellation_clockwise;
BitField<19, 1, u32> logic_op_enable;
BitField<20, 4, u32> logic_op;
BitField<24, 4, Maxwell::PrimitiveTopology> topology;
BitField<28, 4, Tegra::Texture::MsaaMode> msaa_mode;
};
union {
u32 raw2;
BitField<0, 3, u32> alpha_test_func;
BitField<3, 1, u32> early_z;
BitField<0, 1, u32> rasterize_enable;
BitField<1, 3, u32> alpha_test_func;
BitField<4, 1, u32> early_z;
BitField<5, 1, u32> depth_enabled;
BitField<6, 5, u32> depth_format;
BitField<11, 1, u32> y_negate;
BitField<12, 1, u32> provoking_vertex_last;
BitField<13, 1, u32> conservative_raster_enable;
BitField<14, 1, u32> smooth_lines;
};
std::array<u8, Maxwell::NumRenderTargets> color_formats;
u32 alpha_test_ref;
u32 point_size;
std::array<u32, Maxwell::NumVertexArrays> binding_divisors;
std::array<VertexAttribute, Maxwell::NumVertexAttributes> attributes;
std::array<BlendingAttachment, Maxwell::NumRenderTargets> attachments;
std::array<u16, Maxwell::NumViewports> viewport_swizzles;
DynamicState dynamic_state;
union {
u64 attribute_types; // Used with VK_EXT_vertex_input_dynamic_state
u64 enabled_divisors;
};
std::array<VertexAttribute, Maxwell::NumVertexAttributes> attributes;
std::array<u32, Maxwell::NumVertexArrays> binding_divisors;
void Refresh(Tegra::Engines::Maxwell3D& maxwell3d, bool has_extended_dynamic_state);
DynamicState dynamic_state;
VideoCommon::TransformFeedbackState xfb_state;
void Refresh(Tegra::Engines::Maxwell3D& maxwell3d, bool has_extended_dynamic_state,
bool has_dynamic_vertex_input);
size_t Hash() const noexcept;
@ -208,8 +225,24 @@ struct FixedPipelineState {
}
size_t Size() const noexcept {
const size_t total_size = sizeof *this;
return total_size - (no_extended_dynamic_state != 0 ? 0 : sizeof(DynamicState));
if (xfb_enabled) {
// When transform feedback is enabled, use the whole struct
return sizeof(*this);
}
if (dynamic_vertex_input) {
// Exclude dynamic state and attributes
return offsetof(FixedPipelineState, attributes);
}
if (extended_dynamic_state) {
// Exclude dynamic state
return offsetof(FixedPipelineState, dynamic_state);
}
// Default
return offsetof(FixedPipelineState, xfb_state);
}
u32 DynamicAttributeType(size_t index) const noexcept {
return (attribute_types >> (index * 2)) & 0b11;
}
};
static_assert(std::has_unique_object_representations_v<FixedPipelineState>);

54
src/video_core/renderer_vulkan/maxwell_to_vk.cpp
View File

@ -157,7 +157,7 @@ struct FormatTuple {
{VK_FORMAT_R32_SFLOAT, Attachable | Storage}, // R32_FLOAT
{VK_FORMAT_R16_SFLOAT, Attachable | Storage}, // R16_FLOAT
{VK_FORMAT_R16_UNORM, Attachable | Storage}, // R16_UNORM
{VK_FORMAT_UNDEFINED}, // R16_SNORM
{VK_FORMAT_R16_SNORM, Attachable | Storage}, // R16_SNORM
{VK_FORMAT_R16_UINT, Attachable | Storage}, // R16_UINT
{VK_FORMAT_UNDEFINED}, // R16_SINT
{VK_FORMAT_R16G16_UNORM, Attachable | Storage}, // R16G16_UNORM
@ -266,19 +266,20 @@ FormatInfo SurfaceFormat(const Device& device, FormatType format_type, bool with
return {device.GetSupportedFormat(tuple.format, usage, format_type), attachable, storage};
}
VkShaderStageFlagBits ShaderStage(Tegra::Engines::ShaderType stage) {
VkShaderStageFlagBits ShaderStage(Shader::Stage stage) {
switch (stage) {
case Tegra::Engines::ShaderType::Vertex:
case Shader::Stage::VertexA:
case Shader::Stage::VertexB:
return VK_SHADER_STAGE_VERTEX_BIT;
case Tegra::Engines::ShaderType::TesselationControl:
case Shader::Stage::TessellationControl:
return VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT;
case Tegra::Engines::ShaderType::TesselationEval:
case Shader::Stage::TessellationEval:
return VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT;
case Tegra::Engines::ShaderType::Geometry:
case Shader::Stage::Geometry:
return VK_SHADER_STAGE_GEOMETRY_BIT;
case Tegra::Engines::ShaderType::Fragment:
case Shader::Stage::Fragment:
return VK_SHADER_STAGE_FRAGMENT_BIT;
case Tegra::Engines::ShaderType::Compute:
case Shader::Stage::Compute:
return VK_SHADER_STAGE_COMPUTE_BIT;
}
UNIMPLEMENTED_MSG("Unimplemented shader stage={}", stage);
@ -685,6 +686,19 @@ VkCullModeFlagBits CullFace(Maxwell::CullFace cull_face) {
return {};
}
VkPolygonMode PolygonMode(Maxwell::PolygonMode polygon_mode) {
switch (polygon_mode) {
case Maxwell::PolygonMode::Point:
return VK_POLYGON_MODE_POINT;
case Maxwell::PolygonMode::Line:
return VK_POLYGON_MODE_LINE;
case Maxwell::PolygonMode::Fill:
return VK_POLYGON_MODE_FILL;
}
UNIMPLEMENTED_MSG("Unimplemented polygon mode={}", polygon_mode);
return {};
}
VkComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle) {
switch (swizzle) {
case Tegra::Texture::SwizzleSource::Zero:
@ -741,4 +755,28 @@ VkSamplerReductionMode SamplerReduction(Tegra::Texture::SamplerReduction reducti
return VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT;
}
VkSampleCountFlagBits MsaaMode(Tegra::Texture::MsaaMode msaa_mode) {
switch (msaa_mode) {
case Tegra::Texture::MsaaMode::Msaa1x1:
return VK_SAMPLE_COUNT_1_BIT;
case Tegra::Texture::MsaaMode::Msaa2x1:
case Tegra::Texture::MsaaMode::Msaa2x1_D3D:
return VK_SAMPLE_COUNT_2_BIT;
case Tegra::Texture::MsaaMode::Msaa2x2:
case Tegra::Texture::MsaaMode::Msaa2x2_VC4:
case Tegra::Texture::MsaaMode::Msaa2x2_VC12:
return VK_SAMPLE_COUNT_4_BIT;
case Tegra::Texture::MsaaMode::Msaa4x2:
case Tegra::Texture::MsaaMode::Msaa4x2_D3D:
case Tegra::Texture::MsaaMode::Msaa4x2_VC8:
case Tegra::Texture::MsaaMode::Msaa4x2_VC24:
return VK_SAMPLE_COUNT_8_BIT;
case Tegra::Texture::MsaaMode::Msaa4x4:
return VK_SAMPLE_COUNT_16_BIT;
default:
UNREACHABLE_MSG("Invalid msaa_mode={}", static_cast<int>(msaa_mode));
return VK_SAMPLE_COUNT_1_BIT;
}
}
} // namespace Vulkan::MaxwellToVK

7
src/video_core/renderer_vulkan/maxwell_to_vk.h
View File

@ -5,6 +5,7 @@
#pragma once
#include "common/common_types.h"
#include "shader_recompiler/stage.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/surface.h"
#include "video_core/textures/texture.h"
@ -45,7 +46,7 @@ struct FormatInfo {
[[nodiscard]] FormatInfo SurfaceFormat(const Device& device, FormatType format_type, bool with_srgb,
PixelFormat pixel_format);
VkShaderStageFlagBits ShaderStage(Tegra::Engines::ShaderType stage);
VkShaderStageFlagBits ShaderStage(Shader::Stage stage);
VkPrimitiveTopology PrimitiveTopology(const Device& device, Maxwell::PrimitiveTopology topology);
@ -65,10 +66,14 @@ VkFrontFace FrontFace(Maxwell::FrontFace front_face);
VkCullModeFlagBits CullFace(Maxwell::CullFace cull_face);
VkPolygonMode PolygonMode(Maxwell::PolygonMode polygon_mode);
VkComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle);
VkViewportCoordinateSwizzleNV ViewportSwizzle(Maxwell::ViewportSwizzle swizzle);
VkSamplerReductionMode SamplerReduction(Tegra::Texture::SamplerReduction reduction);
VkSampleCountFlagBits MsaaMode(Tegra::Texture::MsaaMode msaa_mode);
} // namespace Vulkan::MaxwellToVK

154
src/video_core/renderer_vulkan/pipeline_helper.h Normal file
View File

@ -0,0 +1,154 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <cstddef>
#include <boost/container/small_vector.hpp>
#include "common/assert.h"
#include "common/common_types.h"
#include "shader_recompiler/shader_info.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/texture_cache/texture_cache.h"
#include "video_core/texture_cache/types.h"
#include "video_core/textures/texture.h"
#include "video_core/vulkan_common/vulkan_device.h"
namespace Vulkan {
class DescriptorLayoutBuilder {
public:
DescriptorLayoutBuilder(const Device& device_) : device{&device_} {}
bool CanUsePushDescriptor() const noexcept {
return device->IsKhrPushDescriptorSupported() &&
num_descriptors <= device->MaxPushDescriptors();
}
vk::DescriptorSetLayout CreateDescriptorSetLayout(bool use_push_descriptor) const {
if (bindings.empty()) {
return nullptr;
}
const VkDescriptorSetLayoutCreateFlags flags =
use_push_descriptor ? VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR : 0;
return device->GetLogical().CreateDescriptorSetLayout({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = flags,
.bindingCount = static_cast<u32>(bindings.size()),
.pBindings = bindings.data(),
});
}
vk::DescriptorUpdateTemplateKHR CreateTemplate(VkDescriptorSetLayout descriptor_set_layout,
VkPipelineLayout pipeline_layout,
bool use_push_descriptor) const {
if (entries.empty()) {
return nullptr;
}
const VkDescriptorUpdateTemplateType type =
use_push_descriptor ? VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_PUSH_DESCRIPTORS_KHR
: VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR;
return device->GetLogical().CreateDescriptorUpdateTemplateKHR({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR,
.pNext = nullptr,
.flags = 0,
.descriptorUpdateEntryCount = static_cast<u32>(entries.size()),
.pDescriptorUpdateEntries = entries.data(),
.templateType = type,
.descriptorSetLayout = descriptor_set_layout,
.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
.pipelineLayout = pipeline_layout,
.set = 0,
});
}
vk::PipelineLayout CreatePipelineLayout(VkDescriptorSetLayout descriptor_set_layout) const {
return device->GetLogical().CreatePipelineLayout({
.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.setLayoutCount = descriptor_set_layout ? 1U : 0U,
.pSetLayouts = bindings.empty() ? nullptr : &descriptor_set_layout,
.pushConstantRangeCount = 0,
.pPushConstantRanges = nullptr,
});
}
void Add(const Shader::Info& info, VkShaderStageFlags stage) {
Add(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, stage, info.constant_buffer_descriptors);
Add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, stage, info.storage_buffers_descriptors);
Add(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, stage, info.texture_buffer_descriptors);
Add(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, stage, info.image_buffer_descriptors);
Add(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, stage, info.texture_descriptors);
Add(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, stage, info.image_descriptors);
}
private:
template <typename Descriptors>
void Add(VkDescriptorType type, VkShaderStageFlags stage, const Descriptors& descriptors) {
const size_t num{descriptors.size()};
for (size_t i = 0; i < num; ++i) {
bindings.push_back({
.binding = binding,
.descriptorType = type,
.descriptorCount = descriptors[i].count,
.stageFlags = stage,
.pImmutableSamplers = nullptr,
});
entries.push_back({
.dstBinding = binding,
.dstArrayElement = 0,
.descriptorCount = descriptors[i].count,
.descriptorType = type,
.offset = offset,
.stride = sizeof(DescriptorUpdateEntry),
});
++binding;
num_descriptors += descriptors[i].count;
offset += sizeof(DescriptorUpdateEntry);
}
}
const Device* device{};
boost::container::small_vector<VkDescriptorSetLayoutBinding, 32> bindings;
boost::container::small_vector<VkDescriptorUpdateTemplateEntryKHR, 32> entries;
u32 binding{};
u32 num_descriptors{};
size_t offset{};
};
inline void PushImageDescriptors(const Shader::Info& info, const VkSampler*& samplers,
const ImageId*& image_view_ids, TextureCache& texture_cache,
VKUpdateDescriptorQueue& update_descriptor_queue) {
for (const auto& desc : info.texture_buffer_descriptors) {
image_view_ids += desc.count;
}
for (const auto& desc : info.image_buffer_descriptors) {
image_view_ids += desc.count;
}
for (const auto& desc : info.texture_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
const VkSampler sampler{*(samplers++)};
ImageView& image_view{texture_cache.GetImageView(*(image_view_ids++))};
const VkImageView vk_image_view{image_view.Handle(desc.type)};
update_descriptor_queue.AddSampledImage(vk_image_view, sampler);
}
}
for (const auto& desc : info.image_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
ImageView& image_view{texture_cache.GetImageView(*(image_view_ids++))};
if (desc.is_written) {
texture_cache.MarkModification(image_view.image_id);
}
const VkImageView vk_image_view{image_view.StorageView(desc.type, desc.format)};
update_descriptor_queue.AddImage(vk_image_view);
}
}
}
} // namespace Vulkan

64
src/video_core/renderer_vulkan/renderer_vulkan.cpp
View File

@ -130,35 +130,45 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
if (!framebuffer) {
return;
}
const auto& layout = render_window.GetFramebufferLayout();
if (layout.width > 0 && layout.height > 0 && render_window.IsShown()) {
const VAddr framebuffer_addr = framebuffer->address + framebuffer->offset;
const bool use_accelerated =
rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride);
const bool is_srgb = use_accelerated && screen_info.is_srgb;
if (swapchain.HasFramebufferChanged(layout) || swapchain.GetSrgbState() != is_srgb) {
swapchain.Create(layout.width, layout.height, is_srgb);
blit_screen.Recreate();
}
scheduler.WaitWorker();
while (!swapchain.AcquireNextImage()) {
swapchain.Create(layout.width, layout.height, is_srgb);
blit_screen.Recreate();
}
const VkSemaphore render_semaphore = blit_screen.Draw(*framebuffer, use_accelerated);
scheduler.Flush(render_semaphore);
if (swapchain.Present(render_semaphore)) {
blit_screen.Recreate();
}
gpu.RendererFrameEndNotify();
rasterizer.TickFrame();
SCOPE_EXIT({ render_window.OnFrameDisplayed(); });
if (!render_window.IsShown()) {
return;
}
const VAddr framebuffer_addr = framebuffer->address + framebuffer->offset;
const bool use_accelerated =
rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride);
const bool is_srgb = use_accelerated && screen_info.is_srgb;
render_window.OnFrameDisplayed();
bool has_been_recreated = false;
const auto recreate_swapchain = [&] {
if (!has_been_recreated) {
has_been_recreated = true;
scheduler.WaitWorker();
}
const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout();
swapchain.Create(layout.width, layout.height, is_srgb);
};
if (swapchain.IsSubOptimal() || swapchain.HasColorSpaceChanged(is_srgb)) {
recreate_swapchain();
}
bool is_outdated;
do {
swapchain.AcquireNextImage();
is_outdated = swapchain.IsOutDated();
if (is_outdated) {
recreate_swapchain();
}
} while (is_outdated);
if (has_been_recreated) {
blit_screen.Recreate();
}
const VkSemaphore render_semaphore = blit_screen.Draw(*framebuffer, use_accelerated);
scheduler.Flush(render_semaphore);
scheduler.WaitWorker();
swapchain.Present(render_semaphore);
gpu.RendererFrameEndNotify();
rasterizer.TickFrame();
}
void RendererVulkan::Report() const {

90
src/video_core/renderer_vulkan/vk_blit_screen.cpp
View File

@ -184,47 +184,43 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
.depth = 1,
},
};
scheduler.Record(
[buffer = *buffer, image = *raw_images[image_index], copy](vk::CommandBuffer cmdbuf) {
const VkImageMemoryBarrier base_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = 0,
.dstAccessMask = 0,
.oldLayout = VK_IMAGE_LAYOUT_GENERAL,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = image,
.subresourceRange =
{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = 0,
.levelCount = 1,
.baseArrayLayer = 0,
.layerCount = 1,
},
};
VkImageMemoryBarrier read_barrier = base_barrier;
read_barrier.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT;
read_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
read_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED;
scheduler.Record([this, copy, image_index](vk::CommandBuffer cmdbuf) {
const VkImage image = *raw_images[image_index];
const VkImageMemoryBarrier base_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = 0,
.dstAccessMask = 0,
.oldLayout = VK_IMAGE_LAYOUT_GENERAL,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = image,
.subresourceRange{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = 0,
.levelCount = 1,
.baseArrayLayer = 0,
.layerCount = 1,
},
};
VkImageMemoryBarrier read_barrier = base_barrier;
read_barrier.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT;
read_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
read_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED;
VkImageMemoryBarrier write_barrier = base_barrier;
write_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
write_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
VkImageMemoryBarrier write_barrier = base_barrier;
write_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
write_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
0, read_barrier);
cmdbuf.CopyBufferToImage(buffer, image, VK_IMAGE_LAYOUT_GENERAL, copy);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier);
});
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
read_barrier);
cmdbuf.CopyBufferToImage(*buffer, image, VK_IMAGE_LAYOUT_GENERAL, copy);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier);
});
}
scheduler.Record([renderpass = *renderpass, framebuffer = *framebuffers[image_index],
descriptor_set = descriptor_sets[image_index], buffer = *buffer,
size = swapchain.GetSize(), pipeline = *pipeline,
layout = *pipeline_layout](vk::CommandBuffer cmdbuf) {
scheduler.Record([this, image_index, size = swapchain.GetSize()](vk::CommandBuffer cmdbuf) {
const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f;
const f32 bg_green = Settings::values.bg_green.GetValue() / 255.0f;
const f32 bg_blue = Settings::values.bg_blue.GetValue() / 255.0f;
@ -234,8 +230,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
const VkRenderPassBeginInfo renderpass_bi{
.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
.pNext = nullptr,
.renderPass = renderpass,
.framebuffer = framebuffer,
.renderPass = *renderpass,
.framebuffer = *framebuffers[image_index],
.renderArea =
{
.offset = {0, 0},
@ -257,12 +253,13 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
.extent = size,
};
cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
cmdbuf.SetViewport(0, viewport);
cmdbuf.SetScissor(0, scissor);
cmdbuf.BindVertexBuffer(0, buffer, offsetof(BufferData, vertices));
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set, {});
cmdbuf.BindVertexBuffer(0, *buffer, offsetof(BufferData, vertices));
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline_layout, 0,
descriptor_sets[image_index], {});
cmdbuf.Draw(4, 1, 0, 0);
cmdbuf.EndRenderPass();
});
@ -304,8 +301,7 @@ void VKBlitScreen::CreateShaders() {
void VKBlitScreen::CreateSemaphores() {
semaphores.resize(image_count);
std::generate(semaphores.begin(), semaphores.end(),
[this] { return device.GetLogical().CreateSemaphore(); });
std::ranges::generate(semaphores, [this] { return device.GetLogical().CreateSemaphore(); });
}
void VKBlitScreen::CreateDescriptorPool() {
@ -633,8 +629,8 @@ void VKBlitScreen::CreateFramebuffers() {
}
void VKBlitScreen::ReleaseRawImages() {
for (std::size_t i = 0; i < raw_images.size(); ++i) {
scheduler.Wait(resource_ticks.at(i));
for (const u64 tick : resource_ticks) {
scheduler.Wait(tick);
}
raw_images.clear();
raw_buffer_commits.clear();

70
src/video_core/renderer_vulkan/vk_buffer_cache.cpp
View File

@ -60,6 +60,27 @@ std::array<T, 6> MakeQuadIndices(u32 quad, u32 first) {
}
return indices;
}
vk::Buffer CreateBuffer(const Device& device, u64 size) {
VkBufferUsageFlags flags =
VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT |
VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT |
VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT;
if (device.IsExtTransformFeedbackSupported()) {
flags |= VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT;
}
return device.GetLogical().CreateBuffer({
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = size,
.usage = flags,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
});
}
} // Anonymous namespace
Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
@ -67,31 +88,46 @@ Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
VAddr cpu_addr_, u64 size_bytes_)
: VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_) {
buffer = runtime.device.GetLogical().CreateBuffer(VkBufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = SizeBytes(),
.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT |
VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
});
: VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_),
device{&runtime.device}, buffer{CreateBuffer(*device, SizeBytes())},
commit{runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal)} {
if (runtime.device.HasDebuggingToolAttached()) {
buffer.SetObjectNameEXT(fmt::format("Buffer 0x{:x}", CpuAddr()).c_str());
}
commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
}
VkBufferView Buffer::View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format) {
if (!device) {
// Null buffer, return a null descriptor
return VK_NULL_HANDLE;
}
const auto it{std::ranges::find_if(views, [offset, size, format](const BufferView& view) {
return offset == view.offset && size == view.size && format == view.format;
})};
if (it != views.end()) {
return *it->handle;
}
views.push_back({
.offset = offset,
.size = size,
.format = format,
.handle = device->GetLogical().CreateBufferView({
.sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.buffer = *buffer,
.format = MaxwellToVK::SurfaceFormat(*device, FormatType::Buffer, false, format).format,
.offset = offset,
.range = size,
}),
});
return *views.back().handle;
}
BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_,
VKScheduler& scheduler_, StagingBufferPool& staging_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
VKDescriptorPool& descriptor_pool)
DescriptorPool& descriptor_pool)
: device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
staging_pool{staging_pool_}, update_descriptor_queue{update_descriptor_queue_},
uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),

22
src/video_core/renderer_vulkan/vk_buffer_cache.h
View File

@ -9,13 +9,14 @@
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/surface.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
class Device;
class VKDescriptorPool;
class DescriptorPool;
class VKScheduler;
class BufferCacheRuntime;
@ -26,6 +27,8 @@ public:
explicit Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
VAddr cpu_addr_, u64 size_bytes_);
[[nodiscard]] VkBufferView View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format);
[[nodiscard]] VkBuffer Handle() const noexcept {
return *buffer;
}
@ -35,8 +38,17 @@ public:
}
private:
struct BufferView {
u32 offset;
u32 size;
VideoCore::Surface::PixelFormat format;
vk::BufferView handle;
};
const Device* device{};
vk::Buffer buffer;
MemoryCommit commit;
std::vector<BufferView> views;
};
class BufferCacheRuntime {
@ -49,7 +61,7 @@ public:
explicit BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_manager_,
VKScheduler& scheduler_, StagingBufferPool& staging_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
VKDescriptorPool& descriptor_pool);
DescriptorPool& descriptor_pool);
void Finish();
@ -87,6 +99,11 @@ public:
BindBuffer(buffer, offset, size);
}
void BindTextureBuffer(Buffer& buffer, u32 offset, u32 size,
VideoCore::Surface::PixelFormat format) {
update_descriptor_queue.AddTexelBuffer(buffer.View(offset, size, format));
}
private:
void BindBuffer(VkBuffer buffer, u32 offset, u32 size) {
update_descriptor_queue.AddBuffer(buffer, offset, size);
@ -124,6 +141,7 @@ struct BufferCacheParams {
static constexpr bool NEEDS_BIND_UNIFORM_INDEX = false;
static constexpr bool NEEDS_BIND_STORAGE_INDEX = false;
static constexpr bool USE_MEMORY_MAPS = true;
static constexpr bool SEPARATE_IMAGE_BUFFER_BINDINGS = false;
};
using BufferCache = VideoCommon::BufferCache<BufferCacheParams>;

272
src/video_core/renderer_vulkan/vk_compute_pass.cpp
View File

@ -41,80 +41,92 @@ constexpr u32 ASTC_BINDING_SWIZZLE_BUFFER = 2;
constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 3;
constexpr size_t ASTC_NUM_BINDINGS = 4;
VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
return {
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.offset = 0,
.size = static_cast<u32>(size),
};
}
template <size_t size>
inline constexpr VkPushConstantRange COMPUTE_PUSH_CONSTANT_RANGE{
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.offset = 0,
.size = static_cast<u32>(size),
};
std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBindings() {
return {{
{
.binding = 0,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
}};
}
std::array<VkDescriptorSetLayoutBinding, ASTC_NUM_BINDINGS> BuildASTCDescriptorSetBindings() {
return {{
{
.binding = ASTC_BINDING_INPUT_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_ENC_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_SWIZZLE_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_OUTPUT_IMAGE,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
}};
}
VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
return {
.dstBinding = 0,
.dstArrayElement = 0,
.descriptorCount = 2,
constexpr std::array<VkDescriptorSetLayoutBinding, 2> INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS{{
{
.binding = 0,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 0,
.stride = sizeof(DescriptorUpdateEntry),
};
}
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
}};
std::array<VkDescriptorUpdateTemplateEntryKHR, ASTC_NUM_BINDINGS>
BuildASTCPassDescriptorUpdateTemplateEntry() {
return {{
constexpr DescriptorBankInfo INPUT_OUTPUT_BANK_INFO{
.uniform_buffers = 0,
.storage_buffers = 2,
.texture_buffers = 0,
.image_buffers = 0,
.textures = 0,
.images = 0,
.score = 2,
};
constexpr std::array<VkDescriptorSetLayoutBinding, 4> ASTC_DESCRIPTOR_SET_BINDINGS{{
{
.binding = ASTC_BINDING_INPUT_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_ENC_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_SWIZZLE_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_OUTPUT_IMAGE,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
}};
constexpr DescriptorBankInfo ASTC_BANK_INFO{
.uniform_buffers = 0,
.storage_buffers = 3,
.texture_buffers = 0,
.image_buffers = 0,
.textures = 0,
.images = 1,
.score = 4,
};
constexpr VkDescriptorUpdateTemplateEntryKHR INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE{
.dstBinding = 0,
.dstArrayElement = 0,
.descriptorCount = 2,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 0,
.stride = sizeof(DescriptorUpdateEntry),
};
constexpr std::array<VkDescriptorUpdateTemplateEntryKHR, ASTC_NUM_BINDINGS>
ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY{{
{
.dstBinding = ASTC_BINDING_INPUT_BUFFER,
.dstArrayElement = 0,
@ -148,7 +160,6 @@ BuildASTCPassDescriptorUpdateTemplateEntry() {
.stride = sizeof(DescriptorUpdateEntry),
},
}};
}
struct AstcPushConstants {
std::array<u32, 2> blocks_dims;
@ -159,14 +170,14 @@ struct AstcPushConstants {
u32 block_height;
u32 block_height_mask;
};
} // Anonymous namespace
VKComputePass::VKComputePass(const Device& device, VKDescriptorPool& descriptor_pool,
vk::Span<VkDescriptorSetLayoutBinding> bindings,
vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates,
vk::Span<VkPushConstantRange> push_constants,
std::span<const u32> code) {
ComputePass::ComputePass(const Device& device_, DescriptorPool& descriptor_pool,
vk::Span<VkDescriptorSetLayoutBinding> bindings,
vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates,
const DescriptorBankInfo& bank_info,
vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code)
: device{device_} {
descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
.pNext = nullptr,
@ -196,8 +207,7 @@ VKComputePass::VKComputePass(const Device& device, VKDescriptorPool& descriptor_
.pipelineLayout = *layout,
.set = 0,
});
descriptor_allocator.emplace(descriptor_pool, *descriptor_set_layout);
descriptor_allocator = descriptor_pool.Allocator(*descriptor_set_layout, bank_info);
}
module = device.GetLogical().CreateShaderModule({
.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
@ -206,43 +216,34 @@ VKComputePass::VKComputePass(const Device& device, VKDescriptorPool& descriptor_
.codeSize = static_cast<u32>(code.size_bytes()),
.pCode = code.data(),
});
device.SaveShader(code);
pipeline = device.GetLogical().CreateComputePipeline({
.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage =
{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage = VK_SHADER_STAGE_COMPUTE_BIT,
.module = *module,
.pName = "main",
.pSpecializationInfo = nullptr,
},
.stage{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage = VK_SHADER_STAGE_COMPUTE_BIT,
.module = *module,
.pName = "main",
.pSpecializationInfo = nullptr,
},
.layout = *layout,
.basePipelineHandle = nullptr,
.basePipelineIndex = 0,
});
}
VKComputePass::~VKComputePass() = default;
ComputePass::~ComputePass() = default;
VkDescriptorSet VKComputePass::CommitDescriptorSet(
VKUpdateDescriptorQueue& update_descriptor_queue) {
if (!descriptor_template) {
return nullptr;
}
const VkDescriptorSet set = descriptor_allocator->Commit();
update_descriptor_queue.Send(*descriptor_template, set);
return set;
}
Uint8Pass::Uint8Pass(const Device& device, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool, StagingBufferPool& staging_buffer_pool_,
Uint8Pass::Uint8Pass(const Device& device_, VKScheduler& scheduler_,
DescriptorPool& descriptor_pool, StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
: VKComputePass(device, descriptor_pool, BuildInputOutputDescriptorSetBindings(),
BuildInputOutputDescriptorUpdateTemplate(), {}, VULKAN_UINT8_COMP_SPV),
: ComputePass(device_, descriptor_pool, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS,
INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, {},
VULKAN_UINT8_COMP_SPV),
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_} {}
@ -256,11 +257,11 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices);
update_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size);
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
const void* const descriptor_data{update_descriptor_queue.UpdateData()};
const VkBuffer buffer{staging.buffer};
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging.buffer, set,
num_vertices](vk::CommandBuffer cmdbuf) {
scheduler.Record([this, buffer, descriptor_data, num_vertices](vk::CommandBuffer cmdbuf) {
static constexpr u32 DISPATCH_SIZE = 1024;
static constexpr VkMemoryBarrier WRITE_BARRIER{
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
@ -268,8 +269,10 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT,
};
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
const VkDescriptorSet set = descriptor_allocator.Commit();
device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
cmdbuf.Dispatch(Common::DivCeil(num_vertices, DISPATCH_SIZE), 1, 1);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, WRITE_BARRIER);
@ -278,12 +281,12 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer
}
QuadIndexedPass::QuadIndexedPass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
DescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
: VKComputePass(device_, descriptor_pool_, BuildInputOutputDescriptorSetBindings(),
BuildInputOutputDescriptorUpdateTemplate(),
BuildComputePushConstantRange(sizeof(u32) * 2), VULKAN_QUAD_INDEXED_COMP_SPV),
: ComputePass(device_, descriptor_pool_, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS,
INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO,
COMPUTE_PUSH_CONSTANT_RANGE<sizeof(u32) * 2>, VULKAN_QUAD_INDEXED_COMP_SPV),
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_} {}
@ -313,11 +316,11 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size);
update_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size);
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
const void* const descriptor_data{update_descriptor_queue.UpdateData()};
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging.buffer, set,
num_tri_vertices, base_vertex, index_shift](vk::CommandBuffer cmdbuf) {
scheduler.Record([this, buffer = staging.buffer, descriptor_data, num_tri_vertices, base_vertex,
index_shift](vk::CommandBuffer cmdbuf) {
static constexpr u32 DISPATCH_SIZE = 1024;
static constexpr VkMemoryBarrier WRITE_BARRIER{
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
@ -325,10 +328,12 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT,
};
const std::array push_constants = {base_vertex, index_shift};
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants),
const std::array push_constants{base_vertex, index_shift};
const VkDescriptorSet set = descriptor_allocator.Commit();
device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants),
&push_constants);
cmdbuf.Dispatch(Common::DivCeil(num_tri_vertices, DISPATCH_SIZE), 1, 1);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
@ -338,15 +343,14 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
}
ASTCDecoderPass::ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
DescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
MemoryAllocator& memory_allocator_)
: VKComputePass(device_, descriptor_pool_, BuildASTCDescriptorSetBindings(),
BuildASTCPassDescriptorUpdateTemplateEntry(),
BuildComputePushConstantRange(sizeof(AstcPushConstants)),
ASTC_DECODER_COMP_SPV),
device{device_}, scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
: ComputePass(device_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS,
ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO,
COMPUTE_PUSH_CONSTANT_RANGE<sizeof(AstcPushConstants)>, ASTC_DECODER_COMP_SPV),
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_}, memory_allocator{memory_allocator_} {}
ASTCDecoderPass::~ASTCDecoderPass() = default;
@ -444,16 +448,14 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
update_descriptor_queue.AddBuffer(*data_buffer, sizeof(ASTC_ENCODINGS_VALUES),
sizeof(SWIZZLE_TABLE));
update_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
const VkPipelineLayout vk_layout = *layout;
const void* const descriptor_data{update_descriptor_queue.UpdateData()};
// To unswizzle the ASTC data
const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
scheduler.Record([vk_layout, num_dispatches_x, num_dispatches_y, num_dispatches_z,
block_dims, params, set](vk::CommandBuffer cmdbuf) {
scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, block_dims,
params, descriptor_data](vk::CommandBuffer cmdbuf) {
const AstcPushConstants uniforms{
.blocks_dims = block_dims,
.bytes_per_block_log2 = params.bytes_per_block_log2,
@ -463,8 +465,10 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
.block_height = params.block_height,
.block_height_mask = params.block_height_mask,
};
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, vk_layout, 0, set, {});
cmdbuf.PushConstants(vk_layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms);
const VkDescriptorSet set = descriptor_allocator.Commit();
device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms);
cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z);
});
}

