GS: Add MultiISA files

pcsx2/CMakeLists.txt

@@ -673,6 +673,7 @@ set(pcsx2GSSources
 	GS/GSTables.cpp
 	GS/GSUtil.cpp
 	GS/GSVector.cpp
+	GS/MultiISA.cpp
 	GS/Renderers/Common/GSDevice.cpp
 	GS/Renderers/Common/GSDirtyRect.cpp
 	GS/Renderers/Common/GSFunctionMap.cpp
@@ -732,6 +733,7 @@ set(pcsx2GSHeaders
 	GS/GSVector4i.h
 	GS/GSVector8.h
 	GS/GSVector8i.h
+	GS/MultiISA.h
 	GS/Renderers/Common/GSDevice.h
 	GS/Renderers/Common/GSDirtyRect.h
 	GS/Renderers/Common/GSFastList.h

pcsx2/GS/GSUtil.cpp

@@ -17,6 +17,7 @@
 #include "GS.h"
 #include "GSExtra.h"
 #include "GSUtil.h"
+#include "MultiISA.h"
 #include "common/StringUtil.h"
 
 #ifdef _WIN32
@@ -29,8 +30,6 @@
 #define SVN_MODS 0
 #endif
 
-Xbyak::util::Cpu g_cpu;
-
 static class GSUtilMaps
 {
 public:
@@ -149,37 +148,30 @@ bool GSUtil::HasCompatibleBits(u32 spsm, u32 dpsm)
 
 bool GSUtil::CheckSSE()
 {
-	bool status = true;
-
 	struct ISA
 	{
-		Xbyak::util::Cpu::Type type;
+		ProcessorFeatures::VectorISA isa;
 		const char* name;
 	};
 
 	ISA checks[] = {
-		{Xbyak::util::Cpu::tSSE41, "SSE41"},
+		{ProcessorFeatures::VectorISA::SSE4, "SSE 4.1"},
 #if _M_SSE >= 0x500
-		{Xbyak::util::Cpu::tAVX, "AVX1"},
+		{ProcessorFeatures::VectorISA::AVX, "AVX"},
 #endif
 #if _M_SSE >= 0x501
-		{Xbyak::util::Cpu::tAVX2, "AVX2"},
+		{ProcessorFeatures::VectorISA::AVX2, "AVX2"},
-		{Xbyak::util::Cpu::tBMI1, "BMI1"},
-		{Xbyak::util::Cpu::tBMI2, "BMI2"},
 #endif
 	};
-
 	for (const ISA& check : checks)
 	{
-		if (!g_cpu.has(check.type))
+		if (g_cpu.vectorISA < check.isa)
 		{
-			fprintf(stderr, "This CPU does not support %s\n", check.name);
+			Console.Error("This CPU does not support %s", check.name);
-
+			return false;
-			status = false;
 		}
 	}
-
+	return true;
-	return status;
 }
 
 CRCHackLevel GSUtil::GetRecommendedCRCHackLevel(GSRendererType type)

pcsx2/GS/GSUtil.h

@@ -22,8 +22,6 @@
 #include <dxgi.h>
 #endif
 
-#include <xbyak/xbyak_util.h>
-
 class GSUtil
 {
 public:
@@ -52,5 +50,3 @@ void GSmkdir(const char* dir);
 std::string GStempdir();
 
 const char* psm_str(int psm);
-
-extern Xbyak::util::Cpu g_cpu;

