lux/yuzu
1
0
Fork 0
mirror of https://git.h3cjp.net/H3cJP/yuzu.git synced 2024-11-15 02:43:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

astc: Various robustness enhancements for the gpu decoder

Browse source 
These changes should help in reducing crashes/drivers panics that may
occur due to synchronization issues between the shader completion and
later access of the decoded texture.
This commit is contained in:
ameerj 2021-06-19 00:55:13 -04:00
parent 3870ba670f
commit 31b125ef57
5 changed files with 16 additions and 47 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
src/video_core/host_shaders/astc_decoder.comp
View file

@ -14,9 +14,8 @@
#define BINDING_6_TO_8_BUFFER 2
#define BINDING_7_TO_8_BUFFER 3
#define BINDING_8_TO_8_BUFFER 4
#define BINDING_BYTE_TO_16_BUFFER 5
#define BINDING_SWIZZLE_BUFFER 6
#define BINDING_OUTPUT_IMAGE 7
#define BINDING_SWIZZLE_BUFFER 5
#define BINDING_OUTPUT_IMAGE 6
#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
@ -29,7 +28,6 @@
#define BINDING_6_TO_8_BUFFER 3
#define BINDING_7_TO_8_BUFFER 4
#define BINDING_8_TO_8_BUFFER 5
#define BINDING_BYTE_TO_16_BUFFER 6
#define BINDING_OUTPUT_IMAGE 0
#endif
@ -86,9 +84,6 @@ layout(binding = BINDING_7_TO_8_BUFFER, std430) readonly buffer REPLICATE_7_BIT_
layout(binding = BINDING_8_TO_8_BUFFER, std430) readonly buffer REPLICATE_8_BIT_TO_8 {
uint REPLICATE_8_BIT_TO_8_TABLE[];
};
layout(binding = BINDING_BYTE_TO_16_BUFFER, std430) readonly buffer REPLICATE_BYTE_TO_16 {
uint REPLICATE_BYTE_TO_16_TABLE[];
};
layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly image2DArray dest_image;
@ -207,8 +202,7 @@ uint Replicate(uint val, uint num_bits, uint to_bit) {
}
uvec4 ReplicateByteTo16(uvec4 value) {
return uvec4(REPLICATE_BYTE_TO_16_TABLE[value.x], REPLICATE_BYTE_TO_16_TABLE[value.y],
REPLICATE_BYTE_TO_16_TABLE[value.z], REPLICATE_BYTE_TO_16_TABLE[value.w]);
return value * 0x101;
}
uint ReplicateBitTo7(uint value) {
@ -1327,6 +1321,9 @@ void main() {
offset += swizzle;
const ivec3 coord = ivec3(gl_GlobalInvocationID * uvec3(block_dims, 1));
if (any(greaterThanEqual(coord, imageSize(dest_image)))) {
return;
}
uint block_index =
pos.z * gl_WorkGroupSize.x * gl_WorkGroupSize.y + pos.y * gl_WorkGroupSize.x + pos.x;

5
src/video_core/renderer_opengl/util_shaders.cpp
View file

@ -83,7 +83,6 @@ void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
static constexpr GLuint BINDING_6_TO_8_BUFFER = 3;
static constexpr GLuint BINDING_7_TO_8_BUFFER = 4;
static constexpr GLuint BINDING_8_TO_8_BUFFER = 5;
static constexpr GLuint BINDING_BYTE_TO_16_BUFFER = 6;
static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
@ -105,9 +104,6 @@ void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_8_TO_8_BUFFER, astc_buffer.handle,
offsetof(AstcBufferData, replicate_8_to_8),
sizeof(AstcBufferData::replicate_8_to_8));
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_BYTE_TO_16_BUFFER, astc_buffer.handle,
offsetof(AstcBufferData, replicate_byte_to_16),
sizeof(AstcBufferData::replicate_byte_to_16));
glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes);
glUniform2ui(1, tile_size.width, tile_size.height);
@ -137,6 +133,7 @@ void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
glDispatchCompute(num_dispatches_x, num_dispatches_y, image.info.resources.layers);
}
glMemoryBarrier(GL_ALL_BARRIER_BITS);
program_manager.RestoreGuestCompute();
}

