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: specialize bounded pointer and packed image tables

Browse Source
This commit is contained in:
nmzik
2026-10-13 01:12:19 +02:00
parent c9ec82e32e
commit 1fa79d5cac
5 changed files with 219 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ResourceTrackingTests.cpp
+72 -10
View File
@@ -570,7 +570,7 @@ void TestGuardedDirectImageTable() {
Finite, FiniteZero, FiniteMissing, FiniteWrong, FiniteSigned, FiniteAdd, FiniteXor,
FiniteShared, FiniteSharedBypass, FiniteLate
};
const auto make_plan = [](Guard guard, uint32_t operand_order = 0u) {
const auto make_plan = [](Guard guard, uint32_t operand_order = 0u, bool packed_key = false) {
const bool finite = guard >= Guard::Finite;
const bool bounded = guard >= Guard::Bounded && !finite;
const bool exec_bound = bounded && guard >= Guard::BoundExec;
@@ -751,11 +751,15 @@ void TestGuardedDirectImageTable() {
fixture.AddMemory(memory, 0x20));
}
fixture.block = descriptor_block;
if (!bounded && !finite) key = fixture.Emit(ValueOpcode::FindILsb32, {mask});
if (!bounded && !finite) key = fixture.Emit(packed_key ? ValueOpcode::ShiftRightLogical32
: ValueOpcode::FindILsb32,
packed_key ? std::initializer_list<Value>{mask, Value(12u)}
: std::initializer_list<Value>{mask});
auto scaled = bounded ? fixture.Emit(ValueOpcode::IMul32, {key, Value(table_stride)})
: fixture.Emit(ValueOpcode::ShiftLeftLogical32, {key, Value(5u)});
if (packed_key) scaled = fixture.Emit(ValueOpcode::BitwiseAnd32, {scaled, Value(992u)});
const auto offset = fixture.Emit(ValueOpcode::IAdd32,
{bounded ? fixture.Emit(ValueOpcode::IMul32, {key, Value(table_stride)})
: fixture.Emit(ValueOpcode::ShiftLeftLogical32, {key, Value(5u)}),
Value(table_offset)});
{scaled, Value(table_offset)});
std::array<Value, 8> words;
for (uint32_t group = 0; group < 2u; ++group) {
// Separate equivalent pointer handles mirror the two scalar x4 loads.
@@ -810,7 +814,7 @@ void TestGuardedDirectImageTable() {
fixture.Emit(ValueOpcode::ImageWrite, {output, fixture.ImageAddress(), texel, Value(true)},
fixture.AddMemory(memory, 0x130));
}
if (exec_bound) ConstantPropagationPass(fixture.program.blocks);
if (exec_bound || packed_key) ConstantPropagationPass(fixture.program.blocks);
if (guard == Guard::BoundExec || guard == Guard::BoundExecFactored) {
std::vector<const Inst *> pending{active.ResolveInstruction()};
uint32_t conjunctions = 0;
@@ -876,6 +880,23 @@ void TestGuardedDirectImageTable() {
.read_specialization_memory = ReadLinearTestMemory};
ResourceSnapshot snapshot;
ResourceSpecialization specialization;
auto packed = make_plan(Guard::Plain, 0u, true);
memory.fail_address = table + 32u * 32u;
Check(MaterializeResources(packed, runtime, snapshot, specialization) &&
memory.descriptor_reads == 32u && snapshot.images.size() == 32u,
"packed five-bit selector did not enumerate exactly its finite domain");
for (uint32_t key = 0; key < 32u; ++key)
memory.words[(table - memory.base) / 4u + key * 8u] = 0x100u;
memory.descriptor_reads = 0u;
Check(MaterializeResources(packed, runtime, snapshot, specialization) &&
memory.descriptor_reads == 32u && snapshot.images.size() == 1u &&
specialization.images[0].indirect_root == ImageResource::NoIndirectImage &&
specialization.images[0].indirect_search_iterations == 0u,
"packed five-bit selector retained duplicate image bindings or lost keys");
memory.fail_address = UINT64_MAX;
memory.reads = 0u;
memory.descriptor_reads = 0u;
fill_table(memory, table);
Check(MaterializeResources(plan, runtime, snapshot, specialization) &&
snapshot.images.size() == 32u && specialization.images.size() == 32u &&
snapshot.flattened_srt[specialization.images[0].indirect_mapping_offset] == 32u &&
@@ -1026,7 +1047,7 @@ void TestBoundedComputeImageLoop() {
Masked, MaskedWrongGuard, MaskResurrection, StatusOverwrite,
GuardedDiamond, GuardedDiamondBypass
};
const auto make_plan = [](Variant variant) {
const auto make_plan = [](Variant variant, bool material = false) {
