diff options
| author | Sebastian Ramacher <sramacher@debian.org> | 2024-12-29 12:50:37 +0000 |
|---|---|---|
| committer | Sebastian Ramacher <sramacher@debian.org> | 2024-12-29 12:50:37 +0000 |
| commit | 873c687c09e0a1a78a8f3601d31330ad1897e484 (patch) | |
| tree | ff2fa2e2d27b55c002a6d46d50aab2c66e0bf21a /src/dsp/x86/loop_restoration_10bit_sse4.cc | |
| parent | 395b1f6877fbc2666383f3428f17400c35aa52a4 (diff) | |
| parent | b9ae809f200982119d16ec9613f3db95eca923b9 (diff) | |
| download | libgav1-873c687c09e0a1a78a8f3601d31330ad1897e484.tar.gz libgav1-873c687c09e0a1a78a8f3601d31330ad1897e484.tar.bz2 libgav1-873c687c09e0a1a78a8f3601d31330ad1897e484.zip | |
Merge branch 'master' into 'scrub-obsolete'
# Conflicts:
# debian/changelog
Diffstat (limited to 'src/dsp/x86/loop_restoration_10bit_sse4.cc')
| -rw-r--r-- | src/dsp/x86/loop_restoration_10bit_sse4.cc | 9 |
1 files changed, 8 insertions, 1 deletions
diff --git a/src/dsp/x86/loop_restoration_10bit_sse4.cc b/src/dsp/x86/loop_restoration_10bit_sse4.cc index 6625d51..029e168 100644 --- a/src/dsp/x86/loop_restoration_10bit_sse4.cc +++ b/src/dsp/x86/loop_restoration_10bit_sse4.cc @@ -1079,7 +1079,14 @@ inline void LookupIntermediate(const __m128i sum, const __m128i index, // general-purpose register to process. Faster than using _mm_extract_epi8(). uint8_t temp[8]; StoreLo8(temp, idx); - *ma = _mm_insert_epi8(*ma, kSgrMaLookup[temp[0]], offset + 0); + // offset == 0 is assumed to be the first call to this function. The value is + // mov'd to avoid -Wuninitialized warnings under gcc. mov should at least + // equivalent if not faster than pinsrb. + if (offset == 0) { + *ma = _mm_cvtsi32_si128(kSgrMaLookup[temp[0]]); + } else { + *ma = _mm_insert_epi8(*ma, kSgrMaLookup[temp[0]], offset + 0); + } *ma = _mm_insert_epi8(*ma, kSgrMaLookup[temp[1]], offset + 1); *ma = _mm_insert_epi8(*ma, kSgrMaLookup[temp[2]], offset + 2); *ma = _mm_insert_epi8(*ma, kSgrMaLookup[temp[3]], offset + 3); |