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: bound material selectors by proven one-wave LDS append lists

Browse Source
This commit is contained in:
nmzik
2026-10-12 19:44:05 +02:00
parent d692d8ee08
commit 0e9c77b0a2
8 changed files with 630 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp
+10 -50
View File
@@ -19,7 +19,6 @@ namespace Libs::Graphics::ShaderRecompiler::IR {
namespace {
constexpr uint64_t AddressMask = 0x0000ffffffffffffull;
constexpr uint64_t MaxIndirectDescriptorProbes = 65536u;
bool SpecializationFail(std::string_view message) {
std::fprintf(stderr, "shader resource specialization failed: %.*s\n",
@@ -782,55 +781,16 @@ static bool IsFullDispatchPredicate(Value value, uint32_t depth = 0) {
return axis.IsImmediate() && axis.U32() < 3 && extent->Arg(0).Resolve() == axis;
}
// Nonnegative affine coefficients for constant, local and workgroup coordinates. Reject modular
// arithmetic that could wrap; runtime coverage also bounds the largest invocation index.
static std::optional<std::array<uint64_t, 3>> FillIndex(Value value, uint32_t axis, Value guard,
uint32_t depth = 0) {
value = ResolveActiveU32(value, guard);
if (depth > 32 || value.GetType() != Type::U32) {
return {};
}
if (value.IsImmediate()) {
return std::array<uint64_t, 3> {value.U32(), 0, 0};
}
const auto* inst = value.TryInstruction();
if (inst == nullptr) {
return {};
}
const auto op = inst->GetOpcode();
if (op == ValueOpcode::GetBuiltin && inst->Arg(1) == Value(axis)) {
if (inst->Arg(0) == Value(static_cast<uint32_t>(StageInputKind::LocalInvocationId))) {
return std::array<uint64_t, 3> {0, 1, 0};
}
if (inst->Arg(0) == Value(static_cast<uint32_t>(StageInputKind::WorkgroupId))) {
return std::array<uint64_t, 3> {0, 0, 1};
}
}
if (op != ValueOpcode::IAdd32 && op != ValueOpcode::IMul32 &&
op != ValueOpcode::ShiftLeftLogical32) {
return {};
}
auto left = FillIndex(inst->Arg(0), axis, guard, depth + 1);
auto right = FillIndex(inst->Arg(1), axis, guard, depth + 1);
if (!left || !right) {
return {};
}
if (op == ValueOpcode::IMul32 && ((*right)[1] != 0 || (*right)[2] != 0)) {
std::swap(left, right);
}
if (op != ValueOpcode::IAdd32 && ((*right)[1] != 0 || (*right)[2] != 0)) {
return {};
}
if (op == ValueOpcode::ShiftLeftLogical32) {
if ((*right)[0] >= 32) return {};
(*right)[0] = uint64_t {1} << (*right)[0];
}
for (uint32_t i = 0; i < left->size(); ++i) {
(*left)[i] =
op == ValueOpcode::IAdd32 ? (*left)[i] + (*right)[i] : (*left)[i] * (*right)[0];
if ((*left)[i] > UINT32_MAX) return {};
}
return left;
static std::optional<std::array<uint64_t, 3>> FillIndex(Value value, uint32_t axis, Value guard) {
return AffineIndex<3>(value, [&](Value at) { return ResolveActiveU32(at, guard); },
[&](Value at) -> std::optional<size_t> {
const auto* inst = at.TryInstruction();
if (inst == nullptr || inst->GetOpcode() != ValueOpcode::GetBuiltin ||
inst->Arg(1) != Value(axis)) return {};
if (inst->Arg(0) == Value(static_cast<uint32_t>(StageInputKind::LocalInvocationId))) return 1u;
if (inst->Arg(0) == Value(static_cast<uint32_t>(StageInputKind::WorkgroupId))) return 2u;
return {};
});
}
static UniformFillPlan AnalyzeUniformFill(const Program& program) {