shader: support RDNA2 V_CMPX_NE_U64

src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.h

@@ -389,6 +389,7 @@ enum class Opcode {
 	V_CMP_EQ_I64,
 	V_CMP_GT_U64,
 	V_CMP_NE_U64,
+	V_CMPX_NE_U64,
 	V_CMPX_LT_U32,
 	V_CMPX_EQ_U32,
 	V_CMPX_LE_U32,

src/graphics/shader/recompiler/frontend/decode/VectorAluOps.cpp

@@ -233,6 +233,7 @@ constexpr VopcOpcodeInfo VOPC_OPCODE_LIST[] = {
     {0xd3u, Opcode::V_CMPX_LE_U32},       {0xd4u, Opcode::V_CMPX_GT_U32},
     {0xd5u, Opcode::V_CMPX_NE_U32},       {0xd6u, Opcode::V_CMPX_GE_U32},
     {0xe4u, Opcode::V_CMP_GT_U64, false}, {0xe5u, Opcode::V_CMP_NE_U64, false},
+    {0xf5u, Opcode::V_CMPX_NE_U64, false},
     {0xc9u, Opcode::V_CMP_LT_F16},        {0xcau, Opcode::V_CMP_EQ_F16},
     {0xcbu, Opcode::V_CMP_LE_F16},        {0xccu, Opcode::V_CMP_GT_F16},
     {0xcdu, Opcode::V_CMP_LG_F16},        {0xceu, Opcode::V_CMP_GE_F16},
@@ -1498,6 +1499,7 @@ bool IsVopcCompareExec(Opcode opcode) {
 		case Opcode::V_CMPX_GT_U32:
 		case Opcode::V_CMPX_NE_U32:
 		case Opcode::V_CMPX_GE_U32:
+		case Opcode::V_CMPX_NE_U64:
 		case Opcode::V_CMPX_LT_F16:
 		case Opcode::V_CMPX_EQ_F16:
 		case Opcode::V_CMPX_LE_F16:

src/graphics/shader/recompiler/ir/opcodes/ShaderIROpcodes.cpp

@@ -338,6 +338,7 @@ constexpr LowerMap LOWER_OPS[] = {
     {Decoder::Opcode::V_CMP_EQ_I64, Opcode::CompareEqU64},
     {Decoder::Opcode::V_CMP_GT_U64, Opcode::CompareGtU64},
     {Decoder::Opcode::V_CMP_NE_U64, Opcode::CompareNeU64},
+    {Decoder::Opcode::V_CMPX_NE_U64, Opcode::CompareNeU64},
     {Decoder::Opcode::V_CMPX_LT_U32, Opcode::CompareMaskLtU32},
     {Decoder::Opcode::V_CMPX_EQ_U32, Opcode::CompareMaskEqU32},
     {Decoder::Opcode::V_CMPX_LE_U32, Opcode::CompareMaskLeU32},

tests/ShaderRecompilerComputeTests.cpp

@@ -14720,6 +14720,34 @@ TestCase VectorVop3CompareNeU64OnGpu() {
            O::S_MOV_B32, O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
 }
 
+TestCase VectorVopcCmpxNeU64CapturedExecMask() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovU32(&code, 7, 0);
+  AppendVMovU32(&code, 8, 0);
+  AppendVMovU32(&code, 31, 0);
+  code.push_back(0x7dea0e80u); // v_cmpx_ne_u64 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(0x7dea0e80u); // v_cmpx_ne_u64 exec, 0, v[7:8]
+  AppendBufferStoreDword(&code, 8, 31);
+  AppendEnd(&code);
+
+  TestCase test{"VectorVopcCmpxNeU64CapturedExecMask",
+                code,
+                {0x11223344u, 0x55667788u},
+                {0x11223344u, 1u},
+                {O::V_MOV_B32, O::V_CMPX_NE_U64, O::BUFFER_STORE_DWORD,
+                 O::S_MOV_B32, O::S_ENDPGM}};
+  test.decoded_counts = {{"V_CMPX_NE_U64 exec_lo, 0, v7", 2}};
+  test.native_ir_counts = {{"CompareNeU64 exec_lo, 0x00000000, v7", 2}};
+  return test;
+}
+
 TestCase VectorVop3CompareGtU64OnGpu() {
   using O = ShaderOpcode;
 
@@ -19564,6 +19592,7 @@ std::vector<TestCase> MakeCases() {
   AddCase(VectorVop3CompareEqI64OnGpu);
   AddCase(VectorVop3CompareGtU64OnGpu);
   AddCase(VectorVop3CompareNeU64OnGpu);
+  AddCase(VectorVopcCmpxNeU64CapturedExecMask);
   AddCase(VectorCompareClassF32);
   AddCase(VectorCompareF16Ops);
   AddCase(Vop2SdwaCndmaskSourceModifier);

tests/shaderCfgTests.cpp

@@ -4043,7 +4043,8 @@ void TestNewShaderDecoderArchitecture() {
 }
 
 void TestNewShaderRecompilerRejectsDppOn64BitCompares() {
-  const uint32_t opcodes[] = {0xa2u, 0xe4u, 0xe5u}; // eq_i64, gt_u64, ne_u64
+  const uint32_t opcodes[] = {0xa2u, 0xe4u, 0xe5u,
+                              0xf5u}; // eq_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
@@ -8360,10 +8361,42 @@ void TestNewShaderRecompilerDispatcherSpillsU32x3() {
 void TestNewShaderRecompilerU64PairLowering() {
   using namespace ShaderRecompiler;
 
-  const uint32_t empty_shader[] = {EncodeSopp(0x01)};
   auto options = MakeCompileOptions(ShaderType::Compute);
+  options.dump_ir = true;
   CompileResult result;
   std::string error;
+
+  const uint32_t cmpx_shader[] = {
+      0x7dea0e80u, // v_cmpx_ne_u64 exec, 0, v[7:8]
+      EncodeSopp(0x01),
+  };
+  Decoder::Instruction decoded_cmpx;
+  Check(Decoder::DecodeInstruction(cmpx_shader, 0u, decoded_cmpx, &error),
+        error.c_str());
+  Check(decoded_cmpx.opcode == Decoder::Opcode::V_CMPX_NE_U64 &&
+            decoded_cmpx.dst.kind == Decoder::OperandKind::ExecLo &&
+            decoded_cmpx.src0.value == 0u &&
+            decoded_cmpx.src1.kind == Decoder::OperandKind::Vgpr &&
+            decoded_cmpx.src1.reg == 7u,
+        "decoder rejected captured V_CMPX_NE_U64 fields");
+  Check(TryRecompile(cmpx_shader, options, result, &error), error.c_str());
+  Decoder::Program decoded_program;
+  CFG::Graph graph;
+  IR::Program native;
+  Check(Decoder::DecodeProgram(cmpx_shader, decoded_program, &error),
+        error.c_str());
+  Check(CFG::BuildGraph(decoded_program, graph, &error) &&
+            CFG::Structurize(graph, &error),
+        error.c_str());
+  Check(IR::LowerProgram(decoded_program, graph, ShaderType::Compute, 64u,
+                         native, &error),
+        error.c_str());
+  Check(Common::ContainsStr(IR::ProgramToString(native),
+                            "CompareNeU64 exec_lo, 0x00000000, v7"),
+        "V_CMPX_NE_U64 did not lower to masked 64-bit inequality");
+  CheckSpirvBinaryValidates(result.spirv);
+
+  const uint32_t empty_shader[] = {EncodeSopp(0x01)};
   Check(TryRecompile(empty_shader, options, result, &error), error.c_str());
 
   auto program = result.program;