Bugfix: Enable SSE4.2 on GCC

fbarriga · 10-09-2011, 01:59 PM #1

You need to compile it with the flag -msse4.2

Code:

diff --git a/Source/Core/Common/Src/Common.h b/Source/Core/Common/Src/Common.h

index 888a206..fd0f730 100644

--- a/Source/Core/Common/Src/Common.h

+++ b/Source/Core/Common/Src/Common.h

@@ -141,6 +141,8 @@ private:

 #define _M_SSE 0x300

 #elif defined __APPLE__ && defined __x86_64__

 #define _M_SSE 0x301

+#elif (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3)) && defined __SSE4_2__

+#define _M_SSE 0x402

 #elif __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3)

 #define _M_SSE 0x301

 #elif (_MSC_VER >= 1500) || __INTEL_COMPILER // Visual Studio 2008

diff --git a/Source/Core/VideoCommon/Src/VertexLoader_TextCoord.cpp b/Source/Core/VideoCommon/Src/VertexLoader_TextCoord.cpp

index de85308..6d1e4e4 100644

--- a/Source/Core/VideoCommon/Src/VertexLoader_TextCoord.cpp

+++ b/Source/Core/VideoCommon/Src/VertexLoader_TextCoord.cpp

@@ -438,7 +438,7 @@ void VertexLoader_TextCoord::Init(void) {

 #if _M_SSE >= 0x401

-    if (cpu_info.bSSE4_1) {

+    if (cpu_info.bSSE4_1 && cpu_info.bSSSE3) {

         tableReadTexCoord[3][3][1] = TexCoord_ReadIndex16_Short2_SSE4;

     }

lamedude · 10-10-2011, 01:09 AM #2

Don't Linux users compile thier stuff with -march=native which would include -msse4.2 if their CPU has it?

fbarriga · 10-10-2011, 01:44 AM #3

(10-10-2011, 01:09 AM)lamedude Wrote: Don't Linux users compile thier stuff with -march=native which would include -msse4.2 if their CPU has it?

No, it enable other flags even if your system support SSE4.2.

fbarriga · 10-10-2011, 03:02 AM #4

(10-10-2011, 01:44 AM)fbarriga Wrote:
(10-10-2011, 01:09 AM)lamedude Wrote: Don't Linux users compile thier stuff with -march=native which would include -msse4.2 if their CPU has it?

No, it enable other flags even if your system support SSE4.2.

My fault, damn SSE* nomenclature, my CPU doesn't support SSE4.2 (so I suppose thst you can ignore my '-msse4.2' comment).