OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2014 The Android Open Source Project | 2 * Copyright 2014 The Android Open Source Project |
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 #ifndef SkColor_opts_SSE2_DEFINED | 8 #ifndef SkColor_opts_SSE2_DEFINED |
9 #define SkColor_opts_SSE2_DEFINED | 9 #define SkColor_opts_SSE2_DEFINED |
10 | 10 |
11 #include <emmintrin.h> | 11 #include <emmintrin.h> |
12 | 12 |
| 13 #define ASSERT_EQ(a,b) SkASSERT(0xffff == _mm_movemask_epi8(_mm_cmpeq_epi8((a),
(b)))) |
| 14 |
13 // Because no _mm_mul_epi32() in SSE2, we emulate it here. | 15 // Because no _mm_mul_epi32() in SSE2, we emulate it here. |
14 // Multiplies 4 32-bit integers from a by 4 32-bit intergers from b. | 16 // Multiplies 4 32-bit integers from a by 4 32-bit intergers from b. |
15 // The 4 multiplication results should be represented within 32-bit | 17 // The 4 multiplication results should be represented within 32-bit |
16 // integers, otherwise they would be overflow. | 18 // integers, otherwise they would be overflow. |
17 static inline __m128i Multiply32_SSE2(const __m128i& a, const __m128i& b) { | 19 static inline __m128i Multiply32_SSE2(const __m128i& a, const __m128i& b) { |
18 // Calculate results of a0 * b0 and a2 * b2. | 20 // Calculate results of a0 * b0 and a2 * b2. |
19 __m128i r1 = _mm_mul_epu32(a, b); | 21 __m128i r1 = _mm_mul_epu32(a, b); |
20 // Calculate results of a1 * b1 and a3 * b3. | 22 // Calculate results of a1 * b1 and a3 * b3. |
21 __m128i r2 = _mm_mul_epu32(_mm_srli_si128(a, 4), _mm_srli_si128(b, 4)); | 23 __m128i r2 = _mm_mul_epu32(_mm_srli_si128(a, 4), _mm_srli_si128(b, 4)); |
22 // Shuffle results to [63..0] and interleave the results. | 24 // Shuffle results to [63..0] and interleave the results. |
(...skipping 22 matching lines...) Expand all Loading... |
45 __m128i mask = _mm_set1_epi32(0xFF00FF); | 47 __m128i mask = _mm_set1_epi32(0xFF00FF); |
46 __m128i s = _mm_or_si128(_mm_slli_epi32(scale, 16), scale); | 48 __m128i s = _mm_or_si128(_mm_slli_epi32(scale, 16), scale); |
47 | 49 |
48 // uint32_t rb = ((c & mask) * scale) >> 8 | 50 // uint32_t rb = ((c & mask) * scale) >> 8 |
49 __m128i rb = _mm_and_si128(mask, c); | 51 __m128i rb = _mm_and_si128(mask, c); |
50 rb = _mm_mullo_epi16(rb, s); | 52 rb = _mm_mullo_epi16(rb, s); |
51 rb = _mm_srli_epi16(rb, 8); | 53 rb = _mm_srli_epi16(rb, 8); |
52 | 54 |
53 // uint32_t ag = ((c >> 8) & mask) * scale | 55 // uint32_t ag = ((c >> 8) & mask) * scale |
54 __m128i ag = _mm_srli_epi16(c, 8); | 56 __m128i ag = _mm_srli_epi16(c, 8); |
55 ag = _mm_and_si128(ag, mask); | 57 ASSERT_EQ(ag, _mm_and_si128(mask, ag)); // ag = _mm_srli_epi16(c, 8) did th
is for us. |
56 ag = _mm_mullo_epi16(ag, s); | 58 ag = _mm_mullo_epi16(ag, s); |
57 | 59 |
58 // (rb & mask) | (ag & ~mask) | 60 // (rb & mask) | (ag & ~mask) |
59 rb = _mm_and_si128(mask, rb); | 61 ASSERT_EQ(rb, _mm_and_si128(mask, rb)); // rb = _mm_srli_epi16(rb, 8) did t
his for us. |
60 ag = _mm_andnot_si128(mask, ag); | 62 ag = _mm_andnot_si128(mask, ag); |
61 return _mm_or_si128(rb, ag); | 63 return _mm_or_si128(rb, ag); |
62 } | 64 } |
63 | 65 |
64 static inline __m128i SkGetPackedA32_SSE2(const __m128i& src) { | 66 static inline __m128i SkGetPackedA32_SSE2(const __m128i& src) { |
65 __m128i a = _mm_slli_epi32(src, (24 - SK_A32_SHIFT)); | 67 __m128i a = _mm_slli_epi32(src, (24 - SK_A32_SHIFT)); |
66 return _mm_srli_epi32(a, 24); | 68 return _mm_srli_epi32(a, 24); |
67 } | 69 } |
68 | 70 |
69 static inline __m128i SkGetPackedR32_SSE2(const __m128i& src) { | 71 static inline __m128i SkGetPackedR32_SSE2(const __m128i& src) { |
(...skipping 106 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
176 SK_B32_SHIFT + (8 - SK_B16_BITS)); | 178 SK_B32_SHIFT + (8 - SK_B16_BITS)); |
177 b2 = _mm_and_si128(b2, _mm_set1_epi32(SK_B16_MASK)); | 179 b2 = _mm_and_si128(b2, _mm_set1_epi32(SK_B16_MASK)); |
178 __m128i b = _mm_packs_epi32(b1, b2); | 180 __m128i b = _mm_packs_epi32(b1, b2); |
179 | 181 |
180 // Store 8 16-bit colors in dst. | 182 // Store 8 16-bit colors in dst. |
181 __m128i d_pixel = SkPackRGB16_SSE2(r, g, b); | 183 __m128i d_pixel = SkPackRGB16_SSE2(r, g, b); |
182 | 184 |
183 return d_pixel; | 185 return d_pixel; |
184 } | 186 } |
185 | 187 |
| 188 #undef ASSERT_EQ |
186 #endif // SkColor_opts_SSE2_DEFINED | 189 #endif // SkColor_opts_SSE2_DEFINED |
OLD | NEW |