shader: execute PS5 BVH intersections and local FLAT stacks

Co-authored-by: Ali Almohaya <AliAlmohaya@gmail.com>
Co-authored-by: Mac <me@itsmemac.com>
Co-authored-by: Senaxx <michelbeld@gmail.com>
Co-authored-by: Jean-Philippe <github@meunier.jp>

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

@@ -225,6 +225,7 @@ Opcode DecodeMimgOpcode(uint32_t opcode, const MimgSampleInfo* sample, const Mim
 		case 0x09u: return Opcode::IMAGE_STORE_MIP;
 		case 0x0eu: return Opcode::IMAGE_GET_RESINFO;
 		case 0x60u: return Opcode::IMAGE_GET_LOD;
+		case 0xe6u: return Opcode::IMAGE_BVH_INTERSECT_RAY;
 		default: return Opcode::UNSUPPORTED;
 	}
 }
@@ -253,6 +254,7 @@ uint32_t DecodeMimgAddressComponents(uint32_t opcode, ImageDimension dimension,
 	}
 
 	switch (opcode) {
+		case 0xe6u: return 11u;
 		case 0x0eu: return 1u;
 		case 0x01u:
 		case 0x09u: return ImageCoordComponents(dimension) + 1u;
@@ -360,6 +362,11 @@ void DecodeMimg(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	if (d16 && !supports_d16) {
 		SetUnsupported(inst, Family::MIMG, opcode, "MIMG opcode does not support D16 data");
 	}
+	if (opcode == 0xe6u &&
+	    (a16 || !r128 || inst.dmask != 0xfu || (nsa_dwords != 0u && nsa_dwords != 3u))) {
+		SetUnsupported(inst, Family::MIMG, opcode,
+		               "BVH intersection requires eleven full-float ray DWORDs and R128/dmask:0xf");
+	}
 
 	DecodeVectorGpr(vdata, inst.dst);
 	DecodeVectorGpr(vaddr, inst.src0);

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

@@ -254,6 +254,8 @@ void DecodeScalarSource(uint32_t code, uint32_t pc, Operand& operand) {
 		case 125u: operand.kind = OperandKind::Null; return;
 		case 126u: operand.kind = OperandKind::ExecLo; return;
 		case 127u: operand.kind = OperandKind::ExecHi; return;
+		case 235u: operand.kind = OperandKind::SharedBase; return;
+		case 237u: operand.kind = OperandKind::PrivateBase; return;
 		case 239u: operand.kind = OperandKind::PopsExitingWaveId; return;
 		case 248u:
 			operand.kind      = OperandKind::FloatInlineConstant;
@@ -404,7 +406,6 @@ void DecodeProgram(std::span<const uint32_t> code, Program& program) {
 	program.instructions.clear();
 	program.instructions.reserve(code.size());
 	program.code = code;
-	program.has_bvh = false;
 
 	std::vector<bool> branch_targets;
 	for (uint32_t word_index = 0; word_index < code.size();) {
@@ -413,10 +414,6 @@ void DecodeProgram(std::span<const uint32_t> code, Program& program) {
 
 		const auto& inst = program.instructions.back();
 		word_index += inst.word_count;
-		if (inst.family == Family::MIMG && (inst.opcode_id == 0xe6u || inst.opcode_id == 0xe7u)) {
-			program.has_bvh = true;
-			return;
-		}
 
 		if (IsDirectBranch(inst.opcode)) {
 			const auto target_index = inst.branch_target / sizeof(uint32_t);
@@ -455,6 +452,8 @@ std::string OperandToString(const Operand& operand) {
 		case OperandKind::Scc: text = "scc"; break;
 		case OperandKind::M0: text = "m0"; break;
 		case OperandKind::PopsExitingWaveId: text = "pops_exiting_wave_id"; break;
+		case OperandKind::SharedBase: text = "shared_base"; break;
+		case OperandKind::PrivateBase: text = "private_base"; break;
 		case OperandKind::Null: text = "null"; break;
 		default: text = "unknown"; break;
 	}

