Merge pull request #537 from wheremyfoodat/accurate-mul

Add accurate shader multiplication option

include/PICA/dynapica/shader_rec.hpp

@@ -22,8 +22,11 @@ class ShaderJIT {
 
 	ShaderCache cache;
 #endif
+	bool accurateMul = false;
 
   public:
+	void setAccurateMul(bool value) { accurateMul = value; }
+
 #ifdef PANDA3DS_SHADER_JIT_SUPPORTED
 	// Call this before starting to process a batch of vertices
 	// This will read the PICA config (uploaded shader and shader operand descriptors) and search if we've already compiled this shader
@@ -36,11 +39,11 @@ class ShaderJIT {
 	static constexpr bool isAvailable() { return true; }
 #else
 	void prepare(PICAShader& shaderUnit) {
-		Helpers::panic("Vertex Loader JIT: Tried to run ShaderJIT::Prepare on platform that does not support shader jit");
+		Helpers::panic("Shader JIT: Tried to run ShaderJIT::Prepare on platform that does not support shader jit");
 	}
 
 	void run(PICAShader& shaderUnit) {
-		Helpers::panic("Vertex Loader JIT: Tried to run ShaderJIT::Run on platform that does not support shader jit");
+		Helpers::panic("Shader JIT: Tried to run ShaderJIT::Run on platform that does not support shader jit");
 	}
 
 	// Define dummy callback. This should never be called if the shader JIT is not supported

include/PICA/dynapica/shader_rec_emitter_arm64.hpp

@@ -37,6 +37,8 @@ class ShaderEmitter : private oaknut::CodeBlock, public oaknut::CodeGenerator {
 	// Shows whether the loaded shader has any log2 and exp2 instructions
 	bool codeHasLog2 = false;
 	bool codeHasExp2 = false;
+	// Whether to compile this shader using accurate, safe, non-IEEE multiplication (slow) or faster but less accurate mul
+	bool useSafeMUL = false;
 
 	oaknut::Label log2Func, exp2Func;
 	oaknut::Label emitLog2Func();
@@ -123,7 +125,7 @@ class ShaderEmitter : private oaknut::CodeBlock, public oaknut::CodeGenerator {
 	PrologueCallback prologueCb = nullptr;
 
 	// Initialize our emitter with "allocSize" bytes of memory allocated for the code buffer
-	ShaderEmitter() : oaknut::CodeBlock(allocSize), oaknut::CodeGenerator(oaknut::CodeBlock::ptr()) {}
+	ShaderEmitter(bool useSafeMUL) : oaknut::CodeBlock(allocSize), oaknut::CodeGenerator(oaknut::CodeBlock::ptr()), useSafeMUL(useSafeMUL) {}
 
 	// PC must be a valid entrypoint here. It doesn't have that much overhead in this case, so we use std::array<>::at() to assert it does
 	InstructionCallback getInstructionCallback(u32 pc) { return getLabelPointer<InstructionCallback>(instructionLabels.at(pc)); }

include/PICA/dynapica/shader_rec_emitter_x64.hpp

@@ -45,6 +45,8 @@ class ShaderEmitter : public Xbyak::CodeGenerator {
 	// Shows whether the loaded shader has any log2 and exp2 instructions
 	bool codeHasLog2 = false;
 	bool codeHasExp2 = false;
+	// Whether to compile this shader using accurate, safe, non-IEEE multiplication (slow) or faster but less accurate mul
+	bool useSafeMUL = false;
 	
 	Xbyak::Label log2Func, exp2Func;
 	Xbyak::Label emitLog2Func();
@@ -130,7 +132,7 @@ class ShaderEmitter : public Xbyak::CodeGenerator {
 	PrologueCallback prologueCb = nullptr;
 
