SSE2 implementation of the PVQ search
authorJean-Marc Valin <jmvalin@jmvalin.ca>
Wed, 10 Aug 2016 03:22:27 +0000 (23:22 -0400)
committerJean-Marc Valin <jmvalin@jmvalin.ca>
Wed, 10 Aug 2016 03:22:27 +0000 (23:22 -0400)
commit76674feae22db03848a40446beb2fcec70d2180d
tree4c6425459ecfed95ba9a3a48ba9aa5ffeac45107
parente806d6a74129c70d2849cf621968f905149c21e1
SSE2 implementation of the PVQ search

We used the SSE reciprocal square root instruction to vectorize the serch rather
than compare one at a time with multiplies. Speeds up the entire encoder by 8-10%.
celt/bands.c
celt/tests/test_unit_mathops.c
celt/tests/test_unit_rotation.c
celt/vq.c
celt/vq.h
celt/x86/vq_sse.h [new file with mode: 0644]
celt/x86/vq_sse2.c [new file with mode: 0644]
celt/x86/x86_celt_map.c
celt_headers.mk
celt_sources.mk