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 // For SkPMFloat(SkPMColor), we widen our 8 bit components (fix8) to 8-bit compo
nents in 16 bits | 8 // For SkPMFloat(SkPMColor), we widen our 8 bit components (fix8) to 8-bit compo
nents in 16 bits |
9 // (fix8_16), then widen those to 8-bit-in-32-bits (fix8_32), and finally conver
t those to floats. | 9 // (fix8_16), then widen those to 8-bit-in-32-bits (fix8_32), and finally conver
t those to floats. |
10 | 10 |
11 // get() and clamped() do the opposite, working from floats to 8-bit-in-32-bit, | 11 // get() and clamped() do the opposite, working from floats to 8-bit-in-32-bit, |
12 // to 8-bit-in-16-bit, back down to 8-bit components. | 12 // to 8-bit-in-16-bit, back down to 8-bit components. |
13 // _mm_packus_epi16() gives us clamping for free while narrowing. | 13 // _mm_packus_epi16() gives us clamping for free while narrowing. |
14 | 14 |
15 inline SkPMFloat::SkPMFloat(SkPMColor c) { | 15 inline SkPMFloat::SkPMFloat(SkPMColor c) { |
16 SkPMColorAssert(c); | 16 SkPMColorAssert(c); |
17 __m128i fix8 = _mm_set_epi32(0,0,0,c), | 17 __m128i fix8 = _mm_set_epi32(0,0,0,c), |
18 fix8_16 = _mm_unpacklo_epi8 (fix8, _mm_setzero_si128()), | 18 fix8_16 = _mm_unpacklo_epi8 (fix8, _mm_setzero_si128()), |
19 fix8_32 = _mm_unpacklo_epi16(fix8_16, _mm_setzero_si128()); | 19 fix8_32 = _mm_unpacklo_epi16(fix8_16, _mm_setzero_si128()); |
20 fColors = _mm_cvtepi32_ps(fix8_32); | 20 fColors = _mm_cvtepi32_ps(fix8_32); |
21 SkASSERT(this->isValid()); | 21 SkASSERT(this->isValid()); |
22 } | 22 } |
23 | 23 |
24 inline SkPMColor SkPMFloat::get() const { | 24 inline SkPMColor SkPMFloat::get() const { |
25 SkASSERT(this->isValid()); | |
26 return this->clamped(); // Haven't beaten this yet. | 25 return this->clamped(); // Haven't beaten this yet. |
27 } | 26 } |
28 | 27 |
29 inline SkPMColor SkPMFloat::clamped() const { | 28 inline SkPMColor SkPMFloat::clamped() const { |
30 // We don't use _mm_cvtps_epi32, because we want precise control over how 0.
5 rounds (up). | 29 // We don't use _mm_cvtps_epi32, because we want precise control over how 0.
5 rounds (up). |
31 __m128i fix8_32 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), fColors.vec
())), | 30 __m128i fix8_32 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), fColors.vec
())), |
32 fix8_16 = _mm_packus_epi16(fix8_32, fix8_32), | 31 fix8_16 = _mm_packus_epi16(fix8_32, fix8_32), |
33 fix8 = _mm_packus_epi16(fix8_16, fix8_16); | 32 fix8 = _mm_packus_epi16(fix8_16, fix8_16); |
34 SkPMColor c = _mm_cvtsi128_si32(fix8); | 33 SkPMColor c = _mm_cvtsi128_si32(fix8); |
35 SkPMColorAssert(c); | 34 SkPMColorAssert(c); |
(...skipping 36 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
72 c2 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), c.fColors.vec())
), | 71 c2 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), c.fColors.vec())
), |
73 c3 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), d.fColors.vec())
); | 72 c3 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), d.fColors.vec())
); |
74 __m128i c3210 = _mm_packus_epi16(_mm_packus_epi16(c0, c1), | 73 __m128i c3210 = _mm_packus_epi16(_mm_packus_epi16(c0, c1), |
75 _mm_packus_epi16(c2, c3)); | 74 _mm_packus_epi16(c2, c3)); |
76 _mm_storeu_si128((__m128i*)colors, c3210); | 75 _mm_storeu_si128((__m128i*)colors, c3210); |
77 SkPMColorAssert(colors[0]); | 76 SkPMColorAssert(colors[0]); |
78 SkPMColorAssert(colors[1]); | 77 SkPMColorAssert(colors[1]); |
79 SkPMColorAssert(colors[2]); | 78 SkPMColorAssert(colors[2]); |
80 SkPMColorAssert(colors[3]); | 79 SkPMColorAssert(colors[3]); |
81 } | 80 } |
OLD | NEW |