From 9aec4229d509fc2247319ca3be5ba56656cf94ad Mon Sep 17 00:00:00 2001
From: Gregory Hainaut <gregory.hainaut@gmail.com>
Date: Sat, 28 Nov 2015 23:39:06 +0100
Subject: [PATCH] x86emitter: support AVX2 (linux only?)

Nah kidding, it is only the cpu detection for the log

x86 Features Detected:
	SSE2.. SSE3.. SSSE3.. SSE4.1.. SSE4.2.. AVX.. AVX2.. FMA
---
 common/include/intrin_x86.h         | 3 ++-
 common/include/x86emitter/tools.h   | 2 ++
 common/src/x86emitter/cpudetect.cpp | 9 +++++++++
 pcsx2/System.cpp                    | 1 +
 4 files changed, 14 insertions(+), 1 deletion(-)

diff --git a/common/include/intrin_x86.h b/common/include/intrin_x86.h
index 2755b2878c..076ae49dfb 100644
--- a/common/include/intrin_x86.h
+++ b/common/include/intrin_x86.h
@@ -96,7 +96,8 @@ static __inline__ __attribute__((always_inline)) s32 _InterlockedIncrement(volat
 /*** System information ***/
 static __inline__ __attribute__((always_inline)) void __cpuid(int CPUInfo[], const int InfoType)
 {
-	__asm__ __volatile__("cpuid" : "=a" (CPUInfo[0]), "=b" (CPUInfo[1]), "=c" (CPUInfo[2]), "=d" (CPUInfo[3]) : "a" (InfoType));
+	// ECX allow to select the leaf. Leaf 0 is the one that you want to get, so I just xor the register
+	__asm__ __volatile__("xor %%ecx, %%ecx\n" "cpuid": "=a" (CPUInfo[0]), "=b" (CPUInfo[1]), "=c" (CPUInfo[2]), "=d" (CPUInfo[3]) : "a" (InfoType));
 }
 
 static __inline__ __attribute__((always_inline)) unsigned long long __xgetbv(unsigned int index)
diff --git a/common/include/x86emitter/tools.h b/common/include/x86emitter/tools.h
index fcbe84a2d6..0ab33fa76c 100644
--- a/common/include/x86emitter/tools.h
+++ b/common/include/x86emitter/tools.h
@@ -44,6 +44,7 @@ public:
 	u32 Flags2;			// More Feature Flags
 	u32 EFlags;			// Extended Feature Flags
 	u32 EFlags2;		// Extended Feature Flags pg2
+	u32 SEFlag;			// Structured Extended Feature Flags Enumeration
 
 	char VendorName[16];	// Vendor/Creator ID
 	char FamilyName[50];    // the original cpu name
@@ -88,6 +89,7 @@ public:
 	u32 hasStreamingSIMD4Extensions					:1;
 	u32 hasStreamingSIMD4Extensions2				:1;
 	u32 hasAVX										:1;
+	u32 hasAVX2										:1;
 	u32 hasFMA										:1;
 
 	// AMD-specific CPU Features
diff --git a/common/src/x86emitter/cpudetect.cpp b/common/src/x86emitter/cpudetect.cpp
index 0e58986660..f6140a6675 100644
--- a/common/src/x86emitter/cpudetect.cpp
+++ b/common/src/x86emitter/cpudetect.cpp
@@ -208,6 +208,14 @@ void x86capabilities::Identify()
 		Flags2		=  regs[ 2 ];
 	}
 
+	if ( cmds >= 0x00000007 )
+	{
+		// Note: ECX must be 0. I did it directly in the __cpuid asm instrinsic
+		__cpuid( regs, 0x00000007 );
+
+		SEFlag = regs[ 1 ];
+	}
+
 	__cpuid( regs, 0x80000000 );
 	cmds = regs[ 0 ];
 	if ( cmds >= 0x80000001 )
@@ -271,6 +279,7 @@ void x86capabilities::Identify()
 		{
 			hasAVX								= ( Flags2 >> 28 ) & 1; //avx
 			hasFMA								= ( Flags2 >> 12 ) & 1; //fma
+			hasAVX2								= ( SEFlag >>  5 ) & 1; //avx2
 		}
 	}
 
diff --git a/pcsx2/System.cpp b/pcsx2/System.cpp
index 96a6df6dea..277312c90c 100644
--- a/pcsx2/System.cpp
+++ b/pcsx2/System.cpp
@@ -267,6 +267,7 @@ void SysLogMachineCaps()
 	if( x86caps.hasStreamingSIMD4Extensions )		features[0].Add( L"SSE4.1" );
 	if( x86caps.hasStreamingSIMD4Extensions2 )		features[0].Add( L"SSE4.2" );
 	if( x86caps.hasAVX )							features[0].Add( L"AVX" );
+	if( x86caps.hasAVX2 )							features[0].Add( L"AVX2" );
 	if( x86caps.hasFMA)								features[0].Add( L"FMA" );
 
 	if( x86caps.hasMultimediaExtensionsExt )		features[1].Add( L"MMX2  " );