34
src/video_core/renderer_vulkan/vk_compute_pass.h
View File

@ -4,7 +4,6 @@
#pragma once
#include <optional>
#include <span>
#include <utility>
@ -27,31 +26,31 @@ class VKUpdateDescriptorQueue;
class Image;
struct StagingBufferRef;
class VKComputePass {
class ComputePass {
public:
explicit VKComputePass(const Device& device, VKDescriptorPool& descriptor_pool,
vk::Span<VkDescriptorSetLayoutBinding> bindings,
vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates,
vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code);
~VKComputePass();
explicit ComputePass(const Device& device, DescriptorPool& descriptor_pool,
vk::Span<VkDescriptorSetLayoutBinding> bindings,
vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates,
const DescriptorBankInfo& bank_info,
vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code);
~ComputePass();
protected:
VkDescriptorSet CommitDescriptorSet(VKUpdateDescriptorQueue& update_descriptor_queue);
const Device& device;
vk::DescriptorUpdateTemplateKHR descriptor_template;
vk::PipelineLayout layout;
vk::Pipeline pipeline;
vk::DescriptorSetLayout descriptor_set_layout;
DescriptorAllocator descriptor_allocator;
private:
vk::DescriptorSetLayout descriptor_set_layout;
std::optional<DescriptorAllocator> descriptor_allocator;
vk::ShaderModule module;
};
class Uint8Pass final : public VKComputePass {
class Uint8Pass final : public ComputePass {
public:
explicit Uint8Pass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_,
DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_);
~Uint8Pass();
@ -66,10 +65,10 @@ private:
VKUpdateDescriptorQueue& update_descriptor_queue;
};
class QuadIndexedPass final : public VKComputePass {
class QuadIndexedPass final : public ComputePass {
public:
explicit QuadIndexedPass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
DescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_);
~QuadIndexedPass();
@ -84,10 +83,10 @@ private:
VKUpdateDescriptorQueue& update_descriptor_queue;
};
class ASTCDecoderPass final : public VKComputePass {
class ASTCDecoderPass final : public ComputePass {
public:
explicit ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
DescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
MemoryAllocator& memory_allocator_);
@ -99,7 +98,6 @@ public:
private:
void MakeDataBuffer();
const Device& device;
VKScheduler& scheduler;
StagingBufferPool& staging_buffer_pool;
VKUpdateDescriptorQueue& update_descriptor_queue;

286
src/video_core/renderer_vulkan/vk_compute_pipeline.cpp
View File

@ -2,152 +2,198 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <vector>
#include <boost/container/small_vector.hpp>
#include "video_core/renderer_vulkan/pipeline_helper.h"
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/shader_notify.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
VKComputePipeline::VKComputePipeline(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
const SPIRVShader& shader_)
: device{device_}, scheduler{scheduler_}, entries{shader_.entries},
descriptor_set_layout{CreateDescriptorSetLayout()},
descriptor_allocator{descriptor_pool_, *descriptor_set_layout},
update_descriptor_queue{update_descriptor_queue_}, layout{CreatePipelineLayout()},
descriptor_template{CreateDescriptorUpdateTemplate()},
shader_module{CreateShaderModule(shader_.code)}, pipeline{CreatePipeline()} {}
using Shader::ImageBufferDescriptor;
using Tegra::Texture::TexturePair;
VKComputePipeline::~VKComputePipeline() = default;
VkDescriptorSet VKComputePipeline::CommitDescriptorSet() {
if (!descriptor_template) {
return {};
ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue_,
Common::ThreadWorker* thread_worker,
VideoCore::ShaderNotify* shader_notify, const Shader::Info& info_,
vk::ShaderModule spv_module_)
: device{device_}, update_descriptor_queue{update_descriptor_queue_}, info{info_},
spv_module(std::move(spv_module_)) {
if (shader_notify) {
shader_notify->MarkShaderBuilding();
}
const VkDescriptorSet set = descriptor_allocator.Commit();
update_descriptor_queue.Send(*descriptor_template, set);
return set;
}
std::copy_n(info.constant_buffer_used_sizes.begin(), uniform_buffer_sizes.size(),
uniform_buffer_sizes.begin());
vk::DescriptorSetLayout VKComputePipeline::CreateDescriptorSetLayout() const {
std::vector<VkDescriptorSetLayoutBinding> bindings;
u32 binding = 0;
const auto add_bindings = [&](VkDescriptorType descriptor_type, std::size_t num_entries) {
// TODO(Rodrigo): Maybe make individual bindings here?
for (u32 bindpoint = 0; bindpoint < static_cast<u32>(num_entries); ++bindpoint) {
bindings.push_back({
.binding = binding++,
.descriptorType = descriptor_type,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
});
}
};
add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, entries.const_buffers.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, entries.global_buffers.size());
add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.uniform_texels.size());
add_bindings(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, entries.samplers.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, entries.storage_texels.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, entries.images.size());
auto func{[this, &descriptor_pool, shader_notify] {
DescriptorLayoutBuilder builder{device};
builder.Add(info, VK_SHADER_STAGE_COMPUTE_BIT);
return device.GetLogical().CreateDescriptorSetLayout({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.bindingCount = static_cast<u32>(bindings.size()),
.pBindings = bindings.data(),
});
}
vk::PipelineLayout VKComputePipeline::CreatePipelineLayout() const {
return device.GetLogical().CreatePipelineLayout({
.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.setLayoutCount = 1,
.pSetLayouts = descriptor_set_layout.address(),
.pushConstantRangeCount = 0,
.pPushConstantRanges = nullptr,
});
}
vk::DescriptorUpdateTemplateKHR VKComputePipeline::CreateDescriptorUpdateTemplate() const {
std::vector<VkDescriptorUpdateTemplateEntryKHR> template_entries;
u32 binding = 0;
u32 offset = 0;
FillDescriptorUpdateTemplateEntries(entries, binding, offset, template_entries);
if (template_entries.empty()) {
// If the shader doesn't use descriptor sets, skip template creation.
return {};
}
return device.GetLogical().CreateDescriptorUpdateTemplateKHR({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR,
.pNext = nullptr,
.flags = 0,
.descriptorUpdateEntryCount = static_cast<u32>(template_entries.size()),
.pDescriptorUpdateEntries = template_entries.data(),
.templateType = VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR,
.descriptorSetLayout = *descriptor_set_layout,
.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
.pipelineLayout = *layout,
.set = DESCRIPTOR_SET,
});
}
vk::ShaderModule VKComputePipeline::CreateShaderModule(const std::vector<u32>& code) const {
device.SaveShader(code);
return device.GetLogical().CreateShaderModule({
.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.codeSize = code.size() * sizeof(u32),
.pCode = code.data(),
});
}
vk::Pipeline VKComputePipeline::CreatePipeline() const {
VkComputePipelineCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage =
{
descriptor_set_layout = builder.CreateDescriptorSetLayout(false);
pipeline_layout = builder.CreatePipelineLayout(*descriptor_set_layout);
descriptor_update_template =
builder.CreateTemplate(*descriptor_set_layout, *pipeline_layout, false);
descriptor_allocator = descriptor_pool.Allocator(*descriptor_set_layout, info);
const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT,
.pNext = nullptr,
.requiredSubgroupSize = GuestWarpSize,
};
pipeline = device.GetLogical().CreateComputePipeline({
.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
.pNext = nullptr,
.pNext = device.IsExtSubgroupSizeControlSupported() ? &subgroup_size_ci : nullptr,
.flags = 0,
.stage = VK_SHADER_STAGE_COMPUTE_BIT,
.module = *shader_module,
.module = *spv_module,
.pName = "main",
.pSpecializationInfo = nullptr,
},
.layout = *layout,
.basePipelineHandle = nullptr,
.basePipelineIndex = 0,
};
.layout = *pipeline_layout,
.basePipelineHandle = 0,
.basePipelineIndex = 0,
});
std::lock_guard lock{build_mutex};
is_built = true;
build_condvar.notify_one();
if (shader_notify) {
shader_notify->MarkShaderComplete();
}
}};
if (thread_worker) {
thread_worker->QueueWork(std::move(func));
} else {
func();
}
}
const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT,
.pNext = nullptr,
.requiredSubgroupSize = GuestWarpSize,
};
void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
Tegra::MemoryManager& gpu_memory, VKScheduler& scheduler,
BufferCache& buffer_cache, TextureCache& texture_cache) {
update_descriptor_queue.Acquire();
if (entries.uses_warps && device.IsGuestWarpSizeSupported(VK_SHADER_STAGE_COMPUTE_BIT)) {
ci.stage.pNext = &subgroup_size_ci;
buffer_cache.SetComputeUniformBufferState(info.constant_buffer_mask, &uniform_buffer_sizes);
buffer_cache.UnbindComputeStorageBuffers();
size_t ssbo_index{};
for (const auto& desc : info.storage_buffers_descriptors) {
ASSERT(desc.count == 1);
buffer_cache.BindComputeStorageBuffer(ssbo_index, desc.cbuf_index, desc.cbuf_offset,
desc.is_written);
++ssbo_index;
}
return device.GetLogical().CreateComputePipeline(ci);
texture_cache.SynchronizeComputeDescriptors();
static constexpr size_t max_elements = 64;
std::array<ImageId, max_elements> image_view_ids;
boost::container::static_vector<u32, max_elements> image_view_indices;
boost::container::static_vector<VkSampler, max_elements> samplers;
const auto& qmd{kepler_compute.launch_description};
const auto& cbufs{qmd.const_buffer_config};
const bool via_header_index{qmd.linked_tsc != 0};
const auto read_handle{[&](const auto& desc, u32 index) {
ASSERT(((qmd.const_buffer_enable_mask >> desc.cbuf_index) & 1) != 0);
const u32 index_offset{index << desc.size_shift};
const u32 offset{desc.cbuf_offset + index_offset};
const GPUVAddr addr{cbufs[desc.cbuf_index].Address() + offset};
if constexpr (std::is_same_v<decltype(desc), const Shader::TextureDescriptor&> ||
std::is_same_v<decltype(desc), const Shader::TextureBufferDescriptor&>) {
if (desc.has_secondary) {
ASSERT(((qmd.const_buffer_enable_mask >> desc.secondary_cbuf_index) & 1) != 0);
const u32 secondary_offset{desc.secondary_cbuf_offset + index_offset};
const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].Address() +
secondary_offset};
const u32 lhs_raw{gpu_memory.Read<u32>(addr)};
const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)};
return TexturePair(lhs_raw | rhs_raw, via_header_index);
}
}
return TexturePair(gpu_memory.Read<u32>(addr), via_header_index);
}};
const auto add_image{[&](const auto& desc) {
for (u32 index = 0; index < desc.count; ++index) {
const auto handle{read_handle(desc, index)};
image_view_indices.push_back(handle.first);
}
}};
std::ranges::for_each(info.texture_buffer_descriptors, add_image);
std::ranges::for_each(info.image_buffer_descriptors, add_image);
for (const auto& desc : info.texture_descriptors) {
for (u32 index = 0; index < desc.count; ++index) {
const auto handle{read_handle(desc, index)};
image_view_indices.push_back(handle.first);
Sampler* const sampler = texture_cache.GetComputeSampler(handle.second);
samplers.push_back(sampler->Handle());
}
}
std::ranges::for_each(info.image_descriptors, add_image);
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
texture_cache.FillComputeImageViews(indices_span, image_view_ids);
buffer_cache.UnbindComputeTextureBuffers();
ImageId* texture_buffer_ids{image_view_ids.data()};
size_t index{};
const auto add_buffer{[&](const auto& desc) {
constexpr bool is_image = std::is_same_v<decltype(desc), const ImageBufferDescriptor&>;
for (u32 i = 0; i < desc.count; ++i) {
bool is_written{false};
if constexpr (is_image) {
is_written = desc.is_written;
}
ImageView& image_view = texture_cache.GetImageView(*texture_buffer_ids);
buffer_cache.BindComputeTextureBuffer(index, image_view.GpuAddr(),
image_view.BufferSize(), image_view.format,
is_written, is_image);
++texture_buffer_ids;
++index;
}
}};
std::ranges::for_each(info.texture_buffer_descriptors, add_buffer);
std::ranges::for_each(info.image_buffer_descriptors, add_buffer);
buffer_cache.UpdateComputeBuffers();
buffer_cache.BindHostComputeBuffers();
const VkSampler* samplers_it{samplers.data()};
const ImageId* views_it{image_view_ids.data()};
PushImageDescriptors(info, samplers_it, views_it, texture_cache, update_descriptor_queue);
if (!is_built.load(std::memory_order::relaxed)) {
// Wait for the pipeline to be built
scheduler.Record([this](vk::CommandBuffer) {
std::unique_lock lock{build_mutex};
build_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); });
});
}
const void* const descriptor_data{update_descriptor_queue.UpdateData()};
scheduler.Record([this, descriptor_data](vk::CommandBuffer cmdbuf) {
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
if (!descriptor_set_layout) {
return;
}
const VkDescriptorSet descriptor_set{descriptor_allocator.Commit()};
const vk::Device& dev{device.GetLogical()};
dev.UpdateDescriptorSet(descriptor_set, *descriptor_update_template, descriptor_data);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline_layout, 0,
descriptor_set, nullptr);
});
}
} // namespace Vulkan

72
src/video_core/renderer_vulkan/vk_compute_pipeline.h
View File

@ -4,61 +4,63 @@
#pragma once
#include <atomic>
#include <condition_variable>
#include <mutex>
#include "common/common_types.h"
#include "common/thread_worker.h"
#include "shader_recompiler/shader_info.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace VideoCore {
class ShaderNotify;
}
namespace Vulkan {
class Device;
class VKScheduler;
class VKUpdateDescriptorQueue;
class VKComputePipeline final {
class ComputePipeline {
public:
explicit VKComputePipeline(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
const SPIRVShader& shader_);
~VKComputePipeline();
explicit ComputePipeline(const Device& device, DescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue,
Common::ThreadWorker* thread_worker,
VideoCore::ShaderNotify* shader_notify, const Shader::Info& info,
vk::ShaderModule spv_module);
VkDescriptorSet CommitDescriptorSet();
ComputePipeline& operator=(ComputePipeline&&) noexcept = delete;
ComputePipeline(ComputePipeline&&) noexcept = delete;
VkPipeline GetHandle() const {
return *pipeline;
}
ComputePipeline& operator=(const ComputePipeline&) = delete;
ComputePipeline(const ComputePipeline&) = delete;
VkPipelineLayout GetLayout() const {
return *layout;
}
const ShaderEntries& GetEntries() const {
return entries;
}
void Configure(Tegra::Engines::KeplerCompute& kepler_compute, Tegra::MemoryManager& gpu_memory,
VKScheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache);
private:
vk::DescriptorSetLayout CreateDescriptorSetLayout() const;
vk::PipelineLayout CreatePipelineLayout() const;
vk::DescriptorUpdateTemplateKHR CreateDescriptorUpdateTemplate() const;
vk::ShaderModule CreateShaderModule(const std::vector<u32>& code) const;
vk::Pipeline CreatePipeline() const;
const Device& device;
VKScheduler& scheduler;
ShaderEntries entries;
VKUpdateDescriptorQueue& update_descriptor_queue;
Shader::Info info;
VideoCommon::ComputeUniformBufferSizes uniform_buffer_sizes{};
vk::ShaderModule spv_module;
vk::DescriptorSetLayout descriptor_set_layout;
DescriptorAllocator descriptor_allocator;
VKUpdateDescriptorQueue& update_descriptor_queue;
vk::PipelineLayout layout;
vk::DescriptorUpdateTemplateKHR descriptor_template;
vk::ShaderModule shader_module;
vk::PipelineLayout pipeline_layout;
vk::DescriptorUpdateTemplateKHR descriptor_update_template;
vk::Pipeline pipeline;
std::condition_variable build_condvar;
std::mutex build_mutex;
std::atomic_bool is_built{false};
};
} // namespace Vulkan

172
src/video_core/renderer_vulkan/vk_descriptor_pool.cpp
View File

@ -2,6 +2,8 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <mutex>
#include <span>
#include <vector>
#include "common/common_types.h"
@ -13,79 +15,149 @@
namespace Vulkan {
// Prefer small grow rates to avoid saturating the descriptor pool with barely used pipelines.
constexpr std::size_t SETS_GROW_RATE = 0x20;
// Prefer small grow rates to avoid saturating the descriptor pool with barely used pipelines
constexpr size_t SETS_GROW_RATE = 16;
constexpr s32 SCORE_THRESHOLD = 3;
constexpr u32 SETS_PER_POOL = 64;
DescriptorAllocator::DescriptorAllocator(VKDescriptorPool& descriptor_pool_,
VkDescriptorSetLayout layout_)
: ResourcePool(descriptor_pool_.master_semaphore, SETS_GROW_RATE),
descriptor_pool{descriptor_pool_}, layout{layout_} {}
struct DescriptorBank {
DescriptorBankInfo info;
std::vector<vk::DescriptorPool> pools;
};
DescriptorAllocator::~DescriptorAllocator() = default;
VkDescriptorSet DescriptorAllocator::Commit() {
const std::size_t index = CommitResource();
return descriptors_allocations[index / SETS_GROW_RATE][index % SETS_GROW_RATE];
bool DescriptorBankInfo::IsSuperset(const DescriptorBankInfo& subset) const noexcept {
return uniform_buffers >= subset.uniform_buffers && storage_buffers >= subset.storage_buffers &&
texture_buffers >= subset.texture_buffers && image_buffers >= subset.image_buffers &&
textures >= subset.textures && images >= subset.image_buffers;
}
void DescriptorAllocator::Allocate(std::size_t begin, std::size_t end) {
descriptors_allocations.push_back(descriptor_pool.AllocateDescriptors(layout, end - begin));
template <typename Descriptors>
static u32 Accumulate(const Descriptors& descriptors) {
u32 count = 0;
for (const auto& descriptor : descriptors) {
count += descriptor.count;
}
return count;
}
VKDescriptorPool::VKDescriptorPool(const Device& device_, VKScheduler& scheduler)
: device{device_}, master_semaphore{scheduler.GetMasterSemaphore()}, active_pool{
AllocateNewPool()} {}
static DescriptorBankInfo MakeBankInfo(std::span<const Shader::Info> infos) {
DescriptorBankInfo bank;
for (const Shader::Info& info : infos) {
bank.uniform_buffers += Accumulate(info.constant_buffer_descriptors);
bank.storage_buffers += Accumulate(info.storage_buffers_descriptors);
bank.texture_buffers += Accumulate(info.texture_buffer_descriptors);
bank.image_buffers += Accumulate(info.image_buffer_descriptors);
bank.textures += Accumulate(info.texture_descriptors);
bank.images += Accumulate(info.image_descriptors);
}
bank.score = bank.uniform_buffers + bank.storage_buffers + bank.texture_buffers +
bank.image_buffers + bank.textures + bank.images;
return bank;
}
VKDescriptorPool::~VKDescriptorPool() = default;
vk::DescriptorPool* VKDescriptorPool::AllocateNewPool() {
static constexpr u32 num_sets = 0x20000;
static constexpr VkDescriptorPoolSize pool_sizes[] = {
{VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, num_sets * 90},
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, num_sets * 60},
{VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, num_sets * 64},
{VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, num_sets * 64},
{VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, num_sets * 64},
{VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, num_sets * 40},
static void AllocatePool(const Device& device, DescriptorBank& bank) {
std::array<VkDescriptorPoolSize, 6> pool_sizes;
size_t pool_cursor{};
const auto add = [&](VkDescriptorType type, u32 count) {
if (count > 0) {
pool_sizes[pool_cursor++] = {
.type = type,
.descriptorCount = count * SETS_PER_POOL,
};
}
};
const VkDescriptorPoolCreateInfo ci{
const auto& info{bank.info};
add(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, info.uniform_buffers);
add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, info.storage_buffers);
add(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, info.texture_buffers);
add(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, info.image_buffers);
add(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, info.textures);
add(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, info.images);
bank.pools.push_back(device.GetLogical().CreateDescriptorPool({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
.pNext = nullptr,
.flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT,
.maxSets = num_sets,
.poolSizeCount = static_cast<u32>(std::size(pool_sizes)),
.maxSets = SETS_PER_POOL,
.poolSizeCount = static_cast<u32>(pool_cursor),
.pPoolSizes = std::data(pool_sizes),
};
return &pools.emplace_back(device.GetLogical().CreateDescriptorPool(ci));
}));
}
vk::DescriptorSets VKDescriptorPool::AllocateDescriptors(VkDescriptorSetLayout layout,
std::size_t count) {
const std::vector layout_copies(count, layout);
VkDescriptorSetAllocateInfo ai{
DescriptorAllocator::DescriptorAllocator(const Device& device_, MasterSemaphore& master_semaphore_,
DescriptorBank& bank_, VkDescriptorSetLayout layout_)
: ResourcePool(master_semaphore_, SETS_GROW_RATE), device{&device_}, bank{&bank_},
layout{layout_} {}
VkDescriptorSet DescriptorAllocator::Commit() {
const size_t index = CommitResource();
return sets[index / SETS_GROW_RATE][index % SETS_GROW_RATE];
}
void DescriptorAllocator::Allocate(size_t begin, size_t end) {
sets.push_back(AllocateDescriptors(end - begin));
}
vk::DescriptorSets DescriptorAllocator::AllocateDescriptors(size_t count) {
const std::vector<VkDescriptorSetLayout> layouts(count, layout);
VkDescriptorSetAllocateInfo allocate_info{
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
.pNext = nullptr,
.descriptorPool = **active_pool,
.descriptorPool = *bank->pools.back(),
.descriptorSetCount = static_cast<u32>(count),
.pSetLayouts = layout_copies.data(),
.pSetLayouts = layouts.data(),
};
vk::DescriptorSets sets = active_pool->Allocate(ai);
if (!sets.IsOutOfPoolMemory()) {
return sets;
vk::DescriptorSets new_sets = bank->pools.back().Allocate(allocate_info);
if (!new_sets.IsOutOfPoolMemory()) {
return new_sets;
}
// Our current pool is out of memory. Allocate a new one and retry
active_pool = AllocateNewPool();
ai.descriptorPool = **active_pool;
sets = active_pool->Allocate(ai);
if (!sets.IsOutOfPoolMemory()) {
return sets;
AllocatePool(*device, *bank);
allocate_info.descriptorPool = *bank->pools.back();
new_sets = bank->pools.back().Allocate(allocate_info);
if (!new_sets.IsOutOfPoolMemory()) {
return new_sets;
}
// After allocating a new pool, we are out of memory again. We can't handle this from here.
throw vk::Exception(VK_ERROR_OUT_OF_POOL_MEMORY);
}
DescriptorPool::DescriptorPool(const Device& device_, VKScheduler& scheduler)
: device{device_}, master_semaphore{scheduler.GetMasterSemaphore()} {}
DescriptorPool::~DescriptorPool() = default;
DescriptorAllocator DescriptorPool::Allocator(VkDescriptorSetLayout layout,
std::span<const Shader::Info> infos) {
return Allocator(layout, MakeBankInfo(infos));
}
DescriptorAllocator DescriptorPool::Allocator(VkDescriptorSetLayout layout,
const Shader::Info& info) {
return Allocator(layout, MakeBankInfo(std::array{info}));
}
DescriptorAllocator DescriptorPool::Allocator(VkDescriptorSetLayout layout,
const DescriptorBankInfo& info) {
return DescriptorAllocator(device, master_semaphore, Bank(info), layout);
}
DescriptorBank& DescriptorPool::Bank(const DescriptorBankInfo& reqs) {
std::shared_lock read_lock{banks_mutex};
const auto it = std::ranges::find_if(bank_infos, [&reqs](const DescriptorBankInfo& bank) {
return std::abs(bank.score - reqs.score) < SCORE_THRESHOLD && bank.IsSuperset(reqs);
});
if (it != bank_infos.end()) {
return *banks[std::distance(bank_infos.begin(), it)].get();
}
read_lock.unlock();
std::unique_lock write_lock{banks_mutex};
bank_infos.push_back(reqs);
auto& bank = *banks.emplace_back(std::make_unique<DescriptorBank>());
bank.info = reqs;
AllocatePool(device, bank);
return bank;
}
} // namespace Vulkan

70
src/video_core/renderer_vulkan/vk_descriptor_pool.h
View File

@ -4,57 +4,85 @@
#pragma once
#include <shared_mutex>
#include <span>
#include <vector>
#include "shader_recompiler/shader_info.h"
#include "video_core/renderer_vulkan/vk_resource_pool.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
class Device;
class VKDescriptorPool;
class VKScheduler;
struct DescriptorBank;
struct DescriptorBankInfo {
[[nodiscard]] bool IsSuperset(const DescriptorBankInfo& subset) const noexcept;
u32 uniform_buffers{}; ///< Number of uniform buffer descriptors
u32 storage_buffers{}; ///< Number of storage buffer descriptors
u32 texture_buffers{}; ///< Number of texture buffer descriptors
u32 image_buffers{}; ///< Number of image buffer descriptors
u32 textures{}; ///< Number of texture descriptors
u32 images{}; ///< Number of image descriptors
s32 score{}; ///< Number of descriptors in total
};
class DescriptorAllocator final : public ResourcePool {
friend class DescriptorPool;
public:
explicit DescriptorAllocator(VKDescriptorPool& descriptor_pool, VkDescriptorSetLayout layout);
~DescriptorAllocator() override;
explicit DescriptorAllocator() = default;
~DescriptorAllocator() override = default;
DescriptorAllocator& operator=(DescriptorAllocator&&) noexcept = default;
DescriptorAllocator(DescriptorAllocator&&) noexcept = default;
DescriptorAllocator& operator=(const DescriptorAllocator&) = delete;
DescriptorAllocator(const DescriptorAllocator&) = delete;
VkDescriptorSet Commit();
protected:
void Allocate(std::size_t begin, std::size_t end) override;
private:
VKDescriptorPool& descriptor_pool;
const VkDescriptorSetLayout layout;
explicit DescriptorAllocator(const Device& device_, MasterSemaphore& master_semaphore_,
DescriptorBank& bank_, VkDescriptorSetLayout layout_);
std::vector<vk::DescriptorSets> descriptors_allocations;
void Allocate(size_t begin, size_t end) override;
vk::DescriptorSets AllocateDescriptors(size_t count);
const Device* device{};
DescriptorBank* bank{};
VkDescriptorSetLayout layout{};
std::vector<vk::DescriptorSets> sets;
};
class VKDescriptorPool final {
friend DescriptorAllocator;
class DescriptorPool {
public:
explicit VKDescriptorPool(const Device& device, VKScheduler& scheduler);
~VKDescriptorPool();
explicit DescriptorPool(const Device& device, VKScheduler& scheduler);
~DescriptorPool();
VKDescriptorPool(const VKDescriptorPool&) = delete;
VKDescriptorPool& operator=(const VKDescriptorPool&) = delete;
DescriptorPool& operator=(const DescriptorPool&) = delete;
DescriptorPool(const DescriptorPool&) = delete;
DescriptorAllocator Allocator(VkDescriptorSetLayout layout,
std::span<const Shader::Info> infos);
DescriptorAllocator Allocator(VkDescriptorSetLayout layout, const Shader::Info& info);
DescriptorAllocator Allocator(VkDescriptorSetLayout layout, const DescriptorBankInfo& info);
private:
vk::DescriptorPool* AllocateNewPool();
vk::DescriptorSets AllocateDescriptors(VkDescriptorSetLayout layout, std::size_t count);
DescriptorBank& Bank(const DescriptorBankInfo& reqs);
const Device& device;
MasterSemaphore& master_semaphore;
std::vector<vk::DescriptorPool> pools;
vk::DescriptorPool* active_pool;
std::shared_mutex banks_mutex;
std::vector<DescriptorBankInfo> bank_infos;
std::vector<std::unique_ptr<DescriptorBank>> banks;
};
} // namespace Vulkan

859
src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
View File

File diff suppressed because it is too large Load Diff

141
src/video_core/renderer_vulkan/vk_graphics_pipeline.h
View File

@ -1,30 +1,36 @@
// Copyright 2019 yuzu Emulator Project
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <algorithm>
#include <array>
#include <optional>
#include <vector>
#include <atomic>
#include <condition_variable>
#include <mutex>
#include <type_traits>
#include "common/common_types.h"
#include "common/thread_worker.h"
#include "shader_recompiler/shader_info.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace VideoCore {
class ShaderNotify;
}
namespace Vulkan {
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
struct GraphicsPipelineCacheKey {
VkRenderPass renderpass;
std::array<GPUVAddr, Maxwell::MaxShaderProgram> shaders;
FixedPipelineState fixed_state;
std::array<u64, 6> unique_hashes;
FixedPipelineState state;
std::size_t Hash() const noexcept;
size_t Hash() const noexcept;
bool operator==(const GraphicsPipelineCacheKey& rhs) const noexcept;
@ -32,72 +38,115 @@ struct GraphicsPipelineCacheKey {
return !operator==(rhs);
}
std::size_t Size() const noexcept {
return sizeof(renderpass) + sizeof(shaders) + fixed_state.Size();
size_t Size() const noexcept {
return sizeof(unique_hashes) + state.Size();
}
};
static_assert(std::has_unique_object_representations_v<GraphicsPipelineCacheKey>);
static_assert(std::is_trivially_copyable_v<GraphicsPipelineCacheKey>);
static_assert(std::is_trivially_constructible_v<GraphicsPipelineCacheKey>);
} // namespace Vulkan
namespace std {
template <>
struct hash<Vulkan::GraphicsPipelineCacheKey> {
size_t operator()(const Vulkan::GraphicsPipelineCacheKey& k) const noexcept {
return k.Hash();
}
};
} // namespace std
namespace Vulkan {
class Device;
class VKDescriptorPool;
class RenderPassCache;
class VKScheduler;
class VKUpdateDescriptorQueue;
using SPIRVProgram = std::array<std::optional<SPIRVShader>, Maxwell::MaxShaderStage>;
class GraphicsPipeline {
static constexpr size_t NUM_STAGES = Tegra::Engines::Maxwell3D::Regs::MaxShaderStage;
class VKGraphicsPipeline final {
public:
explicit VKGraphicsPipeline(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue_,
const GraphicsPipelineCacheKey& key,
vk::Span<VkDescriptorSetLayoutBinding> bindings,
const SPIRVProgram& program, u32 num_color_buffers);
~VKGraphicsPipeline();
explicit GraphicsPipeline(
Tegra::Engines::Maxwell3D& maxwell3d, Tegra::MemoryManager& gpu_memory,
VKScheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache,
VideoCore::ShaderNotify* shader_notify, const Device& device,
DescriptorPool& descriptor_pool, VKUpdateDescriptorQueue& update_descriptor_queue,
Common::ThreadWorker* worker_thread, RenderPassCache& render_pass_cache,
const GraphicsPipelineCacheKey& key, std::array<vk::ShaderModule, NUM_STAGES> stages,
const std::array<const Shader::Info*, NUM_STAGES>& infos);
VkDescriptorSet CommitDescriptorSet();
GraphicsPipeline& operator=(GraphicsPipeline&&) noexcept = delete;
GraphicsPipeline(GraphicsPipeline&&) noexcept = delete;
VkPipeline GetHandle() const {
return *pipeline;
GraphicsPipeline& operator=(const GraphicsPipeline&) = delete;
GraphicsPipeline(const GraphicsPipeline&) = delete;
void AddTransition(GraphicsPipeline* transition);
void Configure(bool is_indexed) {
configure_func(this, is_indexed);
}
VkPipelineLayout GetLayout() const {
return *layout;
[[nodiscard]] GraphicsPipeline* Next(const GraphicsPipelineCacheKey& current_key) noexcept {
if (key == current_key) {
return this;
}
const auto it{std::find(transition_keys.begin(), transition_keys.end(), current_key)};
return it != transition_keys.end() ? transitions[std::distance(transition_keys.begin(), it)]
: nullptr;
}
GraphicsPipelineCacheKey GetCacheKey() const {
return cache_key;
[[nodiscard]] bool IsBuilt() const noexcept {
return is_built.load(std::memory_order::relaxed);
}
template <typename Spec>
static auto MakeConfigureSpecFunc() {
return [](GraphicsPipeline* pl, bool is_indexed) { pl->ConfigureImpl<Spec>(is_indexed); };
}
private:
vk::DescriptorSetLayout CreateDescriptorSetLayout(
vk::Span<VkDescriptorSetLayoutBinding> bindings) const;
template <typename Spec>
void ConfigureImpl(bool is_indexed);
vk::PipelineLayout CreatePipelineLayout() const;
void ConfigureDraw();
vk::DescriptorUpdateTemplateKHR CreateDescriptorUpdateTemplate(
const SPIRVProgram& program) const;
void MakePipeline(VkRenderPass render_pass);
std::vector<vk::ShaderModule> CreateShaderModules(const SPIRVProgram& program) const;
vk::Pipeline CreatePipeline(const SPIRVProgram& program, VkRenderPass renderpass,
u32 num_color_buffers) const;
void Validate();
const GraphicsPipelineCacheKey key;
Tegra::Engines::Maxwell3D& maxwell3d;
Tegra::MemoryManager& gpu_memory;
const Device& device;
TextureCache& texture_cache;
BufferCache& buffer_cache;
VKScheduler& scheduler;
const GraphicsPipelineCacheKey cache_key;
const u64 hash;
VKUpdateDescriptorQueue& update_descriptor_queue;
void (*configure_func)(GraphicsPipeline*, bool){};
std::vector<GraphicsPipelineCacheKey> transition_keys;
std::vector<GraphicsPipeline*> transitions;
std::array<vk::ShaderModule, NUM_STAGES> spv_modules;
std::array<Shader::Info, NUM_STAGES> stage_infos;
std::array<u32, 5> enabled_uniform_buffer_masks{};
VideoCommon::UniformBufferSizes uniform_buffer_sizes{};
vk::DescriptorSetLayout descriptor_set_layout;
DescriptorAllocator descriptor_allocator;
VKUpdateDescriptorQueue& update_descriptor_queue;
vk::PipelineLayout layout;
vk::DescriptorUpdateTemplateKHR descriptor_template;
std::vector<vk::ShaderModule> modules;
vk::PipelineLayout pipeline_layout;
vk::DescriptorUpdateTemplateKHR descriptor_update_template;
vk::Pipeline pipeline;
std::condition_variable build_condvar;
std::mutex build_mutex;
std::atomic_bool is_built{false};
bool uses_push_descriptor{false};
};
} // namespace Vulkan

