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 #ifndef SkNx_sse_DEFINED | 8 #ifndef SkNx_sse_DEFINED |
9 #define SkNx_sse_DEFINED | 9 #define SkNx_sse_DEFINED |
10 | 10 |
(...skipping 114 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
125 SkNx(float a, float b, float c, float d) : fVec(_mm_setr_ps(a,b,c,d)) {} | 125 SkNx(float a, float b, float c, float d) : fVec(_mm_setr_ps(a,b,c,d)) {} |
126 | 126 |
127 void store(float vals[4]) const { _mm_storeu_ps(vals, fVec); } | 127 void store(float vals[4]) const { _mm_storeu_ps(vals, fVec); } |
128 void toBytes(uint8_t bytes[4]) const { | 128 void toBytes(uint8_t bytes[4]) const { |
129 __m128i fix8_32 = _mm_cvttps_epi32(fVec), | 129 __m128i fix8_32 = _mm_cvttps_epi32(fVec), |
130 fix8_16 = _mm_packus_epi16(fix8_32, fix8_32), | 130 fix8_16 = _mm_packus_epi16(fix8_32, fix8_32), |
131 fix8 = _mm_packus_epi16(fix8_16, fix8_16); | 131 fix8 = _mm_packus_epi16(fix8_16, fix8_16); |
132 *(int*)bytes = _mm_cvtsi128_si32(fix8); | 132 *(int*)bytes = _mm_cvtsi128_si32(fix8); |
133 } | 133 } |
134 | 134 |
| 135 static void ToBytes(uint8_t bytes[16], |
| 136 const SkNx& a, const SkNx& b, const SkNx& c, const SkNx&
d) { |
| 137 _mm_storeu_si128((__m128i*)bytes, |
| 138 _mm_packus_epi16(_mm_packus_epi16(_mm_cvttps_epi32(a.fV
ec), |
| 139 _mm_cvttps_epi32(b.fV
ec)), |
| 140 _mm_packus_epi16(_mm_cvttps_epi32(c.fV
ec), |
| 141 _mm_cvttps_epi32(d.fV
ec)))); |
| 142 } |
| 143 |
135 SkNx operator + (const SkNx& o) const { return _mm_add_ps(fVec, o.fVec); } | 144 SkNx operator + (const SkNx& o) const { return _mm_add_ps(fVec, o.fVec); } |
136 SkNx operator - (const SkNx& o) const { return _mm_sub_ps(fVec, o.fVec); } | 145 SkNx operator - (const SkNx& o) const { return _mm_sub_ps(fVec, o.fVec); } |
137 SkNx operator * (const SkNx& o) const { return _mm_mul_ps(fVec, o.fVec); } | 146 SkNx operator * (const SkNx& o) const { return _mm_mul_ps(fVec, o.fVec); } |
138 SkNx operator / (const SkNx& o) const { return _mm_div_ps(fVec, o.fVec); } | 147 SkNx operator / (const SkNx& o) const { return _mm_div_ps(fVec, o.fVec); } |
139 | 148 |
140 SkNx operator == (const SkNx& o) const { return _mm_cmpeq_ps (fVec, o.fVec);
} | 149 SkNx operator == (const SkNx& o) const { return _mm_cmpeq_ps (fVec, o.fVec);
} |
141 SkNx operator != (const SkNx& o) const { return _mm_cmpneq_ps(fVec, o.fVec);
} | 150 SkNx operator != (const SkNx& o) const { return _mm_cmpneq_ps(fVec, o.fVec);
} |
142 SkNx operator < (const SkNx& o) const { return _mm_cmplt_ps (fVec, o.fVec);
} | 151 SkNx operator < (const SkNx& o) const { return _mm_cmplt_ps (fVec, o.fVec);
} |
143 SkNx operator > (const SkNx& o) const { return _mm_cmpgt_ps (fVec, o.fVec);
} | 152 SkNx operator > (const SkNx& o) const { return _mm_cmpgt_ps (fVec, o.fVec);
} |
144 SkNx operator <= (const SkNx& o) const { return _mm_cmple_ps (fVec, o.fVec);
} | 153 SkNx operator <= (const SkNx& o) const { return _mm_cmple_ps (fVec, o.fVec);
} |
(...skipping 141 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
286 | 295 |
287 | 296 |
288 template<> | 297 template<> |
289 inline SkNx<4, int> SkNx_cast<int, float, 4>(const SkNx<4, float>& src) { | 298 inline SkNx<4, int> SkNx_cast<int, float, 4>(const SkNx<4, float>& src) { |
290 return _mm_cvttps_epi32(src.fVec); | 299 return _mm_cvttps_epi32(src.fVec); |
291 } | 300 } |
292 | 301 |
293 } // namespace | 302 } // namespace |
294 | 303 |
295 #endif//SkNx_sse_DEFINED | 304 #endif//SkNx_sse_DEFINED |
OLD | NEW |