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

@@ -10,6 +10,10 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Decoder {
 
+// PS5 flat stack addresses use these SH_MEM_BASES aperture tags in VA[63:32].
+constexpr uint32_t PrivateApertureHigh = 0x70000000u;
+constexpr uint32_t SharedApertureHigh = 0x80000000u;
+
 enum class Family {
 	Unknown,
 	SOP1,
@@ -569,6 +573,7 @@ enum class Opcode {
 	DS_WRITE_B128,
 	DS_WRITE_ADDTID_B32,
 	DS_READ_ADDTID_B32,
+	IMAGE_BVH_INTERSECT_RAY,
 	IMAGE_GET_RESINFO,
 	IMAGE_GET_LOD,
 	IMAGE_LOAD,
@@ -636,6 +641,8 @@ enum class OperandKind {
 	Scc,
 	M0,
 	PopsExitingWaveId,
+	SharedBase,
+	PrivateBase,
 	Null,
 	Vgpr,
 };
@@ -747,7 +754,6 @@ struct Instruction {
 struct Program {
 	std::span<const uint32_t> code;
 	std::vector<Instruction>  instructions;
-	bool                     has_bvh = false;
 };
 
 // Code spans are trusted to contain complete instructions, valid branch targets, and 32-bit PCs.

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

@@ -883,6 +883,15 @@ void Translator::DS_BPERMUTE_B32(const Decoder::Instruction& inst) {
 
 void Translator::EmitMemory(const Decoder::Instruction& inst) {
 	switch (inst.opcode) {
+		case Decoder::Opcode::IMAGE_BVH_INTERSECT_RAY: {
+			const auto result = ir.Emit(IR::ValueOpcode::BvhIntersect,
+			    {ConstructU32x4(inst.src1, 4), MakeImageAddress(inst, inst.src0), ir.GetExec()});
+			for (uint32_t component = 0; component < 4; ++component) {
+				WriteOperand(OffsetOperand(inst.dst, component),
+				    ir.Emit(IR::ValueOpcode::CompositeExtractU32x4, {result, IR::Value(component)}));
+			}
+			return;
+		}
 		case Decoder::Opcode::S_LOAD_DWORD:
 		case Decoder::Opcode::S_LOAD_DWORDX2:
 		case Decoder::Opcode::S_LOAD_DWORDX4:

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

@@ -135,6 +135,8 @@ IR::U32 Translator::ReadRawU32(const Decoder::Operand& operand) {
 		case Decoder::OperandKind::IntegerInlineConstant:
 		case Decoder::OperandKind::FloatInlineConstant: return IR::U32(IR::Value(operand.value));
 		case Decoder::OperandKind::Null:
+		case Decoder::OperandKind::SharedBase:
+		case Decoder::OperandKind::PrivateBase:
 		case Decoder::OperandKind::PopsExitingWaveId: return IR::U32(IR::Value(0u));
 		case Decoder::OperandKind::Sgpr:
 			return ir.GetScalarReg(static_cast<IR::ScalarReg>(operand.reg));
@@ -474,6 +476,12 @@ IR::U32 Translator::ReadU32(const Decoder::Operand& operand) {
 }
 
 std::array<IR::U32, 2> Translator::ReadU32Pair(const Decoder::Operand& operand) {
+	if (operand.kind == Decoder::OperandKind::PrivateBase ||
+	    operand.kind == Decoder::OperandKind::SharedBase) {
+		return {IR::U32(IR::Value(0u)), IR::U32(IR::Value(
+		    operand.kind == Decoder::OperandKind::PrivateBase ? Decoder::PrivateApertureHigh
+		                                                      : Decoder::SharedApertureHigh))};
+	}
 	if (operand.kind == Decoder::OperandKind::ExecLo) {
 		return {ir.GetExecLo(), ir.GetExecHi()};
 	}