6
src/video_core/renderer_vulkan/vk_master_semaphore.h
View File

@ -39,9 +39,9 @@ public:
return KnownGpuTick() >= tick;
}
/// Advance to the logical tick.
void NextTick() noexcept {
++current_tick;
/// Advance to the logical tick and return the old one
[[nodiscard]] u64 NextTick() noexcept {
return current_tick.fetch_add(1, std::memory_order::relaxed);
}
/// Refresh the known GPU tick

911
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View File

@ -4,444 +4,613 @@
#include <algorithm>
#include <cstddef>
#include <fstream>
#include <memory>
#include <thread>
#include <vector>
#include "common/bit_cast.h"
#include "common/cityhash.h"
#include "common/fs/fs.h"
#include "common/fs/path_util.h"
#include "common/microprofile.h"
#include "common/thread_worker.h"
#include "core/core.h"
#include "core/memory.h"
#include "shader_recompiler/backend/spirv/emit_spirv.h"
#include "shader_recompiler/environment.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
#include "shader_recompiler/frontend/maxwell/translate_program.h"
#include "shader_recompiler/program_header.h"
#include "video_core/dirty_flags.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/pipeline_helper.h"
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_util.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/shader/compiler_settings.h"
#include "video_core/shader/memory_util.h"
#include "video_core/shader_cache.h"
#include "video_core/shader_environment.h"
#include "video_core/shader_notify.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
MICROPROFILE_DECLARE(Vulkan_PipelineCache);
using Tegra::Engines::ShaderType;
using VideoCommon::Shader::GetShaderAddress;
using VideoCommon::Shader::GetShaderCode;
using VideoCommon::Shader::KERNEL_MAIN_OFFSET;
using VideoCommon::Shader::ProgramCode;
using VideoCommon::Shader::STAGE_MAIN_OFFSET;
namespace {
using Shader::Backend::SPIRV::EmitSPIRV;
using Shader::Maxwell::MergeDualVertexPrograms;
using Shader::Maxwell::TranslateProgram;
using VideoCommon::ComputeEnvironment;
using VideoCommon::FileEnvironment;
using VideoCommon::GenericEnvironment;
using VideoCommon::GraphicsEnvironment;
constexpr VkDescriptorType UNIFORM_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
constexpr VkDescriptorType STORAGE_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
constexpr VkDescriptorType UNIFORM_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER;
constexpr VkDescriptorType COMBINED_IMAGE_SAMPLER = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
constexpr VkDescriptorType STORAGE_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER;
constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
constexpr u32 CACHE_VERSION = 5;
constexpr VideoCommon::Shader::CompilerSettings compiler_settings{
.depth = VideoCommon::Shader::CompileDepth::FullDecompile,
.disable_else_derivation = true,
};
constexpr std::size_t GetStageFromProgram(std::size_t program) {
return program == 0 ? 0 : program - 1;
template <typename Container>
auto MakeSpan(Container& container) {
return std::span(container.data(), container.size());
}
constexpr ShaderType GetStageFromProgram(Maxwell::ShaderProgram program) {
return static_cast<ShaderType>(GetStageFromProgram(static_cast<std::size_t>(program)));
}
ShaderType GetShaderType(Maxwell::ShaderProgram program) {
switch (program) {
case Maxwell::ShaderProgram::VertexB:
return ShaderType::Vertex;
case Maxwell::ShaderProgram::TesselationControl:
return ShaderType::TesselationControl;
case Maxwell::ShaderProgram::TesselationEval:
return ShaderType::TesselationEval;
case Maxwell::ShaderProgram::Geometry:
return ShaderType::Geometry;
case Maxwell::ShaderProgram::Fragment:
return ShaderType::Fragment;
default:
UNIMPLEMENTED_MSG("program={}", program);
return ShaderType::Vertex;
Shader::CompareFunction MaxwellToCompareFunction(Maxwell::ComparisonOp comparison) {
switch (comparison) {
case Maxwell::ComparisonOp::Never:
case Maxwell::ComparisonOp::NeverOld:
return Shader::CompareFunction::Never;
case Maxwell::ComparisonOp::Less:
case Maxwell::ComparisonOp::LessOld:
return Shader::CompareFunction::Less;
case Maxwell::ComparisonOp::Equal:
case Maxwell::ComparisonOp::EqualOld:
return Shader::CompareFunction::Equal;
case Maxwell::ComparisonOp::LessEqual:
case Maxwell::ComparisonOp::LessEqualOld:
return Shader::CompareFunction::LessThanEqual;
case Maxwell::ComparisonOp::Greater:
case Maxwell::ComparisonOp::GreaterOld:
return Shader::CompareFunction::Greater;
case Maxwell::ComparisonOp::NotEqual:
case Maxwell::ComparisonOp::NotEqualOld:
return Shader::CompareFunction::NotEqual;
case Maxwell::ComparisonOp::GreaterEqual:
case Maxwell::ComparisonOp::GreaterEqualOld:
return Shader::CompareFunction::GreaterThanEqual;
case Maxwell::ComparisonOp::Always:
case Maxwell::ComparisonOp::AlwaysOld:
return Shader::CompareFunction::Always;
}
UNIMPLEMENTED_MSG("Unimplemented comparison op={}", comparison);
return {};
}
template <VkDescriptorType descriptor_type, class Container>
void AddBindings(std::vector<VkDescriptorSetLayoutBinding>& bindings, u32& binding,
VkShaderStageFlags stage_flags, const Container& container) {
const u32 num_entries = static_cast<u32>(std::size(container));
for (std::size_t i = 0; i < num_entries; ++i) {
u32 count = 1;
if constexpr (descriptor_type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) {
// Combined image samplers can be arrayed.
count = container[i].size;
Shader::AttributeType CastAttributeType(const FixedPipelineState::VertexAttribute& attr) {
if (attr.enabled == 0) {
return Shader::AttributeType::Disabled;
}
switch (attr.Type()) {
case Maxwell::VertexAttribute::Type::SignedNorm:
case Maxwell::VertexAttribute::Type::UnsignedNorm:
case Maxwell::VertexAttribute::Type::UnsignedScaled:
case Maxwell::VertexAttribute::Type::SignedScaled:
case Maxwell::VertexAttribute::Type::Float:
return Shader::AttributeType::Float;
case Maxwell::VertexAttribute::Type::SignedInt:
return Shader::AttributeType::SignedInt;
case Maxwell::VertexAttribute::Type::UnsignedInt:
return Shader::AttributeType::UnsignedInt;
}
return Shader::AttributeType::Float;
}
Shader::AttributeType AttributeType(const FixedPipelineState& state, size_t index) {
switch (state.DynamicAttributeType(index)) {
case 0:
return Shader::AttributeType::Disabled;
case 1:
return Shader::AttributeType::Float;
case 2:
return Shader::AttributeType::SignedInt;
case 3:
return Shader::AttributeType::UnsignedInt;
}
return Shader::AttributeType::Disabled;
}
Shader::RuntimeInfo MakeRuntimeInfo(std::span<const Shader::IR::Program> programs,
const GraphicsPipelineCacheKey& key,
const Shader::IR::Program& program,
const Shader::IR::Program* previous_program) {
Shader::RuntimeInfo info;
if (previous_program) {
info.previous_stage_stores = previous_program->info.stores;
if (previous_program->is_geometry_passthrough) {
info.previous_stage_stores.mask |= previous_program->info.passthrough.mask;
}
bindings.push_back({
.binding = binding++,
.descriptorType = descriptor_type,
.descriptorCount = count,
.stageFlags = stage_flags,
.pImmutableSamplers = nullptr,
});
} else {
info.previous_stage_stores.mask.set();
}
const Shader::Stage stage{program.stage};
const bool has_geometry{key.unique_hashes[4] != 0 && !programs[4].is_geometry_passthrough};
const bool gl_ndc{key.state.ndc_minus_one_to_one != 0};
const float point_size{Common::BitCast<float>(key.state.point_size)};
switch (stage) {
case Shader::Stage::VertexB:
if (!has_geometry) {
if (key.state.topology == Maxwell::PrimitiveTopology::Points) {
info.fixed_state_point_size = point_size;
}
if (key.state.xfb_enabled) {
info.xfb_varyings = VideoCommon::MakeTransformFeedbackVaryings(key.state.xfb_state);
}
info.convert_depth_mode = gl_ndc;
}
if (key.state.dynamic_vertex_input) {
for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) {
info.generic_input_types[index] = AttributeType(key.state, index);
}
} else {
std::ranges::transform(key.state.attributes, info.generic_input_types.begin(),
&CastAttributeType);
}
break;
case Shader::Stage::TessellationEval:
// We have to flip tessellation clockwise for some reason...
info.tess_clockwise = key.state.tessellation_clockwise == 0;
info.tess_primitive = [&key] {
const u32 raw{key.state.tessellation_primitive.Value()};
switch (static_cast<Maxwell::TessellationPrimitive>(raw)) {
case Maxwell::TessellationPrimitive::Isolines:
return Shader::TessPrimitive::Isolines;
case Maxwell::TessellationPrimitive::Triangles:
return Shader::TessPrimitive::Triangles;
case Maxwell::TessellationPrimitive::Quads:
return Shader::TessPrimitive::Quads;
}
UNREACHABLE();
return Shader::TessPrimitive::Triangles;
}();
info.tess_spacing = [&] {
const u32 raw{key.state.tessellation_spacing};
switch (static_cast<Maxwell::TessellationSpacing>(raw)) {
case Maxwell::TessellationSpacing::Equal:
return Shader::TessSpacing::Equal;
case Maxwell::TessellationSpacing::FractionalOdd:
return Shader::TessSpacing::FractionalOdd;
case Maxwell::TessellationSpacing::FractionalEven:
return Shader::TessSpacing::FractionalEven;
}
UNREACHABLE();
return Shader::TessSpacing::Equal;
}();
break;
case Shader::Stage::Geometry:
if (program.output_topology == Shader::OutputTopology::PointList) {
info.fixed_state_point_size = point_size;
}
if (key.state.xfb_enabled != 0) {
info.xfb_varyings = VideoCommon::MakeTransformFeedbackVaryings(key.state.xfb_state);
}
info.convert_depth_mode = gl_ndc;
break;
case Shader::Stage::Fragment:
info.alpha_test_func = MaxwellToCompareFunction(
key.state.UnpackComparisonOp(key.state.alpha_test_func.Value()));
info.alpha_test_reference = Common::BitCast<float>(key.state.alpha_test_ref);
break;
default:
break;
}
switch (key.state.topology) {
case Maxwell::PrimitiveTopology::Points:
info.input_topology = Shader::InputTopology::Points;
break;
case Maxwell::PrimitiveTopology::Lines:
case Maxwell::PrimitiveTopology::LineLoop:
case Maxwell::PrimitiveTopology::LineStrip:
info.input_topology = Shader::InputTopology::Lines;
break;
case Maxwell::PrimitiveTopology::Triangles:
case Maxwell::PrimitiveTopology::TriangleStrip:
case Maxwell::PrimitiveTopology::TriangleFan:
case Maxwell::PrimitiveTopology::Quads:
case Maxwell::PrimitiveTopology::QuadStrip:
case Maxwell::PrimitiveTopology::Polygon:
case Maxwell::PrimitiveTopology::Patches:
info.input_topology = Shader::InputTopology::Triangles;
break;
case Maxwell::PrimitiveTopology::LinesAdjacency:
case Maxwell::PrimitiveTopology::LineStripAdjacency:
info.input_topology = Shader::InputTopology::LinesAdjacency;
break;
case Maxwell::PrimitiveTopology::TrianglesAdjacency:
case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
info.input_topology = Shader::InputTopology::TrianglesAdjacency;
break;
}
info.force_early_z = key.state.early_z != 0;
info.y_negate = key.state.y_negate != 0;
return info;
}
u32 FillDescriptorLayout(const ShaderEntries& entries,
std::vector<VkDescriptorSetLayoutBinding>& bindings,
Maxwell::ShaderProgram program_type, u32 base_binding) {
const ShaderType stage = GetStageFromProgram(program_type);
const VkShaderStageFlags flags = MaxwellToVK::ShaderStage(stage);
u32 binding = base_binding;
AddBindings<UNIFORM_BUFFER>(bindings, binding, flags, entries.const_buffers);
AddBindings<STORAGE_BUFFER>(bindings, binding, flags, entries.global_buffers);
AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.uniform_texels);
AddBindings<COMBINED_IMAGE_SAMPLER>(bindings, binding, flags, entries.samplers);
AddBindings<STORAGE_TEXEL_BUFFER>(bindings, binding, flags, entries.storage_texels);
AddBindings<STORAGE_IMAGE>(bindings, binding, flags, entries.images);
return binding;
}
} // Anonymous namespace
std::size_t GraphicsPipelineCacheKey::Hash() const noexcept {
const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), Size());
return static_cast<std::size_t>(hash);
}
bool GraphicsPipelineCacheKey::operator==(const GraphicsPipelineCacheKey& rhs) const noexcept {
return std::memcmp(&rhs, this, Size()) == 0;
}
std::size_t ComputePipelineCacheKey::Hash() const noexcept {
size_t ComputePipelineCacheKey::Hash() const noexcept {
const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this);
return static_cast<std::size_t>(hash);
return static_cast<size_t>(hash);
}
bool ComputePipelineCacheKey::operator==(const ComputePipelineCacheKey& rhs) const noexcept {
return std::memcmp(&rhs, this, sizeof *this) == 0;
}
Shader::Shader(Tegra::Engines::ConstBufferEngineInterface& engine_, ShaderType stage_,
GPUVAddr gpu_addr_, VAddr cpu_addr_, ProgramCode program_code_, u32 main_offset_)
: gpu_addr(gpu_addr_), program_code(std::move(program_code_)), registry(stage_, engine_),
shader_ir(program_code, main_offset_, compiler_settings, registry),
entries(GenerateShaderEntries(shader_ir)) {}
Shader::~Shader() = default;
VKPipelineCache::VKPipelineCache(RasterizerVulkan& rasterizer_, Tegra::GPU& gpu_,
Tegra::Engines::Maxwell3D& maxwell3d_,
Tegra::Engines::KeplerCompute& kepler_compute_,
Tegra::MemoryManager& gpu_memory_, const Device& device_,
VKScheduler& scheduler_, VKDescriptorPool& descriptor_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
: VideoCommon::ShaderCache<Shader>{rasterizer_}, gpu{gpu_}, maxwell3d{maxwell3d_},
kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_}, device{device_},
scheduler{scheduler_}, descriptor_pool{descriptor_pool_}, update_descriptor_queue{
update_descriptor_queue_} {}
VKPipelineCache::~VKPipelineCache() = default;
std::array<Shader*, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
std::array<Shader*, Maxwell::MaxShaderProgram> shaders{};
for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
const auto program{static_cast<Maxwell::ShaderProgram>(index)};
// Skip stages that are not enabled
if (!maxwell3d.regs.IsShaderConfigEnabled(index)) {
continue;
}
const GPUVAddr gpu_addr{GetShaderAddress(maxwell3d, program)};
const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
ASSERT(cpu_addr);
Shader* result = cpu_addr ? TryGet(*cpu_addr) : null_shader.get();
if (!result) {
const u8* const host_ptr{gpu_memory.GetPointer(gpu_addr)};
// No shader found - create a new one
static constexpr u32 stage_offset = STAGE_MAIN_OFFSET;
const auto stage = static_cast<ShaderType>(index == 0 ? 0 : index - 1);
ProgramCode code = GetShaderCode(gpu_memory, gpu_addr, host_ptr, false);
const std::size_t size_in_bytes = code.size() * sizeof(u64);
auto shader = std::make_unique<Shader>(maxwell3d, stage, gpu_addr, *cpu_addr,
std::move(code), stage_offset);
result = shader.get();
if (cpu_addr) {
Register(std::move(shader), *cpu_addr, size_in_bytes);
} else {
null_shader = std::move(shader);
}
}
shaders[index] = result;
}
return last_shaders = shaders;
size_t GraphicsPipelineCacheKey::Hash() const noexcept {
const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), Size());
return static_cast<size_t>(hash);
}
VKGraphicsPipeline* VKPipelineCache::GetGraphicsPipeline(
const GraphicsPipelineCacheKey& key, u32 num_color_buffers,
VideoCommon::Shader::AsyncShaders& async_shaders) {
MICROPROFILE_SCOPE(Vulkan_PipelineCache);
if (last_graphics_pipeline && last_graphics_key == key) {
return last_graphics_pipeline;
}
last_graphics_key = key;
if (device.UseAsynchronousShaders() && async_shaders.IsShaderAsync(gpu)) {
std::unique_lock lock{pipeline_cache};
const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key);
if (is_cache_miss) {
gpu.ShaderNotify().MarkSharderBuilding();
LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
const auto [program, bindings] = DecompileShaders(key.fixed_state);
async_shaders.QueueVulkanShader(this, device, scheduler, descriptor_pool,
update_descriptor_queue, bindings, program, key,
num_color_buffers);
}
last_graphics_pipeline = pair->second.get();
return last_graphics_pipeline;
}
const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key);
auto& entry = pair->second;
if (is_cache_miss) {
gpu.ShaderNotify().MarkSharderBuilding();
LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
const auto [program, bindings] = DecompileShaders(key.fixed_state);
entry = std::make_unique<VKGraphicsPipeline>(device, scheduler, descriptor_pool,
update_descriptor_queue, key, bindings,
program, num_color_buffers);
gpu.ShaderNotify().MarkShaderComplete();
}
last_graphics_pipeline = entry.get();
return last_graphics_pipeline;
bool GraphicsPipelineCacheKey::operator==(const GraphicsPipelineCacheKey& rhs) const noexcept {
return std::memcmp(&rhs, this, Size()) == 0;
}
VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCacheKey& key) {
PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxwell3D& maxwell3d_,
Tegra::Engines::KeplerCompute& kepler_compute_,
Tegra::MemoryManager& gpu_memory_, const Device& device_,
VKScheduler& scheduler_, DescriptorPool& descriptor_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
RenderPassCache& render_pass_cache_, BufferCache& buffer_cache_,
TextureCache& texture_cache_, VideoCore::ShaderNotify& shader_notify_)
: VideoCommon::ShaderCache{rasterizer_, gpu_memory_, maxwell3d_, kepler_compute_},
device{device_}, scheduler{scheduler_}, descriptor_pool{descriptor_pool_},
update_descriptor_queue{update_descriptor_queue_}, render_pass_cache{render_pass_cache_},
buffer_cache{buffer_cache_}, texture_cache{texture_cache_}, shader_notify{shader_notify_},
use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()},
workers(std::max(std::thread::hardware_concurrency(), 2U) - 1, "yuzu:PipelineBuilder"),
serialization_thread(1, "yuzu:PipelineSerialization") {
const auto& float_control{device.FloatControlProperties()};
const VkDriverIdKHR driver_id{device.GetDriverID()};
profile = Shader::Profile{
.supported_spirv = device.IsKhrSpirv1_4Supported() ? 0x00010400U : 0x00010000U,
.unified_descriptor_binding = true,
.support_descriptor_aliasing = true,
.support_int8 = true,
.support_int16 = device.IsShaderInt16Supported(),
.support_int64 = device.IsShaderInt64Supported(),
.support_vertex_instance_id = false,
.support_float_controls = true,
.support_separate_denorm_behavior = float_control.denormBehaviorIndependence ==
VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_ALL_KHR,
.support_separate_rounding_mode =
float_control.roundingModeIndependence == VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_ALL_KHR,
.support_fp16_denorm_preserve = float_control.shaderDenormPreserveFloat16 != VK_FALSE,
.support_fp32_denorm_preserve = float_control.shaderDenormPreserveFloat32 != VK_FALSE,
.support_fp16_denorm_flush = float_control.shaderDenormFlushToZeroFloat16 != VK_FALSE,
.support_fp32_denorm_flush = float_control.shaderDenormFlushToZeroFloat32 != VK_FALSE,
.support_fp16_signed_zero_nan_preserve =
float_control.shaderSignedZeroInfNanPreserveFloat16 != VK_FALSE,
.support_fp32_signed_zero_nan_preserve =
float_control.shaderSignedZeroInfNanPreserveFloat32 != VK_FALSE,
.support_fp64_signed_zero_nan_preserve =
float_control.shaderSignedZeroInfNanPreserveFloat64 != VK_FALSE,
.support_explicit_workgroup_layout = device.IsKhrWorkgroupMemoryExplicitLayoutSupported(),
.support_vote = true,
.support_viewport_index_layer_non_geometry =
device.IsExtShaderViewportIndexLayerSupported(),
.support_viewport_mask = device.IsNvViewportArray2Supported(),
.support_typeless_image_loads = device.IsFormatlessImageLoadSupported(),
.support_demote_to_helper_invocation = true,
.support_int64_atomics = device.IsExtShaderAtomicInt64Supported(),
.support_derivative_control = true,
.support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(),
.warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyBiggerThanGuest(),
.lower_left_origin_mode = false,
.need_declared_frag_colors = false,
.has_broken_spirv_clamp = driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS_KHR,
.has_broken_unsigned_image_offsets = false,
.has_broken_signed_operations = false,
.has_broken_fp16_float_controls = driver_id == VK_DRIVER_ID_NVIDIA_PROPRIETARY_KHR,
.ignore_nan_fp_comparisons = false,
};
host_info = Shader::HostTranslateInfo{
.support_float16 = device.IsFloat16Supported(),
.support_int64 = device.IsShaderInt64Supported(),
};
}
PipelineCache::~PipelineCache() = default;
GraphicsPipeline* PipelineCache::CurrentGraphicsPipeline() {
MICROPROFILE_SCOPE(Vulkan_PipelineCache);
const auto [pair, is_cache_miss] = compute_cache.try_emplace(key);
auto& entry = pair->second;
if (!is_cache_miss) {
return *entry;
if (!RefreshStages(graphics_key.unique_hashes)) {
current_pipeline = nullptr;
return nullptr;
}
LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
graphics_key.state.Refresh(maxwell3d, device.IsExtExtendedDynamicStateSupported(),
device.IsExtVertexInputDynamicStateSupported());
const GPUVAddr gpu_addr = key.shader;
if (current_pipeline) {
GraphicsPipeline* const next{current_pipeline->Next(graphics_key)};
if (next) {
current_pipeline = next;
return BuiltPipeline(current_pipeline);
}
}
return CurrentGraphicsPipelineSlowPath();
}
const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
ASSERT(cpu_addr);
ComputePipeline* PipelineCache::CurrentComputePipeline() {
MICROPROFILE_SCOPE(Vulkan_PipelineCache);
Shader* shader = cpu_addr ? TryGet(*cpu_addr) : null_kernel.get();
const ShaderInfo* const shader{ComputeShader()};
if (!shader) {
// No shader found - create a new one
const auto host_ptr = gpu_memory.GetPointer(gpu_addr);
ProgramCode code = GetShaderCode(gpu_memory, gpu_addr, host_ptr, true);
const std::size_t size_in_bytes = code.size() * sizeof(u64);
auto shader_info = std::make_unique<Shader>(kepler_compute, ShaderType::Compute, gpu_addr,
*cpu_addr, std::move(code), KERNEL_MAIN_OFFSET);
shader = shader_info.get();
if (cpu_addr) {
Register(std::move(shader_info), *cpu_addr, size_in_bytes);
} else {
null_kernel = std::move(shader_info);
}
return nullptr;
}
const Specialization specialization{
.base_binding = 0,
.workgroup_size = key.workgroup_size,
.shared_memory_size = key.shared_memory_size,
.point_size = std::nullopt,
.enabled_attributes = {},
.attribute_types = {},
.ndc_minus_one_to_one = false,
const auto& qmd{kepler_compute.launch_description};
const ComputePipelineCacheKey key{
.unique_hash = shader->unique_hash,
.shared_memory_size = qmd.shared_alloc,
.workgroup_size{qmd.block_dim_x, qmd.block_dim_y, qmd.block_dim_z},
};
const SPIRVShader spirv_shader{Decompile(device, shader->GetIR(), ShaderType::Compute,
shader->GetRegistry(), specialization),
shader->GetEntries()};
entry = std::make_unique<VKComputePipeline>(device, scheduler, descriptor_pool,
update_descriptor_queue, spirv_shader);
return *entry;
const auto [pair, is_new]{compute_cache.try_emplace(key)};
auto& pipeline{pair->second};
if (!is_new) {
return pipeline.get();
}
pipeline = CreateComputePipeline(key, shader);
return pipeline.get();
}
void VKPipelineCache::EmplacePipeline(std::unique_ptr<VKGraphicsPipeline> pipeline) {
gpu.ShaderNotify().MarkShaderComplete();
std::unique_lock lock{pipeline_cache};
graphics_cache.at(pipeline->GetCacheKey()) = std::move(pipeline);
}
void VKPipelineCache::OnShaderRemoval(Shader* shader) {
bool finished = false;
const auto Finish = [&] {
// TODO(Rodrigo): Instead of finishing here, wait for the fences that use this pipeline and
// flush.
if (finished) {
return;
}
finished = true;
scheduler.Finish();
};
const GPUVAddr invalidated_addr = shader->GetGpuAddr();
for (auto it = graphics_cache.begin(); it != graphics_cache.end();) {
auto& entry = it->first;
if (std::find(entry.shaders.begin(), entry.shaders.end(), invalidated_addr) ==
entry.shaders.end()) {
++it;
continue;
}
Finish();
it = graphics_cache.erase(it);
}
for (auto it = compute_cache.begin(); it != compute_cache.end();) {
auto& entry = it->first;
if (entry.shader != invalidated_addr) {
++it;
continue;
}
Finish();
it = compute_cache.erase(it);
}
}
std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>>
VKPipelineCache::DecompileShaders(const FixedPipelineState& fixed_state) {
Specialization specialization;
if (fixed_state.topology == Maxwell::PrimitiveTopology::Points) {
float point_size;
std::memcpy(&point_size, &fixed_state.point_size, sizeof(float));
specialization.point_size = point_size;
ASSERT(point_size != 0.0f);
}
for (std::size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) {
const auto& attribute = fixed_state.attributes[i];
specialization.enabled_attributes[i] = attribute.enabled.Value() != 0;
specialization.attribute_types[i] = attribute.Type();
}
specialization.ndc_minus_one_to_one = fixed_state.ndc_minus_one_to_one;
specialization.early_fragment_tests = fixed_state.early_z;
// Alpha test
specialization.alpha_test_func =
FixedPipelineState::UnpackComparisonOp(fixed_state.alpha_test_func.Value());
specialization.alpha_test_ref = Common::BitCast<float>(fixed_state.alpha_test_ref);
SPIRVProgram program;
std::vector<VkDescriptorSetLayoutBinding> bindings;
for (std::size_t index = 1; index < Maxwell::MaxShaderProgram; ++index) {
const auto program_enum = static_cast<Maxwell::ShaderProgram>(index);
// Skip stages that are not enabled
if (!maxwell3d.regs.IsShaderConfigEnabled(index)) {
continue;
}
const GPUVAddr gpu_addr = GetShaderAddress(maxwell3d, program_enum);
const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
Shader* const shader = cpu_addr ? TryGet(*cpu_addr) : null_shader.get();
const std::size_t stage = index == 0 ? 0 : index - 1; // Stage indices are 0 - 5
const ShaderType program_type = GetShaderType(program_enum);
const auto& entries = shader->GetEntries();
program[stage] = {
Decompile(device, shader->GetIR(), program_type, shader->GetRegistry(), specialization),
entries,
};
const u32 old_binding = specialization.base_binding;
specialization.base_binding =
FillDescriptorLayout(entries, bindings, program_enum, specialization.base_binding);
ASSERT(old_binding + entries.NumBindings() == specialization.base_binding);
}
return {std::move(program), std::move(bindings)};
}
template <VkDescriptorType descriptor_type, class Container>
void AddEntry(std::vector<VkDescriptorUpdateTemplateEntry>& template_entries, u32& binding,
u32& offset, const Container& container) {
static constexpr u32 entry_size = static_cast<u32>(sizeof(DescriptorUpdateEntry));
const u32 count = static_cast<u32>(std::size(container));
if constexpr (descriptor_type == COMBINED_IMAGE_SAMPLER) {
for (u32 i = 0; i < count; ++i) {
const u32 num_samplers = container[i].size;
template_entries.push_back({
.dstBinding = binding,
.dstArrayElement = 0,
.descriptorCount = num_samplers,
.descriptorType = descriptor_type,
.offset = offset,
.stride = entry_size,
});
++binding;
offset += num_samplers * entry_size;
}
void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) {
if (title_id == 0) {
return;
}
if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER ||
descriptor_type == STORAGE_TEXEL_BUFFER) {
// Nvidia has a bug where updating multiple texels at once causes the driver to crash.
// Note: Fixed in driver Windows 443.24, Linux 440.66.15
for (u32 i = 0; i < count; ++i) {
template_entries.push_back({
.dstBinding = binding + i,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = descriptor_type,
.offset = static_cast<std::size_t>(offset + i * entry_size),
.stride = entry_size,
});
}
} else if (count > 0) {
template_entries.push_back({
.dstBinding = binding,
.dstArrayElement = 0,
.descriptorCount = count,
.descriptorType = descriptor_type,
.offset = offset,
.stride = entry_size,
});
const auto shader_dir{Common::FS::GetYuzuPath(Common::FS::YuzuPath::ShaderDir)};
const auto base_dir{shader_dir / fmt::format("{:016x}", title_id)};
if (!Common::FS::CreateDir(shader_dir) || !Common::FS::CreateDir(base_dir)) {
LOG_ERROR(Common_Filesystem, "Failed to create pipeline cache directories");
return;
}
offset += count * entry_size;
binding += count;
pipeline_cache_filename = base_dir / "vulkan.bin";
struct {
std::mutex mutex;
size_t total{};
size_t built{};
bool has_loaded{};
} state;
const auto load_compute{[&](std::ifstream& file, FileEnvironment env) {
ComputePipelineCacheKey key;
file.read(reinterpret_cast<char*>(&key), sizeof(key));
workers.QueueWork([this, key, env = std::move(env), &state, &callback]() mutable {
ShaderPools pools;
auto pipeline{CreateComputePipeline(pools, key, env, false)};
std::lock_guard lock{state.mutex};
if (pipeline) {
compute_cache.emplace(key, std::move(pipeline));
}
++state.built;
if (state.has_loaded) {
callback(VideoCore::LoadCallbackStage::Build, state.built, state.total);
}
});
++state.total;
}};
const bool extended_dynamic_state = device.IsExtExtendedDynamicStateSupported();
const bool dynamic_vertex_input = device.IsExtVertexInputDynamicStateSupported();
const auto load_graphics{[&](std::ifstream& file, std::vector<FileEnvironment> envs) {
GraphicsPipelineCacheKey key;
file.read(reinterpret_cast<char*>(&key), sizeof(key));
if ((key.state.extended_dynamic_state != 0) != extended_dynamic_state ||
(key.state.dynamic_vertex_input != 0) != dynamic_vertex_input) {
return;
}
workers.QueueWork([this, key, envs = std::move(envs), &state, &callback]() mutable {
ShaderPools pools;
boost::container::static_vector<Shader::Environment*, 5> env_ptrs;
for (auto& env : envs) {
env_ptrs.push_back(&env);
}
auto pipeline{CreateGraphicsPipeline(pools, key, MakeSpan(env_ptrs), false)};
std::lock_guard lock{state.mutex};
graphics_cache.emplace(key, std::move(pipeline));
++state.built;
if (state.has_loaded) {
callback(VideoCore::LoadCallbackStage::Build, state.built, state.total);
}
});
++state.total;
}};
VideoCommon::LoadPipelines(stop_loading, pipeline_cache_filename, CACHE_VERSION, load_compute,
load_graphics);
std::unique_lock lock{state.mutex};
callback(VideoCore::LoadCallbackStage::Build, 0, state.total);
state.has_loaded = true;
lock.unlock();
workers.WaitForRequests();
}
void FillDescriptorUpdateTemplateEntries(
const ShaderEntries& entries, u32& binding, u32& offset,
std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries) {
AddEntry<UNIFORM_BUFFER>(template_entries, offset, binding, entries.const_buffers);
AddEntry<STORAGE_BUFFER>(template_entries, offset, binding, entries.global_buffers);
AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.uniform_texels);
AddEntry<COMBINED_IMAGE_SAMPLER>(template_entries, offset, binding, entries.samplers);
AddEntry<STORAGE_TEXEL_BUFFER>(template_entries, offset, binding, entries.storage_texels);
AddEntry<STORAGE_IMAGE>(template_entries, offset, binding, entries.images);
GraphicsPipeline* PipelineCache::CurrentGraphicsPipelineSlowPath() {
const auto [pair, is_new]{graphics_cache.try_emplace(graphics_key)};
auto& pipeline{pair->second};
if (is_new) {
pipeline = CreateGraphicsPipeline();
}
if (!pipeline) {
return nullptr;
}
if (current_pipeline) {
current_pipeline->AddTransition(pipeline.get());
}
current_pipeline = pipeline.get();
return BuiltPipeline(current_pipeline);
}
GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const noexcept {
if (pipeline->IsBuilt()) {
return pipeline;
}
if (!use_asynchronous_shaders) {
return pipeline;
}
// If something is using depth, we can assume that games are not rendering anything which
// will be used one time.
if (maxwell3d.regs.zeta_enable) {
return nullptr;
}
// If games are using a small index count, we can assume these are full screen quads.
// Usually these shaders are only used once for building textures so we can assume they
// can't be built async
if (maxwell3d.regs.index_array.count <= 6 || maxwell3d.regs.vertex_buffer.count <= 6) {
return pipeline;
}
return nullptr;
}
std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
ShaderPools& pools, const GraphicsPipelineCacheKey& key,
std::span<Shader::Environment* const> envs, bool build_in_parallel) try {
LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
size_t env_index{0};
std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs;
const bool uses_vertex_a{key.unique_hashes[0] != 0};
const bool uses_vertex_b{key.unique_hashes[1] != 0};
for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
if (key.unique_hashes[index] == 0) {
continue;
}
Shader::Environment& env{*envs[env_index]};
++env_index;
const u32 cfg_offset{static_cast<u32>(env.StartAddress() + sizeof(Shader::ProgramHeader))};
Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0);
if (!uses_vertex_a || index != 1) {
// Normal path
programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info);
} else {
// VertexB path when VertexA is present.
auto& program_va{programs[0]};
auto program_vb{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
programs[index] = MergeDualVertexPrograms(program_va, program_vb, env);
}
}
std::array<const Shader::Info*, Maxwell::MaxShaderStage> infos{};
std::array<vk::ShaderModule, Maxwell::MaxShaderStage> modules;
const Shader::IR::Program* previous_stage{};
Shader::Backend::Bindings binding;
for (size_t index = uses_vertex_a && uses_vertex_b ? 1 : 0; index < Maxwell::MaxShaderProgram;
++index) {
if (key.unique_hashes[index] == 0) {
continue;
}
UNIMPLEMENTED_IF(index == 0);
Shader::IR::Program& program{programs[index]};
const size_t stage_index{index - 1};
infos[stage_index] = &program.info;
const auto runtime_info{MakeRuntimeInfo(programs, key, program, previous_stage)};
const std::vector<u32> code{EmitSPIRV(profile, runtime_info, program, binding)};
device.SaveShader(code);
modules[stage_index] = BuildShader(device, code);
if (device.HasDebuggingToolAttached()) {
const std::string name{fmt::format("Shader {:016x}", key.unique_hashes[index])};
modules[stage_index].SetObjectNameEXT(name.c_str());
}
previous_stage = &program;
}
Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
return std::make_unique<GraphicsPipeline>(
maxwell3d, gpu_memory, scheduler, buffer_cache, texture_cache, &shader_notify, device,
descriptor_pool, update_descriptor_queue, thread_worker, render_pass_cache, key,
std::move(modules), infos);
} catch (const Shader::Exception& exception) {
LOG_ERROR(Render_Vulkan, "{}", exception.what());
return nullptr;
}
std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() {
GraphicsEnvironments environments;
GetGraphicsEnvironments(environments, graphics_key.unique_hashes);
main_pools.ReleaseContents();
auto pipeline{CreateGraphicsPipeline(main_pools, graphics_key, environments.Span(), true)};
if (!pipeline || pipeline_cache_filename.empty()) {
return pipeline;
}
serialization_thread.QueueWork([this, key = graphics_key, envs = std::move(environments.envs)] {
boost::container::static_vector<const GenericEnvironment*, Maxwell::MaxShaderProgram>
env_ptrs;
for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
if (key.unique_hashes[index] != 0) {
env_ptrs.push_back(&envs[index]);
}
}
SerializePipeline(key, env_ptrs, pipeline_cache_filename, CACHE_VERSION);
});
return pipeline;
}
std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
const ComputePipelineCacheKey& key, const ShaderInfo* shader) {
const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()};
const auto& qmd{kepler_compute.launch_description};
ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start};
env.SetCachedSize(shader->size_bytes);
main_pools.ReleaseContents();
auto pipeline{CreateComputePipeline(main_pools, key, env, true)};
if (!pipeline || pipeline_cache_filename.empty()) {
return pipeline;
}
serialization_thread.QueueWork([this, key, env = std::move(env)] {
SerializePipeline(key, std::array<const GenericEnvironment*, 1>{&env},
pipeline_cache_filename, CACHE_VERSION);
});
return pipeline;
}
std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
ShaderPools& pools, const ComputePipelineCacheKey& key, Shader::Environment& env,
bool build_in_parallel) try {
LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
const std::vector<u32> code{EmitSPIRV(profile, program)};
device.SaveShader(code);
vk::ShaderModule spv_module{BuildShader(device, code)};
if (device.HasDebuggingToolAttached()) {
const auto name{fmt::format("Shader {:016x}", key.unique_hash)};
spv_module.SetObjectNameEXT(name.c_str());
}
Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
return std::make_unique<ComputePipeline>(device, descriptor_pool, update_descriptor_queue,
thread_worker, &shader_notify, program.info,
std::move(spv_module));
} catch (const Shader::Exception& exception) {
LOG_ERROR(Render_Vulkan, "{}", exception.what());
return nullptr;
}
} // namespace Vulkan

