Merge pull request #87 from wheremyfoodat/superskyler

Make colour/depth buffers to also do ringing for now, add exp2/log2 approximations in x87 to the shader JIT

include/PICA/dynapica/shader_rec_emitter_x64.hpp

@@ -82,11 +82,13 @@ class ShaderEmitter : public Xbyak::CodeGenerator {
 	void recDP4(const PICAShader& shader, u32 instruction);
 	void recEMIT(const PICAShader& shader, u32 instruction);
 	void recEND(const PICAShader& shader, u32 instruction);
+	void recEX2(const PICAShader& shader, u32 instruction);
 	void recFLR(const PICAShader& shader, u32 instruction);
 	void recIFC(const PICAShader& shader, u32 instruction);
 	void recIFU(const PICAShader& shader, u32 instruction);
 	void recJMPC(const PICAShader& shader, u32 instruction);
 	void recJMPU(const PICAShader& shader, u32 instruction);
+	void recLG2(const PICAShader& shader, u32 instruction);
 	void recLOOP(const PICAShader& shader, u32 instruction);
 	void recMAD(const PICAShader& shader, u32 instruction);
 	void recMAX(const PICAShader& shader, u32 instruction);

include/renderer_gl/renderer_gl.hpp

@@ -43,8 +43,8 @@ class Renderer {
 	float oldDepthOffset = 0.0;
 	bool oldDepthmapEnable = false;
 
-	SurfaceCache<DepthBuffer, 10> depthBufferCache;
-	SurfaceCache<ColourBuffer, 10> colourBufferCache;
+	SurfaceCache<DepthBuffer, 10, true> depthBufferCache;
+	SurfaceCache<ColourBuffer, 10, true> colourBufferCache;
 	SurfaceCache<Texture, 256, true> textureCache;
 
 	OpenGL::uvec2 fbSize;  // The size of the framebuffer (ie both the colour and depth buffer)'

include/renderer_gl/surface_cache.hpp

@@ -46,7 +46,7 @@ public:
 
     OptionalRef findFromAddress(u32 address) {
         for (auto& e : buffer) {
-            if (e.location == address && e.valid)
+            if (e.location <= address && e.location + e.sizeInBytes() > address && e.valid)
                 return e;
         }
 
@@ -57,6 +57,10 @@ public:
 	SurfaceType& add(const SurfaceType& surface) {
 		if (size >= capacity) {
 			if constexpr (evictOnOverflow) {  // Do a ring buffer if evictOnOverflow is true
+				if constexpr (std::is_same<SurfaceType, ColourBuffer>() || std::is_same<SurfaceType, DepthBuffer>()) {
+					Helpers::panicDev("Colour/Depth buffer cache overflowed, currently stubbed to do a ring-buffer. This might snap in half");
+				}
+
 				auto& e = buffer[evictionIndex];
 				evictionIndex = (evictionIndex + 1) % capacity;
 

include/renderer_gl/surfaces.hpp

@@ -58,11 +58,13 @@ struct ColourBuffer {
     }
 
     void free() {
-        valid = false;
+		valid = false;
 
-        if (texture.exists() || fbo.exists())
-            Helpers::panic("Make this buffer free itself");
-    }
+		if (texture.exists() || fbo.exists()) {
+			texture.free();
+			fbo.free();
+		}
+	}
 
     bool matches(ColourBuffer& other) {
         return location == other.location && format == other.format &&
@@ -128,9 +130,11 @@ struct DepthBuffer {
     }
 
     void free() {
-        valid = false;
-        printf("Make this depth buffer free itself\n");
-    }
+		valid = false;
+		if (texture.exists()) {
+			texture.free();
+		}
+	}
 
     bool matches(DepthBuffer& other) {
         return location == other.location && format == other.format &&

src/core/PICA/dynapica/shader_rec_emitter_x64.cpp

@@ -135,11 +135,13 @@ void ShaderEmitter::compileInstruction(const PICAShader& shaderUnit) {
 		case ShaderOpcodes::DP3: recDP3(shaderUnit, instruction); break;
 		case ShaderOpcodes::DP4: recDP4(shaderUnit, instruction); break;
 		case ShaderOpcodes::END: recEND(shaderUnit, instruction); break;
+		case ShaderOpcodes::EX2: recEX2(shaderUnit, instruction); break;
 		case ShaderOpcodes::FLR: recFLR(shaderUnit, instruction); break;
 		case ShaderOpcodes::IFC: recIFC(shaderUnit, instruction); break;
 		case ShaderOpcodes::IFU: recIFU(shaderUnit, instruction); break;
 		case ShaderOpcodes::JMPC: recJMPC(shaderUnit, instruction); break;
 		case ShaderOpcodes::JMPU: recJMPU(shaderUnit, instruction); break;
+		case ShaderOpcodes::LG2: recLG2(shaderUnit, instruction); break;
 		case ShaderOpcodes::LOOP: recLOOP(shaderUnit, instruction); break;
 		case ShaderOpcodes::MOV: recMOV(shaderUnit, instruction); break;
 		case ShaderOpcodes::MOVA: recMOVA(shaderUnit, instruction); break;
@@ -152,8 +154,6 @@ void ShaderEmitter::compileInstruction(const PICAShader& shaderUnit) {
 
