Unified · Split
| @@ -36,46 +36,13 @@ · 35 unmodified lines | |||
| 36 | 36 | ||
| 37 | #if HAVE_ALTIVEC | 37 | #if HAVE_ALTIVEC |
| 38 | /* next one assumes that ((line_size % 16) == 0) */ | 38 | /* next one assumes that ((line_size % 16) == 0) */ |
| 39 | #if HAVE_VSX | ||
| 40 | void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | 39 | void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) |
| 41 | { | 40 | { |
| 42 | register vector unsigned char pixelsv1; | 41 | register vector unsigned char pixelsv1; |
| 43 | register vector unsigned char pixelsv1B; | 42 | register vector unsigned char pixelsv1B; |
| 44 | register vector unsigned char pixelsv1C; | 43 | register vector unsigned char pixelsv1C; |
| 45 | register vector unsigned char pixelsv1D; | 44 | register vector unsigned char pixelsv1D; |
| 46 | 45 | ||
| 47 | int i; | ||
| 48 | register ptrdiff_t line_size_2 = line_size << 1; | ||
| 49 | register ptrdiff_t line_size_3 = line_size + line_size_2; | ||
| 50 | register ptrdiff_t line_size_4 = line_size << 2; | ||
| 51 | |||
| 52 | // hand-unrolling the loop by 4 gains about 15% | ||
| 53 | // mininum execution time goes from 74 to 60 cycles | ||
| 54 | // it's faster than -funroll-loops, but using | ||
| 55 | // -funroll-loops w/ this is bad - 74 cycles again. | ||
| 56 | // all this is on a 7450, tuning for the 7450 | ||
| 57 | for (i = 0; i < h; i += 4) { | ||
| 58 | pixelsv1 = vec_vsx_ld( 0, pixels); | ||
| 59 | pixelsv1B = vec_vsx_ld(line_size, pixels); | ||
| 60 | pixelsv1C = vec_vsx_ld(line_size_2, pixels); | ||
| 61 | pixelsv1D = vec_vsx_ld(line_size_3, pixels); | ||
| 62 | vec_vsx_st(pixelsv1, 0, (unsigned char*)block); | ||
| 63 | vec_vsx_st(pixelsv1B, line_size, (unsigned char*)block); | ||
| 64 | vec_vsx_st(pixelsv1C, line_size_2, (unsigned char*)block); | ||
| 65 | vec_st(pixelsv1D, line_size_3, (unsigned char*)block); | ||
| 66 | pixels+=line_size_4; | ||
| 67 | block +=line_size_4; | ||
| 68 | } | ||
| 69 | } | ||
| 70 | #else | ||
| 71 | void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | ||
| 72 | { | ||
| 73 | register vector unsigned char pixelsv1, pixelsv2; | ||
| 74 | register vector unsigned char pixelsv1B, pixelsv2B; | ||
| 75 | register vector unsigned char pixelsv1C, pixelsv2C; | ||
| 76 | register vector unsigned char pixelsv1D, pixelsv2D; | ||
| 77 | |||
| 78 | register vector unsigned char perm = vec_lvsl(0, pixels); | ||
| 79 | int i; | 46 | int i; |
| 80 | register ptrdiff_t line_size_2 = line_size << 1; | 47 | register ptrdiff_t line_size_2 = line_size << 1; |
| 81 | register ptrdiff_t line_size_3 = line_size + line_size_2; | 48 | register ptrdiff_t line_size_3 = line_size + line_size_2; |
| @@ -87,42 +54,29 @@ void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t li · 5 unmodified lines | |||
| 87 | // -funroll-loops w/ this is bad - 74 cycles again. | 54 | // -funroll-loops w/ this is bad - 74 cycles again. |
| 88 | // all this is on a 7450, tuning for the 7450 | 55 | // all this is on a 7450, tuning for the 7450 |
| 89 | for (i = 0; i < h; i += 4) { | 56 | for (i = 0; i < h; i += 4) { |
| 90 | pixelsv1 = vec_ld( 0, pixels); | 57 | pixelsv1 = unaligned_load( 0, pixels); |
| 91 | pixelsv2 = vec_ld(15, pixels); | 58 | pixelsv1B = unaligned_load(line_size, pixels); |
| 92 | pixelsv1B = vec_ld(line_size, pixels); | 59 | pixelsv1C = unaligned_load(line_size_2, pixels); |
| 93 | pixelsv2B = vec_ld(15 + line_size, pixels); | 60 | pixelsv1D = unaligned_load(line_size_3, pixels); |