180
src/video_core/renderer_vulkan/vk_pipeline_cache.h
View File

@ -6,24 +6,28 @@
#include <array>
#include <cstddef>
#include <filesystem>
#include <iosfwd>
#include <memory>
#include <type_traits>
#include <unordered_map>
#include <utility>
#include <vector>
#include <boost/functional/hash.hpp>
#include "common/common_types.h"
#include "video_core/engines/const_buffer_engine_interface.h"
#include "common/thread_worker.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
#include "shader_recompiler/host_translate_info.h"
#include "shader_recompiler/object_pool.h"
#include "shader_recompiler/profile.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/shader/async_shaders.h"
#include "video_core/shader/memory_util.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/shader_cache.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -31,23 +35,24 @@ namespace Core {
class System;
}
namespace Vulkan {
namespace Shader::IR {
struct Program;
}
class Device;
class RasterizerVulkan;
class VKComputePipeline;
class VKDescriptorPool;
class VKScheduler;
class VKUpdateDescriptorQueue;
namespace VideoCore {
class ShaderNotify;
}
namespace Vulkan {
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
struct ComputePipelineCacheKey {
GPUVAddr shader;
u64 unique_hash;
u32 shared_memory_size;
std::array<u32, 3> workgroup_size;
std::size_t Hash() const noexcept;
size_t Hash() const noexcept;
bool operator==(const ComputePipelineCacheKey& rhs) const noexcept;
@ -63,16 +68,9 @@ static_assert(std::is_trivially_constructible_v<ComputePipelineCacheKey>);
namespace std {
template <>
struct hash<Vulkan::GraphicsPipelineCacheKey> {
std::size_t operator()(const Vulkan::GraphicsPipelineCacheKey& k) const noexcept {
return k.Hash();
}
};
template <>
struct hash<Vulkan::ComputePipelineCacheKey> {
std::size_t operator()(const Vulkan::ComputePipelineCacheKey& k) const noexcept {
size_t operator()(const Vulkan::ComputePipelineCacheKey& k) const noexcept {
return k.Hash();
}
};
@ -81,94 +79,90 @@ struct hash<Vulkan::ComputePipelineCacheKey> {
namespace Vulkan {
class Shader {
public:
explicit Shader(Tegra::Engines::ConstBufferEngineInterface& engine_,
Tegra::Engines::ShaderType stage_, GPUVAddr gpu_addr, VAddr cpu_addr_,
VideoCommon::Shader::ProgramCode program_code, u32 main_offset_);
~Shader();
class ComputePipeline;
class Device;
class DescriptorPool;
class RasterizerVulkan;
class RenderPassCache;
class VKScheduler;
class VKUpdateDescriptorQueue;
GPUVAddr GetGpuAddr() const {
return gpu_addr;
using VideoCommon::ShaderInfo;
struct ShaderPools {
void ReleaseContents() {
flow_block.ReleaseContents();
block.ReleaseContents();
inst.ReleaseContents();
}
VideoCommon::Shader::ShaderIR& GetIR() {
return shader_ir;
}
const VideoCommon::Shader::ShaderIR& GetIR() const {
return shader_ir;
}
const VideoCommon::Shader::Registry& GetRegistry() const {
return registry;
}
const ShaderEntries& GetEntries() const {
return entries;
}
private:
GPUVAddr gpu_addr{};
VideoCommon::Shader::ProgramCode program_code;
VideoCommon::Shader::Registry registry;
VideoCommon::Shader::ShaderIR shader_ir;
ShaderEntries entries;
Shader::ObjectPool<Shader::IR::Inst> inst;
Shader::ObjectPool<Shader::IR::Block> block;
Shader::ObjectPool<Shader::Maxwell::Flow::Block> flow_block;
};
class VKPipelineCache final : public VideoCommon::ShaderCache<Shader> {
class PipelineCache : public VideoCommon::ShaderCache {
public:
explicit VKPipelineCache(RasterizerVulkan& rasterizer, Tegra::GPU& gpu,
Tegra::Engines::Maxwell3D& maxwell3d,
Tegra::Engines::KeplerCompute& kepler_compute,
Tegra::MemoryManager& gpu_memory, const Device& device,
VKScheduler& scheduler, VKDescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue);
~VKPipelineCache() override;
explicit PipelineCache(RasterizerVulkan& rasterizer, Tegra::Engines::Maxwell3D& maxwell3d,
Tegra::Engines::KeplerCompute& kepler_compute,
Tegra::MemoryManager& gpu_memory, const Device& device,
VKScheduler& scheduler, DescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue,
RenderPassCache& render_pass_cache, BufferCache& buffer_cache,
TextureCache& texture_cache, VideoCore::ShaderNotify& shader_notify_);
~PipelineCache();
std::array<Shader*, Maxwell::MaxShaderProgram> GetShaders();
[[nodiscard]] GraphicsPipeline* CurrentGraphicsPipeline();
VKGraphicsPipeline* GetGraphicsPipeline(const GraphicsPipelineCacheKey& key,
u32 num_color_buffers,
VideoCommon::Shader::AsyncShaders& async_shaders);
[[nodiscard]] ComputePipeline* CurrentComputePipeline();
VKComputePipeline& GetComputePipeline(const ComputePipelineCacheKey& key);
void EmplacePipeline(std::unique_ptr<VKGraphicsPipeline> pipeline);
protected:
void OnShaderRemoval(Shader* shader) final;
void LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback);
private:
std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>> DecompileShaders(
const FixedPipelineState& fixed_state);
[[nodiscard]] GraphicsPipeline* CurrentGraphicsPipelineSlowPath();
Tegra::GPU& gpu;
Tegra::Engines::Maxwell3D& maxwell3d;
Tegra::Engines::KeplerCompute& kepler_compute;
Tegra::MemoryManager& gpu_memory;
[[nodiscard]] GraphicsPipeline* BuiltPipeline(GraphicsPipeline* pipeline) const noexcept;
std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline();
std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(
ShaderPools& pools, const GraphicsPipelineCacheKey& key,
std::span<Shader::Environment* const> envs, bool build_in_parallel);
std::unique_ptr<ComputePipeline> CreateComputePipeline(const ComputePipelineCacheKey& key,
const ShaderInfo* shader);
std::unique_ptr<ComputePipeline> CreateComputePipeline(ShaderPools& pools,
const ComputePipelineCacheKey& key,
Shader::Environment& env,
bool build_in_parallel);
const Device& device;
VKScheduler& scheduler;
VKDescriptorPool& descriptor_pool;
DescriptorPool& descriptor_pool;
VKUpdateDescriptorQueue& update_descriptor_queue;
RenderPassCache& render_pass_cache;
BufferCache& buffer_cache;
TextureCache& texture_cache;
VideoCore::ShaderNotify& shader_notify;
bool use_asynchronous_shaders{};
std::unique_ptr<Shader> null_shader;
std::unique_ptr<Shader> null_kernel;
GraphicsPipelineCacheKey graphics_key{};
GraphicsPipeline* current_pipeline{};
std::array<Shader*, Maxwell::MaxShaderProgram> last_shaders{};
std::unordered_map<ComputePipelineCacheKey, std::unique_ptr<ComputePipeline>> compute_cache;
std::unordered_map<GraphicsPipelineCacheKey, std::unique_ptr<GraphicsPipeline>> graphics_cache;
GraphicsPipelineCacheKey last_graphics_key;
VKGraphicsPipeline* last_graphics_pipeline = nullptr;
ShaderPools main_pools;
std::mutex pipeline_cache;
std::unordered_map<GraphicsPipelineCacheKey, std::unique_ptr<VKGraphicsPipeline>>
graphics_cache;
std::unordered_map<ComputePipelineCacheKey, std::unique_ptr<VKComputePipeline>> compute_cache;
Shader::Profile profile;
Shader::HostTranslateInfo host_info;
std::filesystem::path pipeline_cache_filename;
Common::ThreadWorker workers;
Common::ThreadWorker serialization_thread;
};
void FillDescriptorUpdateTemplateEntries(
const ShaderEntries& entries, u32& binding, u32& offset,
std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries);
} // namespace Vulkan

8
src/video_core/renderer_vulkan/vk_query_cache.cpp
View File

@ -114,14 +114,10 @@ void HostCounter::EndQuery() {
}
u64 HostCounter::BlockingQuery() const {
if (tick >= cache.GetScheduler().CurrentTick()) {
cache.GetScheduler().Flush();
}
cache.GetScheduler().Wait(tick);
u64 data;
const VkResult query_result = cache.GetDevice().GetLogical().GetQueryResults(
query.first, query.second, 1, sizeof(data), &data, sizeof(data),
VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
query.first, query.second, 1, sizeof(data), &data, sizeof(data), VK_QUERY_RESULT_64_BIT);
switch (query_result) {
case VK_SUCCESS:

475
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View File

@ -24,7 +24,6 @@
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
@ -55,11 +54,10 @@ struct DrawParams {
u32 num_instances;
u32 base_vertex;
u32 num_vertices;
u32 first_index;
bool is_indexed;
};
constexpr auto COMPUTE_SHADER_INDEX = static_cast<size_t>(Tegra::Engines::ShaderType::Compute);
VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t index) {
const auto& src = regs.viewport_transform[index];
const float width = src.scale_x * 2.0f;
@ -97,118 +95,6 @@ VkRect2D GetScissorState(const Maxwell& regs, size_t index) {
return scissor;
}
std::array<GPUVAddr, Maxwell::MaxShaderProgram> GetShaderAddresses(
const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders) {
std::array<GPUVAddr, Maxwell::MaxShaderProgram> addresses;
for (size_t i = 0; i < std::size(addresses); ++i) {
addresses[i] = shaders[i] ? shaders[i]->GetGpuAddr() : 0;
}
return addresses;
}
struct TextureHandle {
constexpr TextureHandle(u32 data, bool via_header_index) {
const Tegra::Texture::TextureHandle handle{data};
image = handle.tic_id;
sampler = via_header_index ? image : handle.tsc_id.Value();
}
u32 image;
u32 sampler;
};
template <typename Engine, typename Entry>
TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const Entry& entry,
size_t stage, size_t index = 0) {
const auto shader_type = static_cast<Tegra::Engines::ShaderType>(stage);
if constexpr (std::is_same_v<Entry, SamplerEntry>) {
if (entry.is_separated) {
const u32 buffer_1 = entry.buffer;
const u32 buffer_2 = entry.secondary_buffer;
const u32 offset_1 = entry.offset;
const u32 offset_2 = entry.secondary_offset;
const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1);
const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2);
return TextureHandle(handle_1 | handle_2, via_header_index);
}
}
if (entry.is_bindless) {
const u32 raw = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
return TextureHandle(raw, via_header_index);
}
const u32 buffer = engine.GetBoundBuffer();
const u64 offset = (entry.offset + index) * sizeof(u32);
return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index);
}
ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) {
if (entry.is_buffer) {
return ImageViewType::e2D;
}
switch (entry.type) {
case Tegra::Shader::TextureType::Texture1D:
return entry.is_array ? ImageViewType::e1DArray : ImageViewType::e1D;
case Tegra::Shader::TextureType::Texture2D:
return entry.is_array ? ImageViewType::e2DArray : ImageViewType::e2D;
case Tegra::Shader::TextureType::Texture3D:
return ImageViewType::e3D;
case Tegra::Shader::TextureType::TextureCube:
return entry.is_array ? ImageViewType::CubeArray : ImageViewType::Cube;
}
UNREACHABLE();
return ImageViewType::e2D;
}
ImageViewType ImageViewTypeFromEntry(const ImageEntry& entry) {
switch (entry.type) {
case Tegra::Shader::ImageType::Texture1D:
return ImageViewType::e1D;
case Tegra::Shader::ImageType::Texture1DArray:
return ImageViewType::e1DArray;
case Tegra::Shader::ImageType::Texture2D:
return ImageViewType::e2D;
case Tegra::Shader::ImageType::Texture2DArray:
return ImageViewType::e2DArray;
case Tegra::Shader::ImageType::Texture3D:
return ImageViewType::e3D;
case Tegra::Shader::ImageType::TextureBuffer:
return ImageViewType::Buffer;
}
UNREACHABLE();
return ImageViewType::e2D;
}
void PushImageDescriptors(const ShaderEntries& entries, TextureCache& texture_cache,
VKUpdateDescriptorQueue& update_descriptor_queue,
ImageViewId*& image_view_id_ptr, VkSampler*& sampler_ptr) {
for ([[maybe_unused]] const auto& entry : entries.uniform_texels) {
const ImageViewId image_view_id = *image_view_id_ptr++;
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
update_descriptor_queue.AddTexelBuffer(image_view.BufferView());
}
for (const auto& entry : entries.samplers) {
for (size_t i = 0; i < entry.size; ++i) {
const VkSampler sampler = *sampler_ptr++;
const ImageViewId image_view_id = *image_view_id_ptr++;
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
const VkImageView handle = image_view.Handle(ImageViewTypeFromEntry(entry));
update_descriptor_queue.AddSampledImage(handle, sampler);
}
}
for ([[maybe_unused]] const auto& entry : entries.storage_texels) {
const ImageViewId image_view_id = *image_view_id_ptr++;
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
update_descriptor_queue.AddTexelBuffer(image_view.BufferView());
}
for (const auto& entry : entries.images) {
// TODO: Mark as modified
const ImageViewId image_view_id = *image_view_id_ptr++;
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
const VkImageView handle = image_view.Handle(ImageViewTypeFromEntry(entry));
update_descriptor_queue.AddImage(handle);
}
}
DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_instanced,
bool is_indexed) {
DrawParams params{
@ -216,6 +102,7 @@ DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_instan
.num_instances = is_instanced ? num_instances : 1,
.base_vertex = is_indexed ? regs.vb_element_base : regs.vertex_buffer.first,
.num_vertices = is_indexed ? regs.index_array.count : regs.vertex_buffer.count,
.first_index = is_indexed ? regs.index_array.first : 0,
.is_indexed = is_indexed,
};
if (regs.draw.topology == Maxwell::PrimitiveTopology::Quads) {
@ -243,21 +130,21 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
blit_image(device, scheduler, state_tracker, descriptor_pool),
astc_decoder_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue,
memory_allocator),
texture_cache_runtime{device, scheduler, memory_allocator,
staging_pool, blit_image, astc_decoder_pass},
render_pass_cache(device), texture_cache_runtime{device, scheduler,
memory_allocator, staging_pool,
blit_image, astc_decoder_pass,
render_pass_cache},
texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory),
buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool,
update_descriptor_queue, descriptor_pool),
buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime),
pipeline_cache(*this, gpu, maxwell3d, kepler_compute, gpu_memory, device, scheduler,
descriptor_pool, update_descriptor_queue),
pipeline_cache(*this, maxwell3d, kepler_compute, gpu_memory, device, scheduler,
descriptor_pool, update_descriptor_queue, render_pass_cache, buffer_cache,
texture_cache, gpu.ShaderNotify()),
query_cache{*this, maxwell3d, gpu_memory, device, scheduler}, accelerate_dma{buffer_cache},
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler),
wfi_event(device.GetLogical().CreateEvent()), async_shaders(emu_window_) {
wfi_event(device.GetLogical().CreateEvent()) {
scheduler.SetQueryCache(query_cache);
if (device.UseAsynchronousShaders()) {
async_shaders.AllocateWorkers();
}
}
RasterizerVulkan::~RasterizerVulkan() = default;
@ -270,53 +157,30 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
query_cache.UpdateCounters();
graphics_key.fixed_state.Refresh(maxwell3d, device.IsExtExtendedDynamicStateSupported());
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
texture_cache.SynchronizeGraphicsDescriptors();
texture_cache.UpdateRenderTargets(false);
const auto shaders = pipeline_cache.GetShaders();
graphics_key.shaders = GetShaderAddresses(shaders);
SetupShaderDescriptors(shaders, is_indexed);
const Framebuffer* const framebuffer = texture_cache.GetFramebuffer();
graphics_key.renderpass = framebuffer->RenderPass();
VKGraphicsPipeline* const pipeline = pipeline_cache.GetGraphicsPipeline(
graphics_key, framebuffer->NumColorBuffers(), async_shaders);
if (pipeline == nullptr || pipeline->GetHandle() == VK_NULL_HANDLE) {
// Async graphics pipeline was not ready.
GraphicsPipeline* const pipeline{pipeline_cache.CurrentGraphicsPipeline()};
if (!pipeline) {
return;
}
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
pipeline->Configure(is_indexed);
BeginTransformFeedback();
scheduler.RequestRenderpass(framebuffer);
scheduler.BindGraphicsPipeline(pipeline->GetHandle());
UpdateDynamicStates();
const auto& regs = maxwell3d.regs;
const u32 num_instances = maxwell3d.mme_draw.instance_count;
const DrawParams draw_params = MakeDrawParams(regs, num_instances, is_instanced, is_indexed);
const VkPipelineLayout pipeline_layout = pipeline->GetLayout();
const VkDescriptorSet descriptor_set = pipeline->CommitDescriptorSet();
scheduler.Record([pipeline_layout, descriptor_set, draw_params](vk::CommandBuffer cmdbuf) {
if (descriptor_set) {
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline_layout,
DESCRIPTOR_SET, descriptor_set, nullptr);
}
const auto& regs{maxwell3d.regs};
const u32 num_instances{maxwell3d.mme_draw.instance_count};
const DrawParams draw_params{MakeDrawParams(regs, num_instances, is_instanced, is_indexed)};
scheduler.Record([draw_params](vk::CommandBuffer cmdbuf) {
if (draw_params.is_indexed) {
cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances, 0,
draw_params.base_vertex, draw_params.base_instance);
cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances,
draw_params.first_index, draw_params.base_vertex,
draw_params.base_instance);
} else {
cmdbuf.Draw(draw_params.num_vertices, draw_params.num_instances,
draw_params.base_vertex, draw_params.base_instance);
}
});
EndTransformFeedback();
}
@ -326,6 +190,7 @@ void RasterizerVulkan::Clear() {
if (!maxwell3d.ShouldExecute()) {
return;
}
FlushWork();
query_cache.UpdateCounters();
@ -395,73 +260,20 @@ void RasterizerVulkan::Clear() {
});
}
void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
MICROPROFILE_SCOPE(Vulkan_Compute);
void RasterizerVulkan::DispatchCompute() {
FlushWork();
query_cache.UpdateCounters();
const auto& launch_desc = kepler_compute.launch_description;
auto& pipeline = pipeline_cache.GetComputePipeline({
.shader = code_addr,
.shared_memory_size = launch_desc.shared_alloc,
.workgroup_size{
launch_desc.block_dim_x,
launch_desc.block_dim_y,
launch_desc.block_dim_z,
},
});
// Compute dispatches can't be executed inside a renderpass
scheduler.RequestOutsideRenderPassOperationContext();
image_view_indices.clear();
sampler_handles.clear();
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
const auto& entries = pipeline.GetEntries();
buffer_cache.SetEnabledComputeUniformBuffers(entries.enabled_uniform_buffers);
buffer_cache.UnbindComputeStorageBuffers();
u32 ssbo_index = 0;
for (const auto& buffer : entries.global_buffers) {
buffer_cache.BindComputeStorageBuffer(ssbo_index, buffer.cbuf_index, buffer.cbuf_offset,
buffer.is_written);
++ssbo_index;
ComputePipeline* const pipeline{pipeline_cache.CurrentComputePipeline()};
if (!pipeline) {
return;
}
buffer_cache.UpdateComputeBuffers();
std::scoped_lock lock{texture_cache.mutex, buffer_cache.mutex};
pipeline->Configure(kepler_compute, gpu_memory, scheduler, buffer_cache, texture_cache);
texture_cache.SynchronizeComputeDescriptors();
SetupComputeUniformTexels(entries);
SetupComputeTextures(entries);
SetupComputeStorageTexels(entries);
SetupComputeImages(entries);
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
texture_cache.FillComputeImageViews(indices_span, image_view_ids);
update_descriptor_queue.Acquire();
buffer_cache.BindHostComputeBuffers();
ImageViewId* image_view_id_ptr = image_view_ids.data();
VkSampler* sampler_ptr = sampler_handles.data();
PushImageDescriptors(entries, texture_cache, update_descriptor_queue, image_view_id_ptr,
sampler_ptr);
const VkPipeline pipeline_handle = pipeline.GetHandle();
const VkPipelineLayout pipeline_layout = pipeline.GetLayout();
const VkDescriptorSet descriptor_set = pipeline.CommitDescriptorSet();
scheduler.Record([grid_x = launch_desc.grid_dim_x, grid_y = launch_desc.grid_dim_y,
grid_z = launch_desc.grid_dim_z, pipeline_handle, pipeline_layout,
descriptor_set](vk::CommandBuffer cmdbuf) {
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline_handle);
if (descriptor_set) {
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline_layout,
DESCRIPTOR_SET, descriptor_set, nullptr);
}
cmdbuf.Dispatch(grid_x, grid_y, grid_z);
});
const auto& qmd{kepler_compute.launch_description};
const std::array<u32, 3> dim{qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z};
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([dim](vk::CommandBuffer cmdbuf) { cmdbuf.Dispatch(dim[0], dim[1], dim[2]); });
}
void RasterizerVulkan::ResetCounter(VideoCore::QueryType type) {
@ -626,6 +438,7 @@ void RasterizerVulkan::WaitForIdle() {
void RasterizerVulkan::FragmentBarrier() {
// We already put barriers when a render pass finishes
scheduler.RequestOutsideRenderPassOperationContext();
}
void RasterizerVulkan::TiledCacheBarrier() {
@ -633,10 +446,11 @@ void RasterizerVulkan::TiledCacheBarrier() {
}
void RasterizerVulkan::FlushCommands() {
if (draw_counter > 0) {
draw_counter = 0;
scheduler.Flush();
if (draw_counter == 0) {
return;
}
draw_counter = 0;
scheduler.Flush();
}
void RasterizerVulkan::TickFrame() {
@ -676,13 +490,18 @@ bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config,
if (!image_view) {
return false;
}
screen_info.image_view = image_view->Handle(VideoCommon::ImageViewType::e2D);
screen_info.image_view = image_view->Handle(Shader::TextureType::Color2D);
screen_info.width = image_view->size.width;
screen_info.height = image_view->size.height;
screen_info.is_srgb = VideoCore::Surface::IsPixelFormatSRGB(image_view->format);
return true;
}
void RasterizerVulkan::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) {
pipeline_cache.LoadDiskResources(title_id, stop_loading, callback);
}
void RasterizerVulkan::FlushWork() {
static constexpr u32 DRAWS_TO_DISPATCH = 4096;
@ -691,13 +510,11 @@ void RasterizerVulkan::FlushWork() {
if ((++draw_counter & 7) != 7) {
return;
}
if (draw_counter < DRAWS_TO_DISPATCH) {
// Send recorded tasks to the worker thread
scheduler.DispatchWork();
return;
}
// Otherwise (every certain number of draws) flush execution.
// This submits commands to the Vulkan driver.
scheduler.Flush();
@ -716,52 +533,6 @@ bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64
return buffer_cache.DMACopy(src_address, dest_address, amount);
}
void RasterizerVulkan::SetupShaderDescriptors(
const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders, bool is_indexed) {
image_view_indices.clear();
sampler_handles.clear();
for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
Shader* const shader = shaders[stage + 1];
if (!shader) {
continue;
}
const ShaderEntries& entries = shader->GetEntries();
SetupGraphicsUniformTexels(entries, stage);
SetupGraphicsTextures(entries, stage);
SetupGraphicsStorageTexels(entries, stage);
SetupGraphicsImages(entries, stage);
buffer_cache.SetEnabledUniformBuffers(stage, entries.enabled_uniform_buffers);
buffer_cache.UnbindGraphicsStorageBuffers(stage);
u32 ssbo_index = 0;
for (const auto& buffer : entries.global_buffers) {
buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, buffer.cbuf_index,
buffer.cbuf_offset, buffer.is_written);
++ssbo_index;
}
}
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
buffer_cache.UpdateGraphicsBuffers(is_indexed);
texture_cache.FillGraphicsImageViews(indices_span, image_view_ids);
buffer_cache.BindHostGeometryBuffers(is_indexed);
update_descriptor_queue.Acquire();
ImageViewId* image_view_id_ptr = image_view_ids.data();
VkSampler* sampler_ptr = sampler_handles.data();
for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
// Skip VertexA stage
Shader* const shader = shaders[stage + 1];
if (!shader) {
continue;
}
buffer_cache.BindHostStageBuffers(stage);
PushImageDescriptors(shader->GetEntries(), texture_cache, update_descriptor_queue,
image_view_id_ptr, sampler_ptr);
}
}
void RasterizerVulkan::UpdateDynamicStates() {
auto& regs = maxwell3d.regs;
UpdateViewportsState(regs);
@ -770,6 +541,7 @@ void RasterizerVulkan::UpdateDynamicStates() {
UpdateBlendConstants(regs);
UpdateDepthBounds(regs);
UpdateStencilFaces(regs);
UpdateLineWidth(regs);
if (device.IsExtExtendedDynamicStateSupported()) {
UpdateCullMode(regs);
UpdateDepthBoundsTestEnable(regs);
@ -779,6 +551,9 @@ void RasterizerVulkan::UpdateDynamicStates() {
UpdateFrontFace(regs);
UpdateStencilOp(regs);
UpdateStencilTestEnable(regs);
if (device.IsExtVertexInputDynamicStateSupported()) {
UpdateVertexInput(regs);
}
}
}
@ -810,89 +585,6 @@ void RasterizerVulkan::EndTransformFeedback() {
[](vk::CommandBuffer cmdbuf) { cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr); });
}
void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, size_t stage) {
const auto& regs = maxwell3d.regs;
const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : entries.uniform_texels) {
const TextureHandle handle = GetTextureInfo(maxwell3d, via_header_index, entry, stage);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, size_t stage) {
const auto& regs = maxwell3d.regs;
const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : entries.samplers) {
for (size_t index = 0; index < entry.size; ++index) {
const TextureHandle handle =
GetTextureInfo(maxwell3d, via_header_index, entry, stage, index);
image_view_indices.push_back(handle.image);
Sampler* const sampler = texture_cache.GetGraphicsSampler(handle.sampler);
sampler_handles.push_back(sampler->Handle());
}
}
}
void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, size_t stage) {
const auto& regs = maxwell3d.regs;
const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : entries.storage_texels) {
const TextureHandle handle = GetTextureInfo(maxwell3d, via_header_index, entry, stage);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, size_t stage) {
const auto& regs = maxwell3d.regs;
const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : entries.images) {
const TextureHandle handle = GetTextureInfo(maxwell3d, via_header_index, entry, stage);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : entries.uniform_texels) {
const TextureHandle handle =
GetTextureInfo(kepler_compute, via_header_index, entry, COMPUTE_SHADER_INDEX);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : entries.samplers) {
for (size_t index = 0; index < entry.size; ++index) {
const TextureHandle handle = GetTextureInfo(kepler_compute, via_header_index, entry,
COMPUTE_SHADER_INDEX, index);
image_view_indices.push_back(handle.image);
Sampler* const sampler = texture_cache.GetComputeSampler(handle.sampler);
sampler_handles.push_back(sampler->Handle());
}
}
}
void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : entries.storage_texels) {
const TextureHandle handle =
GetTextureInfo(kepler_compute, via_header_index, entry, COMPUTE_SHADER_INDEX);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : entries.images) {
const TextureHandle handle =
GetTextureInfo(kepler_compute, via_header_index, entry, COMPUTE_SHADER_INDEX);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchViewports()) {
return;
@ -985,6 +677,14 @@ void RasterizerVulkan::UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs)
}
}
void RasterizerVulkan::UpdateLineWidth(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchLineWidth()) {
return;
}
const float width = regs.line_smooth_enable ? regs.line_width_smooth : regs.line_width_aliased;
scheduler.Record([width](vk::CommandBuffer cmdbuf) { cmdbuf.SetLineWidth(width); });
}
void RasterizerVulkan::UpdateCullMode(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchCullMode()) {
return;
@ -999,6 +699,11 @@ void RasterizerVulkan::UpdateDepthBoundsTestEnable(Tegra::Engines::Maxwell3D::Re
if (!state_tracker.TouchDepthBoundsTestEnable()) {
return;
}
bool enabled = regs.depth_bounds_enable;
if (enabled && !device.IsDepthBoundsSupported()) {
LOG_WARNING(Render_Vulkan, "Depth bounds is enabled but not supported");
enabled = false;
}
scheduler.Record([enable = regs.depth_bounds_enable](vk::CommandBuffer cmdbuf) {
cmdbuf.SetDepthBoundsTestEnableEXT(enable);
});
@ -1086,4 +791,62 @@ void RasterizerVulkan::UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs&
});
}
void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs) {
auto& dirty{maxwell3d.dirty.flags};
if (!dirty[Dirty::VertexInput]) {
return;
}
dirty[Dirty::VertexInput] = false;
boost::container::static_vector<VkVertexInputBindingDescription2EXT, 32> bindings;
boost::container::static_vector<VkVertexInputAttributeDescription2EXT, 32> attributes;
// There seems to be a bug on Nvidia's driver where updating only higher attributes ends up
// generating dirty state. Track the highest dirty attribute and update all attributes until
// that one.
size_t highest_dirty_attr{};
for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) {
if (dirty[Dirty::VertexAttribute0 + index]) {
highest_dirty_attr = index;
}
}
for (size_t index = 0; index < highest_dirty_attr; ++index) {
const Maxwell::VertexAttribute attribute{regs.vertex_attrib_format[index]};
const u32 binding{attribute.buffer};
dirty[Dirty::VertexAttribute0 + index] = false;
dirty[Dirty::VertexBinding0 + static_cast<size_t>(binding)] = true;
if (!attribute.constant) {
attributes.push_back({
.sType = VK_STRUCTURE_TYPE_VERTEX_INPUT_ATTRIBUTE_DESCRIPTION_2_EXT,
.pNext = nullptr,
.location = static_cast<u32>(index),
.binding = binding,
.format = MaxwellToVK::VertexFormat(attribute.type, attribute.size),
.offset = attribute.offset,
});
}
}
for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) {
if (!dirty[Dirty::VertexBinding0 + index]) {
continue;
}
dirty[Dirty::VertexBinding0 + index] = false;
const u32 binding{static_cast<u32>(index)};
const auto& input_binding{regs.vertex_array[binding]};
const bool is_instanced{regs.instanced_arrays.IsInstancingEnabled(binding)};
bindings.push_back({
.sType = VK_STRUCTURE_TYPE_VERTEX_INPUT_BINDING_DESCRIPTION_2_EXT,
.pNext = nullptr,
.binding = binding,
.stride = input_binding.stride,
.inputRate = is_instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX,
.divisor = is_instanced ? input_binding.divisor : 1,
});
}
scheduler.Record([bindings, attributes](vk::CommandBuffer cmdbuf) {
cmdbuf.SetVertexInputEXT(bindings, attributes);
});
}
} // namespace Vulkan