 	// Initialize our emitter with "allocSize" bytes of RWX memory
-	ShaderEmitter() : Xbyak::CodeGenerator(allocSize) {
+	ShaderEmitter(bool useSafeMUL) : Xbyak::CodeGenerator(allocSize), useSafeMUL(useSafeMUL) {
 		cpuCaps = Xbyak::util::Cpu();
 
 		haveSSE4_1 = cpuCaps.has(Xbyak::util::Cpu::tSSE41);

include/config.hpp

@@ -15,6 +15,7 @@ struct EmulatorConfig {
 
 	bool shaderJitEnabled = shaderJitDefault;
 	bool discordRpcEnabled = false;
+	bool accurateShaderMul = false;
 	RendererType rendererType = RendererType::OpenGL;
 	Audio::DSPCore::Type dspType = Audio::DSPCore::Type::Null;
 

src/config.cpp

@@ -62,6 +62,7 @@ void EmulatorConfig::load() {
 
 			shaderJitEnabled = toml::find_or<toml::boolean>(gpu, "EnableShaderJIT", shaderJitDefault);
 			vsyncEnabled = toml::find_or<toml::boolean>(gpu, "EnableVSync", true);
+			accurateShaderMul = toml::find_or<toml::boolean>(gpu, "AccurateShaderMultiplication", false);
 		}
 	}
 
@@ -125,6 +126,7 @@ void EmulatorConfig::save() {
 	data["GPU"]["EnableShaderJIT"] = shaderJitEnabled;
 	data["GPU"]["Renderer"] = std::string(Renderer::typeToString(rendererType));
 	data["GPU"]["EnableVSync"] = vsyncEnabled;
+	data["GPU"]["AccurateShaderMultiplication"] = accurateShaderMul;
 	data["Audio"]["DSPEmulation"] = std::string(Audio::DSPCore::typeToString(dspType));
 	data["Audio"]["EnableAudio"] = audioEnabled;
 

src/core/PICA/dynapica/shader_rec.cpp

@@ -16,7 +16,7 @@ void ShaderJIT::prepare(PICAShader& shaderUnit) {
 	auto it = cache.find(hash);
 
 	if (it == cache.end()) { // Block has not been compiled yet
-		auto emitter = std::make_unique<ShaderEmitter>();
+		auto emitter = std::make_unique<ShaderEmitter>(accurateMul);
 		emitter->compile(shaderUnit);
 		// Get pointer to callbacks
 		entrypointCallback = emitter->getInstructionCallback(shaderUnit.entrypoint);

src/core/PICA/dynapica/shader_rec_emitter_arm64.cpp

@@ -7,9 +7,6 @@ using namespace Helpers;
 using namespace oaknut;
 using namespace oaknut::util;
 
-// TODO: Expose safe/unsafe optimizations to the user
-constexpr bool useSafeMUL = true;
-
 // Similar to the x64 recompiler, we use an odd internal ABI, which abuses the fact that we'll very rarely be calling C++ functions
 // So to avoid pushing and popping, we'll be making use of volatile registers as much as possible
 static constexpr QReg src1Vec = Q1;
@@ -491,7 +488,7 @@ void ShaderEmitter::recDP3(const PICAShader& shader, u32 instruction) {
 
 	// Now do a full DP4
 	// Do a piecewise multiplication of the vectors first
-	if constexpr (useSafeMUL) {
+	if (useSafeMUL) {
 		emitSafeMUL(src1Vec, src2Vec, scratch1Vec);
 	} else {
 		FMUL(src1Vec.S4(), src1Vec.S4(), src2Vec.S4());
@@ -518,7 +515,7 @@ void ShaderEmitter::recDP4(const PICAShader& shader, u32 instruction) {
 	loadRegister<2>(src2Vec, shader, src2, 0, operandDescriptor);
 
 	// Do a piecewise multiplication of the vectors first
-	if constexpr (useSafeMUL) {
+	if (useSafeMUL) {
 		emitSafeMUL(src1Vec, src2Vec, scratch1Vec);
 	} else {
 		FMUL(src1Vec.S4(), src1Vec.S4(), src2Vec.S4());
@@ -551,7 +548,7 @@ void ShaderEmitter::recDPH(const PICAShader& shader, u32 instruction) {
 