| 94 | pixelsv1C = vec_ld(line_size_2, pixels); | 61 | VEC_ST(pixelsv1, 0, (unsigned char*)block); |
| 95 | pixelsv2C = vec_ld(15 + line_size_2, pixels); | 62 | VEC_ST(pixelsv1B, line_size, (unsigned char*)block); |
| 96 | pixelsv1D = vec_ld(line_size_3, pixels); | 63 | VEC_ST(pixelsv1C, line_size_2, (unsigned char*)block); |
| 97 | pixelsv2D = vec_ld(15 + line_size_3, pixels); | 64 | VEC_ST(pixelsv1D, line_size_3, (unsigned char*)block); |
| 98 | vec_st(vec_perm(pixelsv1, pixelsv2, perm), | ||
| 99 | 0, (unsigned char*)block); | ||
| 100 | vec_st(vec_perm(pixelsv1B, pixelsv2B, perm), | ||
| 101 | line_size, (unsigned char*)block); | ||
| 102 | vec_st(vec_perm(pixelsv1C, pixelsv2C, perm), | ||
| 103 | line_size_2, (unsigned char*)block); | ||
| 104 | vec_st(vec_perm(pixelsv1D, pixelsv2D, perm), | ||
| 105 | line_size_3, (unsigned char*)block); | ||
| 106 | pixels+=line_size_4; | 65 | pixels+=line_size_4; |
| 107 | block +=line_size_4; | 66 | block +=line_size_4; |
| 108 | } | 67 | } |
| 109 | } | 68 | } |
| 110 | 69 | ||
| 111 | #endif /* HAVE_VSX */ | ||
| 112 | |||
| 113 | /* next one assumes that ((line_size % 16) == 0) */ | 70 | /* next one assumes that ((line_size % 16) == 0) */ |
| 114 | #define op_avg(a,b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) ) | 71 | #define op_avg(a,b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) ) |
| 115 | void ff_avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) | 72 | void ff_avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) |
| 116 | { | 73 | { |
| 117 | register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv; | 74 | register vector unsigned char pixelsv, blockv; |
| 118 | register vector unsigned char perm = vec_lvsl(0, pixels); | ||
| 119 | int i; | ||
| 120 | 75 | ||
| 76 | int i; | ||
| 121 | for (i = 0; i < h; i++) { | 77 | for (i = 0; i < h; i++) { |
| 122 | pixelsv1 = vec_ld( 0, pixels); | ||
| 123 | pixelsv2 = vec_ld(16,pixels); | ||
| 124 | blockv = vec_ld(0, block); | 78 | blockv = vec_ld(0, block); |
| 125 | pixelsv = vec_perm(pixelsv1, pixelsv2, perm); | 79 | pixelsv = VEC_LD( 0, pixels); |
| 126 | blockv = vec_avg(blockv,pixelsv); | 80 | blockv = vec_avg(blockv,pixelsv); |
| 127 | vec_st(blockv, 0, (unsigned char*)block); | 81 | vec_st(blockv, 0, (unsigned char*)block); |
| 128 | pixels+=line_size; | 82 | pixels+=line_size; |
| @@ -142,9 +96,7 @@ static void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff · 13 unmodified lines | |||
| 142 | int rightside = ((unsigned long)block & 0x0000000F); | 96 | int rightside = ((unsigned long)block & 0x0000000F); |
| 143 | 97 | ||
| 144 | blockv = vec_ld(0, block); | 98 | blockv = vec_ld(0, block); |
| 145 | pixelsv1 = vec_ld( 0, pixels); | 99 | pixelsv = VEC_LD( 0, pixels); |
| 146 | pixelsv2 = vec_ld(16, pixels); | ||
| 147 | pixelsv = vec_perm(pixelsv1, pixelsv2, vec_lvsl(0, pixels)); | ||
| 148 | 100 | ||
| 149 | if (rightside) { | 101 | if (rightside) { |
| 150 | pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1)); | 102 | pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1)); |
| @@ -166,21 +118,16 @@ static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi · 15 unmodified lines | |||
| 166 | { | 118 | { |
| 167 | register int i; | 119 | register int i; |
| 168 | register vector unsigned char pixelsv1, pixelsv2, pixelsavg; | 120 | register vector unsigned char pixelsv1, pixelsv2, pixelsavg; |
| 169 | register vector unsigned char blockv, temp1, temp2; | 121 | register vector unsigned char blockv; |
