OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2015 Google Inc. | 2 * Copyright 2015 Google Inc. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license that can be | 4 * Use of this source code is governed by a BSD-style license that can be |
5 * found in the LICENSE file. | 5 * found in the LICENSE file. |
6 */ | 6 */ |
7 | 7 |
8 namespace { // See Sk4px.h | 8 namespace { // See Sk4px.h |
9 | 9 |
10 inline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b(_mm_set1_epi32(px));
} | 10 inline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b(_mm_set1_epi32(px));
} |
(...skipping 27 matching lines...) Expand all Loading... |
38 | 38 |
39 inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const { | 39 inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const { |
40 return this->widenLo() * Sk4px(other).widenLo(); | 40 return this->widenLo() * Sk4px(other).widenLo(); |
41 } | 41 } |
42 | 42 |
43 inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const { | 43 inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const { |
44 Sk4px::Wide r = (*this + other) >> 8; | 44 Sk4px::Wide r = (*this + other) >> 8; |
45 return Sk4px(_mm_packus_epi16(r.fLo.fVec, r.fHi.fVec)); | 45 return Sk4px(_mm_packus_epi16(r.fLo.fVec, r.fHi.fVec)); |
46 } | 46 } |
47 | 47 |
| 48 inline Sk4px Sk4px::Wide::div255() const { |
| 49 // (x + 127) / 255 == ((x+128) * 257)>>16, |
| 50 // and _mm_mulhi_epu16 makes the (_ * 257)>>16 part very convenient. |
| 51 const __m128i _128 = _mm_set1_epi16(128), |
| 52 _257 = _mm_set1_epi16(257); |
| 53 return Sk4px(_mm_packus_epi16(_mm_mulhi_epu16(_mm_add_epi16(fLo.fVec, _128),
_257), |
| 54 _mm_mulhi_epu16(_mm_add_epi16(fHi.fVec, _128),
_257))); |
| 55 } |
| 56 |
48 // Load4Alphas and Load2Alphas use possibly-unaligned loads (SkAlpha[] -> uint16
_t or uint32_t). | 57 // Load4Alphas and Load2Alphas use possibly-unaligned loads (SkAlpha[] -> uint16
_t or uint32_t). |
49 // These are safe on x86, often with no speed penalty. | 58 // These are safe on x86, often with no speed penalty. |
50 | 59 |
51 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3 | 60 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3 |
52 inline Sk4px Sk4px::alphas() const { | 61 inline Sk4px Sk4px::alphas() const { |
53 static_assert(SK_A32_SHIFT == 24, "Intel's always little-endian."); | 62 static_assert(SK_A32_SHIFT == 24, "Intel's always little-endian."); |
54 __m128i splat = _mm_set_epi8(15,15,15,15, 11,11,11,11, 7,7,7,7, 3,3,3,3)
; | 63 __m128i splat = _mm_set_epi8(15,15,15,15, 11,11,11,11, 7,7,7,7, 3,3,3,3)
; |
55 return Sk16b(_mm_shuffle_epi8(this->fVec, splat)); | 64 return Sk16b(_mm_shuffle_epi8(this->fVec, splat)); |
56 } | 65 } |
57 | 66 |
(...skipping 103 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
161 uint32_t dst2 = _mm_cvtsi128_si32(narrow_to_565(this->fVec)); | 170 uint32_t dst2 = _mm_cvtsi128_si32(narrow_to_565(this->fVec)); |
162 dst[0] = dst2; | 171 dst[0] = dst2; |
163 dst[1] = dst2 >> 16; | 172 dst[1] = dst2 >> 16; |
164 } | 173 } |
165 inline void Sk4px::store1(SkPMColor16 dst[1]) const { | 174 inline void Sk4px::store1(SkPMColor16 dst[1]) const { |
166 uint32_t dst2 = _mm_cvtsi128_si32(narrow_to_565(this->fVec)); | 175 uint32_t dst2 = _mm_cvtsi128_si32(narrow_to_565(this->fVec)); |
167 dst[0] = dst2; | 176 dst[0] = dst2; |
168 } | 177 } |
169 | 178 |
170 } // namespace | 179 } // namespace |
OLD | NEW |