Prune SSE2 code out when building specifically for SSSE3

BUG=none
TEST=none
Review URL: https://webrtc-codereview.appspot.com/517001

git-svn-id: http://libyuv.googlecode.com/svn/trunk@250 16f28f9a-4ce2-e073-06de-1de4eb20be90

include/libyuv/planar_functions.h

@@ -160,12 +160,12 @@ int I422ToUYVY(const uint8* src_y, int src_stride_y,
                uint8* dst_frame, int dst_stride_frame,
                int width, int height);
 
-// Convert unattentuated ARGB values to preattenuated ARGB.
+// Convert unattentuated ARGB to preattenuated ARGB.
 int ARGBAttenuate(const uint8* src_argb, int src_stride_argb,
                   uint8* dst_argb, int dst_stride_argb,
                   int width, int height);
 
-// Convert preattentuated ARGB values to unattenuated ARGB.
+// Convert preattentuated ARGB to unattenuated ARGB.
 int ARGBUnattenuate(const uint8* src_argb, int src_stride_argb,
                     uint8* dst_argb, int dst_stride_argb,
                     int width, int height);

source/planar_functions.cc

@@ -10,7 +10,6 @@
 
 #include "libyuv/planar_functions.h"
 
-#include <stdio.h>  // printf()
 #include <string.h>  // for memset()
 
 #include "libyuv/cpu_id.h"
@@ -144,9 +143,10 @@ int ARGBCopy(const uint8* src_argb, int src_stride_argb,
 ARGBBlendRow GetARGBBlend(uint8* dst_argb, int dst_stride_argb, int width) {
   void (*ARGBBlendRow)(const uint8* src_argb, const uint8* src_argb1,
                        uint8* dst_argb, int width) = ARGBBlendRow_C;
-#if defined(HAS_ARGBBLENDROW_SSE2)
+#if defined(HAS_ARGBBLENDROW1_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     ARGBBlendRow = ARGBBlendRow1_SSE2;
+#if defined(HAS_ARGBBLENDROW_SSE2)
     if (width >= 4) {
       ARGBBlendRow = ARGBBlendRow_Any_SSE2;
       if (IS_ALIGNED(width, 4) &&
@@ -154,6 +154,7 @@ ARGBBlendRow GetARGBBlend(uint8* dst_argb, int dst_stride_argb, int width) {
         ARGBBlendRow = ARGBBlendRow_Aligned_SSE2;
       }
     }
+#endif
   }
 #endif
 #if defined(HAS_ARGBBLENDROW_SSSE3)
@@ -864,7 +865,7 @@ int ARGBRect(uint8* dst_argb, int dst_stride_argb,
   return 0;
 }
 
-// Convert unattentuated ARGB values to preattenuated ARGB.
+// Convert unattentuated ARGB to preattenuated ARGB.
 // An unattenutated ARGB alpha blend uses the formula
 // p = a * f + (1 - a) * b
 // where
@@ -910,7 +911,7 @@ int ARGBAttenuate(const uint8* src_argb, int src_stride_argb,
   return 0;
 }
 
-// Convert unattentuated ARGB values to preattenuated ARGB.
+// Convert preattentuated ARGB to unattenuated ARGB.
 int ARGBUnattenuate(const uint8* src_argb, int src_stride_argb,
                     uint8* dst_argb, int dst_stride_argb,
                     int width, int height) {

source/row.h

@@ -25,50 +25,60 @@ extern "C" {
 #if defined(COVERAGE_ENABLED) || defined(TARGET_IPHONE_SIMULATOR)
 #define YUV_DISABLE_ASM
 #endif
+// True if compiling for SSSE3 as a requirement.
+#if defined(__SSSE3__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 3))
+#define LIBYUV_SSSE3_ONLY
+#endif
 