39
src/video_core/renderer_vulkan/vk_compute_pass.cpp
View file

@ -40,9 +40,9 @@ constexpr u32 ASTC_BINDING_ENC_BUFFER = 1;
constexpr u32 ASTC_BINDING_6_TO_8_BUFFER = 2;
constexpr u32 ASTC_BINDING_7_TO_8_BUFFER = 3;
constexpr u32 ASTC_BINDING_8_TO_8_BUFFER = 4;
constexpr u32 ASTC_BINDING_BYTE_TO_16_BUFFER = 5;
constexpr u32 ASTC_BINDING_SWIZZLE_BUFFER = 6;
constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 7;
constexpr u32 ASTC_BINDING_SWIZZLE_BUFFER = 5;
constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 6;
constexpr size_t ASTC_NUM_BINDINGS = 7;
VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
return {
@ -71,7 +71,7 @@ std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBinding
}};
}
std::array<VkDescriptorSetLayoutBinding, 8> BuildASTCDescriptorSetBindings() {
std::array<VkDescriptorSetLayoutBinding, ASTC_NUM_BINDINGS> BuildASTCDescriptorSetBindings() {
return {{
{
.binding = ASTC_BINDING_INPUT_BUFFER,
@ -108,13 +108,6 @@ std::array<VkDescriptorSetLayoutBinding, 8> BuildASTCDescriptorSetBindings() {
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_BYTE_TO_16_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,
@ -143,7 +136,8 @@ VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
};
}
std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateTemplateEntry() {
std::array<VkDescriptorUpdateTemplateEntryKHR, ASTC_NUM_BINDINGS>
BuildASTCPassDescriptorUpdateTemplateEntry() {
return {{
{
.dstBinding = ASTC_BINDING_INPUT_BUFFER,
@ -185,14 +179,6 @@ std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateT
.offset = ASTC_BINDING_8_TO_8_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_BYTE_TO_16_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = ASTC_BINDING_BYTE_TO_16_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_SWIZZLE_BUFFER,
.dstArrayElement = 0,
@ -222,15 +208,6 @@ struct AstcPushConstants {
u32 block_height_mask;
};
struct AstcBufferData {
decltype(SWIZZLE_TABLE) swizzle_table_buffer = SWIZZLE_TABLE;
decltype(EncodingsValues) encoding_values = EncodingsValues;
decltype(REPLICATE_6_BIT_TO_8_TABLE) replicate_6_to_8 = REPLICATE_6_BIT_TO_8_TABLE;
decltype(REPLICATE_7_BIT_TO_8_TABLE) replicate_7_to_8 = REPLICATE_7_BIT_TO_8_TABLE;
decltype(REPLICATE_8_BIT_TO_8_TABLE) replicate_8_to_8 = REPLICATE_8_BIT_TO_8_TABLE;
decltype(REPLICATE_BYTE_TO_16_TABLE) replicate_byte_to_16 = REPLICATE_BYTE_TO_16_TABLE;
} constexpr ASTC_BUFFER_DATA;
} // Anonymous namespace
VKComputePass::VKComputePass(const Device& device, VKDescriptorPool& descriptor_pool,
@ -517,9 +494,6 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
sizeof(AstcBufferData::replicate_7_to_8));
update_descriptor_queue.AddBuffer(*data_buffer, offsetof(AstcBufferData, replicate_8_to_8),
sizeof(AstcBufferData::replicate_8_to_8));
update_descriptor_queue.AddBuffer(*data_buffer,
offsetof(AstcBufferData, replicate_byte_to_16),
sizeof(AstcBufferData::replicate_byte_to_16));
update_descriptor_queue.AddBuffer(*data_buffer, sizeof(AstcBufferData),
sizeof(SWIZZLE_TABLE));
update_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
@ -569,6 +543,7 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, image_barrier);
});
scheduler.Finish();
}
} // namespace Vulkan

2
src/video_core/textures/astc.cpp
View file

@ -551,6 +551,8 @@ static void FillError(std::span<u32> outBuf, u32 blockWidth, u32 blockHeight) {
}
}
}
static constexpr auto REPLICATE_BYTE_TO_16_TABLE = MakeReplicateTable<u32, 8, 16>();
static constexpr u32 ReplicateByteTo16(std::size_t value) {
return REPLICATE_BYTE_TO_16_TABLE[value];
}

2
src/video_core/textures/astc.h
View file

@ -116,7 +116,6 @@ constexpr auto MakeReplicateTable() {
return table;
}
constexpr auto REPLICATE_BYTE_TO_16_TABLE = MakeReplicateTable<u32, 8, 16>();
constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
@ -126,7 +125,6 @@ struct AstcBufferData {
decltype(REPLICATE_6_BIT_TO_8_TABLE) replicate_6_to_8 = REPLICATE_6_BIT_TO_8_TABLE;
decltype(REPLICATE_7_BIT_TO_8_TABLE) replicate_7_to_8 = REPLICATE_7_BIT_TO_8_TABLE;
decltype(REPLICATE_8_BIT_TO_8_TABLE) replicate_8_to_8 = REPLICATE_8_BIT_TO_8_TABLE;
decltype(REPLICATE_BYTE_TO_16_TABLE) replicate_byte_to_16 = REPLICATE_BYTE_TO_16_TABLE;
} constexpr ASTC_BUFFER_DATA;
void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,