pcsx2/GS/MultiISA.cpp

@@ -0,0 +1,92 @@
+/*  PCSX2 - PS2 Emulator for PCs
+ *  Copyright (C) 2002-2021 PCSX2 Dev Team
+ *
+ *  PCSX2 is free software: you can redistribute it and/or modify it under the terms
+ *  of the GNU Lesser General Public License as published by the Free Software Found-
+ *  ation, either version 3 of the License, or (at your option) any later version.
+ *
+ *  PCSX2 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 for more details.
+ *
+ *  You should have received a copy of the GNU General Public License along with PCSX2.
+ *  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "PrecompiledHeader.h"
+#include "MultiISA.h"
+#include <xbyak/xbyak_util.h>
+
+#ifdef _WIN32
+#define strcasecmp _stricmp
+#endif
+
+static Xbyak::util::Cpu s_cpu;
+
+static ProcessorFeatures::VectorISA getCurrentISA()
+{
+	// For debugging
+	if (const char* over = getenv("OVERRIDE_VECTOR_ISA"))
+	{
+		if (strcasecmp(over, "avx2") == 0)
+		{
+			fprintf(stderr, "Vector ISA Override: AVX2\n");
+			return ProcessorFeatures::VectorISA::AVX2;
+		}
+		if (strcasecmp(over, "avx") == 0)
+		{
+			fprintf(stderr, "Vector ISA Override: AVX\n");
+			return ProcessorFeatures::VectorISA::AVX;
+		}
+		if (strcasecmp(over, "sse4") == 0)
+		{
+			fprintf(stderr, "Vector ISA Override: SSE4\n");
+			return ProcessorFeatures::VectorISA::SSE4;
+		}
+	}
+	if (s_cpu.has(Xbyak::util::Cpu::tAVX2) && s_cpu.has(Xbyak::util::Cpu::tBMI1) && s_cpu.has(Xbyak::util::Cpu::tBMI2))
+		return ProcessorFeatures::VectorISA::AVX2;
+	else if (s_cpu.has(Xbyak::util::Cpu::tAVX))
+		return ProcessorFeatures::VectorISA::AVX;
+	else if (s_cpu.has(Xbyak::util::Cpu::tSSE41))
+		return ProcessorFeatures::VectorISA::SSE4;
+	else
+		return ProcessorFeatures::VectorISA::None;
+}
+
+static ProcessorFeatures getProcessorFeatures()
+{
+	ProcessorFeatures features = {};
+	features.vectorISA = getCurrentISA();
+	features.hasFMA = s_cpu.has(Xbyak::util::Cpu::tFMA);
+	if (const char* over = getenv("OVERRIDE_FMA"))
+	{
+		features.hasFMA = over[0] == 'Y' || over[0] == 'y' || over[0] == '1';
+		fprintf(stderr, "Processor FMA override: %s\n", features.hasFMA ? "Supported" : "Unsupported");
+	}
+	features.hasSlowGather = false;
+	if (const char* over = getenv("OVERRIDE_SLOW_GATHER")) // Easy override for comparing on vs off
+	{
+		features.hasSlowGather = over[0] == 'Y' || over[0] == 'y' || over[0] == '1';
+		fprintf(stderr, "Processor gather override: %s\n", features.hasSlowGather ? "Slow" : "Fast");
+	}
+	else if (features.vectorISA == ProcessorFeatures::VectorISA::AVX2)
+	{
+		if (s_cpu.has(Xbyak::util::Cpu::tINTEL))
+		{
+			// Slow on Haswell
+			// CPUID data from https://en.wikichip.org/wiki/intel/cpuid
+			features.hasSlowGather = s_cpu.displayModel == 0x46 || s_cpu.displayModel == 0x45 || s_cpu.displayModel == 0x3c;
+		}
+		else
+		{
+			// Currently no Zen CPUs with fast VPGATHERDD
+			// Check https://uops.info/table.html as new CPUs come out for one that doesn't split it into like 40 µops
+			// Doing it manually is about 28 µops (8x xmm -> gpr, 6x extr, 8x load, 6x insr)
+			features.hasSlowGather = true;
+		}
+	}
+	return features;
+}
+
+const ProcessorFeatures g_cpu = getProcessorFeatures();