59
src/video_core/renderer_vulkan/vk_rasterizer.h
View File

@ -21,14 +21,13 @@
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_fence_manager.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_query_cache.h"
#include "video_core/renderer_vulkan/vk_render_pass_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/shader/async_shaders.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -73,7 +72,7 @@ public:
void Draw(bool is_indexed, bool is_instanced) override;
void Clear() override;
void DispatchCompute(GPUVAddr code_addr) override;
void DispatchCompute() override;
void ResetCounter(VideoCore::QueryType type) override;
void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
@ -102,19 +101,8 @@ public:
Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override;
bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
u32 pixel_stride) override;
VideoCommon::Shader::AsyncShaders& GetAsyncShaders() {
return async_shaders;
}
const VideoCommon::Shader::AsyncShaders& GetAsyncShaders() const {
return async_shaders;
}
/// Maximum supported size that a constbuffer can have in bytes.
static constexpr size_t MaxConstbufferSize = 0x10000;
static_assert(MaxConstbufferSize % (4 * sizeof(float)) == 0,
"The maximum size of a constbuffer must be a multiple of the size of GLvec4");
void LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) override;
private:
static constexpr size_t MAX_TEXTURES = 192;
@ -125,46 +113,19 @@ private:
void FlushWork();
/// Setup descriptors in the graphics pipeline.
void SetupShaderDescriptors(const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders,
bool is_indexed);
void UpdateDynamicStates();
void BeginTransformFeedback();
void EndTransformFeedback();
/// Setup uniform texels in the graphics pipeline.
void SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage);
/// Setup textures in the graphics pipeline.
void SetupGraphicsTextures(const ShaderEntries& entries, std::size_t stage);
/// Setup storage texels in the graphics pipeline.
void SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage);
/// Setup images in the graphics pipeline.
void SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage);
/// Setup texel buffers in the compute pipeline.
void SetupComputeUniformTexels(const ShaderEntries& entries);
/// Setup textures in the compute pipeline.
void SetupComputeTextures(const ShaderEntries& entries);
/// Setup storage texels in the compute pipeline.
void SetupComputeStorageTexels(const ShaderEntries& entries);
/// Setup images in the compute pipeline.
void SetupComputeImages(const ShaderEntries& entries);
void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateBlendConstants(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateDepthBounds(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateLineWidth(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateCullMode(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateDepthBoundsTestEnable(Tegra::Engines::Maxwell3D::Regs& regs);
@ -175,6 +136,8 @@ private:
void UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs);
Tegra::GPU& gpu;
Tegra::MemoryManager& gpu_memory;
Tegra::Engines::Maxwell3D& maxwell3d;
@ -187,24 +150,22 @@ private:
VKScheduler& scheduler;
StagingBufferPool staging_pool;
VKDescriptorPool descriptor_pool;
DescriptorPool descriptor_pool;
VKUpdateDescriptorQueue update_descriptor_queue;
BlitImageHelper blit_image;
ASTCDecoderPass astc_decoder_pass;
GraphicsPipelineCacheKey graphics_key;
RenderPassCache render_pass_cache;
TextureCacheRuntime texture_cache_runtime;
TextureCache texture_cache;
BufferCacheRuntime buffer_cache_runtime;
BufferCache buffer_cache;
VKPipelineCache pipeline_cache;
PipelineCache pipeline_cache;
VKQueryCache query_cache;
AccelerateDMA accelerate_dma;
VKFenceManager fence_manager;
vk::Event wfi_event;
VideoCommon::Shader::AsyncShaders async_shaders;
boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices;
std::array<VideoCommon::ImageViewId, MAX_IMAGE_VIEWS> image_view_ids;

96
src/video_core/renderer_vulkan/vk_render_pass_cache.cpp Normal file
View File

@ -0,0 +1,96 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <unordered_map>
#include <boost/container/static_vector.hpp>
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_render_pass_cache.h"
#include "video_core/surface.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
namespace {
using VideoCore::Surface::PixelFormat;
VkAttachmentDescription AttachmentDescription(const Device& device, PixelFormat format,
VkSampleCountFlagBits samples) {
using MaxwellToVK::SurfaceFormat;
return {
.flags = VK_ATTACHMENT_DESCRIPTION_MAY_ALIAS_BIT,
.format = SurfaceFormat(device, FormatType::Optimal, true, format).format,
.samples = samples,
.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
.storeOp = VK_ATTACHMENT_STORE_OP_STORE,
.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE,
.initialLayout = VK_IMAGE_LAYOUT_GENERAL,
.finalLayout = VK_IMAGE_LAYOUT_GENERAL,
};
}
} // Anonymous namespace
RenderPassCache::RenderPassCache(const Device& device_) : device{&device_} {}
VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
std::lock_guard lock{mutex};
const auto [pair, is_new] = cache.try_emplace(key);
if (!is_new) {
return *pair->second;
}
boost::container::static_vector<VkAttachmentDescription, 9> descriptions;
std::array<VkAttachmentReference, 8> references{};
u32 num_attachments{};
u32 num_colors{};
for (size_t index = 0; index < key.color_formats.size(); ++index) {
const PixelFormat format{key.color_formats[index]};
const bool is_valid{format != PixelFormat::Invalid};
references[index] = VkAttachmentReference{
.attachment = is_valid ? num_colors : VK_ATTACHMENT_UNUSED,
.layout = VK_IMAGE_LAYOUT_GENERAL,
};
if (is_valid) {
descriptions.push_back(AttachmentDescription(*device, format, key.samples));
num_attachments = static_cast<u32>(index + 1);
++num_colors;
}
}
const bool has_depth{key.depth_format != PixelFormat::Invalid};
VkAttachmentReference depth_reference{};
if (key.depth_format != PixelFormat::Invalid) {
depth_reference = VkAttachmentReference{
.attachment = num_colors,
.layout = VK_IMAGE_LAYOUT_GENERAL,
};
descriptions.push_back(AttachmentDescription(*device, key.depth_format, key.samples));
}
const VkSubpassDescription subpass{
.flags = 0,
.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
.inputAttachmentCount = 0,
.pInputAttachments = nullptr,
.colorAttachmentCount = num_attachments,
.pColorAttachments = references.data(),
.pResolveAttachments = nullptr,
.pDepthStencilAttachment = has_depth ? &depth_reference : nullptr,
.preserveAttachmentCount = 0,
.pPreserveAttachments = nullptr,
};
pair->second = device->GetLogical().CreateRenderPass({
.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.attachmentCount = static_cast<u32>(descriptions.size()),
.pAttachments = descriptions.empty() ? nullptr : descriptions.data(),
.subpassCount = 1,
.pSubpasses = &subpass,
.dependencyCount = 0,
.pDependencies = nullptr,
});
return *pair->second;
}
} // namespace Vulkan

55
src/video_core/renderer_vulkan/vk_render_pass_cache.h Normal file
View File

@ -0,0 +1,55 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <mutex>
#include <unordered_map>
#include "video_core/surface.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
struct RenderPassKey {
auto operator<=>(const RenderPassKey&) const noexcept = default;
std::array<VideoCore::Surface::PixelFormat, 8> color_formats;
VideoCore::Surface::PixelFormat depth_format;
VkSampleCountFlagBits samples;
};
} // namespace Vulkan
namespace std {
template <>
struct hash<Vulkan::RenderPassKey> {
[[nodiscard]] size_t operator()(const Vulkan::RenderPassKey& key) const noexcept {
size_t value = static_cast<size_t>(key.depth_format) << 48;
value ^= static_cast<size_t>(key.samples) << 52;
for (size_t i = 0; i < key.color_formats.size(); ++i) {
value ^= static_cast<size_t>(key.color_formats[i]) << (i * 6);
}
return value;
}
};
} // namespace std
namespace Vulkan {
class Device;
class RenderPassCache {
public:
explicit RenderPassCache(const Device& device_);
VkRenderPass Get(const RenderPassKey& key);
private:
const Device* device{};
std::unordered_map<RenderPassKey, vk::RenderPass> cache;
std::mutex mutex;
};
} // namespace Vulkan

12
src/video_core/renderer_vulkan/vk_resource_pool.cpp
View File

@ -10,18 +10,16 @@
namespace Vulkan {
ResourcePool::ResourcePool(MasterSemaphore& master_semaphore_, size_t grow_step_)
: master_semaphore{master_semaphore_}, grow_step{grow_step_} {}
ResourcePool::~ResourcePool() = default;
: master_semaphore{&master_semaphore_}, grow_step{grow_step_} {}
size_t ResourcePool::CommitResource() {
// Refresh semaphore to query updated results
master_semaphore.Refresh();
const u64 gpu_tick = master_semaphore.KnownGpuTick();
master_semaphore->Refresh();
const u64 gpu_tick = master_semaphore->KnownGpuTick();
const auto search = [this, gpu_tick](size_t begin, size_t end) -> std::optional<size_t> {
for (size_t iterator = begin; iterator < end; ++iterator) {
if (gpu_tick >= ticks[iterator]) {
ticks[iterator] = master_semaphore.CurrentTick();
ticks[iterator] = master_semaphore->CurrentTick();
return iterator;
}
}
@ -36,7 +34,7 @@ size_t ResourcePool::CommitResource() {
// Both searches failed, the pool is full; handle it.
const size_t free_resource = ManageOverflow();
ticks[free_resource] = master_semaphore.CurrentTick();
ticks[free_resource] = master_semaphore->CurrentTick();
found = free_resource;
}
}

12
src/video_core/renderer_vulkan/vk_resource_pool.h
View File

@ -18,8 +18,16 @@ class MasterSemaphore;
*/
class ResourcePool {
public:
explicit ResourcePool() = default;
explicit ResourcePool(MasterSemaphore& master_semaphore, size_t grow_step);
virtual ~ResourcePool();
virtual ~ResourcePool() = default;
ResourcePool& operator=(ResourcePool&&) noexcept = default;
ResourcePool(ResourcePool&&) noexcept = default;
ResourcePool& operator=(const ResourcePool&) = default;
ResourcePool(const ResourcePool&) = default;
protected:
size_t CommitResource();
@ -34,7 +42,7 @@ private:
/// Allocates a new page of resources.
void Grow();
MasterSemaphore& master_semaphore;
MasterSemaphore* master_semaphore{};
size_t grow_step = 0; ///< Number of new resources created after an overflow
size_t hint_iterator = 0; ///< Hint to where the next free resources is likely to be found
std::vector<u64> ticks; ///< Ticks for each resource

178
src/video_core/renderer_vulkan/vk_scheduler.cpp
View File

@ -31,7 +31,7 @@ void VKScheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf) {
command->~Command();
command = next;
}
submit = false;
command_offset = 0;
first = nullptr;
last = nullptr;
@ -42,13 +42,16 @@ VKScheduler::VKScheduler(const Device& device_, StateTracker& state_tracker_)
master_semaphore{std::make_unique<MasterSemaphore>(device)},
command_pool{std::make_unique<CommandPool>(*master_semaphore, device)} {
AcquireNewChunk();
AllocateNewContext();
AllocateWorkerCommandBuffer();
worker_thread = std::thread(&VKScheduler::WorkerThread, this);
}
VKScheduler::~VKScheduler() {
quit = true;
cv.notify_all();
{
std::lock_guard lock{work_mutex};
quit = true;
}
work_cv.notify_all();
worker_thread.join();
}
@ -60,6 +63,7 @@ void VKScheduler::Flush(VkSemaphore semaphore) {
void VKScheduler::Finish(VkSemaphore semaphore) {
const u64 presubmit_tick = CurrentTick();
SubmitExecution(semaphore);
WaitWorker();
Wait(presubmit_tick);
AllocateNewContext();
}
@ -68,20 +72,19 @@ void VKScheduler::WaitWorker() {
MICROPROFILE_SCOPE(Vulkan_WaitForWorker);
DispatchWork();
bool finished = false;
do {
cv.notify_all();
std::unique_lock lock{mutex};
finished = chunk_queue.Empty();
} while (!finished);
std::unique_lock lock{work_mutex};
wait_cv.wait(lock, [this] { return work_queue.empty(); });
}
void VKScheduler::DispatchWork() {
if (chunk->Empty()) {
return;
}
chunk_queue.Push(std::move(chunk));
cv.notify_all();
{
std::lock_guard lock{work_mutex};
work_queue.push(std::move(chunk));
}
work_cv.notify_one();
AcquireNewChunk();
}
@ -124,85 +127,41 @@ void VKScheduler::RequestOutsideRenderPassOperationContext() {
EndRenderPass();
}
void VKScheduler::BindGraphicsPipeline(VkPipeline pipeline) {
bool VKScheduler::UpdateGraphicsPipeline(GraphicsPipeline* pipeline) {
if (state.graphics_pipeline == pipeline) {
return;
return false;
}
state.graphics_pipeline = pipeline;
Record([pipeline](vk::CommandBuffer cmdbuf) {
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
});
return true;
}
void VKScheduler::WorkerThread() {
Common::SetCurrentThreadPriority(Common::ThreadPriority::High);
std::unique_lock lock{mutex};
Common::SetCurrentThreadName("yuzu:VulkanWorker");
do {
cv.wait(lock, [this] { return !chunk_queue.Empty() || quit; });
if (quit) {
continue;
if (work_queue.empty()) {
wait_cv.notify_all();
}
auto extracted_chunk = std::move(chunk_queue.Front());
chunk_queue.Pop();
extracted_chunk->ExecuteAll(current_cmdbuf);
chunk_reserve.Push(std::move(extracted_chunk));
std::unique_ptr<CommandChunk> work;
{
std::unique_lock lock{work_mutex};
work_cv.wait(lock, [this] { return !work_queue.empty() || quit; });
if (quit) {
continue;
}
work = std::move(work_queue.front());
work_queue.pop();
}
const bool has_submit = work->HasSubmit();
work->ExecuteAll(current_cmdbuf);
if (has_submit) {
AllocateWorkerCommandBuffer();
}
std::lock_guard reserve_lock{reserve_mutex};
chunk_reserve.push_back(std::move(work));
} while (!quit);
}
void VKScheduler::SubmitExecution(VkSemaphore semaphore) {
EndPendingOperations();
InvalidateState();
WaitWorker();
std::unique_lock lock{mutex};
current_cmdbuf.End();
const VkSemaphore timeline_semaphore = master_semaphore->Handle();
const u32 num_signal_semaphores = semaphore ? 2U : 1U;
const u64 signal_value = master_semaphore->CurrentTick();
const u64 wait_value = signal_value - 1;
const VkPipelineStageFlags wait_stage_mask = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT;
master_semaphore->NextTick();
const std::array signal_values{signal_value, u64(0)};
const std::array signal_semaphores{timeline_semaphore, semaphore};
const VkTimelineSemaphoreSubmitInfoKHR timeline_si{
.sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR,
.pNext = nullptr,
.waitSemaphoreValueCount = 1,
.pWaitSemaphoreValues = &wait_value,
.signalSemaphoreValueCount = num_signal_semaphores,
.pSignalSemaphoreValues = signal_values.data(),
};
const VkSubmitInfo submit_info{
.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
.pNext = &timeline_si,
.waitSemaphoreCount = 1,
.pWaitSemaphores = &timeline_semaphore,
.pWaitDstStageMask = &wait_stage_mask,
.commandBufferCount = 1,
.pCommandBuffers = current_cmdbuf.address(),
.signalSemaphoreCount = num_signal_semaphores,
.pSignalSemaphores = signal_semaphores.data(),
};
switch (const VkResult result = device.GetGraphicsQueue().Submit(submit_info)) {
case VK_SUCCESS:
break;
case VK_ERROR_DEVICE_LOST:
device.ReportLoss();
[[fallthrough]];
default:
vk::Check(result);
}
}
void VKScheduler::AllocateNewContext() {
std::unique_lock lock{mutex};
void VKScheduler::AllocateWorkerCommandBuffer() {
current_cmdbuf = vk::CommandBuffer(command_pool->Commit(), device.GetDispatchLoader());
current_cmdbuf.Begin({
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
@ -210,7 +169,59 @@ void VKScheduler::AllocateNewContext() {
.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
.pInheritanceInfo = nullptr,
});
}
void VKScheduler::SubmitExecution(VkSemaphore semaphore) {
EndPendingOperations();
InvalidateState();
const u64 signal_value = master_semaphore->NextTick();
Record([semaphore, signal_value, this](vk::CommandBuffer cmdbuf) {
cmdbuf.End();
const u32 num_signal_semaphores = semaphore ? 2U : 1U;
const u64 wait_value = signal_value - 1;
const VkPipelineStageFlags wait_stage_mask = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT;
const VkSemaphore timeline_semaphore = master_semaphore->Handle();
const std::array signal_values{signal_value, u64(0)};
const std::array signal_semaphores{timeline_semaphore, semaphore};
const VkTimelineSemaphoreSubmitInfoKHR timeline_si{
.sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR,
.pNext = nullptr,
.waitSemaphoreValueCount = 1,
.pWaitSemaphoreValues = &wait_value,
.signalSemaphoreValueCount = num_signal_semaphores,
.pSignalSemaphoreValues = signal_values.data(),
};
const VkSubmitInfo submit_info{
.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
.pNext = &timeline_si,
.waitSemaphoreCount = 1,
.pWaitSemaphores = &timeline_semaphore,
.pWaitDstStageMask = &wait_stage_mask,
.commandBufferCount = 1,
.pCommandBuffers = cmdbuf.address(),
.signalSemaphoreCount = num_signal_semaphores,
.pSignalSemaphores = signal_semaphores.data(),
};
switch (const VkResult result = device.GetGraphicsQueue().Submit(submit_info)) {
case VK_SUCCESS:
break;
case VK_ERROR_DEVICE_LOST:
device.ReportLoss();
[[fallthrough]];
default:
vk::Check(result);
}
});
chunk->MarkSubmit();
DispatchWork();
}
void VKScheduler::AllocateNewContext() {
// Enable counters once again. These are disabled when a command buffer is finished.
if (query_cache) {
query_cache->UpdateCounters();
@ -265,12 +276,13 @@ void VKScheduler::EndRenderPass() {
}
void VKScheduler::AcquireNewChunk() {
if (chunk_reserve.Empty()) {
std::lock_guard lock{reserve_mutex};
if (chunk_reserve.empty()) {
chunk = std::make_unique<CommandChunk>();
return;
}
chunk = std::move(chunk_reserve.Front());
chunk_reserve.Pop();
chunk = std::move(chunk_reserve.back());
chunk_reserve.pop_back();
}
} // namespace Vulkan

38
src/video_core/renderer_vulkan/vk_scheduler.h
View File

@ -8,12 +8,12 @@
#include <condition_variable>
#include <cstddef>
#include <memory>
#include <stack>
#include <thread>
#include <utility>
#include <queue>
#include "common/alignment.h"
#include "common/common_types.h"
#include "common/threadsafe_queue.h"
#include "video_core/renderer_vulkan/vk_master_semaphore.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -22,6 +22,7 @@ namespace Vulkan {
class CommandPool;
class Device;
class Framebuffer;
class GraphicsPipeline;
class StateTracker;
class VKQueryCache;
@ -52,8 +53,8 @@ public:
/// of a renderpass.
void RequestOutsideRenderPassOperationContext();
/// Binds a pipeline to the current execution context.
void BindGraphicsPipeline(VkPipeline pipeline);
/// Update the pipeline to the current execution context.
bool UpdateGraphicsPipeline(GraphicsPipeline* pipeline);
/// Invalidates current command buffer state except for render passes
void InvalidateState();
@ -85,6 +86,10 @@ public:
/// Waits for the given tick to trigger on the GPU.
void Wait(u64 tick) {
if (tick >= master_semaphore->CurrentTick()) {
// Make sure we are not waiting for the current tick without signalling
Flush();
}
master_semaphore->Wait(tick);
}
@ -154,15 +159,24 @@ private:
return true;
}
void MarkSubmit() {
submit = true;
}
bool Empty() const {
return command_offset == 0;
}
bool HasSubmit() const {
return submit;
}
private:
Command* first = nullptr;
Command* last = nullptr;
size_t command_offset = 0;
bool submit = false;
alignas(std::max_align_t) std::array<u8, 0x8000> data{};
};
@ -170,11 +184,13 @@ private:
VkRenderPass renderpass = nullptr;
VkFramebuffer framebuffer = nullptr;
VkExtent2D render_area = {0, 0};
VkPipeline graphics_pipeline = nullptr;
GraphicsPipeline* graphics_pipeline = nullptr;
};
void WorkerThread();
void AllocateWorkerCommandBuffer();
void SubmitExecution(VkSemaphore semaphore);
void AllocateNewContext();
@ -204,11 +220,13 @@ private:
std::array<VkImage, 9> renderpass_images{};
std::array<VkImageSubresourceRange, 9> renderpass_image_ranges{};
Common::SPSCQueue<std::unique_ptr<CommandChunk>> chunk_queue;
Common::SPSCQueue<std::unique_ptr<CommandChunk>> chunk_reserve;
std::mutex mutex;
std::condition_variable cv;
bool quit = false;
std::queue<std::unique_ptr<CommandChunk>> work_queue;
std::vector<std::unique_ptr<CommandChunk>> chunk_reserve;
std::mutex reserve_mutex;
std::mutex work_mutex;
std::condition_variable work_cv;
std::condition_variable wait_cv;
std::atomic_bool quit{};
};
} // namespace Vulkan

3166
src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
View File

File diff suppressed because it is too large Load Diff

99
src/video_core/renderer_vulkan/vk_shader_decompiler.h
View File

@ -1,99 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <set>
#include <vector>
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace Vulkan {
class Device;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
using UniformTexelEntry = VideoCommon::Shader::SamplerEntry;
using SamplerEntry = VideoCommon::Shader::SamplerEntry;
using StorageTexelEntry = VideoCommon::Shader::ImageEntry;
using ImageEntry = VideoCommon::Shader::ImageEntry;
constexpr u32 DESCRIPTOR_SET = 0;
class ConstBufferEntry : public VideoCommon::Shader::ConstBuffer {
public:
explicit constexpr ConstBufferEntry(const ConstBuffer& entry_, u32 index_)
: ConstBuffer{entry_}, index{index_} {}
constexpr u32 GetIndex() const {
return index;
}
private:
u32 index{};
};
struct GlobalBufferEntry {
u32 cbuf_index{};
u32 cbuf_offset{};
bool is_written{};
};
struct ShaderEntries {
u32 NumBindings() const {
return static_cast<u32>(const_buffers.size() + global_buffers.size() +
uniform_texels.size() + samplers.size() + storage_texels.size() +
images.size());
}
std::vector<ConstBufferEntry> const_buffers;
std::vector<GlobalBufferEntry> global_buffers;
std::vector<UniformTexelEntry> uniform_texels;
std::vector<SamplerEntry> samplers;
std::vector<StorageTexelEntry> storage_texels;
std::vector<ImageEntry> images;
std::set<u32> attributes;
std::array<bool, Maxwell::NumClipDistances> clip_distances{};
std::size_t shader_length{};
u32 enabled_uniform_buffers{};
bool uses_warps{};
};
struct Specialization final {
u32 base_binding{};
// Compute specific
std::array<u32, 3> workgroup_size{};
u32 shared_memory_size{};
// Graphics specific
std::optional<float> point_size;
std::bitset<Maxwell::NumVertexAttributes> enabled_attributes;
std::array<Maxwell::VertexAttribute::Type, Maxwell::NumVertexAttributes> attribute_types{};
bool ndc_minus_one_to_one{};
bool early_fragment_tests{};
float alpha_test_ref{};
Maxwell::ComparisonOp alpha_test_func{};
};
// Old gcc versions don't consider this trivially copyable.
// static_assert(std::is_trivially_copyable_v<Specialization>);
struct SPIRVShader {
std::vector<u32> code;
ShaderEntries entries;
};
ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir);
std::vector<u32> Decompile(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
Tegra::Engines::ShaderType stage,
const VideoCommon::Shader::Registry& registry,
const Specialization& specialization);
} // namespace Vulkan

2
src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
View File

@ -91,7 +91,7 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem
.flags = 0,
.size = STREAM_BUFFER_SIZE,
.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
VK_BUFFER_USAGE_INDEX_BUFFER_BIT,
VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,

