raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: V_CMPX_NE_I64

Browse Source
This commit is contained in:
nmzik
2026-08-23 00:59:44 +02:00
parent 3e25eb0998
commit a610528075
5 changed files with 72 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ShaderRecompilerComputeTests.cpp
+31
View File
@@ -14748,6 +14748,36 @@ TestCase VectorVopcCmpxNeU64CapturedExecMask() {
return test;
}
TestCase VectorVop3CmpxNeI64CapturedExecMask() {
using O = ShaderOpcode;
std::vector<u32> code;
AppendVMovU32(&code, 7, 0);
AppendVMovU32(&code, 8, 0);
AppendVMovU32(&code, 31, 0);
code.push_back(0xd4b5007eu);
code.push_back(0x00020e80u); // v_cmpx_ne_i64 exec, 0, v[7:8]
AppendBufferStoreDword(&code, 8, 31);
code.push_back(EncodeSMovB32(126, InlineU32(1)));
AppendVMovU32(&code, 8, 1);
AppendVMovU32(&code, 31, 4);
code.push_back(0xd4b5007eu);
code.push_back(0x00020e80u); // v_cmpx_ne_i64 exec, 0, v[7:8]
AppendBufferStoreDword(&code, 8, 31);
AppendEnd(&code);
TestCase test{"VectorVop3CmpxNeI64CapturedExecMask",
code,
{0x11223344u, 0x55667788u},
{0x11223344u, 1u},
{O::V_MOV_B32, O::V_CMPX_NE_I64, O::BUFFER_STORE_DWORD,
O::S_MOV_B32, O::S_ENDPGM}};
test.decoded_counts = {{"V_CMPX_NE_I64 exec_lo, 0, v7", 2}};
test.native_ir_counts = {{"CompareNeU64 exec_lo, 0x00000000, v7", 2}};
return test;
}
TestCase VectorVop3CompareGtU64OnGpu() {
using O = ShaderOpcode;
@@ -19593,6 +19623,7 @@ std::vector<TestCase> MakeCases() {
AddCase(VectorVop3CompareGtU64OnGpu);
AddCase(VectorVop3CompareNeU64OnGpu);
AddCase(VectorVopcCmpxNeU64CapturedExecMask);
AddCase(VectorVop3CmpxNeI64CapturedExecMask);
AddCase(VectorCompareClassF32);
AddCase(VectorCompareF16Ops);
AddCase(Vop2SdwaCndmaskSourceModifier);
tests/shaderCfgTests.cpp
+37 -2
View File
@@ -4044,8 +4044,9 @@ void TestNewShaderDecoderArchitecture() {
}
void TestNewShaderRecompilerRejectsDppOn64BitCompares() {
const uint32_t opcodes[] = {0xa2u, 0xe4u, 0xe5u,
0xf5u}; // eq_i64, gt_u64, ne_u64, cmpx_ne_u64
const uint32_t opcodes[] = {
0xa2u, 0xb5u, 0xe4u, 0xe5u,
0xf5u}; // eq_i64, cmpx_ne_i64, gt_u64, ne_u64, cmpx_ne_u64
for (const auto opcode : opcodes) {
const uint32_t shader[] = {
EncodeVopc(opcode, 250u, 0u), // DPP escape in SRC0
@@ -8367,6 +8368,40 @@ void TestNewShaderRecompilerU64PairLowering() {
CompileResult result;
std::string error;
const uint32_t cmpx_i64_shader[] = {
0xd4b5007eu, 0x00020e80u, // v_cmpx_ne_i64 exec, 0, v[7:8]
EncodeSopp(0x01),
};
Decoder::Instruction decoded_cmpx_i64;
Check(Decoder::DecodeInstruction(cmpx_i64_shader, 0u, decoded_cmpx_i64,
&error),
error.c_str());
Check(decoded_cmpx_i64.opcode == Decoder::Opcode::V_CMPX_NE_I64 &&
decoded_cmpx_i64.dst.kind == Decoder::OperandKind::ExecLo &&
decoded_cmpx_i64.src0.value == 0u &&
decoded_cmpx_i64.src1.kind == Decoder::OperandKind::Vgpr &&
decoded_cmpx_i64.src1.reg == 7u,
"decoder rejected captured VOP3 V_CMPX_NE_I64 fields");
Check(TryRecompile(cmpx_i64_shader, options, result, &error), error.c_str());
Check(Common::ContainsStr(result.decoded_dump,
"V_CMPX_NE_I64 exec_lo, 0, v7"),
"captured VOP3 V_CMPX_NE_I64 was not present in the decoded dump");
Decoder::Program decoded_i64_program;
CFG::Graph graph_i64;
IR::Program native_i64;
Check(Decoder::DecodeProgram(cmpx_i64_shader, decoded_i64_program, &error),
error.c_str());
Check(CFG::BuildGraph(decoded_i64_program, graph_i64, &error) &&
CFG::Structurize(graph_i64, &error),
error.c_str());
Check(IR::LowerProgram(decoded_i64_program, graph_i64, ShaderType::Compute,
64u, native_i64, &error),
error.c_str());
Check(Common::ContainsStr(IR::ProgramToString(native_i64),
"CompareNeU64 exec_lo, 0x00000000, v7"),
"V_CMPX_NE_I64 did not lower to masked 64-bit inequality");
CheckSpirvBinaryValidates(result.spirv);
const uint32_t cmpx_shader[] = {
0x7dea0e80u, // v_cmpx_ne_u64 exec, 0, v[7:8]
EncodeSopp(0x01),