| OLD | NEW |
|---|
| (Empty) |
| 1 #include "SkColorPriv.h" | |
| 2 #include "SkColor_opts_SSE2.h" | |
| 3 #include "SkMathPriv.h" | |
| 4 #include "SkXfermode.h" | |
| 5 #include "SkXfermode_opts_SSE2.h" | |
| 6 #include "SkXfermode_proccoeff.h" | |
| 7 | |
| 8 //////////////////////////////////////////////////////////////////////////////// | |
| 9 // 4 pixels SSE2 version functions | |
| 10 //////////////////////////////////////////////////////////////////////////////// | |
| 11 | |
| 12 static inline __m128i SkDiv255Round_SSE2(__m128i prod) { | |
| 13 prod = _mm_add_epi32(prod, _mm_set1_epi32(128)); // prod += 128; | |
| 14 prod = _mm_add_epi32(prod, _mm_srli_epi32(prod, 8)); // prod + (prod >> 8) | |
| 15 prod = _mm_srli_epi32(prod, 8); // >> 8 | |
| 16 | |
| 17 return prod; | |
| 18 } | |
| 19 | |
| 20 static inline __m128i clamp_div255round_SSE2(__m128i prod) { | |
| 21 // test if > 0 | |
| 22 __m128i cmp1 = _mm_cmpgt_epi32(prod, _mm_setzero_si128()); | |
| 23 // test if < 255*255 | |
| 24 __m128i cmp2 = _mm_cmplt_epi32(prod, _mm_set1_epi32(255*255)); | |
| 25 | |
| 26 __m128i ret = _mm_setzero_si128(); | |
| 27 | |
| 28 // if value >= 255*255, value = 255 | |
| 29 ret = _mm_andnot_si128(cmp2, _mm_set1_epi32(255)); | |
| 30 | |
| 31 __m128i div = SkDiv255Round_SSE2(prod); | |
| 32 | |
| 33 // test if > 0 && < 255*255 | |
| 34 __m128i cmp = _mm_and_si128(cmp1, cmp2); | |
| 35 | |
| 36 ret = _mm_or_si128(_mm_and_si128(cmp, div), _mm_andnot_si128(cmp, ret)); | |
| 37 | |
| 38 return ret; | |
| 39 } | |
| 40 | |
| 41 static inline __m128i srcover_byte_SSE2(__m128i a, __m128i b) { | |
| 42 // a + b - SkAlphaMulAlpha(a, b); | |
| 43 return _mm_sub_epi32(_mm_add_epi32(a, b), SkAlphaMulAlpha_SSE2(a, b)); | |
| 44 | |
| 45 } | |
| 46 | |
| 47 static inline __m128i blendfunc_multiply_byte_SSE2(__m128i sc, __m128i dc, | |
| 48 __m128i sa, __m128i da) { | |
| 49 // sc * (255 - da) | |
| 50 __m128i ret1 = _mm_sub_epi32(_mm_set1_epi32(255), da); | |
| 51 ret1 = _mm_mullo_epi16(sc, ret1); | |
| 52 | |
| 53 // dc * (255 - sa) | |
| 54 __m128i ret2 = _mm_sub_epi32(_mm_set1_epi32(255), sa); | |
| 55 ret2 = _mm_mullo_epi16(dc, ret2); | |
| 56 | |
| 57 // sc * dc | |
| 58 __m128i ret3 = _mm_mullo_epi16(sc, dc); | |
| 59 | |
| 60 __m128i ret = _mm_add_epi32(ret1, ret2); | |
| 61 ret = _mm_add_epi32(ret, ret3); | |
| 62 | |
| 63 return clamp_div255round_SSE2(ret); | |
| 64 } | |
| 65 | |
| 66 static __m128i multiply_modeproc_SSE2(__m128i src, __m128i dst) { | |
| 67 __m128i sa = SkGetPackedA32_SSE2(src); | |
| 68 __m128i da = SkGetPackedA32_SSE2(dst); | |
| 69 __m128i a = srcover_byte_SSE2(sa, da); | |
| 70 | |
| 71 __m128i sr = SkGetPackedR32_SSE2(src); | |
| 72 __m128i dr = SkGetPackedR32_SSE2(dst); | |
| 73 __m128i r = blendfunc_multiply_byte_SSE2(sr, dr, sa, da); | |
| 74 | |
| 75 __m128i sg = SkGetPackedG32_SSE2(src); | |
| 76 __m128i dg = SkGetPackedG32_SSE2(dst); | |
| 77 __m128i g = blendfunc_multiply_byte_SSE2(sg, dg, sa, da); | |
| 78 | |
| 79 | |
| 80 __m128i sb = SkGetPackedB32_SSE2(src); | |
| 81 __m128i db = SkGetPackedB32_SSE2(dst); | |
| 82 __m128i b = blendfunc_multiply_byte_SSE2(sb, db, sa, da); | |
