raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 02:47:02 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: preserve folded native constants for descriptor provenance

Browse Source
This commit is contained in:
nmzik
2026-10-15 23:17:01 +02:00
parent e4019af522
commit c8caa27936
11 changed files with 111 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/ShaderRecompiler.cpp
+3 -4
View File
@@ -566,7 +566,7 @@ void PrepareCallTarget(ShaderSource& source, const CompileOptions& options) {
}
RewriteToSsa(query.blocks);
ConstantPropagationPass(query.blocks, query.wave_size);
RemoveIdentities(query.blocks);
RemoveIdentities(query);
EliminateDeadCode(query.blocks);
for (auto& value: source.call->target) value = value.Instruction()->Arg(0).Resolve();
query.value_storage.clear();
@@ -791,8 +791,7 @@ TranslateResult TranslateProgram(const Decoder::Program& decoded, const CompileO
IR::RewriteToSsa(ir.blocks);
const auto simplify = [&ir] {
IR::ConstantPropagationPass(ir.blocks, ir.wave_size);
IR::ResolveControlFlowIdentities(ir);
IR::RemoveIdentities(ir.blocks);
IR::RemoveIdentities(ir);
IR::EliminateDeadCode(ir.blocks);
};
simplify();
@@ -872,7 +871,7 @@ CompileResult CompileProgram(TranslateResult translated, const CompileOptions& o
}
}
ir.value_storage.clear();
IR::RemoveIdentities(ir.blocks);
IR::RemoveIdentities(ir);
IR::EliminateDeadCode(ir.blocks);
IR::CollectShaderInfo(ir, options.input_info);
src/graphics/shader/recompiler/Tessellation.cpp
+1 -1
View File
@@ -318,7 +318,7 @@ void LowerTessellationMemory(IR::Program& program, const CompileOptions& options
}
}
ConstantPropagationPass(program.blocks, program.wave_size);
RemoveIdentities(program.blocks);
RemoveIdentities(program);
EliminateDeadCode(program.blocks);
LOGF("%s tessellation lowering: reads=%u writes=%u factors=%u\n",
options.stage == ShaderType::Local ? "LS"
src/graphics/shader/recompiler/frontend/translate/Translate.cpp
+3 -3
View File
@@ -319,7 +319,7 @@ void Translator::WriteRawU32(const Decoder::Operand& operand, IR::U32 value) {
case Decoder::OperandKind::Sgpr: {
const auto reg = static_cast<IR::ScalarReg>(operand.reg);
ir.SetScalarReg(reg, value);
program.scalar_writes.push_back({current_pc, reg});
program.scalar_writes.push_back({current_pc, reg, value});
ir.SetScalarMaskTag(reg, IR::U1(IR::Value(false)));
if (IR::RegIndex(reg) > 0u) {
ir.SetScalarMaskTag(static_cast<IR::ScalarReg>(IR::RegIndex(reg) - 1u),
@@ -685,13 +685,13 @@ std::array<IR::U32, 2> Translator::WriteMask(const Decoder::Operand& operand, IR
IR::U1(IR::Value(false)));
}
ir.SetScalarReg(reg, mask[0]);
program.scalar_writes.push_back({current_pc, reg});
program.scalar_writes.push_back({current_pc, reg, mask[0]});
// A wave32 VALU mask destination must not overwrite the neighboring SGPR.
if ((write_64 || program.wave_size == 64u) &&
IR::RegIndex(reg) + 1u < IR::NumScalarRegs) {
const auto high = static_cast<IR::ScalarReg>(IR::RegIndex(reg) + 1u);
ir.SetScalarReg(high, mask[1]);
program.scalar_writes.push_back({current_pc, high});
program.scalar_writes.push_back({current_pc, high, mask[1]});
ir.SetThreadBitScalarReg(high, IR::U1(IR::Value(false)));
ir.SetScalarMaskTag(high, IR::U1(IR::Value(false)));
}
src/graphics/shader/recompiler/ir/Program.cpp
-7
View File
@@ -679,13 +679,6 @@ void ValidateProgram(const Program& program, bool require_ssa) {
}
}
void ResolveControlFlowIdentities(Program& program) {
for (auto* block: program.blocks) {
block->condition = block->condition.Resolve();
block->indirect_target = block->indirect_target.Resolve();
}
}
std::string ProgramToString(const Program& program) {
std::map<const Inst*, size_t> ids;
size_t next_id = 1;
src/graphics/shader/recompiler/ir/ShaderIR.h
+1 -2
View File
@@ -701,7 +701,7 @@ struct Program: ResourcePlan {
bool dispatcher_fallback = false;
CFG::FailureKind cfg_failure_kind = CFG::FailureKind::None;
std::string fallback_reason;
struct ScalarWrite { uint32_t pc; ScalarReg reg; };
struct ScalarWrite { uint32_t pc; ScalarReg reg; Value value; };
std::vector<ScalarWrite> scalar_writes;
// Typed memory and export instructions reference shader-local metadata by dense index.
