shader: add BUFFER_STORE_FORMAT_D16_X opcode

src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemory.cpp

@@ -725,9 +725,18 @@ uint32_t FormattedLoad(ValueEmitContext& ctx, const IR::Inst& inst, const IR::Me
 
 uint32_t EncodeFormattedStoreComponent(ValueEmitContext& ctx,
                                        const Format::BufferFormatInfo& info,
-                                       uint32_t component, uint32_t data) {
+                                       uint32_t component, uint32_t data, uint32_t data_bits) {
 	auto& state = ctx.state;
 	const auto bits = info.component_bits[component];
+	if (data_bits == 16u) {
+		if (info.type == Format::ComponentType::Uint) {
+			data = EmitBitFieldUExtract(state, data, ConstantU32(state, 0), ConstantU32(state, 16));
+		} else if (info.type == Format::ComponentType::Sint) {
+			data = EmitBitFieldSExtract(state, data, ConstantU32(state, 0), ConstantU32(state, 16));
+		} else {
+			data = EmitBitCastU32F32(state, EmitF16BitsToF32(state, data));
+		}
+	}
 	const bool normalized = info.type == Format::ComponentType::Unorm ||
 	                        info.type == Format::ComponentType::Snorm;
 	const bool scaled = info.type == Format::ComponentType::Uscaled ||
@@ -779,7 +788,7 @@ void StoreFormattedPrepared(ValueEmitContext& ctx, const IR::Inst& inst,
 					                              value, data, component);
 				}
 			}
-			value = EncodeFormattedStoreComponent(ctx, info, component, value);
+			value = EncodeFormattedStoreComponent(ctx, info, component, value, mem.data_bits);
 			const auto bits = info.component_bits[component];
 			if (info.packed_bitfield) {
 				packed = EmitBitFieldInsert(

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

@@ -63,6 +63,7 @@ constexpr MemoryOpcodeInfo MUBUF_OPCODE_LIST[] = {
     {0x50u, Opcode::BUFFER_ATOMIC_SWAP_X2, 2, 32},
     {0x59u, Opcode::BUFFER_ATOMIC_AND_X2, 2, 32},
     {0x5au, Opcode::BUFFER_ATOMIC_OR_X2, 2, 32},
+    {0x84u, Opcode::BUFFER_STORE_FORMAT_D16_X, 1, 16, false, false, true},
 };
 
 constexpr MemoryOpcodeInfo MTBUF_OPCODE_LIST[] = {

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

@@ -622,6 +622,7 @@ std::string InstructionToString(const Instruction& inst) {
 		case Opcode::BUFFER_STORE_FORMAT_XY:
 		case Opcode::BUFFER_STORE_FORMAT_XYZ:
 		case Opcode::BUFFER_STORE_FORMAT_XYZW:
+		case Opcode::BUFFER_STORE_FORMAT_D16_X:
 		case Opcode::BUFFER_LOAD_UBYTE:
 		case Opcode::BUFFER_LOAD_USHORT:
 		case Opcode::BUFFER_LOAD_DWORD:

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

@@ -509,6 +509,7 @@ enum class Opcode {
 	BUFFER_STORE_FORMAT_XY,
 	BUFFER_STORE_FORMAT_XYZ,
 	BUFFER_STORE_FORMAT_XYZW,
+	BUFFER_STORE_FORMAT_D16_X,
 	BUFFER_LOAD_UBYTE,
 	BUFFER_LOAD_SBYTE,
 	BUFFER_LOAD_USHORT,

src/graphics/shader/recompiler/frontend/translate/Memory.cpp

@@ -187,7 +187,7 @@ Decoder::Operand MemorySourceAt(const Decoder::Instruction& decoded, uint32_t in
 	if (decoded.family == Decoder::Family::MUBUF || decoded.family == Decoder::Family::MTBUF) {
 		const bool store_or_atomic =
 		    (decoded.opcode >= Decoder::Opcode::BUFFER_STORE_FORMAT_X &&
-		     decoded.opcode <= Decoder::Opcode::BUFFER_STORE_FORMAT_XYZW) ||
+		     decoded.opcode <= Decoder::Opcode::BUFFER_STORE_FORMAT_D16_X) ||
 		    (decoded.opcode >= Decoder::Opcode::BUFFER_STORE_BYTE &&
 		     decoded.opcode <= Decoder::Opcode::BUFFER_STORE_DWORDX4) ||
 		    (decoded.opcode >= Decoder::Opcode::TBUFFER_STORE_FORMAT_X &&
@@ -473,7 +473,8 @@ void Translator::BUFFER_STORE(const Decoder::Instruction& inst) {
 	const auto      data     = ReadU32(data_src);
 	IR::ValueOpcode opcode;
 	IR::Value       value;
-	switch (memory.data_bits) {
+	// Formatted stores carry packed VGPR bits; the format determines their conversion.
+	switch (memory.formatted ? 32u : memory.data_bits) {
 		case 8u:
 			opcode = IR::ValueOpcode::StoreBufferU8;
 			value  = NarrowSubdword(data, 8u);
@@ -995,6 +996,7 @@ void Translator::EmitMemory(const Decoder::Instruction& inst) {
 		case Decoder::Opcode::BUFFER_STORE_FORMAT_XY:
 		case Decoder::Opcode::BUFFER_STORE_FORMAT_XYZ:
 		case Decoder::Opcode::BUFFER_STORE_FORMAT_XYZW:
+		case Decoder::Opcode::BUFFER_STORE_FORMAT_D16_X:
 		case Decoder::Opcode::TBUFFER_STORE_FORMAT_X:
 		case Decoder::Opcode::TBUFFER_STORE_FORMAT_XY:
 		case Decoder::Opcode::TBUFFER_STORE_FORMAT_XYZ:

tests/ShaderRecompilerComputeTests.cpp

@@ -712,7 +712,7 @@ constexpr u32 EncodeVopcSdwa(u32 src0, u32 sdst = 0, u32 sd = 0,
 
 constexpr u32 EncodeMubuf0(u32 opcode, u32 offset = 0, bool idxen = false,
                            bool offen = true, bool glc = false) {
-  return (0x38u << 26u) | ((opcode & 0x7fu) << 18u) |
+  return (0x38u << 26u) | ((opcode & 0xffu) << 18u) |
          (offen ? (1u << 12u) : 0u) | (idxen ? (1u << 13u) : 0u) |
          (glc ? (1u << 14u) : 0u) | (offset & 0xfffu);
 }
@@ -28716,6 +28716,44 @@ TestCase BufferStoreFormatXSnorm16ClampsRoundsAndPreservesHalfwords() {
   return test;
 }
 
+std::vector<TestCase> BufferStoreFormatD16XCases() {
+  using F = Prospero::BufferFormat;
+  using O = ShaderOpcode;
+  struct StoreCase {
+    const char *name;
+    F format;
+    u32 input;
+    u32 expected;
+  };
+  constexpr StoreCase inputs[] = {
+      {"BufferStoreFormatD16XFloat16", F::k16Float, 0x7e63c000u, 0xffffc000u},
+      {"BufferStoreFormatD16XFloat32", F::k32Float, 0x7e633800u, 0x3f000000u},
+      {"BufferStoreFormatD16XUint32", F::k32UInt, 0x7e63fffeu, 0x0000fffeu},
+      {"BufferStoreFormatD16XSint32", F::k32SInt, 0x7e63fffeu, 0xfffffffeu},
+      {"BufferStoreFormatD16XUnorm8", F::k8UNorm, 0x7e633800u, 0xffffff80u},
+      {"BufferStoreFormatD16XSnorm16", F::k16SNorm, 0x7e63b800u, 0xffffc000u},
+      {"BufferStoreFormatD16XZeroFill", F::k16_16Float, 0x7e633800u, 0x00003800u},
+  };
+  std::vector<TestCase> cases;
+  for (const auto &input : inputs) {
+    TestCase test;
+    test.name = input.name;
+    AppendVMovLiteral(&test.code, 66, input.input);
+    AppendVMovU32(&test.code, 65, 1);
+    // Match the captured CS824586f35db5c80d instruction at pc 0x484.
+    test.code.push_back(EncodeMubuf0(0x84u, 0, true, false));
+    test.code.push_back(EncodeMubuf1(66, 0, 65));
+    AppendEnd(&test.code);
+    test.initial = {0xdeadbeefu, 0xffffffffu, 0xcafef00du};
+    test.expected = {0xdeadbeefu, input.expected, 0xcafef00du};
+    test.user_data = MakeStructuredStorageBufferData(4, 3, false, BufferFormat(input.format));
+    test.has_user_data = true;
+    test.opcodes = {O::V_MOV_B32, O::BUFFER_STORE_FORMAT_D16_X, O::S_ENDPGM};
+    cases.push_back(std::move(test));
+  }
+  return cases;
+}
+
 std::vector<TestCase> FormattedStoreConversionCases() {
   using F = Prospero::BufferFormat;
   using O = ShaderOpcode;
@@ -35547,6 +35585,9 @@ std::vector<TestCase> MakeCases() {
   AddCase(BufferStoreFormatXyzwFloat16ConvertsComponents);
   AddCase(BufferStoreFormatXyzwSnorm16CapturedSkinningVectors);
   AddCase(BufferStoreFormatXSnorm16ClampsRoundsAndPreservesHalfwords);
+  for (auto &test : BufferStoreFormatD16XCases()) {
+    cases.push_back(std::move(test));
+  }
   for (auto &test : FormattedStoreConversionCases()) {
     cases.push_back(std::move(test));
   }
@@ -40809,6 +40850,9 @@ int main(int argc, char **argv) {
     RunCase(&vulkan, BufferStoreFormatXyzwFloat16ConvertsComponents());
     RunCase(&vulkan, BufferStoreFormatXyzwSnorm16CapturedSkinningVectors());
     RunCase(&vulkan, BufferStoreFormatXSnorm16ClampsRoundsAndPreservesHalfwords());
+    for (const auto &test : BufferStoreFormatD16XCases()) {
+      RunCase(&vulkan, test);
+    }
     for (const auto &test : FormattedStoreConversionCases()) {
       RunCase(&vulkan, test);
     }