raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-21 03:16:54 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: read image descriptors indexed by workgroup

Browse Source 
Co-authored-by: Çağatay Güçlü <33038942+defektu@users.noreply.github.com>
This commit is contained in:
nmzik
2026-10-04 22:06:58 +04:00
parent 0bef3fc0d3
commit 0ee8800905
9 changed files with 187 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/ir/ShaderIR.h
+2
View File
@@ -488,6 +488,8 @@ struct DescriptorSource {
uint32_t selector_stride = 0;
uint32_t selector_offset = 0;
uint32_t table_offset = 0;
uint32_t table_stride = 0;
uint32_t workgroup_axis = UINT32_MAX;
uint32_t selector_shift = 0;
Value key_count;
Value selector_first;
src/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp
+14 -7
View File
@@ -285,12 +285,19 @@ bool MaterializeIndirectDescriptor(const ResourcePlan& p
}
if (indirect.material_source == UINT32_MAX) {
uint32_t key_count = 0;
const bool evaluated = clean.Evaluate(indirect.key_count, key_count);
if (std::bit_cast<int32_t>(key_count) <= 0) key_count = 0;
if (table_value.dword_count != 2u || !evaluated ||
key_count > MaxIndirectDescriptorProbes ||
uint64_t {indirect.table_offset} + uint64_t {key_count} * descriptor_bytes >
UINT32_MAX + 1ull) {
if (indirect.workgroup_axis != UINT32_MAX) {
if (indirect.workgroup_axis >= runtime.workgroup_counts.size()) return false;
key_count = runtime.workgroup_counts[indirect.workgroup_axis];
if (key_count == 0u) return false;
} else {
if (table_value.dword_count != 2u || !clean.Evaluate(indirect.key_count, key_count))
return false;
if (std::bit_cast<int32_t>(key_count) <= 0) key_count = 0;
}
if (indirect.table_stride == 0u || key_count > MaxIndirectDescriptorProbes ||
(key_count != 0u && uint64_t {indirect.table_offset} +
uint64_t {key_count - 1u} * indirect.table_stride + descriptor_bytes >
UINT32_MAX + 1ull)) {
return false;
}
keys.resize(key_count);
@@ -376,7 +383,7 @@ bool MaterializeIndirectDescriptor(const ResourcePlan& p
candidate.dword_count = dword_count;
if (sources.empty()) {
const auto table_offset =
static_cast<uint32_t>(key * descriptor_bytes) + indirect.table_offset;
static_cast<uint32_t>(key * indirect.table_stride) + indirect.table_offset;
if (!ReadScalarTable(table_base, table_size, table_offset, runtime,
std::span(candidate.dwords).first(dword_count))) {
return false;
src/graphics/shader/recompiler/ir/passes/ResourceTracking.cpp
+45 -10
View File
@@ -834,7 +834,8 @@ private:
if (a.material_source != b.material_source || a.table_source != b.table_source ||
a.selector_stride != b.selector_stride || a.selector_offset != b.selector_offset ||
a.selector_shift != b.selector_shift ||
a.table_offset != b.table_offset || a.sources != b.sources ||
a.table_offset != b.table_offset || a.table_stride != b.table_stride ||
a.workgroup_axis != b.workgroup_axis || a.sources != b.sources ||
!EquivalentValue(m_program, a.key_count, b.key_count) ||
a.selector_first.IsEmpty() != b.selector_first.IsEmpty() ||
(!a.selector_first.IsEmpty() &&
@@ -1010,7 +1011,17 @@ private:
return false;
}
bool MatchTableOffset(Value value, Value& key, uint32_t& offset, uint32_t shift = 5u) const {
uint32_t WorkgroupAxis(Value key) const {
const auto* builtin = key.Resolve().TryInstruction();
uint32_t axis = UINT32_MAX;
if (m_program.stage != ShaderType::Compute || builtin == nullptr ||
builtin->GetOpcode() != ValueOpcode::GetBuiltin || builtin->NumArgs() != 2u ||
builtin->Arg(0) != Value(static_cast<uint32_t>(StageInputKind::WorkgroupId)) ||
!ImmediateU32(builtin->Arg(1), axis) || axis >= 3u) return UINT32_MAX;
return axis;
}
bool MatchTableOffset(Value value, Value& key, uint32_t& offset, uint32_t& stride) const {
offset = 0;
for (;;) {
const auto* inst = value.Resolve().TryInstruction();
@@ -1019,10 +1030,17 @@ private:
}
uint32_t immediate;
if (inst->GetOpcode() == ValueOpcode::ShiftLeftLogical32 &&
ImmediateU32(inst->Arg(1), immediate) && immediate == shift) {
ImmediateU32(inst->Arg(1), immediate) && immediate < 32u) {
key = inst->Arg(0).Resolve();
stride = 1u << immediate;
return key.GetType() == Type::U32;
}
if (inst->GetOpcode() == ValueOpcode::IMul32) {
if (ImmediateU32(inst->Arg(0), stride)) key = inst->Arg(1).Resolve();
else if (ImmediateU32(inst->Arg(1), stride)) key = inst->Arg(0).Resolve();
else return false;
return stride != 0u && key.GetType() == Type::U32;