| 170 | register vector unsigned short pixelssum1, pixelssum2, temp3; | 122 | register vector unsigned short pixelssum1, pixelssum2, temp3; |
| 171 | register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); | 123 | register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); |
| 172 | register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); | 124 | register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); |
| 173 | 125 | ||
| 174 | temp1 = vec_ld(0, pixels); | 126 | pixelsv1 = VEC_LD(0, pixels); |
| 175 | temp2 = vec_ld(16, pixels); | 127 | pixelsv2 = VEC_LD(1, pixels); |
| 176 | pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); | 128 | pixelsv1 = VEC_MERGEH(vczero, pixelsv1); |
| 177 | if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) { | 129 | pixelsv2 = VEC_MERGEH(vczero, pixelsv2); |
| 178 | pixelsv2 = temp2; | 130 | |
| 179 | } else { | ||
| 180 | pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels)); | ||
| 181 | } | ||
| 182 | pixelsv1 = vec_mergeh(vczero, pixelsv1); | ||
| 183 | pixelsv2 = vec_mergeh(vczero, pixelsv2); | ||
| 184 | pixelssum1 = vec_add((vector unsigned short)pixelsv1, | 131 | pixelssum1 = vec_add((vector unsigned short)pixelsv1, |
| 185 | (vector unsigned short)pixelsv2); | 132 | (vector unsigned short)pixelsv2); |
| 186 | pixelssum1 = vec_add(pixelssum1, vctwo); | 133 | pixelssum1 = vec_add(pixelssum1, vctwo); |
| @@ -189,17 +136,10 @@ static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi · 2 unmodified lines | |||
| 189 | int rightside = ((unsigned long)block & 0x0000000F); | 136 | int rightside = ((unsigned long)block & 0x0000000F); |
| 190 | blockv = vec_ld(0, block); | 137 | blockv = vec_ld(0, block); |
| 191 | 138 | ||
| 192 | temp1 = vec_ld(line_size, pixels); | 139 | pixelsv1 = unaligned_load(line_size, pixels); |
| 193 | temp2 = vec_ld(line_size + 16, pixels); | 140 | pixelsv2 = unaligned_load(line_size+1, pixels); |
| 194 | pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); | 141 | pixelsv1 = VEC_MERGEH(vczero, pixelsv1); |
| 195 | if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) { | 142 | pixelsv2 = VEC_MERGEH(vczero, pixelsv2); |
| 196 | pixelsv2 = temp2; | ||
| 197 | } else { | ||
| 198 | pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels)); | ||
| 199 | } | ||
| 200 | |||
| 201 | pixelsv1 = vec_mergeh(vczero, pixelsv1); | ||
| 202 | pixelsv2 = vec_mergeh(vczero, pixelsv2); | ||
| 203 | pixelssum2 = vec_add((vector unsigned short)pixelsv1, | 143 | pixelssum2 = vec_add((vector unsigned short)pixelsv1, |
| 204 | (vector unsigned short)pixelsv2); | 144 | (vector unsigned short)pixelsv2); |
| 205 | temp3 = vec_add(pixelssum1, pixelssum2); | 145 | temp3 = vec_add(pixelssum1, pixelssum2); |
| @@ -225,22 +165,16 @@ static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels · 19 unmodified lines | |||
| 225 | { | 165 | { |
| 226 | register int i; | 166 | register int i; |
| 227 | register vector unsigned char pixelsv1, pixelsv2, pixelsavg; | 167 | register vector unsigned char pixelsv1, pixelsv2, pixelsavg; |
| 228 | register vector unsigned char blockv, temp1, temp2; | 168 | register vector unsigned char blockv; |
| 229 | register vector unsigned short pixelssum1, pixelssum2, temp3; | 169 | register vector unsigned short pixelssum1, pixelssum2, temp3; |
| 230 | register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); | 170 | register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); |
