raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-24 12:56:53 +02:00
Code Issues Packages Projects Releases Wiki Activity

Implement RDNA2 per-invocation scratch memory

Browse Source
This commit is contained in:
Senaxx
2026-08-19 17:32:55 +02:00
committed by nmzik
parent 2db0b3cfa7
commit 56a15c2036
20 changed files with 184 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/ir/ShaderIR.cpp
+4 -4
View File
@@ -222,9 +222,9 @@ MemoryInfo MemoryInfoFromDecoded(const Decoder::Instruction& decoded, ResourceKi
mem.image_nsa_addr[i] = decoded.image_nsa_addr[i];
}
mem.memory_segment = decoded.memory_segment;
mem.address_is_full =
kind == ResourceKind::Flat || kind == ResourceKind::Scratch ||
(kind == ResourceKind::Global && decoded.src1.kind == Decoder::OperandKind::Vgpr);
mem.address_is_full = kind == ResourceKind::Flat ||
(kind == ResourceKind::Global &&
decoded.src1.kind == Decoder::OperandKind::Vgpr);
mem.data_signed = decoded.data_signed;
mem.typed = decoded.typed;
mem.formatted = decoded.formatted;
@@ -236,7 +236,7 @@ MemoryInfo MemoryInfoFromDecoded(const Decoder::Instruction& decoded, ResourceKi
mem.offen = decoded.offen;
mem.resource = ResourceIndexFromOperand(decoded.src1);
mem.sampler = ResourceIndexFromOperand(decoded.src2);
if (kind == ResourceKind::Lds || kind == ResourceKind::Gds) {
if (kind == ResourceKind::Lds || kind == ResourceKind::Gds || kind == ResourceKind::Scratch) {
mem.resource = 0;
mem.sampler = 0;
}
src/graphics/shader/recompiler/ir/ShaderIR.h
+2
View File
@@ -430,6 +430,7 @@ struct SpirvRequirements {
bool compute_derivatives = false;
bool image_gather_extended = false;
bool function_lds = false;
bool function_scratch = false;
bool pixel_valid_mask = false;
};
@@ -440,6 +441,7 @@ struct Program {
uint32_t wave_size = 64;
uint32_t user_data_base = 0;
uint32_t user_data_count = 64;
uint32_t scratch_dwords = 0;
bool dispatcher_fallback = false;
CFG::FailureKind cfg_failure_kind = CFG::FailureKind::None;
std::string fallback_reason;
src/graphics/shader/recompiler/ir/opcodes/ValueOpcodes.inc
+1
View File
@@ -192,6 +192,7 @@ VALUE_OPCODE(Permlane16U32, U32, U32, U32, U32, U1)
VALUE_OPCODE(GetSrtResource, SrtResource)
VALUE_OPCODE(GetBufferResource, BufferResource, U32, U32, U32, U32)
VALUE_OPCODE(GetAddressResource, AddressResource, U32, U32)
VALUE_OPCODE(GetScratchResource, AddressResource)
VALUE_OPCODE(GetImageResource, ImageResource, U32, U32, U32, U32, U32, U32, U32, U32)
VALUE_OPCODE(GetSamplerResource, SamplerResource, U32, U32, U32, U32)
VALUE_OPCODE(MakeImageAddress, ImageAddress, U32, U32, U32, U32, U32, U32, U32, U32, U32, U32, U32,
src/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp
+4 -5
View File
@@ -679,11 +679,10 @@ bool MaterializeResources(const Program& program, const SrtRuntime& runtime,
} else {
if (!runtime.flat_memory_base.has_value()) {
if (error != nullptr) {
*error =
fmt::format("unbased {} address at pc 0x{:08x} requires runtime "
"guest-address translation",
address.kind == ResourceKind::Flat ? "FLAT" : "global/scratch",
address.first_use_pc);
*error = fmt::format("unbased {} address at pc 0x{:08x} requires runtime "
"guest-address translation",
address.kind == ResourceKind::Flat ? "FLAT" : "global",
address.first_use_pc);
}
return false;
}
src/graphics/shader/recompiler/ir/passes/ResourceTracking.cpp
+12
View File
@@ -764,6 +764,18 @@ private:
if (!IsAddressResourceKind(memory.kind)) {
return Fail(flags.pc, error, "address operation has invalid resource kind");
}
if (memory.kind == ResourceKind::Scratch) {
handle = inst.Arg(0).Resolve().TryInstruction();
if (handle == nullptr || handle->GetOpcode() != ValueOpcode::GetScratchResource ||
handle->NumArgs() != 0) {
return Fail(flags.pc, error, "scratch operation requires GetScratchResource");
}
if (m_program.scratch_dwords == 0) {
return Fail(flags.pc, error,
"scratch operation requires a nonzero AGC per-thread size");
}
return true;
}
bool unbased = false;
if (!GetAddressHandle(inst, inst.Arg(0), flags.pc, handle, source, unbased, error)) {
return false;