softjit: Add stubbed DrawPixel for x64.

CMakeLists.txt

@@ -1393,6 +1393,7 @@ list(APPEND CoreExtra
 	Core/MIPS/x86/RegCacheFPU.cpp
 	Core/MIPS/x86/RegCacheFPU.h
 	GPU/Common/VertexDecoderX86.cpp
+	GPU/Software/DrawPixelX86.cpp
 	GPU/Software/SamplerX86.cpp
 )
 

GPU/GPU.vcxproj

@@ -631,6 +631,7 @@
     <ClCompile Include="Math3D.cpp" />
     <ClCompile Include="Software\Clipper.cpp" />
     <ClCompile Include="Software\DrawPixel.cpp" />
+    <ClCompile Include="Software\DrawPixelX86.cpp" />
     <ClCompile Include="Software\Lighting.cpp" />
     <ClCompile Include="Software\FuncId.cpp" />
     <ClCompile Include="Software\Rasterizer.cpp" />

GPU/GPU.vcxproj.filters

@@ -542,5 +542,8 @@
     <ClCompile Include="Software\DrawPixel.cpp">
       <Filter>Software</Filter>
     </ClCompile>
+    <ClCompile Include="Software\DrawPixelX86.cpp">
+      <Filter>Software</Filter>
+    </ClCompile>
   </ItemGroup>
 </Project>

GPU/Software/DrawPixel.cpp

@@ -15,6 +15,7 @@
 // Official git repository and contact information can be found at
 // https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
 
+#include <mutex>
 #include "Common/Data/Convert/ColorConv.h"
 #include "GPU/GPUState.h"
 #include "GPU/Software/DrawPixel.h"
@@ -26,14 +27,25 @@ using namespace Math3D;
 
 namespace Rasterizer {
 
+std::mutex jitCacheLock;
+PixelJitCache *jitCache = nullptr;
+
 void Init() {
+	jitCache = new PixelJitCache();
 }
 
 void Shutdown() {
+	delete jitCache;
+	jitCache = nullptr;
 }
 
 bool DescribeCodePtr(const u8 *ptr, std::string &name) {
-	return false;
+	if (!jitCache->IsInSpace(ptr)) {
+		return false;
+	}
+
+	name = jitCache->DescribeCodePtr(ptr);
+	return true;
 }
 
 static inline u8 GetPixelStencil(GEBufferFormat fmt, int x, int y) {
@@ -462,6 +474,11 @@ inline void DrawSinglePixel(int x, int y, int z, int fog, const Vec4<int> &color
 }
 
 SingleFunc GetSingleFunc(const PixelFuncID &id) {
+	SingleFunc jitted = jitCache->GetSingle(id);
+	if (jitted) {
+		return jitted;
+	}
+
 	if (id.clearMode) {
 		switch (id.fbFormat) {
 		case GE_FORMAT_565:
@@ -488,4 +505,67 @@ SingleFunc GetSingleFunc(const PixelFuncID &id) {
 	return nullptr;
 }
 
+PixelJitCache::PixelJitCache()
+#if PPSSPP_ARCH(ARM64)
+	: fp(this)
+#endif
+{
+	AllocCodeSpace(1024 * 256 * 4);
+
+	// Add some random code to "help" MSVC's buggy disassembler :(
+#if defined(_WIN32) && (PPSSPP_ARCH(X86) || PPSSPP_ARCH(AMD64)) && !PPSSPP_PLATFORM(UWP)
+	using namespace Gen;
+	for (int i = 0; i < 100; i++) {
+		MOV(32, R(EAX), R(EBX));
+		RET();
+	}
+#elif PPSSPP_ARCH(ARM)
+	BKPT(0);
+	BKPT(0);
+#endif
+}
+
+void PixelJitCache::Clear() {
+	ClearCodeSpace(0);
+	cache_.clear();
+	addresses_.clear();
+}
+
+std::string PixelJitCache::DescribeCodePtr(const u8 *ptr) {
+	ptrdiff_t dist = 0x7FFFFFFF;
+	PixelFuncID found{};
+	for (const auto &it : addresses_) {
+		ptrdiff_t it_dist = ptr - it.second;
+		if (it_dist >= 0 && it_dist < dist) {
+			found = it.first;
+			dist = it_dist;
+		}
+	}
+
+	return DescribePixelFuncID(found);
+}
+
+SingleFunc PixelJitCache::GetSingle(const PixelFuncID &id) {
+	std::lock_guard<std::mutex> guard(jitCacheLock);
+
+	auto it = cache_.find(id);
+	if (it != cache_.end()) {
+		return it->second;
+	}
+
+	// TODO: What should be the min size?  Can we even hit this?
+	if (GetSpaceLeft() < 65536) {
+		Clear();
+	}
+
+#if PPSSPP_ARCH(AMD64) && !PPSSPP_PLATFORM(UWP)
+	addresses_[id] = GetCodePointer();
+	SingleFunc func = CompileSingle(id);
+	cache_[id] = func;
+	return func;
+#else
+	return nullptr;
+#endif
+}
+
 };

