raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-21 19:36:55 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: memoize bounded material record bit analysis

Browse Source
This commit is contained in:
nmzik
2026-10-13 06:18:34 +02:00
parent d08a6cef64
commit f1577a5a94
2 changed files with 87 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/ir/passes/ResourceTracking.cpp
+45 -22
View File
@@ -13,6 +13,7 @@
#include <optional>
#include <span>
#include <tuple>
#include <unordered_map>
#include <utility>
namespace Libs::Graphics::ShaderRecompiler::IR {
@@ -21,28 +22,44 @@ namespace {
constexpr uint32_t SamplerBorderClampMask = (1u << 2u) | (1u << 5u) | (1u << 8u);
constexpr uint32_t SamplerDword3ReservedMask = 0x3ffff000u;
uint32_t PossibleU32Bits(Value value) {
value = value.Resolve();
if (value.IsImmediate()) {
return value.GetType() == Type::U32 ? value.U32() : UINT32_MAX;
}
const auto* inst = value.TryInstruction();
if (inst == nullptr) {
return UINT32_MAX;
}
switch (inst->GetOpcode()) {
case ValueOpcode::BitwiseAnd32:
return PossibleU32Bits(inst->Arg(0)) & PossibleU32Bits(inst->Arg(1));
case ValueOpcode::BitwiseOr32:
return PossibleU32Bits(inst->Arg(0)) | PossibleU32Bits(inst->Arg(1));
case ValueOpcode::ShiftLeftLogical32: {
const auto shift = inst->Arg(1).Resolve();
return shift.IsImmediate() && shift.GetType() == Type::U32
? PossibleU32Bits(inst->Arg(0)) << (shift.U32() & 31u)
: UINT32_MAX;
uint32_t PossibleU32Bits(Value value, const auto& normalize) {
std::unordered_map<const Inst*, uint32_t> known;
const auto visit = [&](auto&& self, Value at) -> uint32_t {
at = normalize(at);
if (at.IsImmediate()) return at.GetType() == Type::U32 ? at.U32() : UINT32_MAX;
const auto* inst = at.TryInstruction();
if (inst == nullptr) return UINT32_MAX;
// Keep cycles unknown while sharing results across the SSA graph.
const auto [entry, inserted] = known.try_emplace(inst, UINT32_MAX);
if (!inserted) return entry->second;
auto& bits = entry->second;
const auto arg = [&](uint32_t index) { return self(self, inst->Arg(index)); };
switch (inst->GetOpcode()) {
case ValueOpcode::BitwiseAnd32: bits = arg(0) & arg(1); break;
case ValueOpcode::BitwiseOr32:
case ValueOpcode::UMax32: bits = arg(0) | arg(1); break;
case ValueOpcode::SelectU32: bits = arg(1) | arg(2); break;
case ValueOpcode::Phi: {
uint32_t merged = 0u;
for (uint32_t index = 0; index < inst->NumArgs(); ++index) merged |= arg(index);
if (inst->NumArgs() != 0u) bits = merged;
break;
}
case ValueOpcode::ShiftLeftLogical32: {
const auto shift = inst->Arg(1).Resolve();
if (shift.IsImmediate() && shift.GetType() == Type::U32)
bits = arg(0) << (shift.U32() & 31u);
break;
}
default: break;
}
default: return UINT32_MAX;
}
return bits;
};
return visit(visit, value);
}
uint32_t PossibleU32Bits(Value value) {
return PossibleU32Bits(value, [](Value at) { return at.Resolve(); });
}
bool IsNoncanonicalFlatAddress(Value high, Value active) {
@@ -1487,7 +1504,13 @@ private:
if (material_handle == nullptr ||
material_handle->GetOpcode() != ValueOpcode::GetAddressResource ||
!MakeRuntimeTableSource(*read, material_source)) return false;
const auto bits = PossibleU32Bits(ActiveValue(read->Arg(1), active));
const auto bits = PossibleU32Bits(read->Arg(1), [&](Value at) {
at = ActiveValue(at, active);
const auto* inst = at.TryInstruction();
// The current lane predicate cannot constrain a prior loop iteration.
return inst != nullptr && inst->GetOpcode() == ValueOpcode::Phi && InCycle(inst->Parent())
? Value{} : at;
});
const auto shift = std::countr_zero(bits);
const auto count = bits == 0u ? 0u : uint64_t {bits >> shift} + 1u;
if (shift >= 2u && count != 0u && count <= MaxIndirectDescriptorProbes &&