| 231 | register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1); | 171 | register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1); |
| 232 | register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); | 172 | register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); |
| 233 | 173 | ||
| 234 | temp1 = vec_ld(0, pixels); | 174 | pixelsv1 = VEC_LD(0, pixels); |
| 235 | temp2 = vec_ld(16, pixels); | 175 | pixelsv2 = VEC_LD(1, pixels); |
| 236 | pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); | 176 | pixelsv1 = VEC_MERGEH(vczero, pixelsv1); |
| 237 | if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) { | 177 | pixelsv2 = VEC_MERGEH(vczero, pixelsv2); |
| 238 | pixelsv2 = temp2; | ||
| 239 | } else { | ||
| 240 | pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels)); | ||
| 241 | } | ||
| 242 | pixelsv1 = vec_mergeh(vczero, pixelsv1); | ||
| 243 | pixelsv2 = vec_mergeh(vczero, pixelsv2); | ||
| 244 | pixelssum1 = vec_add((vector unsigned short)pixelsv1, | 178 | pixelssum1 = vec_add((vector unsigned short)pixelsv1, |
| 245 | (vector unsigned short)pixelsv2); | 179 | (vector unsigned short)pixelsv2); |
| 246 | pixelssum1 = vec_add(pixelssum1, vcone); | 180 | pixelssum1 = vec_add(pixelssum1, vcone); |
| @@ -249,17 +183,10 @@ static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels · 2 unmodified lines | |||
| 249 | int rightside = ((unsigned long)block & 0x0000000F); | 183 | int rightside = ((unsigned long)block & 0x0000000F); |
| 250 | blockv = vec_ld(0, block); | 184 | blockv = vec_ld(0, block); |
| 251 | 185 | ||
| 252 | temp1 = vec_ld(line_size, pixels); | 186 | pixelsv1 = unaligned_load(line_size, pixels); |
| 253 | temp2 = vec_ld(line_size + 16, pixels); | 187 | pixelsv2 = unaligned_load(line_size+1, pixels); |
| 254 | pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); | 188 | pixelsv1 = VEC_MERGEH(vczero, pixelsv1); |
| 255 | if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) { | 189 | pixelsv2 = VEC_MERGEH(vczero, pixelsv2); |
| 256 | pixelsv2 = temp2; | ||
| 257 | } else { | ||
| 258 | pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels)); | ||
| 259 | } | ||
| 260 | |||
| 261 | pixelsv1 = vec_mergeh(vczero, pixelsv1); | ||
| 262 | pixelsv2 = vec_mergeh(vczero, pixelsv2); | ||
| 263 | pixelssum2 = vec_add((vector unsigned short)pixelsv1, | 190 | pixelssum2 = vec_add((vector unsigned short)pixelsv1, |
| 264 | (vector unsigned short)pixelsv2); | 191 | (vector unsigned short)pixelsv2); |
| 265 | temp3 = vec_add(pixelssum1, pixelssum2); | 192 | temp3 = vec_add(pixelssum1, pixelssum2); |
| @@ -285,24 +212,18 @@ static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, pt · 19 unmodified lines | |||
| 285 | { | 212 | { |
| 286 | register int i; | 213 | register int i; |
| 287 | register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4; | 214 | register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4; |
| 288 | register vector unsigned char blockv, temp1, temp2; | 215 | register vector unsigned char blockv; |
| 289 | register vector unsigned short temp3, temp4, | 216 | register vector unsigned short temp3, temp4, |
| 290 | pixelssum1, pixelssum2, pixelssum3, pixelssum4; | 217 | pixelssum1, pixelssum2, pixelssum3, pixelssum4; |
| 291 | register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); | 218 | register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); |
| 292 | register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); | 219 | register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); |
