trim some fat from SSE2 fixed point alpha code
- extract alpha from a pixel: 5 1-cycle ops to 4 1-cycle ops - load alphas: drop 4 unnecessary ops Should be no pixel diffs. BUG=skia: CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot Review URL: https://codereview.chromium.org/1447273004
This commit is contained in:
parent
32d6b87616
commit
f769c1f80e
@ -63,18 +63,17 @@ inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
|
||||
#else
|
||||
inline Sk4px Sk4px::alphas() const {
|
||||
static_assert(SK_A32_SHIFT == 24, "Intel's always little-endian.");
|
||||
__m128i as = _mm_srli_epi32(this->fVec, 24); // ___3 ___2 ___1 ___0
|
||||
as = _mm_or_si128(as, _mm_slli_si128(as, 1)); // __33 __22 __11 __00
|
||||
as = _mm_or_si128(as, _mm_slli_si128(as, 2)); // 3333 2222 1111 0000
|
||||
// We exploit that A >= rgb for any premul pixel.
|
||||
__m128i as = fVec; // 3xxx 2xxx 1xxx 0xxx
|
||||
as = _mm_max_epu8(as, _mm_srli_epi32(as, 8)); // 33xx 22xx 11xx 00xx
|
||||
as = _mm_max_epu8(as, _mm_srli_epi32(as, 16)); // 3333 2222 1111 0000
|
||||
return Sk16b(as);
|
||||
}
|
||||
|
||||
inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
|
||||
__m128i as = _mm_cvtsi32_si128(*(const uint32_t*)a); // ____ ____ ____ 3210
|
||||
as = _mm_unpacklo_epi8 (as, _mm_setzero_si128()); // ____ ____ _3_2 _1_0
|
||||
as = _mm_unpacklo_epi16(as, _mm_setzero_si128()); // ___3 ___2 ___1 ___0
|
||||
as = _mm_or_si128(as, _mm_slli_si128(as, 1)); // __33 __22 __11 __00
|
||||
as = _mm_or_si128(as, _mm_slli_si128(as, 2)); // 3333 2222 1111 0000
|
||||
as = _mm_unpacklo_epi8 (as, as); // ____ ____ 3322 1100
|
||||
as = _mm_unpacklo_epi16(as, as); // 3333 2222 1111 0000
|
||||
return Sk16b(as);
|
||||
}
|
||||
#endif
|
||||
|
@ -61,10 +61,9 @@ struct SkPx {
|
||||
#if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
|
||||
return _mm_shuffle_epi8(as, _mm_set_epi8(3,3,3,3, 2,2,2,2, 1,1,1,1, 0,0,0,0));
|
||||
#else
|
||||
as = _mm_unpacklo_epi8 (as, _mm_setzero_si128()); // ____ ____ _3_2 _1_0
|
||||
as = _mm_unpacklo_epi16(as, _mm_setzero_si128()); // ___3 ___2 ___1 ___0
|
||||
as = _mm_or_si128(as, _mm_slli_si128(as, 1)); // __33 __22 __11 __00
|
||||
return _mm_or_si128(as, _mm_slli_si128(as, 2)); // 3333 2222 1111 0000
|
||||
as = _mm_unpacklo_epi8 (as, as); // ____ ____ 3322 1100
|
||||
as = _mm_unpacklo_epi16(as, as); // 3333 2222 1111 0000
|
||||
return as;
|
||||
#endif
|
||||
}
|
||||
static Alpha Load(const uint8_t* a, int n) {
|
||||
@ -108,9 +107,11 @@ struct SkPx {
|
||||
#if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
|
||||
return _mm_shuffle_epi8(fVec, _mm_set_epi8(15,15,15,15, 11,11,11,11, 7,7,7,7, 3,3,3,3));
|
||||
#else
|
||||
__m128i as = _mm_srli_epi32(fVec, 24); // ___3 ___2 ___1 ___0
|
||||
as = _mm_or_si128(as, _mm_slli_si128(as, 1)); // __33 __22 __11 __00
|
||||
return _mm_or_si128(as, _mm_slli_si128(as, 2)); // 3333 2222 1111 0000
|
||||
// We exploit that A >= rgb for any premul pixel.
|
||||
__m128i as = fVec; // 3xxx 2xxx 1xxx 0xxx
|
||||
as = _mm_max_epu8(as, _mm_srli_epi32(as, 8)); // 33xx 22xx 11xx 00xx
|
||||
as = _mm_max_epu8(as, _mm_srli_epi32(as, 16)); // 3333 2222 1111 0000
|
||||
return as;
|
||||
#endif
|
||||
}
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user