 		// Unimplemented opcodes that don't seem to actually be used but exist in the binary
 		// EMIT/SETEMIT are used in geometry shaders, however are sometimes found in vertex shaders?
-		case ShaderOpcodes::EX2:
-		case ShaderOpcodes::LG2:
 		case ShaderOpcodes::EMIT:
 		case ShaderOpcodes::SETEMIT:
 			log("[ShaderJIT] Unknown PICA opcode: %02X\n", opcode);
@@ -877,6 +877,74 @@ void ShaderEmitter::recLOOP(const PICAShader& shader, u32 instruction) {
 	loopLevel--;
 }
 
+// SSE does not have a log2 instruction so we temporarily emulate this using x87 FPU
+void ShaderEmitter::recLG2(const PICAShader& shader, u32 instruction) {
+	const u32 operandDescriptor = shader.operandDescriptors[instruction & 0x7f];
+	const u32 src = getBits<12, 7>(instruction);
+	const u32 idx = getBits<19, 2>(instruction);
+	const u32 dest = getBits<21, 5>(instruction);
+	const u32 writeMask = getBits<0, 4>(operandDescriptor);
+
+	// Load swizzled source, push 1.0 to the x87 stack
+	loadRegister<1>(src1_xmm, shader, src, idx, operandDescriptor);
+	fld1();
+
+	// Push source to the x87 stack
+	movd(eax, src1_xmm);
+	push(rax);
+	fld(dword[rsp]);
+
+	// Perform log2, load result to src1_xmm, write it back and undo the previous push rax
+	fyl2x();
+	fstp(dword[rsp]);
+	movss(src1_xmm, dword[rsp]);
+	add(rsp, 8);
+
+	// If we only write back the x component to the result, we needn't perform a shuffle to do res = res.xxxx
+	// Otherwise we do
+	if (writeMask != 0x8) {             // Copy bottom lane to all lanes if we're not simply writing back x
+		shufps(src1_xmm, src1_xmm, 0);  // src1_xmm = src1_xmm.xxxx
+	}
+	storeRegister(src1_xmm, shader, dest, operandDescriptor);
+}
+
+// SSE does not have an exp2 instruction so we temporarily emulate this using x87 FPU
+void ShaderEmitter::recEX2(const PICAShader& shader, u32 instruction) {
+	const u32 operandDescriptor = shader.operandDescriptors[instruction & 0x7f];
+	const u32 src = getBits<12, 7>(instruction);
+	const u32 idx = getBits<19, 2>(instruction);
+	const u32 dest = getBits<21, 5>(instruction);
+	const u32 writeMask = getBits<0, 4>(operandDescriptor);
+
+	loadRegister<1>(src1_xmm, shader, src, idx, operandDescriptor);
+
+	// Push source to the x87 stack, then do some insane compiler-generated x87 math
+	movd(eax, src1_xmm);
+	push(rax);
+	fld(dword[rsp]);
+
+	fld(st0);
+	frndint();
+	fsub(st1, st0);
+	fxch(st1);
+	f2xm1();
+	fadd(dword[rip + onesVector]);
+	fscale();
+
+	// Load result to src1_xmm, write it back and undo the previous push rax
+	fstp(st1);
+	fstp(dword[rsp]);
+	movss(src1_xmm, dword[rsp]);
+	add(rsp, 8);
+
+	// If we only write back the x component to the result, we needn't perform a shuffle to do res = res.xxxx
+	// Otherwise we do
+	if (writeMask != 0x8) {             // Copy bottom lane to all lanes if we're not simply writing back x
+		shufps(src1_xmm, src1_xmm, 0);  // src1_xmm = src1_xmm.xxxx
+	}
+	storeRegister(src1_xmm, shader, dest, operandDescriptor);
+}
+
 void ShaderEmitter::printLog(const PICAShader& shaderUnit) {
 	printf("PC: %04X\n", shaderUnit.pc);
 

src/core/loader/ncch.cpp

@@ -216,10 +216,6 @@ bool NCCH::loadFromHeader(Crypto::AESEngine &aesEngine, IOFile& file, const FSIn
 		printf("RomFS offset: %08llX, size: %08llX\n", romFS.offset, romFS.size);
 	}
 
-	if (stackSize != 0 && stackSize != VirtualAddrs::DefaultStackSize) {
-		Helpers::warn("Requested stack size is %08X bytes. Temporarily emulated as 0x4000 until adjustable sizes are added\n", stackSize);
-	}
-
 	initialized = true;
 	return true;
 }