| 293 | 220 | ||
| 294 | temp1 = vec_ld(0, pixels); | 221 | pixelsv1 = VEC_LD(0, pixels); |
| 295 | temp2 = vec_ld(16, pixels); | 222 | pixelsv2 = VEC_LD(1, pixels); |
| 296 | pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); | 223 | pixelsv3 = VEC_MERGEL(vczero, pixelsv1); |
| 297 | if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) { | 224 | pixelsv4 = VEC_MERGEL(vczero, pixelsv2); |
| 298 | pixelsv2 = temp2; | 225 | pixelsv1 = VEC_MERGEH(vczero, pixelsv1); |
| 299 | } else { | 226 | pixelsv2 = VEC_MERGEH(vczero, pixelsv2); |
| 300 | pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels)); | ||
| 301 | } | ||
| 302 | pixelsv3 = vec_mergel(vczero, pixelsv1); | ||
| 303 | pixelsv4 = vec_mergel(vczero, pixelsv2); | ||
| 304 | pixelsv1 = vec_mergeh(vczero, pixelsv1); | ||
| 305 | pixelsv2 = vec_mergeh(vczero, pixelsv2); | ||
| 306 | pixelssum3 = vec_add((vector unsigned short)pixelsv3, | 227 | pixelssum3 = vec_add((vector unsigned short)pixelsv3, |
| 307 | (vector unsigned short)pixelsv4); | 228 | (vector unsigned short)pixelsv4); |
| 308 | pixelssum3 = vec_add(pixelssum3, vctwo); | 229 | pixelssum3 = vec_add(pixelssum3, vctwo); |
| @@ -313,20 +234,13 @@ static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, pt · 4 unmodified lines | |||
| 313 | for (i = 0; i < h ; i++) { | 234 | for (i = 0; i < h ; i++) { |
| 314 | blockv = vec_ld(0, block); | 235 | blockv = vec_ld(0, block); |
| 315 | 236 | ||
| 316 | temp1 = vec_ld(line_size, pixels); | 237 | pixelsv1 = unaligned_load(line_size, pixels); |
| 317 | temp2 = vec_ld(line_size + 16, pixels); | 238 | pixelsv2 = unaligned_load(line_size+1, pixels); |
| 318 | pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); | ||
| 319 | if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) { | ||
| 320 | pixelsv2 = temp2; | ||
| 321 | } else { | ||
| 322 | pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels)); | ||
| 323 | } | ||
| 324 | |||
| 325 | pixelsv3 = vec_mergel(vczero, pixelsv1); | ||
| 326 | pixelsv4 = vec_mergel(vczero, pixelsv2); | ||
| 327 | pixelsv1 = vec_mergeh(vczero, pixelsv1); | ||
| 328 | pixelsv2 = vec_mergeh(vczero, pixelsv2); | ||
| 329 | 239 | ||
| 240 | pixelsv3 = VEC_MERGEL(vczero, pixelsv1); | ||
| 241 | pixelsv4 = VEC_MERGEL(vczero, pixelsv2); | ||
| 242 | pixelsv1 = VEC_MERGEH(vczero, pixelsv1); | ||
| 243 | pixelsv2 = VEC_MERGEH(vczero, pixelsv2); | ||
| 330 | pixelssum4 = vec_add((vector unsigned short)pixelsv3, | 244 | pixelssum4 = vec_add((vector unsigned short)pixelsv3, |
| 331 | (vector unsigned short)pixelsv4); | 245 | (vector unsigned short)pixelsv4); |
| 332 | pixelssum2 = vec_add((vector unsigned short)pixelsv1, | 246 | pixelssum2 = vec_add((vector unsigned short)pixelsv1, |
| @@ -353,25 +267,19 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix · 20 unmodified lines | |||
| 353 | { | 267 | { |
| 354 | register int i; | 268 | register int i; |
| 355 | register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4; | 269 | register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4; |
| 356 | register vector unsigned char blockv, temp1, temp2; | 270 | register vector unsigned char blockv; |
| 357 | register vector unsigned short temp3, temp4, | 271 | register vector unsigned short temp3, temp4, |
| 358 | pixelssum1, pixelssum2, pixelssum3, pixelssum4; | 272 | pixelssum1, pixelssum2, pixelssum3, pixelssum4; |