 	// Now perform a DP4
 	// Do a piecewise multiplication of the vectors first
-	if constexpr (useSafeMUL) {
+	if (useSafeMUL) {
 		emitSafeMUL(src1Vec, src2Vec, scratch1Vec);
 	} else {
 		FMUL(src1Vec.S4(), src1Vec.S4(), src2Vec.S4());
@@ -834,7 +831,7 @@ void ShaderEmitter::recMUL(const PICAShader& shader, u32 instruction) {
 	loadRegister<1>(src1Vec, shader, src1, idx, operandDescriptor);
 	loadRegister<2>(src2Vec, shader, src2, 0, operandDescriptor);
 
-	if constexpr (useSafeMUL) {
+	if (useSafeMUL) {
 		emitSafeMUL(src1Vec, src2Vec, scratch1Vec);
 	} else {
 		FMUL(src1Vec.S4(), src1Vec.S4(), src2Vec.S4());
@@ -907,7 +904,7 @@ void ShaderEmitter::recMAD(const PICAShader& shader, u32 instruction) {
 	loadRegister<2>(src2Vec, shader, src2, isMADI ? 0 : idx, operandDescriptor);
 	loadRegister<3>(src3Vec, shader, src3, isMADI ? idx : 0, operandDescriptor);
 
-	if constexpr (useSafeMUL) {
+	if (useSafeMUL) {
 		emitSafeMUL(src1Vec, src2Vec, scratch1Vec);
 		FADD(src3Vec.S4(), src3Vec.S4(), src1Vec.S4());
 	} else {

src/core/PICA/dynapica/shader_rec_emitter_x64.cpp

@@ -12,9 +12,6 @@ using namespace Xbyak;
 using namespace Xbyak::util;
 using namespace Helpers;
 
-// TODO: Expose safe/unsafe optimizations to the user
-constexpr bool useSafeMUL = false;
-
 // The shader recompiler uses quite an odd internal ABI
 // We make use of the fact that in regular conditions, we should pretty much never be calling C++ code from recompiled shader code
 // This allows us to establish an ABI that's optimized for this sort of workflow, statically allocating volatile host registers

src/core/PICA/gpu.cpp

@@ -64,6 +64,8 @@ void GPU::reset() {
 	regs.fill(0);
 	shaderUnit.reset();
 	shaderJIT.reset();
+	shaderJIT.setAccurateMul(config.accurateShaderMul);
+
 	std::memset(vram, 0, vramSize);
 	lightingLUT.fill(0);
 	lightingLUTDirty = true;

src/libretro_core.cpp

@@ -146,6 +146,7 @@ static bool FetchVariableBool(std::string key, bool def) {
 static void configInit() {
 	static const retro_variable values[] = {
 		{"panda3ds_use_shader_jit", "Enable shader JIT; enabled|disabled"},
+		{"panda3ds_accurate_shader_mul", "Enable accurate shader multiplication; disabled|enabled"},
 		{"panda3ds_use_vsync", "Enable VSync; enabled|disabled"},
 		{"panda3ds_dsp_emulation", "DSP emulation; Null|HLE|LLE"},
 		{"panda3ds_use_audio", "Enable audio; disabled|enabled"},
@@ -153,7 +154,7 @@ static void configInit() {
 		{"panda3ds_write_protect_virtual_sd", "Write protect virtual SD card; disabled|enabled"},
 		{"panda3ds_battery_level", "Battery percentage; 5|10|20|30|50|70|90|100"},
 		{"panda3ds_use_charger", "Charger plugged; enabled|disabled"},
-		{nullptr, nullptr}
+		{nullptr, nullptr},
 	};
 
 	envCallbacks(RETRO_ENVIRONMENT_SET_VARIABLES, (void*)values);
@@ -171,6 +172,7 @@ static void configUpdate() {
 	config.audioEnabled = FetchVariableBool("panda3ds_use_audio", false);
 	config.sdCardInserted = FetchVariableBool("panda3ds_use_virtual_sd", true);
 	config.sdWriteProtected = FetchVariableBool("panda3ds_write_protect_virtual_sd", false);
+	config.accurateShaderMul = FetchVariableBool("panda3ds_accurate_shader_mul", false);
 	config.discordRpcEnabled = false;
 
 	config.save();