pcsx2/GS/MultiISA.h

@@ -0,0 +1,85 @@
+/*  PCSX2 - PS2 Emulator for PCs
+ *  Copyright (C) 2002-2021 PCSX2 Dev Team
+ *
+ *  PCSX2 is free software: you can redistribute it and/or modify it under the terms
+ *  of the GNU Lesser General Public License as published by the Free Software Found-
+ *  ation, either version 3 of the License, or (at your option) any later version.
+ *
+ *  PCSX2 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 for more details.
+ *
+ *  You should have received a copy of the GNU General Public License along with PCSX2.
+ *  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include "PCSX2Base.h"
+
+// For multiple-isa compilation
+#ifdef MULTI_ISA_UNSHARED_COMPILATION
+	// Preprocessor should have MULTI_ISA_UNSHARED_COMPILATION defined to `isa_sse4`, `isa_avx`, or `isa_avx2`
+	#define CURRENT_ISA MULTI_ISA_UNSHARED_COMPILATION
+#else
+	// Define to isa_native in shared section in addition to multi-isa-off so if someone tries to use it they'll hopefully get a linker error and notice
+	#define CURRENT_ISA isa_native
+#endif
+
+#if defined(MULTI_ISA_UNSHARED_COMPILATION)
+	// Preprocessor should have MULTI_ISA_IS_FIRST defined to 0 or 1
+	/// Use with `#if MULTI_ISA_COMPILE_ONCE` to make a single definition of something in an otherwise multi-isa-compiled file.
+	#define MULTI_ISA_COMPILE_ONCE MULTI_ISA_IS_FIRST
+#elif !defined(MULTI_ISA_SHARED_COMPILATION)
+	#define MULTI_ISA_COMPILE_ONCE 1
+#else
+	#define MULTI_ISA_COMPILE_ONCE static_assert(0, "You don't need MULTI_ISA_COMPILE_ONCE in a non-multi-isa file!");
+#endif
+
+#ifndef MULTI_ISA_SHARED_COMPILATION
+	/// Mark the start of a header defining code that will be compiled multiple times in multi-isa mode
+	/// Anything between this and a `MULTI_ISA_UNSHARED_END` will be placed in a different namespace for each of the multilple compilations
+	#define MULTI_ISA_UNSHARED_START namespace CURRENT_ISA {
+	/// Mark the end of a header defining code that will be compiled multiple times in multi-isa mode
+	#define MULTI_ISA_UNSHARED_END }
+	/// Mark the beginning of a file implementing things that will be compiled multiple times in multi-isa mode
+	/// Takes advantage of the fact that a `using namespace` declaration will also affect any implementations of things as long as they're not valid without it
+	/// Fully global variables are valid as-is, however, and will need to have `CURRENT_ISA::` manually prepended to them.
+	/// If you forget to do this, it will show up as a linker error (either multiple definitions of the function/variable, or a "failed to find isa_native::function")
+	#define MULTI_ISA_UNSHARED_IMPL using namespace CURRENT_ISA
+#else
+	#define MULTI_ISA_UNSHARED_START static_assert(0, "This file should not be included by multi-isa shared compilation!");
+	#define MULTI_ISA_UNSHARED_IMPL static_assert(0, "This file should be compiled unshared in multi-isa mode!");
+	#define MULTI_ISA_UNSHARED_END
+#endif
+
+struct ProcessorFeatures
+{
+	enum class VectorISA { None, SSE4, AVX, AVX2 };
+	VectorISA vectorISA;
+	bool hasFMA;
+	bool hasSlowGather;
+};
+
+extern const ProcessorFeatures g_cpu;
+
+#if defined(MULTI_ISA_UNSHARED_COMPILATION) || defined(MULTI_ISA_SHARED_COMPILATION)
+	#define MULTI_ISA_DEF(...) \
+		namespace isa_sse4 { __VA_ARGS__ } \
+		namespace isa_avx  { __VA_ARGS__ } \
+		namespace isa_avx2 { __VA_ARGS__ }
+
+	#define MULTI_ISA_FRIEND(klass) \
+		friend class isa_sse4::klass; \
+		friend class isa_avx ::klass; \
+		friend class isa_avx2::klass;
+
+	#define MULTI_ISA_SELECT(fn) (\
+		::g_cpu.vectorISA == ProcessorFeatures::VectorISA::AVX2 ? isa_avx2::fn : \
+		::g_cpu.vectorISA == ProcessorFeatures::VectorISA::AVX  ? isa_avx ::fn : \
+		                                                          isa_sse4::fn)
+#else
+	#define MULTI_ISA_DEF(...) namespace isa_native { __VA_ARGS__ }
+	#define MULTI_ISA_FRIEND(klass) friend class isa_native::klass;
+	#define MULTI_ISA_SELECT(fn) (isa_native::fn)
+#endif