// Decoder-only details (such as NSA register numbers) have already become IR operands.
@@ -718,7 +718,6 @@ std::string ProgramToString(const Program& program);
bool HasShaderMemoryWrites(const Program& program);
void ValidateProgram(const Program& program, bool require_ssa);
void ResolveControlFlowIdentities(Program& program);
bool EquivalentValue(const ResourcePlan& program, Value left, Value right);
bool EquivalentSelectorDomain(const ResourcePlan& program,
const DescriptorSource::IndirectDescriptor& left,
src/graphics/shader/recompiler/ir/passes/DeadCodeElimination.cpp
+12 -2
View File
@@ -1,11 +1,21 @@
#include "graphics/shader/recompiler/ir/passes/DeadCodeElimination.h"
#include "graphics/shader/recompiler/ir/ShaderIR.h"
#include <algorithm>
namespace Libs::Graphics::ShaderRecompiler::IR {
void RemoveIdentities(const BlockList& blocks) {
for (auto* block: blocks) {
void RemoveIdentities(Program& program) {
// Resolve metadata before deleting its identity nodes; retain only folded native constants.
for (auto& write: program.scalar_writes) {
const auto value = write.value.Resolve();
write.value = value.IsImmediate() && value.GetType() == Type::U32 ? value : Value {};
}
for (auto* block: program.blocks) {
block->condition = block->condition.Resolve();
block->indirect_target = block->indirect_target.Resolve();
}
for (auto* block: program.blocks) {
auto& instructions = block->Instructions();
for (auto inst = instructions.begin(); inst != instructions.end();) {
if (inst->GetOpcode() != ValueOpcode::Identity) {
src/graphics/shader/recompiler/ir/passes/DeadCodeElimination.h
+3 -1
View File
@@ -4,7 +4,9 @@
namespace Libs::Graphics::ShaderRecompiler::IR {
void RemoveIdentities(const BlockList& blocks);
struct Program;
void RemoveIdentities(Program& program);
void EliminateDeadCode(const BlockList& blocks);
} // namespace Libs::Graphics::ShaderRecompiler::IR
src/graphics/shader/recompiler/ir/passes/ResourceTracking.cpp
+15 -3
View File
@@ -604,9 +604,20 @@ private:
candidates.push_back(inst);
}
}
if (candidates.empty()) return value;
const auto source_at = [&](uint32_t pc) {
Value source;
bool dynamic = false;
for (auto write = std::ranges::lower_bound(m_scalar_writes, pc, {}, &Program::ScalarWrite::pc);
write != m_scalar_writes.end() && write->pc == pc; ++write) {
if (RegIndex(write->reg) != reg) continue;
if (!write->value.IsImmediate() || write->value.GetType() != Type::U32) {
dynamic = true;
continue;
}
if (!source.IsEmpty() && source != write->value) return Value {};
source = write->value;
}
if (!source.IsEmpty()) return dynamic ? Value {} : source;
const auto native = std::ranges::lower_bound(m_decoded.instructions, pc, {},
&Decoder::Instruction::pc);
for (const auto* candidate: candidates) {
@@ -644,8 +655,9 @@ private:
const auto source = source_at(pc);
if (source.IsEmpty()) return false;
if (!selected.IsEmpty()) {
const auto op = source.TryInstruction()->GetOpcode();
if ((op == ValueOpcode::LoadAddressU32 || op == ValueOpcode::ReadConstBuffer) &&
const auto* inst = source.TryInstruction();
if (inst != nullptr &&
(inst->GetOpcode() == ValueOpcode::LoadAddressU32 || inst->GetOpcode() == ValueOpcode::ReadConstBuffer) &&
selected_pc != pc) return false;
if (!EquivalentValue(m_program, selected, source)) return false;
}