tree 62639c72f5324e3ed132d6ada659133ab6ecb759
parent b59712ce5092ac5f48ef9ef3e9c20b80bc5da3cb
author Yunqing Wang <yunqingwang@google.com> 1555104579 -0700
committer Yunqing Wang <yunqingwang@google.com> 1555367874 +0000

Speed up wiener search with aligned loads/stores

Speed 4 encoder profiling result showed:
3.15%--acc_stat_avx2 (inlined)
        |
        |--1.44%--yy_loadu_256 (inlined)
        |          _mm256_loadu_si256 (inlined)
        |
         --1.18%--yy_storeu_256 (inlined)
                   _mm256_storeu_si256 (inlined)

Forced the buffer alignment to use aligned loads/stores. Now, we have:
1.39%--acc_stat_avx2 (inlined)
        |
         --0.58%--yy_store_256 (inlined)
                   _mm256_store_si256 (inlined)

The speed test showed ~1.3% speedup at speed 4.

Change-Id: Ib7e4ad9b1ff0c602a2cd8bf61a6ffd5f3e26152d