58
src/video_core/renderer_vulkan/vk_state_tracker.cpp
View File

@ -29,9 +29,10 @@ using Flags = Maxwell3D::DirtyState::Flags;
Flags MakeInvalidationFlags() {
static constexpr int INVALIDATION_FLAGS[]{
Viewports, Scissors, DepthBias, BlendConstants, DepthBounds,
StencilProperties, CullMode, DepthBoundsEnable, DepthTestEnable, DepthWriteEnable,
DepthCompareOp, FrontFace, StencilOp, StencilTestEnable, VertexBuffers,
Viewports, Scissors, DepthBias, BlendConstants, DepthBounds,
StencilProperties, LineWidth, CullMode, DepthBoundsEnable, DepthTestEnable,
DepthWriteEnable, DepthCompareOp, FrontFace, StencilOp, StencilTestEnable,
VertexBuffers, VertexInput,
};
Flags flags{};
for (const int flag : INVALIDATION_FLAGS) {
@ -40,6 +41,12 @@ Flags MakeInvalidationFlags() {
for (int index = VertexBuffer0; index <= VertexBuffer31; ++index) {
flags[index] = true;
}
for (int index = VertexAttribute0; index <= VertexAttribute31; ++index) {
flags[index] = true;
}
for (int index = VertexBinding0; index <= VertexBinding31; ++index) {
flags[index] = true;
}
return flags;
}
@ -79,6 +86,11 @@ void SetupDirtyStencilProperties(Tables& tables) {
table[OFF(stencil_back_func_mask)] = StencilProperties;
}
void SetupDirtyLineWidth(Tables& tables) {
tables[0][OFF(line_width_smooth)] = LineWidth;
tables[0][OFF(line_width_aliased)] = LineWidth;
}
void SetupDirtyCullMode(Tables& tables) {
auto& table = tables[0];
table[OFF(cull_face)] = CullMode;
@ -134,19 +146,6 @@ void SetupDirtyBlending(Tables& tables) {
FillBlock(tables[0], OFF(independent_blend), NUM(independent_blend), Blending);
}
void SetupDirtyInstanceDivisors(Tables& tables) {
static constexpr size_t divisor_offset = 3;
for (size_t index = 0; index < Regs::NumVertexArrays; ++index) {
tables[0][OFF(instanced_arrays) + index] = InstanceDivisors;
tables[0][OFF(vertex_array) + index * NUM(vertex_array[0]) + divisor_offset] =
InstanceDivisors;
}
}
void SetupDirtyVertexAttributes(Tables& tables) {
FillBlock(tables[0], OFF(vertex_attrib_format), NUM(vertex_attrib_format), VertexAttributes);
}
void SetupDirtyViewportSwizzles(Tables& tables) {
static constexpr size_t swizzle_offset = 6;
for (size_t index = 0; index < Regs::NumViewports; ++index) {
@ -154,11 +153,31 @@ void SetupDirtyViewportSwizzles(Tables& tables) {
ViewportSwizzles;
}
}
void SetupDirtyVertexAttributes(Tables& tables) {
for (size_t i = 0; i < Regs::NumVertexAttributes; ++i) {
const size_t offset = OFF(vertex_attrib_format) + i * NUM(vertex_attrib_format[0]);
FillBlock(tables[0], offset, NUM(vertex_attrib_format[0]), VertexAttribute0 + i);
}
FillBlock(tables[1], OFF(vertex_attrib_format), Regs::NumVertexAttributes, VertexInput);
}
void SetupDirtyVertexBindings(Tables& tables) {
// Do NOT include stride here, it's implicit in VertexBuffer
static constexpr size_t divisor_offset = 3;
for (size_t i = 0; i < Regs::NumVertexArrays; ++i) {
const u8 flag = static_cast<u8>(VertexBinding0 + i);
tables[0][OFF(instanced_arrays) + i] = VertexInput;
tables[1][OFF(instanced_arrays) + i] = flag;
tables[0][OFF(vertex_array) + i * NUM(vertex_array[0]) + divisor_offset] = VertexInput;
tables[1][OFF(vertex_array) + i * NUM(vertex_array[0]) + divisor_offset] = flag;
}
}
} // Anonymous namespace
StateTracker::StateTracker(Tegra::GPU& gpu)
: flags{gpu.Maxwell3D().dirty.flags}, invalidation_flags{MakeInvalidationFlags()} {
auto& tables = gpu.Maxwell3D().dirty.tables;
auto& tables{gpu.Maxwell3D().dirty.tables};
SetupDirtyFlags(tables);
SetupDirtyViewports(tables);
SetupDirtyScissors(tables);
@ -166,6 +185,7 @@ StateTracker::StateTracker(Tegra::GPU& gpu)
SetupDirtyBlendConstants(tables);
SetupDirtyDepthBounds(tables);
SetupDirtyStencilProperties(tables);
SetupDirtyLineWidth(tables);
SetupDirtyCullMode(tables);
SetupDirtyDepthBoundsEnable(tables);
SetupDirtyDepthTestEnable(tables);
@ -175,9 +195,9 @@ StateTracker::StateTracker(Tegra::GPU& gpu)
SetupDirtyStencilOp(tables);
SetupDirtyStencilTestEnable(tables);
SetupDirtyBlending(tables);
SetupDirtyInstanceDivisors(tables);
SetupDirtyVertexAttributes(tables);
SetupDirtyViewportSwizzles(tables);
SetupDirtyVertexAttributes(tables);
SetupDirtyVertexBindings(tables);
}
} // namespace Vulkan

15
src/video_core/renderer_vulkan/vk_state_tracker.h
View File

@ -19,12 +19,19 @@ namespace Dirty {
enum : u8 {
First = VideoCommon::Dirty::LastCommonEntry,
VertexInput,
VertexAttribute0,
VertexAttribute31 = VertexAttribute0 + 31,
VertexBinding0,
VertexBinding31 = VertexBinding0 + 31,
Viewports,
Scissors,
DepthBias,
BlendConstants,
DepthBounds,
StencilProperties,
LineWidth,
CullMode,
DepthBoundsEnable,
@ -36,11 +43,9 @@ enum : u8 {
StencilTestEnable,
Blending,
InstanceDivisors,
VertexAttributes,
ViewportSwizzles,
Last
Last,
};
static_assert(Last <= std::numeric_limits<u8>::max());
@ -89,6 +94,10 @@ public:
return Exchange(Dirty::StencilProperties, false);
}
bool TouchLineWidth() const {
return Exchange(Dirty::LineWidth, false);
}
bool TouchCullMode() {
return Exchange(Dirty::CullMode, false);
}

59
src/video_core/renderer_vulkan/vk_swapchain.cpp
View File

@ -65,6 +65,9 @@ VKSwapchain::VKSwapchain(VkSurfaceKHR surface_, const Device& device_, VKSchedul
VKSwapchain::~VKSwapchain() = default;
void VKSwapchain::Create(u32 width, u32 height, bool srgb) {
is_outdated = false;
is_suboptimal = false;
const auto physical_device = device.GetPhysical();
const auto capabilities{physical_device.GetSurfaceCapabilitiesKHR(surface)};
if (capabilities.maxImageExtent.width == 0 || capabilities.maxImageExtent.height == 0) {
@ -82,21 +85,31 @@ void VKSwapchain::Create(u32 width, u32 height, bool srgb) {
resource_ticks.resize(image_count);
}
bool VKSwapchain::AcquireNextImage() {
const VkResult result =
device.GetLogical().AcquireNextImageKHR(*swapchain, std::numeric_limits<u64>::max(),
*present_semaphores[frame_index], {}, &image_index);
void VKSwapchain::AcquireNextImage() {
const VkResult result = device.GetLogical().AcquireNextImageKHR(
*swapchain, std::numeric_limits<u64>::max(), *present_semaphores[frame_index],
VK_NULL_HANDLE, &image_index);
switch (result) {
case VK_SUCCESS:
break;
case VK_SUBOPTIMAL_KHR:
is_suboptimal = true;
break;
case VK_ERROR_OUT_OF_DATE_KHR:
is_outdated = true;
break;
default:
LOG_ERROR(Render_Vulkan, "vkAcquireNextImageKHR returned {}", vk::ToString(result));
break;
}
scheduler.Wait(resource_ticks[image_index]);
return result == VK_SUCCESS || result == VK_SUBOPTIMAL_KHR;
resource_ticks[image_index] = scheduler.CurrentTick();
}
bool VKSwapchain::Present(VkSemaphore render_semaphore) {
void VKSwapchain::Present(VkSemaphore render_semaphore) {
const VkSemaphore present_semaphore{*present_semaphores[frame_index]};
const std::array<VkSemaphore, 2> semaphores{present_semaphore, render_semaphore};
const auto present_queue{device.GetPresentQueue()};
bool recreated = false;
const VkPresentInfoKHR present_info{
.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR,
.pNext = nullptr,
@ -107,7 +120,6 @@ bool VKSwapchain::Present(VkSemaphore render_semaphore) {
.pImageIndices = &image_index,
.pResults = nullptr,
};
switch (const VkResult result = present_queue.Present(present_info)) {
case VK_SUCCESS:
break;
@ -115,24 +127,16 @@ bool VKSwapchain::Present(VkSemaphore render_semaphore) {
LOG_DEBUG(Render_Vulkan, "Suboptimal swapchain");
break;
case VK_ERROR_OUT_OF_DATE_KHR:
if (current_width > 0 && current_height > 0) {
Create(current_width, current_height, current_srgb);
recreated = true;
}
is_outdated = true;
break;
default:
LOG_CRITICAL(Render_Vulkan, "Failed to present with error {}", vk::ToString(result));
break;
}
resource_ticks[image_index] = scheduler.CurrentTick();
frame_index = (frame_index + 1) % static_cast<u32>(image_count);
return recreated;
}
bool VKSwapchain::HasFramebufferChanged(const Layout::FramebufferLayout& framebuffer) const {
// TODO(Rodrigo): Handle framebuffer pixel format changes
return framebuffer.width != current_width || framebuffer.height != current_height;
++frame_index;
if (frame_index >= image_count) {
frame_index = 0;
}
}
void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u32 width,
@ -148,7 +152,6 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities,
if (capabilities.maxImageCount > 0 && requested_image_count > capabilities.maxImageCount) {
requested_image_count = capabilities.maxImageCount;
}
VkSwapchainCreateInfoKHR swapchain_ci{
.sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR,
.pNext = nullptr,
@ -169,7 +172,6 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities,
.clipped = VK_FALSE,
.oldSwapchain = nullptr,
};
const u32 graphics_family{device.GetGraphicsFamily()};
const u32 present_family{device.GetPresentFamily()};
const std::array<u32, 2> queue_indices{graphics_family, present_family};
@ -178,7 +180,6 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities,
swapchain_ci.queueFamilyIndexCount = static_cast<u32>(queue_indices.size());
swapchain_ci.pQueueFamilyIndices = queue_indices.data();
}
// Request the size again to reduce the possibility of a TOCTOU race condition.
const auto updated_capabilities = physical_device.GetSurfaceCapabilitiesKHR(surface);
swapchain_ci.imageExtent = ChooseSwapExtent(updated_capabilities, width, height);
@ -186,8 +187,6 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities,
swapchain = device.GetLogical().CreateSwapchainKHR(swapchain_ci);
extent = swapchain_ci.imageExtent;
current_width = extent.width;
current_height = extent.height;
current_srgb = srgb;
images = swapchain.GetImages();
@ -197,8 +196,8 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities,
void VKSwapchain::CreateSemaphores() {
present_semaphores.resize(image_count);
std::generate(present_semaphores.begin(), present_semaphores.end(),
[this] { return device.GetLogical().CreateSemaphore(); });
std::ranges::generate(present_semaphores,
[this] { return device.GetLogical().CreateSemaphore(); });
}
void VKSwapchain::CreateImageViews() {

31
src/video_core/renderer_vulkan/vk_swapchain.h
View File

@ -28,14 +28,25 @@ public:
void Create(u32 width, u32 height, bool srgb);
/// Acquires the next image in the swapchain, waits as needed.
bool AcquireNextImage();
void AcquireNextImage();
/// Presents the rendered image to the swapchain. Returns true when the swapchains had to be
/// recreated. Takes responsability for the ownership of fence.
bool Present(VkSemaphore render_semaphore);
/// Presents the rendered image to the swapchain.
void Present(VkSemaphore render_semaphore);
/// Returns true when the framebuffer layout has changed.
bool HasFramebufferChanged(const Layout::FramebufferLayout& framebuffer) const;
/// Returns true when the color space has changed.
bool HasColorSpaceChanged(bool is_srgb) const {
return current_srgb != is_srgb;
}
/// Returns true when the swapchain is outdated.
bool IsOutDated() const {
return is_outdated;
}
/// Returns true when the swapchain is suboptimal.
bool IsSubOptimal() const {
return is_suboptimal;
}
VkExtent2D GetSize() const {
return extent;
@ -61,10 +72,6 @@ public:
return image_format;
}
bool GetSrgbState() const {
return current_srgb;
}
private:
void CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, u32 height,
bool srgb);
@ -92,9 +99,9 @@ private:
VkFormat image_format{};
VkExtent2D extent{};
u32 current_width{};
u32 current_height{};
bool current_srgb{};
bool is_outdated{};
bool is_suboptimal{};
};
} // namespace Vulkan

243
src/video_core/renderer_vulkan/vk_texture_cache.cpp
View File

@ -15,6 +15,7 @@
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_render_pass_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
@ -34,19 +35,6 @@ using VideoCommon::SubresourceRange;
using VideoCore::Surface::IsPixelFormatASTC;
namespace {
constexpr std::array ATTACHMENT_REFERENCES{
VkAttachmentReference{0, VK_IMAGE_LAYOUT_GENERAL},
VkAttachmentReference{1, VK_IMAGE_LAYOUT_GENERAL},
VkAttachmentReference{2, VK_IMAGE_LAYOUT_GENERAL},
VkAttachmentReference{3, VK_IMAGE_LAYOUT_GENERAL},
VkAttachmentReference{4, VK_IMAGE_LAYOUT_GENERAL},
VkAttachmentReference{5, VK_IMAGE_LAYOUT_GENERAL},
VkAttachmentReference{6, VK_IMAGE_LAYOUT_GENERAL},
VkAttachmentReference{7, VK_IMAGE_LAYOUT_GENERAL},
VkAttachmentReference{8, VK_IMAGE_LAYOUT_GENERAL},
};
constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
if (color == std::array<float, 4>{0, 0, 0, 0}) {
return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
@ -174,25 +162,6 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
return device.GetLogical().CreateImage(MakeImageCreateInfo(device, info));
}
[[nodiscard]] vk::Buffer MakeBuffer(const Device& device, const ImageInfo& info) {
if (info.type != ImageType::Buffer) {
return vk::Buffer{};
}
const size_t bytes_per_block = VideoCore::Surface::BytesPerBlock(info.format);
return device.GetLogical().CreateBuffer(VkBufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = info.size.width * bytes_per_block,
.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT |
VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
});
}
[[nodiscard]] VkImageAspectFlags ImageAspectMask(PixelFormat format) {
switch (VideoCore::Surface::GetFormatType(format)) {
case VideoCore::Surface::SurfaceType::ColorTexture:
@ -226,23 +195,6 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
}
}
[[nodiscard]] VkAttachmentDescription AttachmentDescription(const Device& device,
const ImageView* image_view) {
using MaxwellToVK::SurfaceFormat;
const PixelFormat pixel_format = image_view->format;
return VkAttachmentDescription{
.flags = VK_ATTACHMENT_DESCRIPTION_MAY_ALIAS_BIT,
.format = SurfaceFormat(device, FormatType::Optimal, true, pixel_format).format,
.samples = image_view->Samples(),
.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
.storeOp = VK_ATTACHMENT_STORE_OP_STORE,
.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE,
.initialLayout = VK_IMAGE_LAYOUT_GENERAL,
.finalLayout = VK_IMAGE_LAYOUT_GENERAL,
};
}
[[nodiscard]] VkComponentSwizzle ComponentSwizzle(SwizzleSource swizzle) {
switch (swizzle) {
case SwizzleSource::Zero:
@ -263,6 +215,30 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
return VK_COMPONENT_SWIZZLE_ZERO;
}
[[nodiscard]] VkImageViewType ImageViewType(Shader::TextureType type) {
switch (type) {
case Shader::TextureType::Color1D:
return VK_IMAGE_VIEW_TYPE_1D;
case Shader::TextureType::Color2D:
return VK_IMAGE_VIEW_TYPE_2D;
case Shader::TextureType::ColorCube:
return VK_IMAGE_VIEW_TYPE_CUBE;
case Shader::TextureType::Color3D:
return VK_IMAGE_VIEW_TYPE_3D;
case Shader::TextureType::ColorArray1D:
return VK_IMAGE_VIEW_TYPE_1D_ARRAY;
case Shader::TextureType::ColorArray2D:
return VK_IMAGE_VIEW_TYPE_2D_ARRAY;
case Shader::TextureType::ColorArrayCube:
return VK_IMAGE_VIEW_TYPE_CUBE_ARRAY;
case Shader::TextureType::Buffer:
UNREACHABLE_MSG("Texture buffers can't be image views");
return VK_IMAGE_VIEW_TYPE_1D;
}
UNREACHABLE_MSG("Invalid image view type={}", type);
return VK_IMAGE_VIEW_TYPE_2D;
}
[[nodiscard]] VkImageViewType ImageViewType(VideoCommon::ImageViewType type) {
switch (type) {
case VideoCommon::ImageViewType::e1D:
@ -280,7 +256,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
case VideoCommon::ImageViewType::CubeArray:
return VK_IMAGE_VIEW_TYPE_CUBE_ARRAY;
case VideoCommon::ImageViewType::Rect:
LOG_WARNING(Render_Vulkan, "Unnormalized image view type not supported");
UNIMPLEMENTED_MSG("Rect image view");
return VK_IMAGE_VIEW_TYPE_2D;
case VideoCommon::ImageViewType::Buffer:
UNREACHABLE_MSG("Texture buffers can't be image views");
@ -327,7 +303,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
};
}
[[nodiscard]] std::vector<VkBufferCopy> TransformBufferCopies(
[[maybe_unused]] [[nodiscard]] std::vector<VkBufferCopy> TransformBufferCopies(
std::span<const VideoCommon::BufferCopy> copies, size_t buffer_offset) {
std::vector<VkBufferCopy> result(copies.size());
std::ranges::transform(
@ -587,6 +563,28 @@ struct RangedBarrierRange {
}
};
[[nodiscard]] VkFormat Format(Shader::ImageFormat format) {
switch (format) {
case Shader::ImageFormat::Typeless:
break;
case Shader::ImageFormat::R8_SINT:
return VK_FORMAT_R8_SINT;
case Shader::ImageFormat::R8_UINT:
return VK_FORMAT_R8_UINT;
case Shader::ImageFormat::R16_UINT:
return VK_FORMAT_R16_UINT;
case Shader::ImageFormat::R16_SINT:
return VK_FORMAT_R16_SINT;
case Shader::ImageFormat::R32_UINT:
return VK_FORMAT_R32_UINT;
case Shader::ImageFormat::R32G32_UINT:
return VK_FORMAT_R32G32_UINT;
case Shader::ImageFormat::R32G32B32A32_UINT:
return VK_FORMAT_R32G32B32A32_UINT;
}
UNREACHABLE_MSG("Invalid image format={}", format);
return VK_FORMAT_R32_UINT;
}
} // Anonymous namespace
void TextureCacheRuntime::Finish() {
@ -625,7 +623,7 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
return;
}
}
ASSERT(src.ImageFormat() == dst.ImageFormat());
ASSERT(src.format == dst.format);
ASSERT(!(is_dst_msaa && !is_src_msaa));
ASSERT(operation == Fermi2D::Operation::SrcCopy);
@ -842,13 +840,9 @@ u64 TextureCacheRuntime::GetDeviceLocalMemory() const {
Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_addr_,
VAddr cpu_addr_)
: VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime.scheduler},
image(MakeImage(runtime.device, info)), buffer(MakeBuffer(runtime.device, info)),
image(MakeImage(runtime.device, info)),
commit(runtime.memory_allocator.Commit(image, MemoryUsage::DeviceLocal)),
aspect_mask(ImageAspectMask(info.format)) {
if (image) {
commit = runtime.memory_allocator.Commit(image, MemoryUsage::DeviceLocal);
} else {
commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
}
if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) {
if (Settings::values.accelerate_astc.GetValue()) {
flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
@ -857,11 +851,7 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_
}
}
if (runtime.device.HasDebuggingToolAttached()) {
if (image) {
image.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
} else {
buffer.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
}
image.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
}
static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
@ -913,19 +903,6 @@ void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImag
});
}
void Image::UploadMemory(const StagingBufferRef& map,
std::span<const VideoCommon::BufferCopy> copies) {
// TODO: Move this to another API
scheduler->RequestOutsideRenderPassOperationContext();
std::vector vk_copies = TransformBufferCopies(copies, map.offset);
const VkBuffer src_buffer = map.buffer;
const VkBuffer dst_buffer = *buffer;
scheduler->Record([src_buffer, dst_buffer, vk_copies](vk::CommandBuffer cmdbuf) {
// TODO: Barriers
cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies);
});
}
void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask);
scheduler->RequestOutsideRenderPassOperationContext();
@ -984,8 +961,9 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferIm
ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewInfo& info,
ImageId image_id_, Image& image)
: VideoCommon::ImageViewBase{info, image.info, image_id_}, device{&runtime.device},
image_handle{image.Handle()}, image_format{image.info.format}, samples{ConvertSampleCount(
image.info.num_samples)} {
image_handle{image.Handle()}, samples{ConvertSampleCount(image.info.num_samples)} {
using Shader::TextureType;
const VkImageAspectFlags aspect_mask = ImageViewAspectMask(info);
std::array<SwizzleSource, 4> swizzle{
SwizzleSource::R,
@ -1023,57 +1001,54 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
},
.subresourceRange = MakeSubresourceRange(aspect_mask, info.range),
};
const auto create = [&](VideoCommon::ImageViewType view_type, std::optional<u32> num_layers) {
const auto create = [&](TextureType tex_type, std::optional<u32> num_layers) {
VkImageViewCreateInfo ci{create_info};
ci.viewType = ImageViewType(view_type);
ci.viewType = ImageViewType(tex_type);
if (num_layers) {
ci.subresourceRange.layerCount = *num_layers;
}
vk::ImageView handle = device->GetLogical().CreateImageView(ci);
if (device->HasDebuggingToolAttached()) {
handle.SetObjectNameEXT(VideoCommon::Name(*this, view_type).c_str());
handle.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
}
image_views[static_cast<size_t>(view_type)] = std::move(handle);
image_views[static_cast<size_t>(tex_type)] = std::move(handle);
};
switch (info.type) {
case VideoCommon::ImageViewType::e1D:
case VideoCommon::ImageViewType::e1DArray:
create(VideoCommon::ImageViewType::e1D, 1);
create(VideoCommon::ImageViewType::e1DArray, std::nullopt);
render_target = Handle(VideoCommon::ImageViewType::e1DArray);
create(TextureType::Color1D, 1);
create(TextureType::ColorArray1D, std::nullopt);
render_target = Handle(TextureType::ColorArray1D);
break;
case VideoCommon::ImageViewType::e2D:
case VideoCommon::ImageViewType::e2DArray:
create(VideoCommon::ImageViewType::e2D, 1);
create(VideoCommon::ImageViewType::e2DArray, std::nullopt);
render_target = Handle(VideoCommon::ImageViewType::e2DArray);
create(TextureType::Color2D, 1);
create(TextureType::ColorArray2D, std::nullopt);
render_target = Handle(Shader::TextureType::ColorArray2D);
break;
case VideoCommon::ImageViewType::e3D:
create(VideoCommon::ImageViewType::e3D, std::nullopt);
render_target = Handle(VideoCommon::ImageViewType::e3D);
create(TextureType::Color3D, std::nullopt);
render_target = Handle(Shader::TextureType::Color3D);
break;
case VideoCommon::ImageViewType::Cube:
case VideoCommon::ImageViewType::CubeArray:
create(VideoCommon::ImageViewType::Cube, 6);
create(VideoCommon::ImageViewType::CubeArray, std::nullopt);
create(TextureType::ColorCube, 6);
create(TextureType::ColorArrayCube, std::nullopt);
break;
case VideoCommon::ImageViewType::Rect:
UNIMPLEMENTED();
break;
case VideoCommon::ImageViewType::Buffer:
buffer_view = device->GetLogical().CreateBufferView(VkBufferViewCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.buffer = image.Buffer(),
.format = format_info.format,
.offset = 0, // TODO: Redesign buffer cache to support this
.range = image.guest_size_bytes,
});
UNREACHABLE();
break;
}
}
ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info,
const VideoCommon::ImageViewInfo& view_info, GPUVAddr gpu_addr_)
: VideoCommon::ImageViewBase{info, view_info}, gpu_addr{gpu_addr_},
buffer_size{VideoCommon::CalculateGuestSizeInBytes(info)} {}
ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams& params)
: VideoCommon::ImageViewBase{params} {}
@ -1081,7 +1056,8 @@ VkImageView ImageView::DepthView() {
if (depth_view) {
return *depth_view;
}
depth_view = MakeDepthStencilView(VK_IMAGE_ASPECT_DEPTH_BIT);
const auto& info = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format);
depth_view = MakeView(info.format, VK_IMAGE_ASPECT_DEPTH_BIT);
return *depth_view;
}
@ -1089,18 +1065,38 @@ VkImageView ImageView::StencilView() {
if (stencil_view) {
return *stencil_view;
}
stencil_view = MakeDepthStencilView(VK_IMAGE_ASPECT_STENCIL_BIT);
const auto& info = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format);
stencil_view = MakeView(info.format, VK_IMAGE_ASPECT_STENCIL_BIT);
return *stencil_view;
}
vk::ImageView ImageView::MakeDepthStencilView(VkImageAspectFlags aspect_mask) {
VkImageView ImageView::StorageView(Shader::TextureType texture_type,
Shader::ImageFormat image_format) {
if (image_format == Shader::ImageFormat::Typeless) {
return Handle(texture_type);
}
const bool is_signed{image_format == Shader::ImageFormat::R8_SINT ||
image_format == Shader::ImageFormat::R16_SINT};
if (!storage_views) {
storage_views = std::make_unique<StorageViews>();
}
auto& views{is_signed ? storage_views->signeds : storage_views->unsigneds};
auto& view{views[static_cast<size_t>(texture_type)]};
if (view) {
return *view;
}
view = MakeView(Format(image_format), VK_IMAGE_ASPECT_COLOR_BIT);
return *view;
}
vk::ImageView ImageView::MakeView(VkFormat vk_format, VkImageAspectFlags aspect_mask) {
return device->GetLogical().CreateImageView({
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.image = image_handle,
.viewType = ImageViewType(type),
.format = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format).format,
.format = vk_format,
.components{
.r = VK_COMPONENT_SWIZZLE_IDENTITY,
.g = VK_COMPONENT_SWIZZLE_IDENTITY,
@ -1164,7 +1160,6 @@ Sampler::Sampler(TextureCacheRuntime& runtime, const Tegra::Texture::TSCEntry& t
Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers,
ImageView* depth_buffer, const VideoCommon::RenderTargets& key) {
std::vector<VkAttachmentDescription> descriptions;
std::vector<VkImageView> attachments;
RenderPassKey renderpass_key{};
s32 num_layers = 1;
@ -1175,7 +1170,6 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM
renderpass_key.color_formats[index] = PixelFormat::Invalid;
continue;
}
descriptions.push_back(AttachmentDescription(runtime.device, color_buffer));
attachments.push_back(color_buffer->RenderTarget());
renderpass_key.color_formats[index] = color_buffer->format;
num_layers = std::max(num_layers, color_buffer->range.extent.layers);
@ -1185,10 +1179,7 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM
++num_images;
}
const size_t num_colors = attachments.size();
const VkAttachmentReference* depth_attachment =
depth_buffer ? &ATTACHMENT_REFERENCES[num_colors] : nullptr;
if (depth_buffer) {
descriptions.push_back(AttachmentDescription(runtime.device, depth_buffer));
attachments.push_back(depth_buffer->RenderTarget());
renderpass_key.depth_format = depth_buffer->format;
num_layers = std::max(num_layers, depth_buffer->range.extent.layers);
@ -1201,40 +1192,14 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM
}
renderpass_key.samples = samples;
const auto& device = runtime.device.GetLogical();
const auto [cache_pair, is_new] = runtime.renderpass_cache.try_emplace(renderpass_key);
if (is_new) {
const VkSubpassDescription subpass{
.flags = 0,
.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
.inputAttachmentCount = 0,
.pInputAttachments = nullptr,
.colorAttachmentCount = static_cast<u32>(num_colors),
.pColorAttachments = num_colors != 0 ? ATTACHMENT_REFERENCES.data() : nullptr,
.pResolveAttachments = nullptr,
.pDepthStencilAttachment = depth_attachment,
.preserveAttachmentCount = 0,
.pPreserveAttachments = nullptr,
};
cache_pair->second = device.CreateRenderPass(VkRenderPassCreateInfo{
.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.attachmentCount = static_cast<u32>(descriptions.size()),
.pAttachments = descriptions.data(),
.subpassCount = 1,
.pSubpasses = &subpass,
.dependencyCount = 0,
.pDependencies = nullptr,
});
}
renderpass = *cache_pair->second;
renderpass = runtime.render_pass_cache.Get(renderpass_key);
render_area = VkExtent2D{
.width = key.size.width,
.height = key.size.height,
};
num_color_buffers = static_cast<u32>(num_colors);
framebuffer = device.CreateFramebuffer(VkFramebufferCreateInfo{
framebuffer = runtime.device.GetLogical().CreateFramebuffer({
.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,

80
src/video_core/renderer_vulkan/vk_texture_cache.h
View File

@ -7,6 +7,7 @@
#include <compare>
#include <span>
#include "shader_recompiler/shader_info.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/texture_cache/texture_cache.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
@ -26,35 +27,10 @@ class Device;
class Image;
class ImageView;
class Framebuffer;
class RenderPassCache;
class StagingBufferPool;
class VKScheduler;
struct RenderPassKey {
constexpr auto operator<=>(const RenderPassKey&) const noexcept = default;
std::array<PixelFormat, NUM_RT> color_formats;
PixelFormat depth_format;
VkSampleCountFlagBits samples;
};
} // namespace Vulkan
namespace std {
template <>
struct hash<Vulkan::RenderPassKey> {
[[nodiscard]] constexpr size_t operator()(const Vulkan::RenderPassKey& key) const noexcept {
size_t value = static_cast<size_t>(key.depth_format) << 48;
value ^= static_cast<size_t>(key.samples) << 52;
for (size_t i = 0; i < key.color_formats.size(); ++i) {
value ^= static_cast<size_t>(key.color_formats[i]) << (i * 6);
}
return value;
}
};
} // namespace std
namespace Vulkan {
struct TextureCacheRuntime {
const Device& device;
VKScheduler& scheduler;
@ -62,13 +38,13 @@ struct TextureCacheRuntime {
StagingBufferPool& staging_buffer_pool;
BlitImageHelper& blit_image_helper;
ASTCDecoderPass& astc_decoder_pass;
std::unordered_map<RenderPassKey, vk::RenderPass> renderpass_cache{};
RenderPassCache& render_pass_cache;
void Finish();
[[nodiscard]] StagingBufferRef UploadStagingBuffer(size_t size);
StagingBufferRef UploadStagingBuffer(size_t size);
[[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size);
StagingBufferRef DownloadStagingBuffer(size_t size);
void BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src,
const Region2D& dst_region, const Region2D& src_region,
@ -79,7 +55,7 @@ struct TextureCacheRuntime {
void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view);
[[nodiscard]] bool CanAccelerateImageUpload(Image&) const noexcept {
bool CanAccelerateImageUpload(Image&) const noexcept {
return false;
}
@ -117,8 +93,6 @@ public:
void UploadMemory(const StagingBufferRef& map,
std::span<const VideoCommon::BufferImageCopy> copies);
void UploadMemory(const StagingBufferRef& map, std::span<const VideoCommon::BufferCopy> copies);
void DownloadMemory(const StagingBufferRef& map,
std::span<const VideoCommon::BufferImageCopy> copies);
@ -126,10 +100,6 @@ public:
return *image;
}
[[nodiscard]] VkBuffer Buffer() const noexcept {
return *buffer;
}
[[nodiscard]] VkImageAspectFlags AspectMask() const noexcept {
return aspect_mask;
}
@ -146,7 +116,6 @@ public:
private:
VKScheduler* scheduler;
vk::Image image;
vk::Buffer buffer;
MemoryCommit commit;
vk::ImageView image_view;
std::vector<vk::ImageView> storage_image_views;
@ -157,18 +126,19 @@ private:
class ImageView : public VideoCommon::ImageViewBase {
public:
explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageViewInfo&, ImageId, Image&);
explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo&,
const VideoCommon::ImageViewInfo&, GPUVAddr);
explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams&);
[[nodiscard]] VkImageView DepthView();
[[nodiscard]] VkImageView StencilView();
[[nodiscard]] VkImageView Handle(VideoCommon::ImageViewType query_type) const noexcept {
return *image_views[static_cast<size_t>(query_type)];
}
[[nodiscard]] VkImageView StorageView(Shader::TextureType texture_type,
Shader::ImageFormat image_format);
[[nodiscard]] VkBufferView BufferView() const noexcept {
return *buffer_view;
[[nodiscard]] VkImageView Handle(Shader::TextureType texture_type) const noexcept {
return *image_views[static_cast<size_t>(texture_type)];
}
[[nodiscard]] VkImage ImageHandle() const noexcept {
@ -179,26 +149,36 @@ public:
return render_target;
}
[[nodiscard]] PixelFormat ImageFormat() const noexcept {
return image_format;
}
[[nodiscard]] VkSampleCountFlagBits Samples() const noexcept {
return samples;
}
[[nodiscard]] GPUVAddr GpuAddr() const noexcept {
return gpu_addr;
}
[[nodiscard]] u32 BufferSize() const noexcept {
return buffer_size;
}
private:
[[nodiscard]] vk::ImageView MakeDepthStencilView(VkImageAspectFlags aspect_mask);
struct StorageViews {
std::array<vk::ImageView, Shader::NUM_TEXTURE_TYPES> signeds;
std::array<vk::ImageView, Shader::NUM_TEXTURE_TYPES> unsigneds;
};
[[nodiscard]] vk::ImageView MakeView(VkFormat vk_format, VkImageAspectFlags aspect_mask);
const Device* device = nullptr;
std::array<vk::ImageView, VideoCommon::NUM_IMAGE_VIEW_TYPES> image_views;
std::array<vk::ImageView, Shader::NUM_TEXTURE_TYPES> image_views;
std::unique_ptr<StorageViews> storage_views;
vk::ImageView depth_view;
vk::ImageView stencil_view;
vk::BufferView buffer_view;
VkImage image_handle = VK_NULL_HANDLE;
VkImageView render_target = VK_NULL_HANDLE;
PixelFormat image_format = PixelFormat::Invalid;
VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT;
GPUVAddr gpu_addr = 0;
u32 buffer_size = 0;
};
class ImageAlloc : public VideoCommon::ImageAllocBase {};