| 83 | |
| 84 return SkPackARGB32_SSE2(a, r, g, b); | |
| 85 } | |
| 86 | |
| 87 //////////////////////////////////////////////////////////////////////////////// | |
| 88 | |
| 89 extern SkXfermodeProcSIMD gSSE2XfermodeProcs[]; | |
| 90 | |
| 91 SkSSE2ProcCoeffXfermode::SkSSE2ProcCoeffXfermode(SkReadBuffer& buffer) | |
| 92 : INHERITED(buffer) { | |
| 93 fProcSIMD = gSSE2XfermodeProcs[this->getMode()]; | |
| 94 } | |
| 95 | |
| 96 void SkSSE2ProcCoeffXfermode::xfer32(SkPMColor dst[], const SkPMColor src[], | |
| 97 int count, const SkAlpha aa[]) const { | |
| 98 SkASSERT(dst && src && count >= 0); | |
| 99 | |
| 100 SkXfermodeProc proc = this->getProc(); | |
| 101 SkXfermodeProcSIMD procSIMD = fProcSIMD; | |
| 102 SkASSERT(procSIMD != NULL); | |
| 103 | |
| 104 if (NULL == aa) { | |
| 105 if (count >= 4) { | |
| 106 while (((size_t)dst & 0x0F) != 0) { | |
| 107 *dst = proc(*src, *dst); | |
| 108 dst++; | |
| 109 src++; | |
| 110 count--; | |
| 111 } | |
| 112 | |
| 113 const __m128i* s = reinterpret_cast<const __m128i*>(src); | |
| 114 __m128i* d = reinterpret_cast<__m128i*>(dst); | |
| 115 | |
| 116 while (count >= 4) { | |
| 117 __m128i src_pixel = _mm_loadu_si128(s++); | |
| 118 __m128i dst_pixel = _mm_load_si128(d); | |
| 119 | |
| 120 dst_pixel = procSIMD(src_pixel, dst_pixel); | |
| 121 _mm_store_si128(d++, dst_pixel); | |
| 122 count -= 4; | |
| 123 } | |
| 124 | |
| 125 src = reinterpret_cast<const SkPMColor*>(s); | |
| 126 dst = reinterpret_cast<SkPMColor*>(d); | |
| 127 } | |
| 128 | |
| 129 for (int i = count - 1; i >= 0; --i) { | |
| 130 *dst = proc(*src, *dst); | |
| 131 dst++; | |
| 132 src++; | |
| 133 } | |
| 134 } else { | |
| 135 for (int i = count - 1; i >= 0; --i) { | |
| 136 unsigned a = aa[i]; | |
| 137 if (0 != a) { | |
| 138 SkPMColor dstC = dst[i]; | |
| 139 SkPMColor C = proc(src[i], dstC); | |
| 140 if (a != 0xFF) { | |
| 141 C = SkFourByteInterp(C, dstC, a); | |
| 142 } | |
| 143 dst[i] = C; | |
| 144 } | |
| 145 } | |
| 146 } | |
| 147 } | |
| 148 | |
| 149 void SkSSE2ProcCoeffXfermode::xfer16(uint16_t* SK_RESTRICT dst, | |
| 150 const SkPMColor* SK_RESTRICT src, int count
, | |
| 151 const SkAlpha* SK_RESTRICT aa) const { | |
| 152 SkASSERT(dst && src && count >= 0); | |
| 153 | |
| 154 SkXfermodeProc proc = this->getProc(); | |
| 155 SkXfermodeProcSIMD procSIMD = fProcSIMD; | |
| 156 SkASSERT(procSIMD != NULL); | |
| 157 | |
| 158 if (NULL == aa) { | |
| 159 if (count >= 8) { | |
| 160 while (((size_t)dst & 0x0F) != 0) { | |
| 161 SkPMColor dstC = SkPixel16ToPixel32(*dst); | |
| 162 *dst = SkPixel32ToPixel16_ToU16(proc(*src, dstC)); | |
| 163 dst++; | |
| 164 src++; | |
| 165 count--; | |
| 166 } | |
| 167 | |
| 168 const __m128i* s = reinterpret_cast<const __m128i*>(src); | |
| 169 __m128i* d = reinterpret_cast<__m128i*>(dst); | |
| 170 | |
| 171 while (count >= 8) { | |
| 172 __m128i src_pixel1 = _mm_loadu_si128(s++); | |
| 173 __m128i src_pixel2 = _mm_loadu_si128(s++); | |
| 174 __m128i dst_pixel = _mm_load_si128(d); | |
| 175 | |
| 176 __m128i dst_pixel1 = _mm_unpacklo_epi16(dst_pixel, _mm_setzero_s
i128()); | |