const bool diamond = variant >= Variant::GuardedDiamond;
Fixture fixture(diamond ? ShaderType::Vertex : ShaderType::Compute);
fixture.program.wave_size = 64u;
@@ -1102,7 +1123,7 @@ void TestBoundedComputeImageLoop() {
{Value(static_cast<uint32_t>(diamond ? StageInputKind::VertexIndex
: StageInputKind::LocalInvocationId)), Value(0u)});
const auto count = variant == Variant::DivergentBound ? local : fixture.UserData(2);
const auto in_range = fixture.Emit(ValueOpcode::SLessThan32,
const auto in_range = fixture.Emit(material ? ValueOpcode::ULessThan32 : ValueOpcode::SLessThan32,
{variant == Variant::WrongGuard
? Value(0u) : key,
count}, 0, compare);
@@ -1249,8 +1270,17 @@ void TestBoundedComputeImageLoop() {
fixture.block = variant == Variant::GuardBlock ? header : body;
const auto table = fixture.Address(fixture.UserData(0), fixture.UserData(1));
auto image_key = key;
if (material) {
MemoryInfo memory;
memory.kind = ResourceKind::ScalarAddress;
memory.offset = 2756u;
image_key = fixture.Emit(ValueOpcode::LoadAddressU32,
{table, fixture.Emit(ValueOpcode::IMul32, {key, Value(48u)}), Value(0u), Value(true)},
fixture.AddMemory(memory, 0x2ff0));
}
const auto offset = fixture.Emit(ValueOpcode::IAdd32,
{fixture.Emit(ValueOpcode::ShiftLeftLogical32, {key, Value(5u)}),
{fixture.Emit(ValueOpcode::ShiftLeftLogical32, {image_key, Value(5u)}),
Value(0x6b0u)});
std::array<Value, 8> words;
for (uint32_t word = 0; word < words.size(); ++word) {
@@ -1273,10 +1303,14 @@ void TestBoundedComputeImageLoop() {
fixture.PlanAndTrack();
const auto source = fixture.program.info.images[0].source;
const auto &indirect = fixture.program.descriptor_sources[source].indirect_descriptor;
Check(indirect && !indirect->selector.has_value() &&
Check(indirect && indirect->selector.has_value() == material &&
indirect->table_offset == 0x6b0u &&
indirect->key_count.Resolve() == count.Resolve(),
"bounded compute loop lost its runtime image count");
if (material) Check(indirect->selector->stride == 48u &&
indirect->selector->offset == 2756u &&
fixture.program.descriptor_sources[indirect->selector->source].dword_count == 2u,
"bounded scalar selector lost its pointer or record layout");
return ExtractResourcePlan(fixture.program);
};
@@ -1360,6 +1394,34 @@ void TestBoundedComputeImageLoop() {
user_data[2] = 65537u;
Check(!MaterializeResources(plan, runtime, snapshot, specialization),
"oversized compute loop bound was accepted for image enumeration");
auto material = make_plan(Variant::Bounded, true);
for (const auto variant : {Variant::WrongGuard, Variant::EntryBypass, Variant::WrongStep})
CheckFatal([&] { make_plan(variant, true); }, "not a valid runtime value",
"raw pointer selector accepted an unbounded source");
const auto first_key = uint64_t{user_data[0]} + 2756u;
memory.words[(first_key - memory.base) / 4u] = 0u;
memory.words[(first_key + 48u - memory.base) / 4u] = 2u;
user_data[2] = 0u;
memory.reads = 0u;
memory.fail_address = first_key;
Check(MaterializeResources(material, runtime, snapshot, specialization) && memory.reads == 0u &&
snapshot.images.size() == 1u && snapshot.images[0].dwords[0] == 0u,
"empty raw pointer selector read an unreachable record");
user_data[2] = 2u;
memory.fail_address = first_key + 96u;
memory.descriptor_reads = 0u;
Check(MaterializeResources(material, runtime, snapshot, specialization) &&
memory.descriptor_reads == 2u && snapshot.images.size() == 2u &&
snapshot.images[0].dwords[0] == 0x100u && snapshot.images[1].dwords[0] == 0x102u,
"bounded raw pointer selector lost sparse keys or read an extra record");
memory.fail_address = first_key + 48u;
Check(!MaterializeResources(material, runtime, snapshot, specialization),
"raw pointer selector accepted an unreadable selected record");
memory.fail_address = UINT64_MAX;
memory.words[(first_key - memory.base) / 4u] = UINT32_MAX - 1000u;
Check(!MaterializeResources(material, runtime, snapshot, specialization),
"raw pointer selector silently discarded an invalid negative key");
}
void TestDescendingImageRange() {