13
src/video_core/renderer_vulkan/vk_update_descriptor.cpp
View File

@ -15,7 +15,9 @@
namespace Vulkan {
VKUpdateDescriptorQueue::VKUpdateDescriptorQueue(const Device& device_, VKScheduler& scheduler_)
: device{device_}, scheduler{scheduler_} {}
: device{device_}, scheduler{scheduler_} {
payload_cursor = payload.data();
}
VKUpdateDescriptorQueue::~VKUpdateDescriptorQueue() = default;
@ -36,13 +38,4 @@ void VKUpdateDescriptorQueue::Acquire() {
upload_start = payload_cursor;
}
void VKUpdateDescriptorQueue::Send(VkDescriptorUpdateTemplateKHR update_template,
VkDescriptorSet set) {
const void* const data = upload_start;
const vk::Device* const logical = &device.GetLogical();
scheduler.Record([data, logical, set, update_template](vk::CommandBuffer) {
logical->UpdateDescriptorSet(set, update_template, data);
});
}
} // namespace Vulkan

4
src/video_core/renderer_vulkan/vk_update_descriptor.h
View File

@ -39,7 +39,9 @@ public:
void Acquire();
void Send(VkDescriptorUpdateTemplateKHR update_template, VkDescriptorSet set);
const DescriptorUpdateEntry* UpdateData() const noexcept {
return upload_start;
}
void AddSampledImage(VkImageView image_view, VkSampler sampler) {
*(payload_cursor++) = VkDescriptorImageInfo{

752
src/video_core/shader/ast.cpp
View File

@ -1,752 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <string>
#include <string_view>
#include <fmt/format.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/shader/ast.h"
#include "video_core/shader/expr.h"
namespace VideoCommon::Shader {
ASTZipper::ASTZipper() = default;
void ASTZipper::Init(const ASTNode new_first, const ASTNode parent) {
ASSERT(new_first->manager == nullptr);
first = new_first;
last = new_first;
ASTNode current = first;
while (current) {
current->manager = this;
current->parent = parent;
last = current;
current = current->next;
}
}
void ASTZipper::PushBack(const ASTNode new_node) {
ASSERT(new_node->manager == nullptr);
new_node->previous = last;
if (last) {
last->next = new_node;
}
new_node->next.reset();
last = new_node;
if (!first) {
first = new_node;
}
new_node->manager = this;
}
void ASTZipper::PushFront(const ASTNode new_node) {
ASSERT(new_node->manager == nullptr);
new_node->previous.reset();
new_node->next = first;
if (first) {
first->previous = new_node;
}
if (last == first) {
last = new_node;
}
first = new_node;
new_node->manager = this;
}
void ASTZipper::InsertAfter(const ASTNode new_node, const ASTNode at_node) {
ASSERT(new_node->manager == nullptr);
if (!at_node) {
PushFront(new_node);
return;
}
const ASTNode next = at_node->next;
if (next) {
next->previous = new_node;
}
new_node->previous = at_node;
if (at_node == last) {
last = new_node;
}
new_node->next = next;
at_node->next = new_node;
new_node->manager = this;
}
void ASTZipper::InsertBefore(const ASTNode new_node, const ASTNode at_node) {
ASSERT(new_node->manager == nullptr);
if (!at_node) {
PushBack(new_node);
return;
}
const ASTNode previous = at_node->previous;
if (previous) {
previous->next = new_node;
}
new_node->next = at_node;
if (at_node == first) {
first = new_node;
}
new_node->previous = previous;
at_node->previous = new_node;
new_node->manager = this;
}
void ASTZipper::DetachTail(ASTNode node) {
ASSERT(node->manager == this);
if (node == first) {
first.reset();
last.reset();
return;
}
last = node->previous;
last->next.reset();
node->previous.reset();
ASTNode current = std::move(node);
while (current) {
current->manager = nullptr;
current->parent.reset();
current = current->next;
}
}
void ASTZipper::DetachSegment(const ASTNode start, const ASTNode end) {
ASSERT(start->manager == this && end->manager == this);
if (start == end) {
DetachSingle(start);
return;
}
const ASTNode prev = start->previous;
const ASTNode post = end->next;
if (!prev) {
first = post;
} else {
prev->next = post;
}
if (!post) {
last = prev;
} else {
post->previous = prev;
}
start->previous.reset();
end->next.reset();
ASTNode current = start;
bool found = false;
while (current) {
current->manager = nullptr;
current->parent.reset();
found |= current == end;
current = current->next;
}
ASSERT(found);
}
void ASTZipper::DetachSingle(const ASTNode node) {
ASSERT(node->manager == this);
const ASTNode prev = node->previous;
const ASTNode post = node->next;
node->previous.reset();
node->next.reset();
if (!prev) {
first = post;
} else {
prev->next = post;
}
if (!post) {
last = prev;
} else {
post->previous = prev;
}
node->manager = nullptr;
node->parent.reset();
}
void ASTZipper::Remove(const ASTNode node) {
ASSERT(node->manager == this);
const ASTNode next = node->next;
const ASTNode previous = node->previous;
if (previous) {
previous->next = next;
}
if (next) {
next->previous = previous;
}
node->parent.reset();
node->manager = nullptr;
if (node == last) {
last = previous;
}
if (node == first) {
first = next;
}
}
class ExprPrinter final {
public:
void operator()(const ExprAnd& expr) {
inner += "( ";
std::visit(*this, *expr.operand1);
inner += " && ";
std::visit(*this, *expr.operand2);
inner += ')';
}
void operator()(const ExprOr& expr) {
inner += "( ";
std::visit(*this, *expr.operand1);
inner += " || ";
std::visit(*this, *expr.operand2);
inner += ')';
}
void operator()(const ExprNot& expr) {
inner += "!";
std::visit(*this, *expr.operand1);
}
void operator()(const ExprPredicate& expr) {
inner += fmt::format("P{}", expr.predicate);
}
void operator()(const ExprCondCode& expr) {
inner += fmt::format("CC{}", expr.cc);
}
void operator()(const ExprVar& expr) {
inner += fmt::format("V{}", expr.var_index);
}
void operator()(const ExprBoolean& expr) {
inner += expr.value ? "true" : "false";
}
void operator()(const ExprGprEqual& expr) {
inner += fmt::format("(gpr_{} == {})", expr.gpr, expr.value);
}
const std::string& GetResult() const {
return inner;
}
private:
std::string inner;
};
class ASTPrinter {
public:
void operator()(const ASTProgram& ast) {
scope++;
inner += "program {\n";
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
inner += "}\n";
scope--;
}
void operator()(const ASTIfThen& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner += fmt::format("{}if ({}) {{\n", Indent(), expr_parser.GetResult());
scope++;
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
scope--;
inner += fmt::format("{}}}\n", Indent());
}
void operator()(const ASTIfElse& ast) {
inner += Indent();
inner += "else {\n";
scope++;
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
scope--;
inner += Indent();
inner += "}\n";
}
void operator()(const ASTBlockEncoded& ast) {
inner += fmt::format("{}Block({}, {});\n", Indent(), ast.start, ast.end);
}
void operator()([[maybe_unused]] const ASTBlockDecoded& ast) {
inner += Indent();
inner += "Block;\n";
}
void operator()(const ASTVarSet& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner += fmt::format("{}V{} := {};\n", Indent(), ast.index, expr_parser.GetResult());
}
void operator()(const ASTLabel& ast) {
inner += fmt::format("Label_{}:\n", ast.index);
}
void operator()(const ASTGoto& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner +=
fmt::format("{}({}) -> goto Label_{};\n", Indent(), expr_parser.GetResult(), ast.label);
}
void operator()(const ASTDoWhile& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner += fmt::format("{}do {{\n", Indent());
scope++;
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
scope--;
inner += fmt::format("{}}} while ({});\n", Indent(), expr_parser.GetResult());
}
void operator()(const ASTReturn& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner += fmt::format("{}({}) -> {};\n", Indent(), expr_parser.GetResult(),
ast.kills ? "discard" : "exit");
}
void operator()(const ASTBreak& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner += fmt::format("{}({}) -> break;\n", Indent(), expr_parser.GetResult());
}
void Visit(const ASTNode& node) {
std::visit(*this, *node->GetInnerData());
}
const std::string& GetResult() const {
return inner;
}
private:
std::string_view Indent() {
if (space_segment_scope == scope) {
return space_segment;
}
// Ensure that we don't exceed our view.
ASSERT(scope * 2 < spaces.size());
space_segment = spaces.substr(0, scope * 2);
space_segment_scope = scope;
return space_segment;
}
std::string inner{};
std::string_view space_segment;
u32 scope{};
u32 space_segment_scope{};
static constexpr std::string_view spaces{" "};
};
std::string ASTManager::Print() const {
ASTPrinter printer{};
printer.Visit(main_node);
return printer.GetResult();
}
ASTManager::ASTManager(bool do_full_decompile, bool disable_else_derivation_)
: full_decompile{do_full_decompile}, disable_else_derivation{disable_else_derivation_} {}
ASTManager::~ASTManager() {
Clear();
}
void ASTManager::Init() {
main_node = ASTBase::Make<ASTProgram>(ASTNode{});
program = std::get_if<ASTProgram>(main_node->GetInnerData());
false_condition = MakeExpr<ExprBoolean>(false);
}
void ASTManager::DeclareLabel(u32 address) {
const auto pair = labels_map.emplace(address, labels_count);
if (pair.second) {
labels_count++;
labels.resize(labels_count);
}
}
void ASTManager::InsertLabel(u32 address) {
const u32 index = labels_map[address];
const ASTNode label = ASTBase::Make<ASTLabel>(main_node, index);
labels[index] = label;
program->nodes.PushBack(label);
}
void ASTManager::InsertGoto(Expr condition, u32 address) {
const u32 index = labels_map[address];
const ASTNode goto_node = ASTBase::Make<ASTGoto>(main_node, std::move(condition), index);
gotos.push_back(goto_node);
program->nodes.PushBack(goto_node);
}
void ASTManager::InsertBlock(u32 start_address, u32 end_address) {
ASTNode block = ASTBase::Make<ASTBlockEncoded>(main_node, start_address, end_address);
program->nodes.PushBack(std::move(block));
}
void ASTManager::InsertReturn(Expr condition, bool kills) {
ASTNode node = ASTBase::Make<ASTReturn>(main_node, std::move(condition), kills);
program->nodes.PushBack(std::move(node));
}
// The decompile algorithm is based on
// "Taming control flow: A structured approach to eliminating goto statements"
// by AM Erosa, LJ Hendren 1994. In general, the idea is to get gotos to be
// on the same structured level as the label which they jump to. This is done,
// through outward/inward movements and lifting. Once they are at the same
// level, you can enclose them in an "if" structure or a "do-while" structure.
void ASTManager::Decompile() {
auto it = gotos.begin();
while (it != gotos.end()) {
const ASTNode goto_node = *it;
const auto label_index = goto_node->GetGotoLabel();
if (!label_index) {
return;
}
const ASTNode label = labels[*label_index];
if (!full_decompile) {
// We only decompile backward jumps
if (!IsBackwardsJump(goto_node, label)) {
it++;
continue;
}
}
if (IndirectlyRelated(goto_node, label)) {
while (!DirectlyRelated(goto_node, label)) {
MoveOutward(goto_node);
}
}
if (DirectlyRelated(goto_node, label)) {
u32 goto_level = goto_node->GetLevel();
const u32 label_level = label->GetLevel();
while (label_level < goto_level) {
MoveOutward(goto_node);
goto_level--;
}
// TODO(Blinkhawk): Implement Lifting and Inward Movements
}
if (label->GetParent() == goto_node->GetParent()) {
bool is_loop = false;
ASTNode current = goto_node->GetPrevious();
while (current) {
if (current == label) {
is_loop = true;
break;
}
current = current->GetPrevious();
}
if (is_loop) {
EncloseDoWhile(goto_node, label);
} else {
EncloseIfThen(goto_node, label);
}
it = gotos.erase(it);
continue;
}
it++;
}
if (full_decompile) {
for (const ASTNode& label : labels) {
auto& manager = label->GetManager();
manager.Remove(label);
}
labels.clear();
} else {
auto label_it = labels.begin();
while (label_it != labels.end()) {
bool can_remove = true;
ASTNode label = *label_it;
for (const ASTNode& goto_node : gotos) {
const auto label_index = goto_node->GetGotoLabel();
if (!label_index) {
return;
}
ASTNode& glabel = labels[*label_index];
if (glabel == label) {
can_remove = false;
break;
}
}
if (can_remove) {
label->MarkLabelUnused();
}
}
}
}
bool ASTManager::IsBackwardsJump(ASTNode goto_node, ASTNode label_node) const {
u32 goto_level = goto_node->GetLevel();
u32 label_level = label_node->GetLevel();
while (goto_level > label_level) {
goto_level--;
goto_node = goto_node->GetParent();
}
while (label_level > goto_level) {
label_level--;
label_node = label_node->GetParent();
}
while (goto_node->GetParent() != label_node->GetParent()) {
goto_node = goto_node->GetParent();
label_node = label_node->GetParent();
}
ASTNode current = goto_node->GetPrevious();
while (current) {
if (current == label_node) {
return true;
}
current = current->GetPrevious();
}
return false;
}
bool ASTManager::IndirectlyRelated(const ASTNode& first, const ASTNode& second) const {
return !(first->GetParent() == second->GetParent() || DirectlyRelated(first, second));
}
bool ASTManager::DirectlyRelated(const ASTNode& first, const ASTNode& second) const {
if (first->GetParent() == second->GetParent()) {
return false;
}
const u32 first_level = first->GetLevel();
const u32 second_level = second->GetLevel();
u32 min_level;
u32 max_level;
ASTNode max;
ASTNode min;
if (first_level > second_level) {
min_level = second_level;
min = second;
max_level = first_level;
max = first;
} else {
min_level = first_level;
min = first;
max_level = second_level;
max = second;
}
while (max_level > min_level) {
max_level--;
max = max->GetParent();
}
return min->GetParent() == max->GetParent();
}
void ASTManager::ShowCurrentState(std::string_view state) const {
LOG_CRITICAL(HW_GPU, "\nState {}:\n\n{}\n", state, Print());
SanityCheck();
}
void ASTManager::SanityCheck() const {
for (const auto& label : labels) {
if (!label->GetParent()) {
LOG_CRITICAL(HW_GPU, "Sanity Check Failed");
}
}
}
void ASTManager::EncloseDoWhile(ASTNode goto_node, ASTNode label) {
ASTZipper& zipper = goto_node->GetManager();
const ASTNode loop_start = label->GetNext();
if (loop_start == goto_node) {
zipper.Remove(goto_node);
return;
}
const ASTNode parent = label->GetParent();
const Expr condition = goto_node->GetGotoCondition();
zipper.DetachSegment(loop_start, goto_node);
const ASTNode do_while_node = ASTBase::Make<ASTDoWhile>(parent, condition);
ASTZipper* sub_zipper = do_while_node->GetSubNodes();
sub_zipper->Init(loop_start, do_while_node);
zipper.InsertAfter(do_while_node, label);
sub_zipper->Remove(goto_node);
}
void ASTManager::EncloseIfThen(ASTNode goto_node, ASTNode label) {
ASTZipper& zipper = goto_node->GetManager();
const ASTNode if_end = label->GetPrevious();
if (if_end == goto_node) {
zipper.Remove(goto_node);
return;
}
const ASTNode prev = goto_node->GetPrevious();
const Expr condition = goto_node->GetGotoCondition();
bool do_else = false;
if (!disable_else_derivation && prev->IsIfThen()) {
const Expr if_condition = prev->GetIfCondition();
do_else = ExprAreEqual(if_condition, condition);
}
const ASTNode parent = label->GetParent();
zipper.DetachSegment(goto_node, if_end);
ASTNode if_node;
if (do_else) {
if_node = ASTBase::Make<ASTIfElse>(parent);
} else {
Expr neg_condition = MakeExprNot(condition);
if_node = ASTBase::Make<ASTIfThen>(parent, neg_condition);
}
ASTZipper* sub_zipper = if_node->GetSubNodes();
sub_zipper->Init(goto_node, if_node);
zipper.InsertAfter(if_node, prev);
sub_zipper->Remove(goto_node);
}
void ASTManager::MoveOutward(ASTNode goto_node) {
ASTZipper& zipper = goto_node->GetManager();
const ASTNode parent = goto_node->GetParent();
ASTZipper& zipper2 = parent->GetManager();
const ASTNode grandpa = parent->GetParent();
const bool is_loop = parent->IsLoop();
const bool is_else = parent->IsIfElse();
const bool is_if = parent->IsIfThen();
const ASTNode prev = goto_node->GetPrevious();
const ASTNode post = goto_node->GetNext();
const Expr condition = goto_node->GetGotoCondition();
zipper.DetachSingle(goto_node);
if (is_loop) {
const u32 var_index = NewVariable();
const Expr var_condition = MakeExpr<ExprVar>(var_index);
const ASTNode var_node = ASTBase::Make<ASTVarSet>(parent, var_index, condition);
const ASTNode var_node_init = ASTBase::Make<ASTVarSet>(parent, var_index, false_condition);
zipper2.InsertBefore(var_node_init, parent);
zipper.InsertAfter(var_node, prev);
goto_node->SetGotoCondition(var_condition);
const ASTNode break_node = ASTBase::Make<ASTBreak>(parent, var_condition);
zipper.InsertAfter(break_node, var_node);
} else if (is_if || is_else) {
const u32 var_index = NewVariable();
const Expr var_condition = MakeExpr<ExprVar>(var_index);
const ASTNode var_node = ASTBase::Make<ASTVarSet>(parent, var_index, condition);
const ASTNode var_node_init = ASTBase::Make<ASTVarSet>(parent, var_index, false_condition);
if (is_if) {
zipper2.InsertBefore(var_node_init, parent);
} else {
zipper2.InsertBefore(var_node_init, parent->GetPrevious());
}
zipper.InsertAfter(var_node, prev);
goto_node->SetGotoCondition(var_condition);
if (post) {
zipper.DetachTail(post);
const ASTNode if_node = ASTBase::Make<ASTIfThen>(parent, MakeExprNot(var_condition));
ASTZipper* sub_zipper = if_node->GetSubNodes();
sub_zipper->Init(post, if_node);
zipper.InsertAfter(if_node, var_node);
}
} else {
UNREACHABLE();
}
const ASTNode next = parent->GetNext();
if (is_if && next && next->IsIfElse()) {
zipper2.InsertAfter(goto_node, next);
goto_node->SetParent(grandpa);
return;
}
zipper2.InsertAfter(goto_node, parent);
goto_node->SetParent(grandpa);
}
class ASTClearer {
public:
ASTClearer() = default;
void operator()(const ASTProgram& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()(const ASTIfThen& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()(const ASTIfElse& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()([[maybe_unused]] const ASTBlockEncoded& ast) {}
void operator()(ASTBlockDecoded& ast) {
ast.nodes.clear();
}
void operator()([[maybe_unused]] const ASTVarSet& ast) {}
void operator()([[maybe_unused]] const ASTLabel& ast) {}
void operator()([[maybe_unused]] const ASTGoto& ast) {}
void operator()(const ASTDoWhile& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()([[maybe_unused]] const ASTReturn& ast) {}
void operator()([[maybe_unused]] const ASTBreak& ast) {}
void Visit(const ASTNode& node) {
std::visit(*this, *node->GetInnerData());
node->Clear();
}
};
void ASTManager::Clear() {
if (!main_node) {
return;
}
ASTClearer clearer{};
clearer.Visit(main_node);
main_node.reset();
program = nullptr;
labels_map.clear();
labels.clear();
gotos.clear();
}
} // namespace VideoCommon::Shader

398
src/video_core/shader/ast.h
View File

@ -1,398 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <functional>
#include <list>
#include <memory>
#include <optional>
#include <string>
#include <unordered_map>
#include <vector>
#include "video_core/shader/expr.h"
#include "video_core/shader/node.h"
namespace VideoCommon::Shader {
class ASTBase;
class ASTBlockDecoded;
class ASTBlockEncoded;
class ASTBreak;
class ASTDoWhile;
class ASTGoto;
class ASTIfElse;
class ASTIfThen;
class ASTLabel;
class ASTProgram;
class ASTReturn;
class ASTVarSet;
using ASTData = std::variant<ASTProgram, ASTIfThen, ASTIfElse, ASTBlockEncoded, ASTBlockDecoded,
ASTVarSet, ASTGoto, ASTLabel, ASTDoWhile, ASTReturn, ASTBreak>;
using ASTNode = std::shared_ptr<ASTBase>;
enum class ASTZipperType : u32 {
Program,
IfThen,
IfElse,
Loop,
};
class ASTZipper final {
public:
explicit ASTZipper();
void Init(ASTNode first, ASTNode parent);
ASTNode GetFirst() const {
return first;
}
ASTNode GetLast() const {
return last;
}
void PushBack(ASTNode new_node);
void PushFront(ASTNode new_node);
void InsertAfter(ASTNode new_node, ASTNode at_node);
void InsertBefore(ASTNode new_node, ASTNode at_node);
void DetachTail(ASTNode node);
void DetachSingle(ASTNode node);
void DetachSegment(ASTNode start, ASTNode end);
void Remove(ASTNode node);
ASTNode first;
ASTNode last;
};
class ASTProgram {
public:
ASTZipper nodes{};
};
class ASTIfThen {
public:
explicit ASTIfThen(Expr condition_) : condition{std::move(condition_)} {}
Expr condition;
ASTZipper nodes{};
};
class ASTIfElse {
public:
ASTZipper nodes{};
};
class ASTBlockEncoded {
public:
explicit ASTBlockEncoded(u32 start_, u32 _) : start{start_}, end{_} {}
u32 start;
u32 end;
};
class ASTBlockDecoded {
public:
explicit ASTBlockDecoded(NodeBlock&& new_nodes_) : nodes(std::move(new_nodes_)) {}
NodeBlock nodes;
};
class ASTVarSet {
public:
explicit ASTVarSet(u32 index_, Expr condition_)
: index{index_}, condition{std::move(condition_)} {}
u32 index;
Expr condition;
};
class ASTLabel {
public:
explicit ASTLabel(u32 index_) : index{index_} {}
u32 index;
bool unused{};
};
class ASTGoto {
public:
explicit ASTGoto(Expr condition_, u32 label_)
: condition{std::move(condition_)}, label{label_} {}
Expr condition;
u32 label;
};
class ASTDoWhile {
public:
explicit ASTDoWhile(Expr condition_) : condition{std::move(condition_)} {}
Expr condition;
ASTZipper nodes{};
};
class ASTReturn {
public:
explicit ASTReturn(Expr condition_, bool kills_)
: condition{std::move(condition_)}, kills{kills_} {}
Expr condition;
bool kills;
};
class ASTBreak {
public:
explicit ASTBreak(Expr condition_) : condition{std::move(condition_)} {}
Expr condition;
};
class ASTBase {
public:
explicit ASTBase(ASTNode parent_, ASTData data_)
: data{std::move(data_)}, parent{std::move(parent_)} {}
template <class U, class... Args>
static ASTNode Make(ASTNode parent, Args&&... args) {
return std::make_shared<ASTBase>(std::move(parent),
ASTData(U(std::forward<Args>(args)...)));
}
void SetParent(ASTNode new_parent) {
parent = std::move(new_parent);
}
ASTNode& GetParent() {
return parent;
}
const ASTNode& GetParent() const {
return parent;
}
u32 GetLevel() const {
u32 level = 0;
auto next_parent = parent;
while (next_parent) {
next_parent = next_parent->GetParent();
level++;
}
return level;
}
ASTData* GetInnerData() {
return &data;
}
const ASTData* GetInnerData() const {
return &data;
}
ASTNode GetNext() const {
return next;
}
ASTNode GetPrevious() const {
return previous;
}
ASTZipper& GetManager() {
return *manager;
}
const ASTZipper& GetManager() const {
return *manager;
}
std::optional<u32> GetGotoLabel() const {
if (const auto* inner = std::get_if<ASTGoto>(&data)) {
return {inner->label};
}
return std::nullopt;
}
Expr GetGotoCondition() const {
if (const auto* inner = std::get_if<ASTGoto>(&data)) {
return inner->condition;
}
return nullptr;
}
void MarkLabelUnused() {
if (auto* inner = std::get_if<ASTLabel>(&data)) {
inner->unused = true;
}
}
bool IsLabelUnused() const {
if (const auto* inner = std::get_if<ASTLabel>(&data)) {
return inner->unused;
}
return true;
}
std::optional<u32> GetLabelIndex() const {
if (const auto* inner = std::get_if<ASTLabel>(&data)) {
return {inner->index};
}
return std::nullopt;
}
Expr GetIfCondition() const {
if (const auto* inner = std::get_if<ASTIfThen>(&data)) {
return inner->condition;
}
return nullptr;
}
void SetGotoCondition(Expr new_condition) {
if (auto* inner = std::get_if<ASTGoto>(&data)) {
inner->condition = std::move(new_condition);
}
}
bool IsIfThen() const {
return std::holds_alternative<ASTIfThen>(data);
}
bool IsIfElse() const {
return std::holds_alternative<ASTIfElse>(data);
}
bool IsBlockEncoded() const {
return std::holds_alternative<ASTBlockEncoded>(data);
}
void TransformBlockEncoded(NodeBlock&& nodes) {
data = ASTBlockDecoded(std::move(nodes));
}
bool IsLoop() const {
return std::holds_alternative<ASTDoWhile>(data);
}
ASTZipper* GetSubNodes() {
if (std::holds_alternative<ASTProgram>(data)) {
return &std::get_if<ASTProgram>(&data)->nodes;
}
if (std::holds_alternative<ASTIfThen>(data)) {
return &std::get_if<ASTIfThen>(&data)->nodes;
}
if (std::holds_alternative<ASTIfElse>(data)) {
return &std::get_if<ASTIfElse>(&data)->nodes;
}
if (std::holds_alternative<ASTDoWhile>(data)) {
return &std::get_if<ASTDoWhile>(&data)->nodes;
}
return nullptr;
}
void Clear() {
next.reset();
previous.reset();
parent.reset();
manager = nullptr;
}
private:
friend class ASTZipper;
ASTData data;
ASTNode parent;
ASTNode next;
ASTNode previous;
ASTZipper* manager{};
};
class ASTManager final {
public:
explicit ASTManager(bool do_full_decompile, bool disable_else_derivation_);
~ASTManager();
ASTManager(const ASTManager& o) = delete;
ASTManager& operator=(const ASTManager& other) = delete;
ASTManager(ASTManager&& other) noexcept = default;
ASTManager& operator=(ASTManager&& other) noexcept = default;
void Init();
void DeclareLabel(u32 address);
void InsertLabel(u32 address);
void InsertGoto(Expr condition, u32 address);
void InsertBlock(u32 start_address, u32 end_address);
void InsertReturn(Expr condition, bool kills);
std::string Print() const;
void Decompile();
void ShowCurrentState(std::string_view state) const;
void SanityCheck() const;
void Clear();
bool IsFullyDecompiled() const {
if (full_decompile) {
return gotos.empty();
}
for (ASTNode goto_node : gotos) {
auto label_index = goto_node->GetGotoLabel();
if (!label_index) {
return false;
}
ASTNode glabel = labels[*label_index];
if (IsBackwardsJump(goto_node, glabel)) {
return false;
}
}
return true;
}
ASTNode GetProgram() const {
return main_node;
}
u32 GetVariables() const {
return variables;
}
const std::vector<ASTNode>& GetLabels() const {
return labels;
}
private:
bool IsBackwardsJump(ASTNode goto_node, ASTNode label_node) const;
bool IndirectlyRelated(const ASTNode& first, const ASTNode& second) const;
bool DirectlyRelated(const ASTNode& first, const ASTNode& second) const;
void EncloseDoWhile(ASTNode goto_node, ASTNode label);
void EncloseIfThen(ASTNode goto_node, ASTNode label);
void MoveOutward(ASTNode goto_node);
u32 NewVariable() {
return variables++;
}
bool full_decompile{};
bool disable_else_derivation{};
std::unordered_map<u32, u32> labels_map{};
u32 labels_count{};
std::vector<ASTNode> labels{};
std::list<ASTNode> gotos{};
u32 variables{};
ASTProgram* program{};
ASTNode main_node{};
Expr false_condition{};
};
} // namespace VideoCommon::Shader

234
src/video_core/shader/async_shaders.cpp
View File

@ -1,234 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <condition_variable>
#include <mutex>
#include <thread>
#include <vector>
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_base.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/shader/async_shaders.h"
namespace VideoCommon::Shader {
AsyncShaders::AsyncShaders(Core::Frontend::EmuWindow& emu_window_) : emu_window(emu_window_) {}
AsyncShaders::~AsyncShaders() {
KillWorkers();
}
void AsyncShaders::AllocateWorkers() {
// Use at least one thread
u32 num_workers = 1;
// Deduce how many more threads we can use
const u32 thread_count = std::thread::hardware_concurrency();
if (thread_count >= 8) {
// Increase async workers by 1 for every 2 threads >= 8
num_workers += 1 + (thread_count - 8) / 2;
}
// If we already have workers queued, ignore
if (num_workers == worker_threads.size()) {
return;
}
// If workers already exist, clear them
if (!worker_threads.empty()) {
FreeWorkers();
}
// Create workers
for (std::size_t i = 0; i < num_workers; i++) {
context_list.push_back(emu_window.CreateSharedContext());
worker_threads.emplace_back(&AsyncShaders::ShaderCompilerThread, this,
context_list[i].get());
}
}
void AsyncShaders::FreeWorkers() {
// Mark all threads to quit
is_thread_exiting.store(true);
cv.notify_all();
for (auto& thread : worker_threads) {
thread.join();
}
// Clear our shared contexts
context_list.clear();
// Clear our worker threads
worker_threads.clear();
}
void AsyncShaders::KillWorkers() {
is_thread_exiting.store(true);
cv.notify_all();
for (auto& thread : worker_threads) {
thread.detach();
}
// Clear our shared contexts
context_list.clear();
// Clear our worker threads
worker_threads.clear();
}
bool AsyncShaders::HasWorkQueued() const {
return !pending_queue.empty();
}
bool AsyncShaders::HasCompletedWork() const {
std::shared_lock lock{completed_mutex};
return !finished_work.empty();
}
bool AsyncShaders::IsShaderAsync(const Tegra::GPU& gpu) const {
const auto& regs = gpu.Maxwell3D().regs;
// If something is using depth, we can assume that games are not rendering anything which will
// be used one time.
if (regs.zeta_enable) {
return true;
}
// If games are using a small index count, we can assume these are full screen quads. Usually
// these shaders are only used once for building textures so we can assume they can't be built
// async
if (regs.index_array.count <= 6 || regs.vertex_buffer.count <= 6) {
return false;
}
return true;
}
std::vector<AsyncShaders::Result> AsyncShaders::GetCompletedWork() {
std::vector<Result> results;
{
std::unique_lock lock{completed_mutex};
results = std::move(finished_work);
finished_work.clear();
}
return results;
}
void AsyncShaders::QueueOpenGLShader(const OpenGL::Device& device,
Tegra::Engines::ShaderType shader_type, u64 uid,
std::vector<u64> code, std::vector<u64> code_b,
u32 main_offset, CompilerSettings compiler_settings,
const Registry& registry, VAddr cpu_addr) {
std::unique_lock lock(queue_mutex);
pending_queue.push({
.backend = device.UseAssemblyShaders() ? Backend::GLASM : Backend::OpenGL,
.device = &device,
.shader_type = shader_type,
.uid = uid,
.code = std::move(code),
.code_b = std::move(code_b),
.main_offset = main_offset,
.compiler_settings = compiler_settings,
.registry = registry,
.cpu_address = cpu_addr,
.pp_cache = nullptr,
.vk_device = nullptr,
.scheduler = nullptr,
.descriptor_pool = nullptr,
.update_descriptor_queue = nullptr,
.bindings{},
.program{},
.key{},
.num_color_buffers = 0,
});
cv.notify_one();
}
void AsyncShaders::QueueVulkanShader(Vulkan::VKPipelineCache* pp_cache,
const Vulkan::Device& device, Vulkan::VKScheduler& scheduler,
Vulkan::VKDescriptorPool& descriptor_pool,
Vulkan::VKUpdateDescriptorQueue& update_descriptor_queue,
std::vector<VkDescriptorSetLayoutBinding> bindings,
Vulkan::SPIRVProgram program,
Vulkan::GraphicsPipelineCacheKey key, u32 num_color_buffers) {
std::unique_lock lock(queue_mutex);
pending_queue.push({
.backend = Backend::Vulkan,
.device = nullptr,
.shader_type{},
.uid = 0,
.code{},
.code_b{},
.main_offset = 0,
.compiler_settings{},
.registry{},
.cpu_address = 0,
.pp_cache = pp_cache,
.vk_device = &device,
.scheduler = &scheduler,
.descriptor_pool = &descriptor_pool,
.update_descriptor_queue = &update_descriptor_queue,
.bindings = std::move(bindings),
.program = std::move(program),
.key = key,
.num_color_buffers = num_color_buffers,
});
cv.notify_one();
}
void AsyncShaders::ShaderCompilerThread(Core::Frontend::GraphicsContext* context) {
while (!is_thread_exiting.load(std::memory_order_relaxed)) {
std::unique_lock lock{queue_mutex};
cv.wait(lock, [this] { return HasWorkQueued() || is_thread_exiting; });
if (is_thread_exiting) {
return;
}
// Partial lock to allow all threads to read at the same time
if (!HasWorkQueued()) {
continue;
}
// Another thread beat us, just unlock and wait for the next load
if (pending_queue.empty()) {
continue;
}
// Pull work from queue
WorkerParams work = std::move(pending_queue.front());
pending_queue.pop();
lock.unlock();
if (work.backend == Backend::OpenGL || work.backend == Backend::GLASM) {
const ShaderIR ir(work.code, work.main_offset, work.compiler_settings, *work.registry);
const auto scope = context->Acquire();
auto program =
OpenGL::BuildShader(*work.device, work.shader_type, work.uid, ir, *work.registry);
Result result{};
result.backend = work.backend;
result.cpu_address = work.cpu_address;
result.uid = work.uid;
result.code = std::move(work.code);
result.code_b = std::move(work.code_b);
result.shader_type = work.shader_type;
if (work.backend == Backend::OpenGL) {
result.program.opengl = std::move(program->source_program);
} else if (work.backend == Backend::GLASM) {
result.program.glasm = std::move(program->assembly_program);
}
{
std::unique_lock complete_lock(completed_mutex);
finished_work.push_back(std::move(result));
}
} else if (work.backend == Backend::Vulkan) {
auto pipeline = std::make_unique<Vulkan::VKGraphicsPipeline>(
*work.vk_device, *work.scheduler, *work.descriptor_pool,
*work.update_descriptor_queue, work.key, work.bindings, work.program,
work.num_color_buffers);
work.pp_cache->EmplacePipeline(std::move(pipeline));
}
}
}
} // namespace VideoCommon::Shader

