raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-22 20:06:54 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: combine image execution masks with descriptor branch bounds

Browse Source
This commit is contained in:
nmzik
2026-10-13 01:27:57 +02:00
parent 1fa79d5cac
commit 9a3aca963a
3 changed files with 87 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/ir/passes/ResourceTracking.cpp
+18 -7
View File
@@ -1655,16 +1655,24 @@ private:
}
Value UnsignedBound(Value guard, Value value, bool positive = true,
bool finite_bitscan = false) const {
bool finite_bitscan = false, Value witness = {}) const {
const auto* test = SimplifyGuard(guard).TryInstruction();
if (test == nullptr) return {};
const auto op = test->GetOpcode();
if (op == ValueOpcode::LogicalNot)
return UnsignedBound(test->Arg(0), value, !positive, finite_bitscan);
return UnsignedBound(test->Arg(0), value, !positive, finite_bitscan, witness);
if ((op == ValueOpcode::LogicalAnd && positive) ||
(op == ValueOpcode::LogicalOr && !positive)) {
const auto left = UnsignedBound(test->Arg(0), value, positive, finite_bitscan);
return left.IsEmpty() ? UnsignedBound(test->Arg(1), value, positive, finite_bitscan) : left;
const auto left = UnsignedBound(test->Arg(0), value, positive, finite_bitscan, witness);
return left.IsEmpty() ? UnsignedBound(test->Arg(1), value, positive, finite_bitscan, witness) : left;
}
if (op == ValueOpcode::LogicalAnd && !positive && !witness.IsEmpty()) {
// !(A & Q) proves !Q only for accesses whose EXEC also proves A.
for (uint32_t arm = 0; arm < 2u; ++arm) {
if (!Implies(witness, test->Arg(arm))) continue;
const auto bound = UnsignedBound(test->Arg(arm ^ 1u), value, false, finite_bitscan, witness);
if (!bound.IsEmpty()) return bound;
}
}
if ((positive ? op == ValueOpcode::ULessThan32 : op == ValueOpcode::UGreaterThanEqual32) &&
EquivalentValue(m_program, test->Arg(0), value)) return test->Arg(1);
@@ -1684,7 +1692,8 @@ private:
return {};
}
Value UnsignedBoundOnEntry(Value key, const Block* block, bool finite_bitscan = false) const {
Value UnsignedBoundOnEntry(Value key, const Block* block, bool finite_bitscan = false,
Value witness = {}) const {
const auto* definition = key.Resolve().TryInstruction();
// A mask-nonzero guard may precede FF1. Its proof lasts until the mask's
// definition, including a loop Phi that changes it for the next iteration.
@@ -1695,7 +1704,7 @@ private:
return definition != nullptr && current == definition->Parent();
}, [&](const EdgePredicate& edge) {
if (edge.lanes != LaneQuantifier::All) return false;
const auto candidate = UnsignedBound(edge.condition, key, edge.positive, finite_bitscan);
const auto candidate = UnsignedBound(edge.condition, key, edge.positive, finite_bitscan, witness);
if (candidate.IsEmpty() ||
!ValidateRuntimeValue(m_program, candidate, RuntimeValueType::Integer)) return false;
if (bound.IsEmpty()) bound = candidate;
@@ -1714,7 +1723,9 @@ private:
op != ValueOpcode::ImageGatherRaw && op != ValueOpcode::ImageRead)) return {};
// The image's own EXEC predicate proves the range for every accessing lane;
// an EXEC-nonzero branch alone only proves that some lane is active.
const auto candidate = UnsignedBound(image.Arg(image.NumArgs() - 1u), key);
const auto active = image.Arg(image.NumArgs() - 1u);
auto candidate = UnsignedBound(active, key);
if (candidate.IsEmpty()) candidate = UnsignedBoundOnEntry(key, image.Parent(), false, active);
if (candidate.IsEmpty() ||
!ValidateRuntimeValue(m_program, candidate, RuntimeValueType::Integer)) return {};
if (bound.IsEmpty()) bound = candidate;