| 359 | register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); | 273 | register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); |
| 360 | register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1); | 274 | register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1); |
| 361 | register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); | 275 | register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); |
| 362 | 276 | ||
| 363 | temp1 = vec_ld(0, pixels); | 277 | pixelsv1 = VEC_LD(0, pixels); |
| 364 | temp2 = vec_ld(16, pixels); | 278 | pixelsv2 = VEC_LD(1, pixels); |
| 365 | pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); | 279 | pixelsv3 = VEC_MERGEL(vczero, pixelsv1); |
| 366 | if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) { | 280 | pixelsv4 = VEC_MERGEL(vczero, pixelsv2); |
| 367 | pixelsv2 = temp2; | 281 | pixelsv1 = VEC_MERGEH(vczero, pixelsv1); |
| 368 | } else { | 282 | pixelsv2 = VEC_MERGEH(vczero, pixelsv2); |
| 369 | pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels)); | ||
| 370 | } | ||
| 371 | pixelsv3 = vec_mergel(vczero, pixelsv1); | ||
| 372 | pixelsv4 = vec_mergel(vczero, pixelsv2); | ||
| 373 | pixelsv1 = vec_mergeh(vczero, pixelsv1); | ||
| 374 | pixelsv2 = vec_mergeh(vczero, pixelsv2); | ||
| 375 | pixelssum3 = vec_add((vector unsigned short)pixelsv3, | 283 | pixelssum3 = vec_add((vector unsigned short)pixelsv3, |
| 376 | (vector unsigned short)pixelsv4); | 284 | (vector unsigned short)pixelsv4); |
| 377 | pixelssum3 = vec_add(pixelssum3, vcone); | 285 | pixelssum3 = vec_add(pixelssum3, vcone); |
| @@ -380,22 +288,13 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix · 2 unmodified lines | |||
| 380 | pixelssum1 = vec_add(pixelssum1, vcone); | 288 | pixelssum1 = vec_add(pixelssum1, vcone); |
| 381 | 289 | ||
| 382 | for (i = 0; i < h ; i++) { | 290 | for (i = 0; i < h ; i++) { |
| 383 | blockv = vec_ld(0, block); | 291 | pixelsv1 = unaligned_load(line_size, pixels); |
| 384 | 292 | pixelsv2 = unaligned_load(line_size+1, pixels); | |
| 385 | temp1 = vec_ld(line_size, pixels); | ||
| 386 | temp2 = vec_ld(line_size + 16, pixels); | ||
| 387 | pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); | ||
| 388 | if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) { | ||
| 389 | pixelsv2 = temp2; | ||
| 390 | } else { | ||
| 391 | pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels)); | ||
| 392 | } | ||
| 393 | |||
| 394 | pixelsv3 = vec_mergel(vczero, pixelsv1); | ||
| 395 | pixelsv4 = vec_mergel(vczero, pixelsv2); | ||
| 396 | pixelsv1 = vec_mergeh(vczero, pixelsv1); | ||
| 397 | pixelsv2 = vec_mergeh(vczero, pixelsv2); | ||
| 398 | 293 | ||
| 294 | pixelsv3 = VEC_MERGEL(vczero, pixelsv1); | ||
| 295 | pixelsv4 = VEC_MERGEL(vczero, pixelsv2); | ||
| 296 | pixelsv1 = VEC_MERGEH(vczero, pixelsv1); | ||
| 297 | pixelsv2 = VEC_MERGEH(vczero, pixelsv2); | ||
| 399 | pixelssum4 = vec_add((vector unsigned short)pixelsv3, | 298 | pixelssum4 = vec_add((vector unsigned short)pixelsv3, |
| 400 | (vector unsigned short)pixelsv4); | 299 | (vector unsigned short)pixelsv4); |
| 401 | pixelssum2 = vec_add((vector unsigned short)pixelsv1, | 300 | pixelssum2 = vec_add((vector unsigned short)pixelsv1, |
| @@ -410,7 +309,7 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix · 8 unmodified lines | |||
| 410 | 309 | ||
| 411 | blockv = vec_packsu(temp3, temp4); | 310 | blockv = vec_packsu(temp3, temp4); |
| 412 | 311 | ||
| 413 | vec_st(blockv, 0, block); | 312 | VEC_ST(blockv, 0, block); |
| 414 | 313 | ||
