raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 19:07:01 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader_recompiler: allow GDS append/consume offsets

Browse Source
This commit is contained in:
nmzik
2026-07-29 21:30:41 +02:00
parent e91dd39cb0
commit c508c4a9c0
3 changed files with 26 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/decompiler/MemoryOps.cpp
+7 -12
View File
@@ -421,13 +421,13 @@ bool DecodeDs(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
const uint32_t data0 = (word1 >> 8u) & 0xffu; const uint32_t data0 = (word1 >> 8u) & 0xffu;
const uint32_t addr = word1 & 0xffu; const uint32_t addr = word1 & 0xffu;
inst.pc = pc; inst.pc = pc;
inst.word = word0; inst.word = word0;
inst.word_count = 2; inst.word_count = 2;
inst.offset = offset0 | (offset1 << 8u); inst.offset = offset0 | (offset1 << 8u);
inst.gds = ((word0 >> 17u) & 1u) != 0u; inst.gds = ((word0 >> 17u) & 1u) != 0u;
inst.family = Family::DS; inst.family = Family::DS;
inst.opcode_id = opcode; inst.opcode_id = opcode;
const auto* info = LookupMemoryOpcode(DS_OPS, static_cast<uint32_t>(std::size(DS_OPS)), opcode); const auto* info = LookupMemoryOpcode(DS_OPS, static_cast<uint32_t>(std::size(DS_OPS)), opcode);
ApplyMemoryInfo(inst, info); ApplyMemoryInfo(inst, info);
SetRawWords(inst, code, word_index, 2); SetRawWords(inst, code, word_index, 2);
@@ -442,11 +442,6 @@ bool DecodeDs(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
inst.opcode == Opcode::DsReadAddtidB32)) { inst.opcode == Opcode::DsReadAddtidB32)) {
SetUnsupported(inst, Family::DS, opcode, "DS swizzle/addtid is available only for LDS"); SetUnsupported(inst, Family::DS, opcode, "DS swizzle/addtid is available only for LDS");
} }
if (inst.gds && (inst.opcode == Opcode::DsAppend || inst.opcode == Opcode::DsConsume) &&
inst.offset != 0u) {
SetUnsupported(inst, Family::DS, opcode,
"GDS append/consume requires a zero instruction offset");
}
if (inst.opcode == Opcode::DsWriteAddtidB32 && data1 != 0u) { if (inst.opcode == Opcode::DsWriteAddtidB32 && data1 != 0u) {
SetUnsupported(inst, Family::DS, opcode, SetUnsupported(inst, Family::DS, opcode,
"DS write addtid data1 operand is not implemented"); "DS write addtid data1 operand is not implemented");
src/graphics/shader/recompiler/emitter/SpirvEmitter.cpp
+2 -3
View File
@@ -1,7 +1,7 @@
#include "graphics/shader/recompiler/emitter/SpirvEmitter.h" #include "graphics/shader/recompiler/emitter/SpirvEmitter.h"
#include "graphics/shader/recompiler/ir/SrtWalker.h"
#include "graphics/shader/recompiler/emitter/spirvEmitterInternal.h" #include "graphics/shader/recompiler/emitter/spirvEmitterInternal.h"
#include "graphics/shader/recompiler/ir/SrtWalker.h"
#include <algorithm> #include <algorithm>
#include <array> #include <array>
@@ -160,8 +160,7 @@ bool ValidateInstructionContract(const IR::Instruction& inst, std::string* error
inst.dst.kind != IR::OperandKind::Null)) || inst.dst.kind != IR::OperandKind::Null)) ||
((inst.op == IR::Opcode::DsAppend || inst.op == IR::Opcode::DsConsume) && ((inst.op == IR::Opcode::DsAppend || inst.op == IR::Opcode::DsConsume) &&
(!ds_kind || !ds_resource || inst.src_count != 1 || (!ds_kind || !ds_resource || inst.src_count != 1 ||
inst.dst.kind != IR::OperandKind::Register || inst.dst.kind != IR::OperandKind::Register)) ||
(kind == IR::ResourceKind::Gds && inst.memory.offset != 0))) ||
((inst.op == IR::Opcode::DsMinF32 || inst.op == IR::Opcode::DsMaxF32) && ((inst.op == IR::Opcode::DsMinF32 || inst.op == IR::Opcode::DsMaxF32) &&
(!ds_kind || !ds_resource || inst.src_count != 3 || (!ds_kind || !ds_resource || inst.src_count != 3 ||
inst.dst.kind != IR::OperandKind::Null)) || inst.dst.kind != IR::OperandKind::Null)) ||
tests/ShaderRecompilerComputeTests.cpp
+17 -4
View File
@@ -13593,23 +13593,36 @@ TestCase DsAppendUsesEncodedGdsSelector() {
using O = ShaderOpcode; using O = ShaderOpcode;
std::vector<u32> code; std::vector<u32> code;
AppendSMovLiteral(&code, 124, 0x00000001u); AppendSMovLiteral(&code, 124, 0x00000008u);
code.push_back(EncodeDs0(0x3e, 0, true)); code.push_back(EncodeDs0(0x3e, 0, true));
code.push_back(EncodeDs1(0, 0, 0)); code.push_back(EncodeDs1(0, 0, 0));
code.push_back(EncodeDs0(0x3d, 0, true)); code.push_back(EncodeDs0(0x3d, 0, true));
code.push_back(EncodeDs1(1, 0, 0)); code.push_back(EncodeDs1(1, 0, 0));
code.push_back(EncodeDs0(0x3e, 4, true));
code.push_back(EncodeDs1(2, 0, 0));
code.push_back(EncodeDs0(0x3d, 4, true));
code.push_back(EncodeDs1(3, 0, 0));
AppendSMovLiteral(&code, 124, 0x00080008u);
code.push_back(EncodeDs0(0x3e, 4, true));
code.push_back(EncodeDs1(4, 0, 0));
code.push_back(EncodeDs0(0x3d, 4, true));
code.push_back(EncodeDs1(5, 0, 0));
AppendStoreVgpr(&code, 0, 0); AppendStoreVgpr(&code, 0, 0);
AppendStoreVgpr(&code, 1, 1); AppendStoreVgpr(&code, 1, 1);
AppendStoreVgpr(&code, 2, 2);
AppendStoreVgpr(&code, 3, 3);
AppendStoreVgpr(&code, 4, 4);
AppendStoreVgpr(&code, 5, 5);
AppendEnd(&code); AppendEnd(&code);
TestCase test{ TestCase test{
"DsAppendGdsSelector", "DsAppendGdsSelector",
code, code,
{}, {},
{10, 74}, {10, 74, 20, 84, 40, 104},
{O::SMovB32, O::DsAppend, O::DsConsume, O::BufferStoreDword, O::SEndpgm}}; {O::SMovB32, O::DsAppend, O::DsConsume, O::BufferStoreDword, O::SEndpgm}};
test.gds_initial = {10}; test.gds_initial = {10, 20, 30, 40};
test.expected_gds = {10}; test.expected_gds = {10, 20, 30, 40};
return test; return test;
} }