Skip to content

fix(agc): drop layer I/O when linking mesh pipelines in driver to prevent AMD crash #6379

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
Feuris wants to merge 1 commit into from
Closed

fix(agc): drop layer I/O when linking mesh pipelines in driver to prevent AMD crash #6379

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
155 changes: 153 additions & 2 deletions core/libs/prx/libSceAgcDriver/Graphics/src/Pipeline.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@
#include "prx/libSceAgcDriver/Graphics/include/PipelineSpecialization.hpp"
#include "prx/libSceAgcDriver/Graphics/include/VertexInput.hpp"
#include "prx/libc/include/HostThreadLocal.hpp"
#include <spirv/unified1/spirv.hpp>
#include <algorithm>
#include <array>
#include <chrono>
Expand All @@ -18,6 +19,7 @@
#include <string_view>
#include <type_traits>
#include <unordered_map>
#include <unordered_set>
#include <vector>

namespace AgcDriver::Graphics {
Expand Down Expand Up @@ -287,6 +289,148 @@ bool libraryKeys(PipelineLibraryKeys& keys, const Context& context, const State&
return true;
}

std::vector<std::uint32_t> DropFragmentLayerInput(std::span<const std::uint32_t> spirv) {
std::uint32_t layerId = 0;
std::uint32_t zeroId = 0;
std::unordered_map<std::uint32_t, std::uint32_t> zeros;
for (std::size_t i = 5; i < spirv.size();) {
const auto word = spirv[i];
const auto count = word >> 16u;
const auto op = word & 0xffffu;
if (count == 0 || i + count > spirv.size()) break;
if (op == spv::OpDecorate && count == 4 && spirv[i + 2] == spv::DecorationBuiltIn && spirv[i + 3] == spv::BuiltInLayer) {
layerId = spirv[i + 1];
}
if (op == spv::OpConstant && count == 4 && spirv[i + 3] == 0) {
zeros[spirv[i + 1]] = spirv[i + 2];
if (zeroId == 0) zeroId = spirv[i + 2];
} else if (op == spv::OpConstantNull && count == 3) {
zeros[spirv[i + 1]] = spirv[i + 2];
if (zeroId == 0) zeroId = spirv[i + 2];
}
i += count;
}
if (layerId == 0 || zeroId == 0) return {};
std::vector<std::uint32_t> words(spirv.begin(), spirv.end());
for (std::size_t i = 5; i < words.size();) {
const auto word = words[i];
const auto count = word >> 16u;
const auto op = word & 0xffffu;
if (count == 0 || i + count > words.size()) break;
if (op == spv::OpEntryPoint && count >= 4) {
const auto* nameStart = reinterpret_cast<const char*>(&words[i + 3]);
const auto maxLen = (count - 3) * sizeof(std::uint32_t);
std::size_t len = 0;
while (len < maxLen && nameStart[len] != '\0') ++len;
const auto nameWords = (len + 4) / 4;
const auto interfaceStart = words.begin() + i + 3 + nameWords;
if (interfaceStart <= words.begin() + i + count) {
const auto it = std::find(interfaceStart, words.begin() + i + count, layerId);
if (it != words.begin() + i + count) {
std::move(it + 1, words.begin() + i + count, it);
words[i] = ((count - 1u) << 16u) | spv::OpEntryPoint;
words[i + count - 1u] = 1u << 16u;
}
}
} else if (op == spv::OpLoad && count == 4 && words[i + 3] == layerId) {
const auto typeId = words[i + 1];
const auto it = zeros.find(typeId);
const auto copyId = it != zeros.end() ? it->second : zeroId;
words[i] = (4u << 16u) | spv::OpCopyObject;
words[i + 3] = copyId;
} else if (op == spv::OpVariable && count >= 3 && words[i + 2] == layerId) {
std::fill_n(words.begin() + i, count, 0u);
words[i] = count << 16u;
} else if ((op == spv::OpDecorate || op == spv::OpName) && count >= 2 && words[i + 1] == layerId) {
std::fill_n(words.begin() + i, count, 0u);
words[i] = count << 16u;
}
i += count;
}
return words;
}

