shader: 64-bit buffer atomic add (BUFFER_ATOMIC_ADD_X2)

tests/ShaderRecompilerComputeTests.cpp

@@ -20991,6 +20991,7 @@ CoverageClass ClassifyOpcode(ShaderOpcode opcode,
   case Opcode::BUFFER_ATOMIC_SMAX:
   case Opcode::BUFFER_ATOMIC_UMAX:
   case Opcode::BUFFER_ATOMIC_AND:
+  case Opcode::BUFFER_ATOMIC_ADD_X2:
   case Opcode::BUFFER_ATOMIC_AND_X2:
   case Opcode::BUFFER_ATOMIC_OR:
   case Opcode::BUFFER_ATOMIC_XOR:
@@ -35712,6 +35713,40 @@ TestCase BufferAtomicAndX2GlcAndExec() {
   return test;
 }
 
+TestCase BufferAtomicAddX2GlcAndExec() {
+  using O = ShaderOpcode;
+  TestCase test;
+  test.name = "BufferAtomicAddX2GlcAndExec";
+  auto &code = test.code;
+  code.push_back(EncodeSop1(0x04, 12, 126));
+  AppendVMovU32(&code, 20, 0);
+  const u32 values[][2] = {{0x0ff00ff0u, 0xf00ff00fu},
+                         {0xff00ff00u, 0x00ff00ffu},
+                         {0x76543210u, 0xfedcba98u}};
+  for (u32 i = 0; i < std::size(values); ++i) {
+    AppendVMovLiteral(&code, 2, values[i][0]);
+    AppendVMovLiteral(&code, 3, values[i][1]);
+    if (i == 2) {
+      code.push_back(EncodeSop1(0x04, 126, InlineU32(0)));
+    }
+    AppendBufferStoreOpcode(&code, 0x52, 2, 20, i != 1);
+    code.push_back(EncodeSop1(0x04, 126, 12));
+    AppendStoreVgpr(&code, 2, 2 + i * 2);
+    AppendStoreVgpr(&code, 3, 3 + i * 2);
+  }
+  AppendEnd(&code);
+  test.initial = {0x12345678u, 0x87654321u, 0, 0, 0, 0, 0, 0};
+  // The first add wraps at 64 bits; the second carries from the low dword.
+  // GLC returns the first old value, preserves the second operand, and EXEC = 0
+  // suppresses the third add and its writeback despite its nonzero operand.
+  test.expected = {0x21256568u, 0x78743430u, 0x12345678u, 0x87654321u,
+                   0xff00ff00u, 0x00ff00ffu, 0x76543210u, 0xfedcba98u};
+  test.opcodes = {O::V_MOV_B32, O::S_MOV_B64, O::BUFFER_ATOMIC_ADD_X2,
+                 O::BUFFER_STORE_DWORD, O::S_ENDPGM};
+  test.required_spirv = {"OpAtomicIAdd %ulong", "Int64Atomics"};
+  return test;
+}
+
 TestCase BufferAtomicCmpSwapExactRaw() {
   using O = ShaderOpcode;
 
@@ -42248,6 +42283,7 @@ std::vector<TestCase> MakeCases() {
     }
   }
   AddCase(BufferAtomicAndX2GlcAndExec);
+  AddCase(BufferAtomicAddX2GlcAndExec);
   AddCase(ImageAtomicSignedMinMax<false>);
   AddCase(ImageAtomicSignedMinMax<true>);
   AddCase(ImageAtomicSignedMinMax<false, true>);
@@ -47786,6 +47822,7 @@ int main(int argc, char **argv) {
     }
     RunCase(&vulkan, DsAtomicReturn64());
     RunCase(&vulkan, BufferAtomicAndX2GlcAndExec());
+    RunCase(&vulkan, BufferAtomicAddX2GlcAndExec());
     RunCase(&vulkan, ImageAtomicSignedMinMax<false>());
     RunCase(&vulkan, ImageAtomicSignedMinMax<true>());
     return 0;

tests/shaderCfgTests.cpp

@@ -10539,6 +10539,8 @@ void TestCapturedBufferAtomicsX2() {
     ValueOpcode ir_opcode;
   };
   const Case cases[] = {
+      {{0xe1480018u, 0x80000000u}, 0u, 234u, "BUFFER_ATOMIC_ADD_X2",
+       "BufferAtomicIAdd64", "OpAtomicIAdd", ValueOpcode::BufferAtomicIAdd64},
       {{0xe1680018u, 0x80000000u}, 0u, 241u, "BUFFER_ATOMIC_OR_X2",
        "BufferAtomicOr64", "OpAtomicOr", ValueOpcode::BufferAtomicOr64},
       {{0xe1402000u, 0x80000913u}, 9u, 229u, "BUFFER_ATOMIC_SWAP_X2",