shithub: libvpx

Info  •  Files  •  Log  •  Branches

Merge "vpx_int_pro_col_sse2: use unaligned loads"

--- a/vpx_dsp/x86/avg_intrin_sse2.c

+++ b/vpx_dsp/x86/avg_intrin_sse2.c

@@ -515,7 +515,7 @@

 int16_t vpx_int_pro_col_sse2(const uint8_t *ref, const int width) {

   __m128i zero = _mm_setzero_si128();

-  __m128i src_line = _mm_load_si128((const __m128i *)ref);

+  __m128i src_line = _mm_loadu_si128((const __m128i *)ref);

   __m128i s0 = _mm_sad_epu8(src_line, zero);

   __m128i s1;

   int i;

@@ -522,7 +522,7 @@

   for (i = 16; i < width; i += 16) {

     ref += 16;

-    src_line = _mm_load_si128((const __m128i *)ref);

+    src_line = _mm_loadu_si128((const __m128i *)ref);

     s1 = _mm_sad_epu8(src_line, zero);

     s0 = _mm_adds_epu16(s0, s1);

   }