138
src/video_core/shader/async_shaders.h
View File

@ -1,138 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <condition_variable>
#include <memory>
#include <shared_mutex>
#include <thread>
#include <glad/glad.h>
#include "common/common_types.h"
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/vulkan_common/vulkan_device.h"
namespace Core::Frontend {
class EmuWindow;
class GraphicsContext;
} // namespace Core::Frontend
namespace Tegra {
class GPU;
}
namespace Vulkan {
class VKPipelineCache;
}
namespace VideoCommon::Shader {
class AsyncShaders {
public:
enum class Backend {
OpenGL,
GLASM,
Vulkan,
};
struct ResultPrograms {
OpenGL::OGLProgram opengl;
OpenGL::OGLAssemblyProgram glasm;
};
struct Result {
u64 uid;
VAddr cpu_address;
Backend backend;
ResultPrograms program;
std::vector<u64> code;
std::vector<u64> code_b;
Tegra::Engines::ShaderType shader_type;
};
explicit AsyncShaders(Core::Frontend::EmuWindow& emu_window_);
~AsyncShaders();
/// Start up shader worker threads
void AllocateWorkers();
/// Clear the shader queue and kill all worker threads
void FreeWorkers();
// Force end all threads
void KillWorkers();
/// Check to see if any shaders have actually been compiled
[[nodiscard]] bool HasCompletedWork() const;
/// Deduce if a shader can be build on another thread of MUST be built in sync. We cannot build
/// every shader async as some shaders are only built and executed once. We try to "guess" which
/// shader would be used only once
[[nodiscard]] bool IsShaderAsync(const Tegra::GPU& gpu) const;
/// Pulls completed compiled shaders
[[nodiscard]] std::vector<Result> GetCompletedWork();
void QueueOpenGLShader(const OpenGL::Device& device, Tegra::Engines::ShaderType shader_type,
u64 uid, std::vector<u64> code, std::vector<u64> code_b, u32 main_offset,
CompilerSettings compiler_settings, const Registry& registry,
VAddr cpu_addr);
void QueueVulkanShader(Vulkan::VKPipelineCache* pp_cache, const Vulkan::Device& device,
Vulkan::VKScheduler& scheduler,
Vulkan::VKDescriptorPool& descriptor_pool,
Vulkan::VKUpdateDescriptorQueue& update_descriptor_queue,
std::vector<VkDescriptorSetLayoutBinding> bindings,
Vulkan::SPIRVProgram program, Vulkan::GraphicsPipelineCacheKey key,
u32 num_color_buffers);
private:
void ShaderCompilerThread(Core::Frontend::GraphicsContext* context);
/// Check our worker queue to see if we have any work queued already
[[nodiscard]] bool HasWorkQueued() const;
struct WorkerParams {
Backend backend;
// For OGL
const OpenGL::Device* device;
Tegra::Engines::ShaderType shader_type;
u64 uid;
std::vector<u64> code;
std::vector<u64> code_b;
u32 main_offset;
CompilerSettings compiler_settings;
std::optional<Registry> registry;
VAddr cpu_address;
// For Vulkan
Vulkan::VKPipelineCache* pp_cache;
const Vulkan::Device* vk_device;
Vulkan::VKScheduler* scheduler;
Vulkan::VKDescriptorPool* descriptor_pool;
Vulkan::VKUpdateDescriptorQueue* update_descriptor_queue;
std::vector<VkDescriptorSetLayoutBinding> bindings;
Vulkan::SPIRVProgram program;
Vulkan::GraphicsPipelineCacheKey key;
u32 num_color_buffers;
};
std::condition_variable cv;
mutable std::mutex queue_mutex;
mutable std::shared_mutex completed_mutex;
std::atomic<bool> is_thread_exiting{};
std::vector<std::unique_ptr<Core::Frontend::GraphicsContext>> context_list;
std::vector<std::thread> worker_threads;
std::queue<WorkerParams> pending_queue;
std::vector<Result> finished_work;
Core::Frontend::EmuWindow& emu_window;
};
} // namespace VideoCommon::Shader

26
src/video_core/shader/compiler_settings.cpp
View File

@ -1,26 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "video_core/shader/compiler_settings.h"
namespace VideoCommon::Shader {
std::string CompileDepthAsString(const CompileDepth cd) {
switch (cd) {
case CompileDepth::BruteForce:
return "Brute Force Compile";
case CompileDepth::FlowStack:
return "Simple Flow Stack Mode";
case CompileDepth::NoFlowStack:
return "Remove Flow Stack";
case CompileDepth::DecompileBackwards:
return "Decompile Backward Jumps";
case CompileDepth::FullDecompile:
return "Full Decompilation";
default:
return "Unknown Compiler Process";
}
}
} // namespace VideoCommon::Shader

26
src/video_core/shader/compiler_settings.h
View File

@ -1,26 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "video_core/engines/shader_bytecode.h"
namespace VideoCommon::Shader {
enum class CompileDepth : u32 {
BruteForce = 0,
FlowStack = 1,
NoFlowStack = 2,
DecompileBackwards = 3,
FullDecompile = 4,
};
std::string CompileDepthAsString(CompileDepth cd);
struct CompilerSettings {
CompileDepth depth{CompileDepth::NoFlowStack};
bool disable_else_derivation{true};
};
} // namespace VideoCommon::Shader

751
src/video_core/shader/control_flow.cpp
View File

@ -1,751 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <list>
#include <map>
#include <set>
#include <stack>
#include <unordered_map>
#include <vector>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/shader/ast.h"
#include "video_core/shader/control_flow.h"
#include "video_core/shader/memory_util.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
namespace {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
constexpr s32 unassigned_branch = -2;
struct Query {
u32 address{};
std::stack<u32> ssy_stack{};
std::stack<u32> pbk_stack{};
};
struct BlockStack {
BlockStack() = default;
explicit BlockStack(const Query& q) : ssy_stack{q.ssy_stack}, pbk_stack{q.pbk_stack} {}
std::stack<u32> ssy_stack{};
std::stack<u32> pbk_stack{};
};
template <typename T, typename... Args>
BlockBranchInfo MakeBranchInfo(Args&&... args) {
static_assert(std::is_convertible_v<T, BranchData>);
return std::make_shared<BranchData>(T(std::forward<Args>(args)...));
}
bool BlockBranchIsIgnored(BlockBranchInfo first) {
bool ignore = false;
if (std::holds_alternative<SingleBranch>(*first)) {
const auto branch = std::get_if<SingleBranch>(first.get());
ignore = branch->ignore;
}
return ignore;
}
struct BlockInfo {
u32 start{};
u32 end{};
bool visited{};
BlockBranchInfo branch{};
bool IsInside(const u32 address) const {
return start <= address && address <= end;
}
};
struct CFGRebuildState {
explicit CFGRebuildState(const ProgramCode& program_code_, u32 start_, Registry& registry_)
: program_code{program_code_}, registry{registry_}, start{start_} {}
const ProgramCode& program_code;
Registry& registry;
u32 start{};
std::vector<BlockInfo> block_info;
std::list<u32> inspect_queries;
std::list<Query> queries;
std::unordered_map<u32, u32> registered;
std::set<u32> labels;
std::map<u32, u32> ssy_labels;
std::map<u32, u32> pbk_labels;
std::unordered_map<u32, BlockStack> stacks;
ASTManager* manager{};
};
enum class BlockCollision : u32 { None, Found, Inside };
std::pair<BlockCollision, u32> TryGetBlock(CFGRebuildState& state, u32 address) {
const auto& blocks = state.block_info;
for (u32 index = 0; index < blocks.size(); index++) {
if (blocks[index].start == address) {
return {BlockCollision::Found, index};
}
if (blocks[index].IsInside(address)) {
return {BlockCollision::Inside, index};
}
}
return {BlockCollision::None, 0xFFFFFFFF};
}
struct ParseInfo {
BlockBranchInfo branch_info{};
u32 end_address{};
};
BlockInfo& CreateBlockInfo(CFGRebuildState& state, u32 start, u32 end) {
auto& it = state.block_info.emplace_back();
it.start = start;
it.end = end;
const u32 index = static_cast<u32>(state.block_info.size() - 1);
state.registered.insert({start, index});
return it;
}
Pred GetPredicate(u32 index, bool negated) {
return static_cast<Pred>(static_cast<u64>(index) + (negated ? 8ULL : 0ULL));
}
enum class ParseResult : u32 {
ControlCaught,
BlockEnd,
AbnormalFlow,
};
struct BranchIndirectInfo {
u32 buffer{};
u32 offset{};
u32 entries{};
s32 relative_position{};
};
struct BufferInfo {
u32 index;
u32 offset;
};
std::optional<std::pair<s32, u64>> GetBRXInfo(const CFGRebuildState& state, u32& pos) {
const Instruction instr = state.program_code[pos];
const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() != OpCode::Id::BRX) {
return std::nullopt;
}
if (instr.brx.constant_buffer != 0) {
return std::nullopt;
}
--pos;
return std::make_pair(instr.brx.GetBranchExtend(), instr.gpr8.Value());
}
template <typename Result, typename TestCallable, typename PackCallable>
// requires std::predicate<TestCallable, Instruction, const OpCode::Matcher&>
// requires std::invocable<PackCallable, Instruction, const OpCode::Matcher&>
std::optional<Result> TrackInstruction(const CFGRebuildState& state, u32& pos, TestCallable test,
PackCallable pack) {
for (; pos >= state.start; --pos) {
if (IsSchedInstruction(pos, state.start)) {
continue;
}
const Instruction instr = state.program_code[pos];
const auto opcode = OpCode::Decode(instr);
if (!opcode) {
continue;
}
if (test(instr, opcode->get())) {
--pos;
return std::make_optional(pack(instr, opcode->get()));
}
}
return std::nullopt;
}
std::optional<std::pair<BufferInfo, u64>> TrackLDC(const CFGRebuildState& state, u32& pos,
u64 brx_tracked_register) {
return TrackInstruction<std::pair<BufferInfo, u64>>(
state, pos,
[brx_tracked_register](auto instr, const auto& opcode) {
return opcode.GetId() == OpCode::Id::LD_C &&
instr.gpr0.Value() == brx_tracked_register &&
instr.ld_c.type.Value() == Tegra::Shader::UniformType::Single;
},
[](auto instr, const auto& opcode) {
const BufferInfo info = {static_cast<u32>(instr.cbuf36.index.Value()),
static_cast<u32>(instr.cbuf36.GetOffset())};
return std::make_pair(info, instr.gpr8.Value());
});
}
std::optional<u64> TrackSHLRegister(const CFGRebuildState& state, u32& pos,
u64 ldc_tracked_register) {
return TrackInstruction<u64>(
state, pos,
[ldc_tracked_register](auto instr, const auto& opcode) {
return opcode.GetId() == OpCode::Id::SHL_IMM &&
instr.gpr0.Value() == ldc_tracked_register;
},
[](auto instr, const auto&) { return instr.gpr8.Value(); });
}
std::optional<u32> TrackIMNMXValue(const CFGRebuildState& state, u32& pos,
u64 shl_tracked_register) {
return TrackInstruction<u32>(
state, pos,
[shl_tracked_register](auto instr, const auto& opcode) {
return opcode.GetId() == OpCode::Id::IMNMX_IMM &&
instr.gpr0.Value() == shl_tracked_register;
},
[](auto instr, const auto&) {
return static_cast<u32>(instr.alu.GetSignedImm20_20() + 1);
});
}
std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState& state, u32 pos) {
const auto brx_info = GetBRXInfo(state, pos);
if (!brx_info) {
return std::nullopt;
}
const auto [relative_position, brx_tracked_register] = *brx_info;
const auto ldc_info = TrackLDC(state, pos, brx_tracked_register);
if (!ldc_info) {
return std::nullopt;
}
const auto [buffer_info, ldc_tracked_register] = *ldc_info;
const auto shl_tracked_register = TrackSHLRegister(state, pos, ldc_tracked_register);
if (!shl_tracked_register) {
return std::nullopt;
}
const auto entries = TrackIMNMXValue(state, pos, *shl_tracked_register);
if (!entries) {
return std::nullopt;
}
return BranchIndirectInfo{buffer_info.index, buffer_info.offset, *entries, relative_position};
}
std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address) {
u32 offset = static_cast<u32>(address);
const u32 end_address = static_cast<u32>(state.program_code.size());
ParseInfo parse_info{};
SingleBranch single_branch{};
const auto insert_label = [](CFGRebuildState& rebuild_state, u32 label_address) {
const auto pair = rebuild_state.labels.emplace(label_address);
if (pair.second) {
rebuild_state.inspect_queries.push_back(label_address);
}
};
while (true) {
if (offset >= end_address) {
// ASSERT_OR_EXECUTE can't be used, as it ignores the break
ASSERT_MSG(false, "Shader passed the current limit!");
single_branch.address = exit_branch;
single_branch.ignore = false;
break;
}
if (state.registered.contains(offset)) {
single_branch.address = offset;
single_branch.ignore = true;
break;
}
if (IsSchedInstruction(offset, state.start)) {
offset++;
continue;
}
const Instruction instr = {state.program_code[offset]};
const auto opcode = OpCode::Decode(instr);
if (!opcode || opcode->get().GetType() != OpCode::Type::Flow) {
offset++;
continue;
}
switch (opcode->get().GetId()) {
case OpCode::Id::EXIT: {
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
if (single_branch.condition.predicate == Pred::NeverExecute) {
offset++;
continue;
}
const ConditionCode cc = instr.flow_condition_code;
single_branch.condition.cc = cc;
if (cc == ConditionCode::F) {
offset++;
continue;
}
single_branch.address = exit_branch;
single_branch.kill = false;
single_branch.is_sync = false;
single_branch.is_brk = false;
single_branch.ignore = false;
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill,
single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
return {ParseResult::ControlCaught, parse_info};
}
case OpCode::Id::BRA: {
if (instr.bra.constant_buffer != 0) {
return {ParseResult::AbnormalFlow, parse_info};
}
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
if (single_branch.condition.predicate == Pred::NeverExecute) {
offset++;
continue;
}
const ConditionCode cc = instr.flow_condition_code;
single_branch.condition.cc = cc;
if (cc == ConditionCode::F) {
offset++;
continue;
}
const u32 branch_offset = offset + instr.bra.GetBranchTarget();
if (branch_offset == 0) {
single_branch.address = exit_branch;
} else {
single_branch.address = branch_offset;
}
insert_label(state, branch_offset);
single_branch.kill = false;
single_branch.is_sync = false;
single_branch.is_brk = false;
single_branch.ignore = false;
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill,
single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
return {ParseResult::ControlCaught, parse_info};
}
case OpCode::Id::SYNC: {
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
if (single_branch.condition.predicate == Pred::NeverExecute) {
offset++;
continue;
}
const ConditionCode cc = instr.flow_condition_code;
single_branch.condition.cc = cc;
if (cc == ConditionCode::F) {
offset++;
continue;
}
single_branch.address = unassigned_branch;
single_branch.kill = false;
single_branch.is_sync = true;
single_branch.is_brk = false;
single_branch.ignore = false;
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill,
single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
return {ParseResult::ControlCaught, parse_info};
}
case OpCode::Id::BRK: {
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
if (single_branch.condition.predicate == Pred::NeverExecute) {
offset++;
continue;
}
const ConditionCode cc = instr.flow_condition_code;
single_branch.condition.cc = cc;
if (cc == ConditionCode::F) {
offset++;
continue;
}
single_branch.address = unassigned_branch;
single_branch.kill = false;
single_branch.is_sync = false;
single_branch.is_brk = true;
single_branch.ignore = false;
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill,
single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
return {ParseResult::ControlCaught, parse_info};
}
case OpCode::Id::KIL: {
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
if (single_branch.condition.predicate == Pred::NeverExecute) {
offset++;
continue;
}
const ConditionCode cc = instr.flow_condition_code;
single_branch.condition.cc = cc;
if (cc == ConditionCode::F) {
offset++;
continue;
}
single_branch.address = exit_branch;
single_branch.kill = true;
single_branch.is_sync = false;
single_branch.is_brk = false;
single_branch.ignore = false;
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill,
single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
return {ParseResult::ControlCaught, parse_info};
}
case OpCode::Id::SSY: {
const u32 target = offset + instr.bra.GetBranchTarget();
insert_label(state, target);
state.ssy_labels.emplace(offset, target);
break;
}
case OpCode::Id::PBK: {
const u32 target = offset + instr.bra.GetBranchTarget();
insert_label(state, target);
state.pbk_labels.emplace(offset, target);
break;
}
case OpCode::Id::BRX: {
const auto tmp = TrackBranchIndirectInfo(state, offset);
if (!tmp) {
LOG_WARNING(HW_GPU, "BRX Track Unsuccesful");
return {ParseResult::AbnormalFlow, parse_info};
}
const auto result = *tmp;
const s32 pc_target = offset + result.relative_position;
std::vector<CaseBranch> branches;
for (u32 i = 0; i < result.entries; i++) {
auto key = state.registry.ObtainKey(result.buffer, result.offset + i * 4);
if (!key) {
return {ParseResult::AbnormalFlow, parse_info};
}
u32 value = *key;
u32 target = static_cast<u32>((value >> 3) + pc_target);
insert_label(state, target);
branches.emplace_back(value, target);
}
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<MultiBranch>(
static_cast<u32>(instr.gpr8.Value()), std::move(branches));
return {ParseResult::ControlCaught, parse_info};
}
default:
break;
}
offset++;
}
single_branch.kill = false;
single_branch.is_sync = false;
single_branch.is_brk = false;
parse_info.end_address = offset - 1;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill, single_branch.is_sync,
single_branch.is_brk, single_branch.ignore);
return {ParseResult::BlockEnd, parse_info};
}
bool TryInspectAddress(CFGRebuildState& state) {
if (state.inspect_queries.empty()) {
return false;
}
const u32 address = state.inspect_queries.front();
state.inspect_queries.pop_front();
const auto [result, block_index] = TryGetBlock(state, address);
switch (result) {
case BlockCollision::Found: {
return true;
}
case BlockCollision::Inside: {
// This case is the tricky one:
// We need to split the block into 2 separate blocks
const u32 end = state.block_info[block_index].end;
BlockInfo& new_block = CreateBlockInfo(state, address, end);
BlockInfo& current_block = state.block_info[block_index];
current_block.end = address - 1;
new_block.branch = std::move(current_block.branch);
BlockBranchInfo forward_branch = MakeBranchInfo<SingleBranch>();
const auto branch = std::get_if<SingleBranch>(forward_branch.get());
branch->address = address;
branch->ignore = true;
current_block.branch = std::move(forward_branch);
return true;
}
default:
break;
}
const auto [parse_result, parse_info] = ParseCode(state, address);
if (parse_result == ParseResult::AbnormalFlow) {
// if it's AbnormalFlow, we end it as false, ending the CFG reconstruction
return false;
}
BlockInfo& block_info = CreateBlockInfo(state, address, parse_info.end_address);
block_info.branch = parse_info.branch_info;
if (std::holds_alternative<SingleBranch>(*block_info.branch)) {
const auto branch = std::get_if<SingleBranch>(block_info.branch.get());
if (branch->condition.IsUnconditional()) {
return true;
}
const u32 fallthrough_address = parse_info.end_address + 1;
state.inspect_queries.push_front(fallthrough_address);
return true;
}
return true;
}
bool TryQuery(CFGRebuildState& state) {
const auto gather_labels = [](std::stack<u32>& cc, std::map<u32, u32>& labels,
BlockInfo& block) {
auto gather_start = labels.lower_bound(block.start);
const auto gather_end = labels.upper_bound(block.end);
while (gather_start != gather_end) {
cc.push(gather_start->second);
++gather_start;
}
};
if (state.queries.empty()) {
return false;
}
Query& q = state.queries.front();
const u32 block_index = state.registered[q.address];
BlockInfo& block = state.block_info[block_index];
// If the block is visited, check if the stacks match, else gather the ssy/pbk
// labels into the current stack and look if the branch at the end of the block
// consumes a label. Schedule new queries accordingly
if (block.visited) {
BlockStack& stack = state.stacks[q.address];
const bool all_okay = (stack.ssy_stack.empty() || q.ssy_stack == stack.ssy_stack) &&
(stack.pbk_stack.empty() || q.pbk_stack == stack.pbk_stack);
state.queries.pop_front();
return all_okay;
}
block.visited = true;
state.stacks.insert_or_assign(q.address, BlockStack{q});
Query q2(q);
state.queries.pop_front();
gather_labels(q2.ssy_stack, state.ssy_labels, block);
gather_labels(q2.pbk_stack, state.pbk_labels, block);
if (std::holds_alternative<SingleBranch>(*block.branch)) {
auto* branch = std::get_if<SingleBranch>(block.branch.get());
if (!branch->condition.IsUnconditional()) {
q2.address = block.end + 1;
state.queries.push_back(q2);
}
auto& conditional_query = state.queries.emplace_back(q2);
if (branch->is_sync) {
if (branch->address == unassigned_branch) {
branch->address = conditional_query.ssy_stack.top();
}
conditional_query.ssy_stack.pop();
}
if (branch->is_brk) {
if (branch->address == unassigned_branch) {
branch->address = conditional_query.pbk_stack.top();
}
conditional_query.pbk_stack.pop();
}
conditional_query.address = branch->address;
return true;
}
const auto* multi_branch = std::get_if<MultiBranch>(block.branch.get());
for (const auto& branch_case : multi_branch->branches) {
auto& conditional_query = state.queries.emplace_back(q2);
conditional_query.address = branch_case.address;
}
return true;
}
void InsertBranch(ASTManager& mm, const BlockBranchInfo& branch_info) {
const auto get_expr = [](const Condition& cond) -> Expr {
Expr result;
if (cond.cc != ConditionCode::T) {
result = MakeExpr<ExprCondCode>(cond.cc);
}
if (cond.predicate != Pred::UnusedIndex) {
u32 pred = static_cast<u32>(cond.predicate);
bool negate = false;
if (pred > 7) {
negate = true;
pred -= 8;
}
Expr extra = MakeExpr<ExprPredicate>(pred);
if (negate) {
extra = MakeExpr<ExprNot>(std::move(extra));
}
if (result) {
return MakeExpr<ExprAnd>(std::move(extra), std::move(result));
}
return extra;
}
if (result) {
return result;
}
return MakeExpr<ExprBoolean>(true);
};
if (std::holds_alternative<SingleBranch>(*branch_info)) {
const auto* branch = std::get_if<SingleBranch>(branch_info.get());
if (branch->address < 0) {
if (branch->kill) {
mm.InsertReturn(get_expr(branch->condition), true);
return;
}
mm.InsertReturn(get_expr(branch->condition), false);
return;
}
mm.InsertGoto(get_expr(branch->condition), branch->address);
return;
}
const auto* multi_branch = std::get_if<MultiBranch>(branch_info.get());
for (const auto& branch_case : multi_branch->branches) {
mm.InsertGoto(MakeExpr<ExprGprEqual>(multi_branch->gpr, branch_case.cmp_value),
branch_case.address);
}
}
void DecompileShader(CFGRebuildState& state) {
state.manager->Init();
for (auto label : state.labels) {
state.manager->DeclareLabel(label);
}
for (const auto& block : state.block_info) {
if (state.labels.contains(block.start)) {
state.manager->InsertLabel(block.start);
}
const bool ignore = BlockBranchIsIgnored(block.branch);
const u32 end = ignore ? block.end + 1 : block.end;
state.manager->InsertBlock(block.start, end);
if (!ignore) {
InsertBranch(*state.manager, block.branch);
}
}
state.manager->Decompile();
}
} // Anonymous namespace
std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
const CompilerSettings& settings,
Registry& registry) {
auto result_out = std::make_unique<ShaderCharacteristics>();
if (settings.depth == CompileDepth::BruteForce) {
result_out->settings.depth = CompileDepth::BruteForce;
return result_out;
}
CFGRebuildState state{program_code, start_address, registry};
// Inspect Code and generate blocks
state.labels.clear();
state.labels.emplace(start_address);
state.inspect_queries.push_back(state.start);
while (!state.inspect_queries.empty()) {
if (!TryInspectAddress(state)) {
result_out->settings.depth = CompileDepth::BruteForce;
return result_out;
}
}
bool use_flow_stack = true;
bool decompiled = false;
if (settings.depth != CompileDepth::FlowStack) {
// Decompile Stacks
state.queries.push_back(Query{state.start, {}, {}});
decompiled = true;
while (!state.queries.empty()) {
if (!TryQuery(state)) {
decompiled = false;
break;
}
}
}
use_flow_stack = !decompiled;
// Sort and organize results
std::sort(state.block_info.begin(), state.block_info.end(),
[](const BlockInfo& a, const BlockInfo& b) -> bool { return a.start < b.start; });
if (decompiled && settings.depth != CompileDepth::NoFlowStack) {
ASTManager manager{settings.depth != CompileDepth::DecompileBackwards,
settings.disable_else_derivation};
state.manager = &manager;
DecompileShader(state);
decompiled = state.manager->IsFullyDecompiled();
if (!decompiled) {
if (settings.depth == CompileDepth::FullDecompile) {
LOG_CRITICAL(HW_GPU, "Failed to remove all the gotos!:");
} else {
LOG_CRITICAL(HW_GPU, "Failed to remove all backward gotos!:");
}
state.manager->ShowCurrentState("Of Shader");
state.manager->Clear();
} else {
auto characteristics = std::make_unique<ShaderCharacteristics>();
characteristics->start = start_address;
characteristics->settings.depth = settings.depth;
characteristics->manager = std::move(manager);
characteristics->end = state.block_info.back().end + 1;
return characteristics;
}
}
result_out->start = start_address;
result_out->settings.depth =
use_flow_stack ? CompileDepth::FlowStack : CompileDepth::NoFlowStack;
result_out->blocks.clear();
for (auto& block : state.block_info) {
ShaderBlock new_block{};
new_block.start = block.start;
new_block.end = block.end;
new_block.ignore_branch = BlockBranchIsIgnored(block.branch);
if (!new_block.ignore_branch) {
new_block.branch = block.branch;
}
result_out->end = std::max(result_out->end, block.end);
result_out->blocks.push_back(new_block);
}
if (!use_flow_stack) {
result_out->labels = std::move(state.labels);
return result_out;
}
auto back = result_out->blocks.begin();
auto next = std::next(back);
while (next != result_out->blocks.end()) {
if (!state.labels.contains(next->start) && next->start == back->end + 1) {
back->end = next->end;
next = result_out->blocks.erase(next);
continue;
}
back = next;
++next;
}
return result_out;
}
} // namespace VideoCommon::Shader

117
src/video_core/shader/control_flow.h
View File

@ -1,117 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <list>
#include <optional>
#include <set>
#include <variant>
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/ast.h"
#include "video_core/shader/compiler_settings.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::ConditionCode;
using Tegra::Shader::Pred;
constexpr s32 exit_branch = -1;
struct Condition {
Pred predicate{Pred::UnusedIndex};
ConditionCode cc{ConditionCode::T};
bool IsUnconditional() const {
return predicate == Pred::UnusedIndex && cc == ConditionCode::T;
}
bool operator==(const Condition& other) const {
return std::tie(predicate, cc) == std::tie(other.predicate, other.cc);
}
bool operator!=(const Condition& other) const {
return !operator==(other);
}
};
class SingleBranch {
public:
SingleBranch() = default;
explicit SingleBranch(Condition condition_, s32 address_, bool kill_, bool is_sync_,
bool is_brk_, bool ignore_)
: condition{condition_}, address{address_}, kill{kill_}, is_sync{is_sync_}, is_brk{is_brk_},
ignore{ignore_} {}
bool operator==(const SingleBranch& b) const {
return std::tie(condition, address, kill, is_sync, is_brk, ignore) ==
std::tie(b.condition, b.address, b.kill, b.is_sync, b.is_brk, b.ignore);
}
bool operator!=(const SingleBranch& b) const {
return !operator==(b);
}
Condition condition{};
s32 address{exit_branch};
bool kill{};
bool is_sync{};
bool is_brk{};
bool ignore{};
};
struct CaseBranch {
explicit CaseBranch(u32 cmp_value_, u32 address_) : cmp_value{cmp_value_}, address{address_} {}
u32 cmp_value;
u32 address;
};
class MultiBranch {
public:
explicit MultiBranch(u32 gpr_, std::vector<CaseBranch>&& branches_)
: gpr{gpr_}, branches{std::move(branches_)} {}
u32 gpr{};
std::vector<CaseBranch> branches{};
};
using BranchData = std::variant<SingleBranch, MultiBranch>;
using BlockBranchInfo = std::shared_ptr<BranchData>;
bool BlockBranchInfoAreEqual(BlockBranchInfo first, BlockBranchInfo second);
struct ShaderBlock {
u32 start{};
u32 end{};
bool ignore_branch{};
BlockBranchInfo branch{};
bool operator==(const ShaderBlock& sb) const {
return std::tie(start, end, ignore_branch) ==
std::tie(sb.start, sb.end, sb.ignore_branch) &&
BlockBranchInfoAreEqual(branch, sb.branch);
}
bool operator!=(const ShaderBlock& sb) const {
return !operator==(sb);
}
};
struct ShaderCharacteristics {
std::list<ShaderBlock> blocks{};
std::set<u32> labels{};
u32 start{};
u32 end{};
ASTManager manager{true, true};
CompilerSettings settings{};
};
std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
const CompilerSettings& settings,
Registry& registry);
} // namespace VideoCommon::Shader

Some files were not shown because too many files have changed in this diff Show More