mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2025-08-05 00:42:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

control_flow: Avoid adding unnecessary divergant blocks

Browse Source
This commit is contained in:
IndecisiveTurtle 2024-12-05 02:30:31 +02:00
parent 6244d00795
commit 29b51567ea
2 changed files with 28 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
src/shader_recompiler/frontend/control_flow_graph.cpp
View File

@ -47,6 +47,15 @@ static IR::Condition MakeCondition(const GcnInst& inst) {
} }
} }
static bool IgnoresExecMask(Opcode opcode) {
switch (opcode) {
case Opcode::V_WRITELANE_B32:
return true;
default:
return false;
}
}
static constexpr size_t LabelReserveSize = 32; static constexpr size_t LabelReserveSize = 32;
CFG::CFG(Common::ObjectPool<Block>& block_pool_, std::span<const GcnInst> inst_list_) CFG::CFG(Common::ObjectPool<Block>& block_pool_, std::span<const GcnInst> inst_list_)
@ -133,6 +142,9 @@ void CFG::EmitDivergenceLabels() {
curr_begin = -1; curr_begin = -1;
continue; continue;
} }
// If all instructions in the scope ignore exec masking, we shouldn't insert a scope.
const auto start = inst_list.begin() + curr_begin + 1;
if (!std::ranges::all_of(start, inst_list.begin() + index, IgnoresExecMask, &GcnInst::opcode)) {
// Add a label to the instruction right after the open scope call. // Add a label to the instruction right after the open scope call.
// It is the start of a new basic block. // It is the start of a new basic block.
const auto& save_inst = inst_list[curr_begin]; const auto& save_inst = inst_list[curr_begin];
@ -148,6 +160,7 @@ void CFG::EmitDivergenceLabels() {
if (is_close) { if (is_close) {
AddLabel(index_to_pc[index]); AddLabel(index_to_pc[index]);
} }
}
// Reset scope begin. // Reset scope begin.
curr_begin = -1; curr_begin = -1;
} }

2
src/video_core/amdgpu/liverpool.cpp
View File

@ -702,7 +702,7 @@ Liverpool::Task Liverpool::ProcessCompute(std::span<const u32> acb, int vqid) {
false); false);
} else if (dma_data->src_sel == DmaDataSrc::Gds && } else if (dma_data->src_sel == DmaDataSrc::Gds &&
dma_data->dst_sel == DmaDataDst::Memory) { dma_data->dst_sel == DmaDataDst::Memory) {
LOG_WARNING(Render_Vulkan, "GDS memory read"); // LOG_WARNING(Render_Vulkan, "GDS memory read");
} else if (dma_data->src_sel == DmaDataSrc::Memory && } else if (dma_data->src_sel == DmaDataSrc::Memory &&
dma_data->dst_sel == DmaDataDst::Memory) { dma_data->dst_sel == DmaDataDst::Memory) {
rasterizer->InlineData(dma_data->DstAddress<VAddr>(), rasterizer->InlineData(dma_data->DstAddress<VAddr>(),