pcsx2/pcsx2.vcxproj

@@ -475,6 +475,7 @@
     <ClCompile Include="GS\Renderers\Common\GSVertexList.cpp" />
     <ClCompile Include="GS\Renderers\SW\GSVertexSW.cpp" />
     <ClCompile Include="GS\Renderers\Common\GSVertexTrace.cpp" />
+    <ClCompile Include="GS\MultiISA.cpp" />
     <ClCompile Include="Dump.cpp" />
     <ClCompile Include="Pcsx2Config.cpp" />
     <ClCompile Include="windows\FlatFileReaderWindows.cpp" />
@@ -875,6 +876,7 @@
     <ClInclude Include="GS\Renderers\SW\GSVertexSW.h" />
     <ClInclude Include="GS\Renderers\Common\GSVertexTrace.h" />
     <ClInclude Include="GS\resource.h" />
+    <ClInclude Include="GS\MultiISA.h" />
     <ClInclude Include="IPU\IPUdma.h" />
     <ClInclude Include="Mdec.h" />
     <ClInclude Include="Patch.h" />

pcsx2/pcsx2.vcxproj.filters

@@ -1520,6 +1520,9 @@
     <ClCompile Include="GS\GSLzma.cpp">
       <Filter>System\Ps2\GS</Filter>
     </ClCompile>
+    <ClCompile Include="GS\MultiISA.cpp">
+      <Filter>System\Ps2\GS</Filter>
+    </ClCompile>
     <ClCompile Include="gui\Dialogs\PINEDialog.cpp">
       <Filter>AppHost\Dialogs</Filter>
     </ClCompile>
@@ -2722,6 +2725,9 @@
     <ClInclude Include="GS\GSLzma.h">
       <Filter>System\Ps2\GS</Filter>
     </ClInclude>
+    <ClInclude Include="GS\MultiISA.h">
+      <Filter>System\Ps2\GS</Filter>
+    </ClInclude>
     <ClInclude Include="GS\Renderers\Common\GSFastList.h">
       <Filter>System\Ps2\GS</Filter>
     </ClInclude>

pcsx2/pcsx2core.vcxproj

@@ -328,6 +328,7 @@
     <ClCompile Include="GS\Renderers\Common\GSVertexList.cpp" />
     <ClCompile Include="GS\Renderers\SW\GSVertexSW.cpp" />
     <ClCompile Include="GS\Renderers\Common\GSVertexTrace.cpp" />
+    <ClCompile Include="GS\MultiISA.cpp" />
     <ClCompile Include="SPU2\Windows\SndOut_XAudio2.cpp" />
     <ClCompile Include="USB\USBNull.cpp" />
     <ClCompile Include="Dump.cpp" />
@@ -651,6 +652,7 @@
     <ClInclude Include="GS\Renderers\Common\GSVertexList.h" />
     <ClInclude Include="GS\Renderers\SW\GSVertexSW.h" />
     <ClInclude Include="GS\Renderers\Common\GSVertexTrace.h" />
+    <ClInclude Include="GS\MultiISA.h" />
     <ClInclude Include="GS\resource.h" />
     <ClInclude Include="IPU\IPUdma.h" />
     <ClInclude Include="Mdec.h" />

pcsx2/pcsx2core.vcxproj.filters

@@ -1013,6 +1013,9 @@
     <ClCompile Include="GS\GSVector.cpp">
       <Filter>System\Ps2\GS</Filter>
     </ClCompile>
+    <ClCompile Include="GS\MultiISA.cpp">
+      <Filter>System\Ps2\GS</Filter>
+    </ClCompile>
     <ClCompile Include="GS\GSPng.cpp">
       <Filter>System\Ps2\GS</Filter>
     </ClCompile>
@@ -1871,6 +1874,9 @@
     <ClInclude Include="GS\GSVector8.h">
       <Filter>System\Ps2\GS</Filter>
     </ClInclude>
+    <ClInclude Include="GS\MultiISA.h">
+      <Filter>System\Ps2\GS</Filter>
+    </ClInclude>
     <ClInclude Include="GS\config.h">
       <Filter>System\Ps2\GS</Filter>
     </ClInclude>