GPU/Software/DrawPixel.h

@@ -45,4 +45,35 @@ void Shutdown();
 
 bool DescribeCodePtr(const u8 *ptr, std::string &name);
 
+#if PPSSPP_ARCH(ARM)
+class PixelJitCache : public ArmGen::ARMXCodeBlock {
+#elif PPSSPP_ARCH(ARM64)
+class PixelJitCache : public Arm64Gen::ARM64CodeBlock {
+#elif PPSSPP_ARCH(X86) || PPSSPP_ARCH(AMD64)
+class PixelJitCache : public Gen::XCodeBlock {
+#elif PPSSPP_ARCH(MIPS)
+class PixelJitCache : public MIPSGen::MIPSCodeBlock {
+#else
+class PixelJitCache : public FakeGen::FakeXCodeBlock {
+#endif
+public:
+	PixelJitCache();
+
+	// Returns a pointer to the code to run.
+	SingleFunc GetSingle(const PixelFuncID &id);
+	void Clear();
+
+	std::string DescribeCodePtr(const u8 *ptr);
+
+private:
+	SingleFunc CompileSingle(const PixelFuncID &id);
+
+#if PPSSPP_ARCH(ARM64)
+	Arm64Gen::ARM64FloatEmitter fp;
+#endif
+
+	std::unordered_map<PixelFuncID, SingleFunc> cache_;
+	std::unordered_map<PixelFuncID, const u8 *> addresses_;
+};
+
 };

GPU/Software/DrawPixelX86.cpp

@@ -0,0 +1,38 @@
+// Copyright (c) 2017- PPSSPP Project.
+
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, version 2.0 or later versions.
+
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License 2.0 for more details.
+
+// A copy of the GPL 2.0 should have been included with the program.
+// If not, see http://www.gnu.org/licenses/
+
+// Official git repository and contact information can be found at
+// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
+
+#include "ppsspp_config.h"
+#if PPSSPP_ARCH(AMD64)
+
+#include <emmintrin.h>
+#include "Common/x64Emitter.h"
+#include "Common/CPUDetect.h"
+#include "GPU/GPUState.h"
+#include "GPU/Software/DrawPixel.h"
+#include "GPU/ge_constants.h"
+
+using namespace Gen;
+
+namespace Rasterizer {
+
+SingleFunc PixelJitCache::CompileSingle(const PixelFuncID &id) {
+	return nullptr;
+}
+
+};
+
+#endif

android/jni/Android.mk

@@ -30,6 +30,7 @@ ARCH_FILES := \
   $(SRC)/Core/MIPS/x86/RegCache.cpp \
   $(SRC)/Core/MIPS/x86/RegCacheFPU.cpp \
   $(SRC)/GPU/Common/VertexDecoderX86.cpp \
+  $(SRC)/GPU/Software/DrawPixelX86.cpp \
   $(SRC)/GPU/Software/SamplerX86.cpp
 endif
 
@@ -52,6 +53,7 @@ ARCH_FILES := \
   $(SRC)/Core/MIPS/x86/RegCache.cpp \
   $(SRC)/Core/MIPS/x86/RegCacheFPU.cpp \
   $(SRC)/GPU/Common/VertexDecoderX86.cpp \
+  $(SRC)/GPU/Software/DrawPixelX86.cpp \
   $(SRC)/GPU/Software/SamplerX86.cpp
 endif
 

libretro/Makefile.common

@@ -679,7 +679,7 @@ ifeq ($(WITH_DYNAREC),1)
             CPUFLAGS += -m32
          endif
       endif
-	   SOURCES_CXX += $(GPUDIR)/Software/SamplerX86.cpp
+	   SOURCES_CXX += $(GPUDIR)/Software/DrawPixelX86.cpp $(GPUDIR)/Software/SamplerX86.cpp
 	   SOURCES_CXX += $(COMMONDIR)/x64Emitter.cpp \
 						$(COMMONDIR)/x64Analyzer.cpp \
 						$(COMMONDIR)/ABI.cpp \