| 415 | block += line_size; | 314 | block += line_size; |
| 416 | pixels += line_size; | 315 | pixels += line_size; |
| @@ -422,7 +321,7 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi · 5 unmodified lines | |||
| 422 | { | 321 | { |
| 423 | register int i; | 322 | register int i; |
| 424 | register vector unsigned char pixelsv1, pixelsv2, pixelsavg; | 323 | register vector unsigned char pixelsv1, pixelsv2, pixelsavg; |
| 425 | register vector unsigned char blockv, temp1, temp2, blocktemp; | 324 | register vector unsigned char blockv, blocktemp; |
| 426 | register vector unsigned short pixelssum1, pixelssum2, temp3; | 325 | register vector unsigned short pixelssum1, pixelssum2, temp3; |
| 427 | 326 | ||
| 428 | register const vector unsigned char vczero = (const vector unsigned char) | 327 | register const vector unsigned char vczero = (const vector unsigned char) |
| @@ -430,16 +329,10 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi · 1 unmodified line | |||
| 430 | register const vector unsigned short vctwo = (const vector unsigned short) | 329 | register const vector unsigned short vctwo = (const vector unsigned short) |
| 431 | vec_splat_u16(2); | 330 | vec_splat_u16(2); |
| 432 | 331 | ||
| 433 | temp1 = vec_ld(0, pixels); | 332 | pixelsv1 = VEC_LD(0, pixels); |
| 434 | temp2 = vec_ld(16, pixels); | 333 | pixelsv2 = VEC_LD(1, pixels); |
| 435 | pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); | 334 | pixelsv1 = VEC_MERGEH(vczero, pixelsv1); |
| 436 | if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) { | 335 | pixelsv2 = VEC_MERGEH(vczero, pixelsv2); |
| 437 | pixelsv2 = temp2; | ||
| 438 | } else { | ||
| 439 | pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels)); | ||
| 440 | } | ||
| 441 | pixelsv1 = vec_mergeh(vczero, pixelsv1); | ||
| 442 | pixelsv2 = vec_mergeh(vczero, pixelsv2); | ||
| 443 | pixelssum1 = vec_add((vector unsigned short)pixelsv1, | 336 | pixelssum1 = vec_add((vector unsigned short)pixelsv1, |
| 444 | (vector unsigned short)pixelsv2); | 337 | (vector unsigned short)pixelsv2); |
| 445 | pixelssum1 = vec_add(pixelssum1, vctwo); | 338 | pixelssum1 = vec_add(pixelssum1, vctwo); |
| @@ -448,17 +341,11 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi · 2 unmodified lines | |||
| 448 | int rightside = ((unsigned long)block & 0x0000000F); | 341 | int rightside = ((unsigned long)block & 0x0000000F); |
| 449 | blockv = vec_ld(0, block); | 342 | blockv = vec_ld(0, block); |
| 450 | 343 | ||
| 451 | temp1 = vec_ld(line_size, pixels); | 344 | pixelsv1 = unaligned_load(line_size, pixels); |
| 452 | temp2 = vec_ld(line_size + 16, pixels); | 345 | pixelsv2 = unaligned_load(line_size+1, pixels); |
| 453 | pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); | ||
| 454 | if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) { | ||
| 455 | pixelsv2 = temp2; | ||
| 456 | } else { | ||
| 457 | pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels)); | ||
| 458 | } | ||
| 459 | 346 | ||
| 460 | pixelsv1 = vec_mergeh(vczero, pixelsv1); | 347 | pixelsv1 = VEC_MERGEH(vczero, pixelsv1); |
| 461 | pixelsv2 = vec_mergeh(vczero, pixelsv2); | 348 | pixelsv2 = VEC_MERGEH(vczero, pixelsv2); |
| 462 | pixelssum2 = vec_add((vector unsigned short)pixelsv1, | 349 | pixelssum2 = vec_add((vector unsigned short)pixelsv1, |
| 463 | (vector unsigned short)pixelsv2); | 350 | (vector unsigned short)pixelsv2); |
| 464 | temp3 = vec_add(pixelssum1, pixelssum2); | 351 | temp3 = vec_add(pixelssum1, pixelssum2); |