| 177 __m128i dst_pixel2 = _mm_unpackhi_epi16(dst_pixel, _mm_setzero_s
i128()); | |
| 178 | |
| 179 __m128i dstC1 = SkPixel16ToPixel32_SSE2(dst_pixel1); | |
| 180 __m128i dstC2 = SkPixel16ToPixel32_SSE2(dst_pixel2); | |
| 181 | |
| 182 dst_pixel1 = procSIMD(src_pixel1, dstC1); | |
| 183 dst_pixel2 = procSIMD(src_pixel2, dstC2); | |
| 184 dst_pixel = SkPixel32ToPixel16_ToU16_SSE2(dst_pixel1, dst_pixel2
); | |
| 185 | |
| 186 _mm_store_si128(d++, dst_pixel); | |
| 187 count -= 8; | |
| 188 } | |
| 189 | |
| 190 src = reinterpret_cast<const SkPMColor*>(s); | |
| 191 dst = reinterpret_cast<uint16_t*>(d); | |
| 192 } | |
| 193 | |
| 194 for (int i = count - 1; i >= 0; --i) { | |
| 195 SkPMColor dstC = SkPixel16ToPixel32(*dst); | |
| 196 *dst = SkPixel32ToPixel16_ToU16(proc(*src, dstC)); | |
| 197 dst++; | |
| 198 src++; | |
| 199 } | |
| 200 } else { | |
| 201 for (int i = count - 1; i >= 0; --i) { | |
| 202 unsigned a = aa[i]; | |
| 203 if (0 != a) { | |
| 204 SkPMColor dstC = SkPixel16ToPixel32(dst[i]); | |
| 205 SkPMColor C = proc(src[i], dstC); | |
| 206 if (0xFF != a) { | |
| 207 C = SkFourByteInterp(C, dstC, a); | |
| 208 } | |
| 209 dst[i] = SkPixel32ToPixel16_ToU16(C); | |
| 210 } | |
| 211 } | |
| 212 } | |
| 213 } | |
| 214 | |
| 215 #ifndef SK_IGNORE_TO_STRING | |
| 216 void SkSSE2ProcCoeffXfermode::toString(SkString* str) const { | |
| 217 this->INHERITED::toString(str); | |
| 218 } | |
| 219 #endif | |
| 220 | |
| 221 //////////////////////////////////////////////////////////////////////////////// | |
| 222 | |
| 223 // 4 pixels modeprocs with SSE2 | |
| 224 SkXfermodeProcSIMD gSSE2XfermodeProcs[] = { | |
| 225 NULL, // kClear_Mode | |
| 226 NULL, // kSrc_Mode | |
| 227 NULL, // kDst_Mode | |
| 228 NULL, // kSrcOver_Mode | |
| 229 NULL, // kDstOver_Mode | |
| 230 NULL, // kSrcIn_Mode | |
| 231 NULL, // kDstIn_Mode | |
| 232 NULL, // kSrcOut_Mode | |
| 233 NULL, // kDstOut_Mode | |
| 234 NULL, // kSrcATop_Mode | |
| 235 NULL, // kDstATop_Mode | |
| 236 NULL, // kXor_Mode | |
| 237 NULL, // kPlus_Mode | |
| 238 NULL, // kModulate_Mode | |
| 239 NULL, // kScreen_Mode | |
| 240 | |
| 241 NULL, // kOverlay_Mode | |
| 242 NULL, // kDarken_Mode | |
| 243 NULL, // kLighten_Mode | |
| 244 NULL, // kColorDodge_Mode | |
| 245 NULL, // kColorBurn_Mode | |
| 246 NULL, // kHardLight_Mode | |
| 247 NULL, // kSoftLight_Mode | |
| 248 NULL, // kDifference_Mode | |
| 249 NULL, // kExclusion_Mode | |
| 250 multiply_modeproc_SSE2, | |
| 251 | |
| 252 NULL, // kHue_Mode | |
| 253 NULL, // kSaturation_Mode | |
| 254 NULL, // kColor_Mode | |
| 255 NULL, // kLuminosity_Mode | |
| 256 }; | |
| 257 | |
| 258 SkProcCoeffXfermode* SkPlatformXfermodeFactory_impl_SSE2(const ProcCoeff& rec, | |
| 259 SkXfermode::Mode mode)
{ | |
| 260 SkXfermodeProcSIMD procSIMD = gSSE2XfermodeProcs[mode]; | |
| 261 | |
| 262 if (procSIMD != NULL) { | |
| 263 return SkNEW_ARGS(SkSSE2ProcCoeffXfermode, (rec, mode, procSIMD)); | |
| 264 } | |
| 265 return NULL; | |
| 266 } | |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 219243009:
Revert of Xfermode: SSE2 implementation of multiply_modeproc (Closed)
Base URL: https://skia.googlesource.com/skia.git@xfermode