dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2025-07-16 05:49:51 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

SSE optimization for vector mismatch.

Browse Source 
Change-Id: I564b822033b59d86635230f29ed6197e306a2c4f
This commit is contained in:
Vincent Rabaud
2016-01-07 17:23:48 +01:00
parent 7db53831a9
commit 8ce975ac82
4 changed files with 78 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
src/dsp/lossless_enc_sse2.c
View File

@ -324,6 +324,57 @@ static float CombinedShannonEntropy(const int X[256], const int Y[256]) {
#undef ANALYZE_X_OR_Y
#undef ANALYZE_XY
//------------------------------------------------------------------------------
static int VectorMismatch(const uint32_t* const array1,
const uint32_t* const array2, int length) {
int match_len;
if (12 <= length) {
__m128i A0 = _mm_loadu_si128((const __m128i*)&array1[0]);
__m128i A1 = _mm_loadu_si128((const __m128i*)&array2[0]);
match_len = 0;
do {
// Loop unrolling and early load both provide a speedup of 10% for the
// current function. Also, max_limit can be MAX_LENGTH=4096 at most.
const __m128i cmpA = _mm_cmpeq_epi32(A0, A1);
const __m128i B0 =
_mm_loadu_si128((const __m128i*)&array1[match_len + 4]);
const __m128i B1 =
_mm_loadu_si128((const __m128i*)&array2[match_len + 4]);
if (_mm_movemask_epi8(cmpA) != 0xffff) break;
match_len += 4;
{
const __m128i cmpB = _mm_cmpeq_epi32(B0, B1);
A0 = _mm_loadu_si128((const __m128i*)&array1[match_len + 4]);
A1 = _mm_loadu_si128((const __m128i*)&array2[match_len + 4]);
if (_mm_movemask_epi8(cmpB) != 0xffff) break;
match_len += 4;
}
} while (match_len + 12 < length);
} else {
match_len = 0;
// Unroll the potential first two loops.
if (4 <= length &&
_mm_movemask_epi8(_mm_cmpeq_epi32(
_mm_loadu_si128((const __m128i*)&array1[0]),
_mm_loadu_si128((const __m128i*)&array2[0]))) == 0xffff) {
match_len = 4;
if (8 <= length &&
_mm_movemask_epi8(_mm_cmpeq_epi32(
_mm_loadu_si128((const __m128i*)&array1[4]),
_mm_loadu_si128((const __m128i*)&array2[4]))) == 0xffff)
match_len = 8;
}
}
while (match_len < length && array1[match_len] == array2[match_len]) {
++match_len;
}
return match_len;
}
//------------------------------------------------------------------------------
// Entry point
@ -336,6 +387,7 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitSSE2(void) {
VP8LCollectColorRedTransforms = CollectColorRedTransforms;
VP8LHistogramAdd = HistogramAdd;
VP8LCombinedShannonEntropy = CombinedShannonEntropy;
VP8LVectorMismatch = VectorMismatch;
}
#else // !WEBP_USE_SSE2