libwebp: update to 0.5.0

third_party/libwebp/dsp/upsampling_sse2.c

 // Copyright 2011 Google Inc. All Rights Reserved.
 //
 // Use of this source code is governed by a BSD-style license
 // that can be found in the COPYING file in the root of the source
 // tree. An additional intellectual property rights grant can be found
 // in the file PATENTS. All contributing project authors may
 // be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
 // SSE2 version of YUV to RGB upsampling functions.
@@ skipping 42 matching lines @@
   const __m128i t_b = _mm_avg_epu8(b, db);  /* (3a + 9b +  c + 3d + 8) / 16 */ \
   const __m128i t_1 = _mm_unpacklo_epi8(t_a, t_b);                             \
   const __m128i t_2 = _mm_unpackhi_epi8(t_a, t_b);                             \
   _mm_store_si128(((__m128i*)(out)) + 0, t_1);                                 \
   _mm_store_si128(((__m128i*)(out)) + 1, t_2);                                 \
 } while (0)
 
 // Loads 17 pixels each from rows r1 and r2 and generates 32 pixels.
 #define UPSAMPLE_32PIXELS(r1, r2, out) {                                       \
   const __m128i one = _mm_set1_epi8(1);                                        \
-  const __m128i a = _mm_loadu_si128((__m128i*)&(r1)[0]);                       \
-  const __m128i b = _mm_loadu_si128((__m128i*)&(r1)[1]);                       \
-  const __m128i c = _mm_loadu_si128((__m128i*)&(r2)[0]);                       \
-  const __m128i d = _mm_loadu_si128((__m128i*)&(r2)[1]);                       \
+  const __m128i a = _mm_loadu_si128((const __m128i*)&(r1)[0]);                 \
+  const __m128i b = _mm_loadu_si128((const __m128i*)&(r1)[1]);                 \
+  const __m128i c = _mm_loadu_si128((const __m128i*)&(r2)[0]);                 \
+  const __m128i d = _mm_loadu_si128((const __m128i*)&(r2)[1]);                 \
                                                                                \
   const __m128i s = _mm_avg_epu8(a, d);        /* s = (a + d + 1) / 2 */       \
   const __m128i t = _mm_avg_epu8(b, c);        /* t = (b + c + 1) / 2 */       \
   const __m128i st = _mm_xor_si128(s, t);      /* st = s^t */                  \
                                                                                \
   const __m128i ad = _mm_xor_si128(a, d);      /* ad = a^d */                  \
   const __m128i bc = _mm_xor_si128(b, c);      /* bc = b^c */                  \
                                                                                \
   const __m128i t1 = _mm_or_si128(ad, bc);     /* (a^d) | (b^c) */             \
   const __m128i t2 = _mm_or_si128(t1, st);     /* (a^d) | (b^c) | (s^t) */     \
@@ skipping 89 matching lines @@
     CONVERT2RGB(FUNC, XSTEP, top_y, bottom_y, top_dst, bottom_dst,             \
                 pos, len - pos);                                               \
   }                                                                            \
 }
 
 // SSE2 variants of the fancy upsampler.
 SSE2_UPSAMPLE_FUNC(UpsampleRgbLinePair,  VP8YuvToRgb,  3)
 SSE2_UPSAMPLE_FUNC(UpsampleBgrLinePair,  VP8YuvToBgr,  3)
 SSE2_UPSAMPLE_FUNC(UpsampleRgbaLinePair, VP8YuvToRgba, 4)
 SSE2_UPSAMPLE_FUNC(UpsampleBgraLinePair, VP8YuvToBgra, 4)
+SSE2_UPSAMPLE_FUNC(UpsampleArgbLinePair, VP8YuvToArgb, 4)
+SSE2_UPSAMPLE_FUNC(UpsampleRgba4444LinePair, VP8YuvToRgba4444, 2)
+SSE2_UPSAMPLE_FUNC(UpsampleRgb565LinePair, VP8YuvToRgb565, 2)
 
 #undef GET_M
 #undef PACK_AND_STORE
 #undef UPSAMPLE_32PIXELS
 #undef UPSAMPLE_LAST_BLOCK
 #undef CONVERT2RGB
 #undef CONVERT2RGB_32
 #undef SSE2_UPSAMPLE_FUNC
 
-#endif  // FANCY_UPSAMPLING
+//------------------------------------------------------------------------------
+// Entry point
 
-#endif   // WEBP_USE_SSE2
-
-//------------------------------------------------------------------------------
+extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
 
 extern void WebPInitUpsamplersSSE2(void);
 
-#ifdef FANCY_UPSAMPLING
-
-extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
-
-void WebPInitUpsamplersSSE2(void) {
-#if defined(WEBP_USE_SSE2)
-  VP8YUVInitSSE2();
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplersSSE2(void) {
   WebPUpsamplers[MODE_RGB]  = UpsampleRgbLinePair;
   WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair;
   WebPUpsamplers[MODE_BGR]  = UpsampleBgrLinePair;
   WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair;
+  WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair;
   WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair;
   WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair;
-#endif   // WEBP_USE_SSE2
+  WebPUpsamplers[MODE_Argb] = UpsampleArgbLinePair;
+  WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair;
+  WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
+  WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
+}
+
+#endif  // FANCY_UPSAMPLING
+
+//------------------------------------------------------------------------------
+
+extern WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */];
+extern void WebPInitYUV444ConvertersSSE2(void);
+
+#define YUV444_FUNC(FUNC_NAME, CALL, XSTEP) \
+extern void WebP##FUNC_NAME##C(const uint8_t* y, const uint8_t* u,             \
+                               const uint8_t* v, uint8_t* dst, int len);       \
+static void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v,    \
+                      uint8_t* dst, int len) {                                 \
+  int i;                                                                       \
+  const int max_len = len & ~31;                                               \
+  for (i = 0; i < max_len; i += 32) CALL(y + i, u + i, v + i, dst + i * XSTEP);\
+  if (i < len) {  /* C-fallback */                                             \
+    WebP##FUNC_NAME##C(y + i, u + i, v + i, dst + i * XSTEP, len - i);         \
+  }                                                                            \
+}
+
+YUV444_FUNC(Yuv444ToRgba, VP8YuvToRgba32, 4);
+YUV444_FUNC(Yuv444ToBgra, VP8YuvToBgra32, 4);
+YUV444_FUNC(Yuv444ToRgb, VP8YuvToRgb32, 3);
+YUV444_FUNC(Yuv444ToBgr, VP8YuvToBgr32, 3);
+
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitYUV444ConvertersSSE2(void) {
+  WebPYUV444Converters[MODE_RGBA] = Yuv444ToRgba;
+  WebPYUV444Converters[MODE_BGRA] = Yuv444ToBgra;
+  WebPYUV444Converters[MODE_RGB]  = Yuv444ToRgb;
+  WebPYUV444Converters[MODE_BGR]  = Yuv444ToBgr;
 }
 
 #else
 
-// this empty function is to avoid an empty .o
-void WebPInitUpsamplersSSE2(void) {}
+WEBP_DSP_INIT_STUB(WebPInitYUV444ConvertersSSE2)
 
-#endif  // FANCY_UPSAMPLING
+#endif  // WEBP_USE_SSE2
+
+#if !(defined(FANCY_UPSAMPLING) && defined(WEBP_USE_SSE2))
+WEBP_DSP_INIT_STUB(WebPInitUpsamplersSSE2)
+#endif