}
if (inst->GetOpcode() != ValueOpcode::IAdd32) {
return false;
}
@@ -1563,7 +1581,8 @@ private:
bool MatchDescriptorTable(Inst& handle, const DescriptorSource& descriptor,
IndirectDescriptorPlan& plan,
DescriptorSource& table_source, Value& key, uint32_t& table_offset) {
DescriptorSource& table_source, Value& key, uint32_t& table_offset,
uint32_t& table_stride) {
Inst* table_handle = nullptr;
for (uint32_t dword = 0; dword < handle.NumArgs(); ++dword) {
auto* read = descriptor.dwords[dword].Resolve().TryInstruction();
@@ -1579,6 +1598,7 @@ private:
auto* current_handle = read->Arg(0).Resolve().TryInstruction();
Value current_key;
uint32_t offset = 0;
uint32_t stride = 0;
if (current_handle == nullptr ||
current_handle->GetOpcode() != (memory->kind == ResourceKind::ScalarAddress
? ValueOpcode::GetAddressResource
@@ -1586,7 +1606,7 @@ private:
(memory->kind == ResourceKind::ScalarAddress && read->Parent() != handle.Parent()) ||
(table_handle != nullptr &&
!EquivalentValue(m_program, Value(table_handle), Value(current_handle))) ||
!MatchTableOffset(read->Arg(1), current_key, offset, handle.NumArgs() == 4u ? 4u : 5u) ||
!MatchTableOffset(read->Arg(1), current_key, offset, stride) ||
memory->offset > UINT32_MAX - offset) {
return false;
}
@@ -1594,12 +1614,14 @@ private:
if (dword == 0u) {
key = current_key;
table_offset = offset;
} else if (!EquivalentValue(m_program, key, current_key) ||
table_stride = stride;
} else if (table_stride != stride || !EquivalentValue(m_program, key, current_key) ||
static_cast<uint64_t>(table_offset) + dword * sizeof(uint32_t) != offset) {
return false;
}
table_handle = current_handle;
if (handle.NumArgs() == 8u ? !UsesOnlyImageDescriptors(*read) :
if (handle.NumArgs() == 8u
? WorkgroupAxis(key) == UINT32_MAX && !UsesOnlyImageDescriptors(*read) :
std::ranges::any_of(read->Uses(), [](const Use& use) {
return use.user->GetOpcode() != ValueOpcode::GetBufferResource ||
std::ranges::any_of(use.user->Uses(), [](const Use& consumer) {
@@ -1654,11 +1676,21 @@ private:
DescriptorSource table_source;
Value key;
uint32_t table_offset = 0;
if (!MatchDescriptorTable(handle, descriptor, plan, table_source, key, table_offset)) return false;
uint32_t table_stride = 0;
if (!MatchDescriptorTable(handle, descriptor, plan, table_source, key, table_offset,
table_stride)) return false;
DescriptorSource material_source;
DescriptorSource::IndirectDescriptor indirect;
indirect.table_offset = table_offset;
if (table_source.dword_count == 2u) {
indirect.table_stride = table_stride;
indirect.workgroup_axis = WorkgroupAxis(key);
if (indirect.workgroup_axis != UINT32_MAX) {
// The descriptor also supplies dimensions to shader arithmetic. Keep its reads;
// only the image handle is projected onto the bounded workgroup key.
plan.reads.fill(nullptr);
} else if (table_stride != 32u) {
return false;
} else if (table_source.dword_count == 2u) {
const auto* selector = key.Resolve().TryInstruction();
const bool bitscan = selector != nullptr && selector->GetOpcode() == ValueOpcode::FindILsb32 &&
selector->NumArgs() == 1u && !m_shader_writes &&
@@ -1733,7 +1765,9 @@ private:
DescriptorSource table_source;
Value key;
uint32_t table_offset = 0;
if (!MatchDescriptorTable(handle, descriptor, plan, table_source, key, table_offset) ||
uint32_t table_stride = 0;
if (!MatchDescriptorTable(handle, descriptor, plan, table_source, key, table_offset,
table_stride) || table_stride != 16u ||
table_source.dword_count != 4u || table_offset != 0u) return false;
const auto* lane_read = key.Resolve().TryInstruction();
if (lane_read == nullptr || lane_read->GetOpcode() != ValueOpcode::ReadFirstLane)
@@ -1763,6 +1797,7 @@ private:
!EquivalentValue(m_program, selected->Arg(0), read->Arg(4))) return false;
DescriptorSource material_source;
DescriptorSource::IndirectDescriptor indirect;
indirect.table_stride = table_stride;
if (!MakeRuntimeTableSource(*read, material_source) ||
!BoundedSelectedIndex(read->Arg(1), read->Arg(4), indirect)) return false;
indirect.material_source = InternSource(material_source);
src/graphics/shader/recompiler/ir/passes/SrtWalker.h
+1
View File
@@ -17,6 +17,7 @@ struct SrtRuntime {
SrtMemoryReader read_memory = nullptr;
void* userdata = nullptr;
SrtMemoryReader read_specialization_memory = nullptr;
std::span<const uint32_t> workgroup_counts;
};
enum class RuntimeValueType { Any, Integer };