raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-21 03:16:54 +02:00
Code Issues Packages Projects Releases Wiki Activity

gpu: store large compute shared memory in device buffers

Browse Source
This commit is contained in:
nmzik
2026-10-04 22:07:07 +04:00
parent 0ee8800905
commit b0bbaef1eb
17 changed files with 269 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/host_gpu/renderer/pipeline/descriptors.cpp
+6 -1
View File
@@ -68,7 +68,8 @@ vk::DescriptorType NativeDescriptorType(BindingKind kind) {
case BindingKind::BdaPagetable:
case BindingKind::FaultBuffer:
case BindingKind::FlattenedSrt:
case BindingKind::ShaderData: return vk::DescriptorType::eStorageBuffer;
case BindingKind::ShaderData:
case BindingKind::SharedMemory: return vk::DescriptorType::eStorageBuffer;
case BindingKind::Count: EXIT("invalid native descriptor binding kind");
}
EXIT("invalid native descriptor binding kind");
@@ -779,6 +780,7 @@ void RenderExecutor::PrepareBindings(const ShaderStageRuntime& runtime,
prepared.gds = {nullptr, 0, VK_WHOLE_SIZE};
prepared.flattened_srt = {};
prepared.shader_data_buffer = {};
prepared.shared_memory = {};
prepared.images.resize(program.info.images.size());
prepared.samplers.clear();
prepared.shader_data.clear();
@@ -1120,12 +1122,15 @@ void RenderExecutor::CommitBindings(CommandBuffer& buffer,
}
case BindingKind::FlattenedSrt:
case BindingKind::ShaderData:
case BindingKind::SharedMemory:
case BindingKind::Gds: {
const vk::DescriptorBufferInfo* view = &descriptors.gds;
if (binding.kind == BindingKind::FlattenedSrt) {
view = &descriptors.flattened_srt;
} else if (binding.kind == BindingKind::ShaderData) {
view = &descriptors.shader_data_buffer;
} else if (binding.kind == BindingKind::SharedMemory) {
view = &descriptors.shared_memory;
}
EXIT_IF(view->buffer == nullptr);
m_descriptor_buffers.push_back(*view);
src/graphics/host_gpu/renderer/pipeline/descriptors.h
+1
View File
@@ -44,6 +44,7 @@ struct PreparedBindings {
vk::DescriptorBufferInfo gds {nullptr, 0, VK_WHOLE_SIZE};
vk::DescriptorBufferInfo flattened_srt;
vk::DescriptorBufferInfo shader_data_buffer;
vk::DescriptorBufferInfo shared_memory;
std::vector<uint32_t> shader_data;
};
src/graphics/host_gpu/renderer/pipeline/pipelineCache.cpp
+2 -13
View File
@@ -680,19 +680,8 @@ ShaderProgram PipelineCache::GetComputeProgram(const HW::ComputeShaderInfo& regs
ShaderComputeInputInfo& input_info) {
input_info.host_subgroup_size = m_graphics.SupportsComputeWave64() ? 64u : 32u;
const auto params = PrepareProgram(regs, sh, input_info);
// Use one effective size for the cache key, LDS declaration, and access bounds.
const auto max_lds_dwords =
m_graphics.GetPhysicalDeviceProperties().limits.maxComputeSharedMemorySize / 4u;
if (input_info.lds_size_dwords > max_lds_dwords) {
static std::atomic_bool warned = false;
if (!warned.exchange(true, std::memory_order_relaxed)) {
PipelineCacheLog("GPU warning: game compute shader requests {} bytes of LDS, but "
"the Vulkan device limit is {} bytes. Clamping LDS; rendering may "
"be incorrect.",
input_info.lds_size_dwords * 4u, max_lds_dwords * 4u);
}
}
input_info.lds_size_dwords = std::min(input_info.lds_size_dwords, max_lds_dwords);
input_info.lds_storage = input_info.lds_size_dwords * 4u >
m_graphics.GetPhysicalDeviceProperties().limits.maxComputeSharedMemorySize;
uint32_t push_data_cursor = 0;
return m_program_cache->Get(params, input_info, push_data_cursor);
}