 // The following are available on all x86 platforms
 #if !defined(YUV_DISABLE_ASM) && \
     (defined(_M_IX86) || defined(__x86_64__) || defined(__i386__))
-
 #define HAS_ABGRTOARGBROW_SSSE3
-#define HAS_BGRATOARGBROW_SSSE3
-#define HAS_RGB24TOARGBROW_SSSE3
-#define HAS_RAWTOARGBROW_SSSE3
-#define HAS_RGB565TOARGBROW_SSE2
+#define HAS_ABGRTOUVROW_SSSE3
+#define HAS_ABGRTOYROW_SSSE3
 #define HAS_ARGB1555TOARGBROW_SSE2
 #define HAS_ARGB4444TOARGBROW_SSE2
-#define HAS_ARGBTORGB24ROW_SSSE3
-#define HAS_ARGBTORAWROW_SSSE3
-#define HAS_ARGBTORGB565ROW_SSE2
+#define HAS_ARGBATTENUATE_SSSE3
+#define HAS_ARGBBLENDROW_SSSE3
+#define HAS_ARGBBLENDROW1_SSE2
 #define HAS_ARGBTOARGB1555ROW_SSE2
 #define HAS_ARGBTOARGB4444ROW_SSE2
-#define HAS_ARGBTOYROW_SSSE3
-#define HAS_BGRATOYROW_SSSE3
-#define HAS_ABGRTOYROW_SSSE3
+#define HAS_ARGBTORAWROW_SSSE3
+#define HAS_ARGBTORGB24ROW_SSSE3
+#define HAS_ARGBTORGB565ROW_SSE2
 #define HAS_ARGBTOUVROW_SSSE3
+#define HAS_ARGBTOYROW_SSSE3
+#define HAS_ARGBUNATTENUATE_SSE2
+#define HAS_BGRATOARGBROW_SSSE3
 #define HAS_BGRATOUVROW_SSSE3
-#define HAS_ABGRTOUVROW_SSSE3
+#define HAS_BGRATOYROW_SSSE3
+#define HAS_COPYROW_SSE2
+#define HAS_COPYROW_X86
 #define HAS_I400TOARGBROW_SSE2
-#define HAS_YTOARGBROW_SSE2
+#define HAS_I420TOABGRROW_SSSE3
 #define HAS_I420TOARGBROW_SSSE3
 #define HAS_I420TOBGRAROW_SSSE3
-#define HAS_I420TOABGRROW_SSSE3
 #define HAS_I444TOARGBROW_SSSE3
 #define HAS_MIRRORROW_SSSE3
-#define HAS_MIRRORROW_SSE2
 #define HAS_MIRRORROWUV_SSSE3
+#define HAS_RAWTOARGBROW_SSSE3
+#define HAS_RGB24TOARGBROW_SSSE3
+#define HAS_RGB565TOARGBROW_SSE2
 #define HAS_SPLITUV_SSE2
-#define HAS_COPYROW_SSE2
-#define HAS_COPYROW_X86
-#define HAS_YUY2TOYROW_SSE2
+#define HAS_UYVYTOUVROW_SSE2
 #define HAS_UYVYTOYROW_SSE2
+#define HAS_YTOARGBROW_SSE2
 #define HAS_YUY2TOUVROW_SSE2
-#define HAS_UYVYTOUVROW_SSE2
-#define HAS_ARGBBLENDROW_SSE2
-#define HAS_ARGBBLENDROW_SSSE3
+#define HAS_YUY2TOYROW_SSE2
+#endif
+
+// The following are available only useful when SSSE3 is unavailable.
+#if !defined(YUV_DISABLE_ASM) && \
+    (defined(_M_IX86) || defined(__x86_64__) || defined(__i386__)) && \
+    !defined(LIBYUV_SSSE3_ONLY)
+#define HAS_MIRRORROW_SSE2
 #define HAS_ARGBATTENUATE_SSE2
-#define HAS_ARGBATTENUATE_SSSE3
-#define HAS_ARGBUNATTENUATE_SSE2
+#define HAS_ARGBBLENDROW_SSE2
 #endif
 
 // The following are available on Neon platforms

source/row_posix.cc

@@ -2105,7 +2105,9 @@ void ARGBBlendRow_Aligned_SSE2(const uint8* src_argb0, const uint8* src_argb1,
 #endif
   );
 }
+#endif  // HAS_ARGBBLENDROW_SSE2
 
+#ifdef HAS_ARGBBLENDROW1_SSE2
 // Blend 1 pixel at a time, unaligned
 void ARGBBlendRow1_SSE2(const uint8* src_argb0, const uint8* src_argb1,
                         uint8* dst_argb, int width) {
@@ -2156,7 +2158,7 @@ void ARGBBlendRow1_SSE2(const uint8* src_argb0, const uint8* src_argb1,
 #endif
   );
 }
-#endif  // HAS_ARGBBLENDROW_SSE2
+#endif  // HAS_ARGBBLENDROW1_SSE2
 
 #ifdef HAS_ARGBBLENDROW_SSSE3
 // Shuffle table for reversing the bytes.

source/row_win.cc

@@ -2151,7 +2151,9 @@ void ARGBBlendRow_Aligned_SSE2(const uint8* src_argb0, const uint8* src_argb1,
     ret
   }
 }
+#endif  // HAS_ARGBBLENDROW_SSE2
 
+#ifdef HAS_ARGBBLENDROW1_SSE2
 // Blend 1 pixel at a time, unaligned.
 __declspec(naked) __declspec(align(16))
 void ARGBBlendRow1_SSE2(const uint8* src_argb0, const uint8* src_argb1,
@@ -2202,8 +2204,8 @@ void ARGBBlendRow1_SSE2(const uint8* src_argb0, const uint8* src_argb1,
     ret
   }
 }
+#endif  // HAS_ARGBBLENDROW1_SSE2
 
-#endif  // HAS_ARGBBLENDROW_SSE2
 #ifdef HAS_ARGBBLENDROW_SSSE3
 // Shuffle table for reversing the bytes.
 static const uvec8 kShuffleAlpha = {