RGB565ToARGB_SSE2

BUG=none
TEST=media_unittest
Review URL: http://webrtc-codereview.appspot.com/347011

git-svn-id: http://libyuv.googlecode.com/svn/trunk@132 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 131
+Version: 132
 License: BSD
 License File: LICENSE
 

source/row.h

@@ -61,6 +61,7 @@
 // The following are available on Windows platforms
 #if defined(_M_IX86) && !defined(YUV_DISABLE_ASM)
 #define HAS_ARGB4444TOARGBROW_SSE2
+#define HAS_RGB565TOARGBROW_SSE2
 #endif
 
 // The following are available on Neon platforms
@@ -213,10 +214,12 @@ void ABGRToARGBRow_SSSE3(const uint8* src_abgr, uint8* dst_argb, int pix);
 void BGRAToARGBRow_SSSE3(const uint8* src_bgra, uint8* dst_argb, int pix);
 void RGB24ToARGBRow_SSSE3(const uint8* src_rgb24, uint8* dst_argb, int pix);
 void RAWToARGBRow_SSSE3(const uint8* src_rgb24, uint8* dst_argb, int pix);
-// TODO(fbarchard): SSE2 565 etc
+// TODO(fbarchard): SSE2 555
-//void RGB565ToARGBRow_SSE2(const uint8* src_rgb, uint8* dst_argb, int pix);
 //void ARGB1555ToARGBRow_SSE2(const uint8* src_argb, uint8* dst_argb, int pix);
 #endif
+#ifdef HAS_RGB565TOARGBROW_SSE2
+void RGB565ToARGBRow_SSE2(const uint8* src_argb, uint8* dst_argb, int pix);
+#endif
 #ifdef HAS_ARGB4444TOARGBROW_SSE2
 void ARGB4444ToARGBRow_SSE2(const uint8* src_argb, uint8* dst_argb, int pix);
 #endif

source/row_common.cc

@@ -318,7 +318,11 @@ void RAWToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) {
 // TODO(fbarchard): RGB565ToARGBRow_SSE2 for rgb conversion
 void RGB565ToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int pix) {
   SIMD_ALIGNED(uint8 row[kMaxStride]);
+#ifdef HAS_RGB565TOARGBROW_SSE2
+  RGB565ToARGBRow_SSE2(src_argb, row, pix);
+#else
   RGB565ToARGBRow_C(src_argb, row, pix);
+#endif
   ARGBToYRow_SSSE3(row, dst_y, pix);
 }
 
@@ -361,8 +365,13 @@ void RAWToUVRow_SSSE3(const uint8* src_argb, int src_stride_argb,
 void RGB565ToUVRow_SSSE3(const uint8* src_argb, int src_stride_argb,
                          uint8* dst_u, uint8* dst_v, int pix) {
   SIMD_ALIGNED(uint8 row[kMaxStride * 2]);
+#ifdef HAS_RGB565TOARGBROW_SSE2
+  RGB565ToARGBRow_SSE2(src_argb, row, pix);
+  RGB565ToARGBRow_SSE2(src_argb + src_stride_argb, row + kMaxStride, pix);
+#else
   RGB565ToARGBRow_C(src_argb, row, pix);
   RGB565ToARGBRow_C(src_argb + src_stride_argb, row + kMaxStride, pix);
+#endif
   ARGBToUVRow_SSSE3(row, kMaxStride, dst_u, dst_v, pix);
 }
 

source/row_win.cc

@@ -229,6 +229,65 @@ __asm {
   }
 }
 
+// TODO(fbarchard): Port RGB565ToARGBRow_SSE2 to gcc
+__declspec(naked)
+void RGB565ToARGBRow_SSE2(const uint8* src_rgb565, uint8* dst_argb,
+                          int pix) {
+__asm {
+    mov       eax, [esp + 4]   // src_rgb565
+    mov       edx, [esp + 8]   // dst_argb
+    mov       ecx, [esp + 12]  // pix
+    pcmpeqb   xmm5, xmm5       // generate mask 0xff000000
+    pslld     xmm5, 24
+    pcmpeqb   xmm4, xmm4       // generate mask 0xf800f800
+    psllw     xmm4, 11
+    pcmpeqb   xmm6, xmm6       // generate mask 0x001f001f
+    psrlw     xmm6, 11
+    pcmpeqb   xmm7, xmm7       // generate mask 0x00fc00fc
+    psrlw     xmm7, 10
+    psllw     xmm7, 2
+
+
+ convertloop:
+    movdqa    xmm0, [eax] // fetch 8 pixels of bgr565
+    lea       eax, [eax + 16]
+
+    movdqa    xmm1, xmm0
+    movdqa    xmm2, xmm0
+    pand      xmm1, xmm4    // R in upper 5 bits
+    psrlw     xmm2, 13      // R 3 bits
+    psllw     xmm2, 8
+    por       xmm1, xmm2
+
+    movdqa    xmm2, xmm0
+    pand      xmm2, xmm6    // mask B 5 bits
+    movdqa    xmm3, xmm2
+    psllw     xmm2, 3
+    psrlw     xmm3, 2
+    por       xmm2, xmm3
+
+    por       xmm1, xmm2    // RB
+
+    psrlw     xmm0, 3       // G in top 6 bits of lower byte
+    pand      xmm0, xmm7    // mask G 6 bits
+    movdqa    xmm2, xmm0
+    psrlw     xmm2, 6
+    por       xmm0, xmm2
+
+    por       xmm0, xmm5   // AG
+
+    movdqa    xmm2, xmm1
+    punpcklbw xmm1, xmm0
+    punpckhbw xmm2, xmm0
+    movdqa    [edx], xmm1  // store 4 pixels of ARGB
+    movdqa    [edx + 16], xmm2  // store next 4 pixels of ARGB
+    lea       edx, [edx + 32]
+    sub       ecx, 8
+    ja        convertloop
+    ret
+  }
+}
+
 // TODO(fbarchard): Port ARGB4444ToARGBRow_SSE2 to gcc
 __declspec(naked)
 void ARGB4444ToARGBRow_SSE2(const uint8* src_argb4444, uint8* dst_argb,