std::vector<std::uint32_t> DropMeshLayerOutput(std::span<const std::uint32_t> spirv) {
std::uint32_t layerId = 0;
for (std::size_t i = 5; i < spirv.size();) {
const auto word = spirv[i];
const auto count = word >> 16u;
const auto op = word & 0xffffu;
if (count == 0 || i + count > spirv.size()) break;
if (op == spv::OpDecorate && count == 4 && spirv[i + 2] == spv::DecorationBuiltIn && spirv[i + 3] == spv::BuiltInLayer) {
layerId = spirv[i + 1];
break;
}
i += count;
}
if (layerId == 0) return {};
std::vector<std::uint32_t> words(spirv.begin(), spirv.end());
std::unordered_set<std::uint32_t> pointerIds;
for (std::size_t i = 5; i < words.size();) {
const auto word = words[i];
const auto count = word >> 16u;
const auto op = word & 0xffffu;
if (count == 0 || i + count > words.size()) break;
if (op == spv::OpEntryPoint && count >= 4) {
const auto* nameStart = reinterpret_cast<const char*>(&words[i + 3]);
const auto maxLen = (count - 3) * sizeof(std::uint32_t);
std::size_t len = 0;
while (len < maxLen && nameStart[len] != '\0') ++len;
const auto nameWords = (len + 4) / 4;
const auto interfaceStart = words.begin() + i + 3 + nameWords;
if (interfaceStart <= words.begin() + i + count) {
const auto it = std::find(interfaceStart, words.begin() + i + count, layerId);
if (it != words.begin() + i + count) {
std::move(it + 1, words.begin() + i + count, it);
words[i] = ((count - 1u) << 16u) | spv::OpEntryPoint;
words[i + count - 1u] = 1u << 16u;
}
}
} else if ((op == spv::OpDecorate || op == spv::OpName) && count >= 2 && words[i + 1] == layerId) {
std::fill_n(words.begin() + i, count, 0u);
words[i] = count << 16u;
} else if (op == spv::OpVariable && count >= 3 && words[i + 2] == layerId) {
std::fill_n(words.begin() + i, count, 0u);
words[i] = count << 16u;
} else if (op == spv::OpAccessChain && count >= 4 && words[i + 3] == layerId) {
pointerIds.insert(words[i + 2]);
std::fill_n(words.begin() + i, count, 0u);
words[i] = count << 16u;
} else if (op == spv::OpStore && count >= 3 && pointerIds.contains(words[i + 1])) {
std::fill_n(words.begin() + i, count, 0u);
words[i] = count << 16u;
}
i += count;
}
return words;
}

bool StageDeclaresLayer(std::span<const std::uint32_t> spirv) {
for (std::size_t i = 5; i < spirv.size();) {
const auto word = spirv[i];
const auto count = word >> 16u;
const auto op = word & 0xffffu;
if (count == 0 || i + count > spirv.size()) break;
if (op == spv::OpDecorate && count == 4 && spirv[i + 2] == spv::DecorationBuiltIn && spirv[i + 3] == spv::BuiltInLayer) {
return true;
}
if (op == spv::OpMemberDecorate && count == 5 && spirv[i + 3] == spv::DecorationBuiltIn && spirv[i + 4] == spv::BuiltInLayer) {
return true;
}
i += count;
}
return false;
}

bool PrecedingStageExportsLayer(std::span<const CompiledShader> shaders, const State& state) {
if (state.renderLayers <= 1 || state.layerExports == 0) return false;
for (const auto& shader : shaders) {
if (shader.stage == ShaderRecompiler::ShaderStage::Fragment) break;
if (StageDeclaresLayer(shader.program->spirv)) return true;
}
return false;
}

}

Pipeline::Pipeline(const Context& context, const State& state, const VertexInputLayout& vertexInput, const ShaderResources& resources, std::span<const CompiledShader> shaders, VkImageLayout attachmentLayout) : Pipeline(context, state, vertexInput, resources.SetLayouts(), shaders, attachmentLayout) {}
Expand Down Expand Up @@ -334,9 +478,16 @@ Pipeline::Pipeline(const Context& context, const State& state, const VertexInput
const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT meshSubgroup{VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT, nullptr, 32u};
for (std::uint32_t i = 0; i < shaders.size(); ++i) {
const auto& shader = *shaders[i].program;
std::vector<std::uint32_t> sanitizedSpirv;
if (shaders[i].stage == ShaderRecompiler::ShaderStage::Mesh) {
sanitizedSpirv = DropMeshLayerOutput(shader.spirv);
} else if (shaders[i].stage == ShaderRecompiler::ShaderStage::Fragment && (state.stages.mesh.has_value() || !PrecedingStageExportsLayer(shaders, state))) {
sanitizedSpirv = DropFragmentLayerInput(shader.spirv);
}
const auto& spirv = sanitizedSpirv.empty() ? shader.spirv : sanitizedSpirv;
VkShaderModuleCreateInfo module{VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO};
module.codeSize = shader.spirv.size() * sizeof(std::uint32_t);
module.pCode = shader.spirv.data();
module.codeSize = spirv.size() * sizeof(std::uint32_t);
module.pCode = spirv.data();
Check(context.Function<PFN_vkCreateShaderModule>("vkCreateShaderModule")(context.device, &module, nullptr, &_modules[i]), "vkCreateShaderModule graphics");
const auto stage = VulkanStage(shaders[i].stage);
stages[i].sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
Expand Down
Loading