commit 5cbeca8c515141a30126d2ca6a41d1028413677f
author Ravi Chaudhary <ravi.chaudhary@ittiam.com> Wed Oct 17 12:34:08 2018 +0530
committer Yunqing Wang <yunqingwang@google.com> Wed Oct 17 21:00:45 2018 +0000
tree 799d39f6636d7cef1b6ec27f0f472cfb3b8058b5
parent 1996a35dc0ccac5709b6f92d64ee59068fab49e7

Fix thread sanitizer issue in hbd modules

Read/write race condition of dst.buf for width=4 has been fixed.

Change-Id: Iaf6a8ed06f86a179ba331205ea26239ccbee14fe

aom_dsp/x86/highbd_subtract_sse2.c

diff --git a/aom_dsp/x86/highbd_subtract_sse2.c b/aom_dsp/x86/highbd_subtract_sse2.c
index 18eb03d..b72d1cf 100644
--- a/aom_dsp/x86/highbd_subtract_sse2.c
+++ b/aom_dsp/x86/highbd_subtract_sse2.c
@@ -29,15 +29,15 @@
   __m128i x0, x1, x2, x3;
   int64_t *store_diff = (int64_t *)(diff + 0 * diff_stride);
 
-  u0 = _mm_loadu_si128((__m128i const *)(src + 0 * src_stride));
-  u1 = _mm_loadu_si128((__m128i const *)(src + 1 * src_stride));
-  u2 = _mm_loadu_si128((__m128i const *)(src + 2 * src_stride));
-  u3 = _mm_loadu_si128((__m128i const *)(src + 3 * src_stride));
+  u0 = _mm_loadl_epi64((__m128i const *)(src + 0 * src_stride));
+  u1 = _mm_loadl_epi64((__m128i const *)(src + 1 * src_stride));
+  u2 = _mm_loadl_epi64((__m128i const *)(src + 2 * src_stride));
+  u3 = _mm_loadl_epi64((__m128i const *)(src + 3 * src_stride));
 
-  v0 = _mm_loadu_si128((__m128i const *)(pred + 0 * pred_stride));
-  v1 = _mm_loadu_si128((__m128i const *)(pred + 1 * pred_stride));
-  v2 = _mm_loadu_si128((__m128i const *)(pred + 2 * pred_stride));
-  v3 = _mm_loadu_si128((__m128i const *)(pred + 3 * pred_stride));
+  v0 = _mm_loadl_epi64((__m128i const *)(pred + 0 * pred_stride));
+  v1 = _mm_loadl_epi64((__m128i const *)(pred + 1 * pred_stride));
+  v2 = _mm_loadl_epi64((__m128i const *)(pred + 2 * pred_stride));
+  v3 = _mm_loadl_epi64((__m128i const *)(pred + 3 * pred_stride));
 
   x0 = _mm_sub_epi16(u0, v0);
   x1 = _mm_sub_epi16(u1, v1);
@@ -61,23 +61,23 @@
   __m128i x0, x1, x2, x3, x4, x5, x6, x7;
   int64_t *store_diff = (int64_t *)(diff + 0 * diff_stride);
 
-  u0 = _mm_loadu_si128((__m128i const *)(src + 0 * src_stride));
-  u1 = _mm_loadu_si128((__m128i const *)(src + 1 * src_stride));
-  u2 = _mm_loadu_si128((__m128i const *)(src + 2 * src_stride));
-  u3 = _mm_loadu_si128((__m128i const *)(src + 3 * src_stride));
-  u4 = _mm_loadu_si128((__m128i const *)(src + 4 * src_stride));
-  u5 = _mm_loadu_si128((__m128i const *)(src + 5 * src_stride));
-  u6 = _mm_loadu_si128((__m128i const *)(src + 6 * src_stride));
-  u7 = _mm_loadu_si128((__m128i const *)(src + 7 * src_stride));
+  u0 = _mm_loadl_epi64((__m128i const *)(src + 0 * src_stride));
+  u1 = _mm_loadl_epi64((__m128i const *)(src + 1 * src_stride));
+  u2 = _mm_loadl_epi64((__m128i const *)(src + 2 * src_stride));
+  u3 = _mm_loadl_epi64((__m128i const *)(src + 3 * src_stride));
+  u4 = _mm_loadl_epi64((__m128i const *)(src + 4 * src_stride));
+  u5 = _mm_loadl_epi64((__m128i const *)(src + 5 * src_stride));
+  u6 = _mm_loadl_epi64((__m128i const *)(src + 6 * src_stride));
+  u7 = _mm_loadl_epi64((__m128i const *)(src + 7 * src_stride));
 
-  v0 = _mm_loadu_si128((__m128i const *)(pred + 0 * pred_stride));
-  v1 = _mm_loadu_si128((__m128i const *)(pred + 1 * pred_stride));
-  v2 = _mm_loadu_si128((__m128i const *)(pred + 2 * pred_stride));
-  v3 = _mm_loadu_si128((__m128i const *)(pred + 3 * pred_stride));
-  v4 = _mm_loadu_si128((__m128i const *)(pred + 4 * pred_stride));
-  v5 = _mm_loadu_si128((__m128i const *)(pred + 5 * pred_stride));
-  v6 = _mm_loadu_si128((__m128i const *)(pred + 6 * pred_stride));
-  v7 = _mm_loadu_si128((__m128i const *)(pred + 7 * pred_stride));
+  v0 = _mm_loadl_epi64((__m128i const *)(pred + 0 * pred_stride));
+  v1 = _mm_loadl_epi64((__m128i const *)(pred + 1 * pred_stride));
+  v2 = _mm_loadl_epi64((__m128i const *)(pred + 2 * pred_stride));
+  v3 = _mm_loadl_epi64((__m128i const *)(pred + 3 * pred_stride));
+  v4 = _mm_loadl_epi64((__m128i const *)(pred + 4 * pred_stride));
+  v5 = _mm_loadl_epi64((__m128i const *)(pred + 5 * pred_stride));
+  v6 = _mm_loadl_epi64((__m128i const *)(pred + 6 * pred_stride));
+  v7 = _mm_loadl_epi64((__m128i const *)(pred + 7 * pred_stride));
 
   x0 = _mm_sub_epi16(u0, v0);
   x1 = _mm_sub_epi16(u1, v1);