OLD | NEW |
1 // qcms | 1 // qcms |
2 // Copyright (C) 2009 Mozilla Foundation | 2 // Copyright (C) 2009 Mozilla Foundation |
3 // | 3 // |
4 // Permission is hereby granted, free of charge, to any person obtaining | 4 // Permission is hereby granted, free of charge, to any person obtaining |
5 // a copy of this software and associated documentation files (the "Software"), | 5 // a copy of this software and associated documentation files (the "Software"), |
6 // to deal in the Software without restriction, including without limitation | 6 // to deal in the Software without restriction, including without limitation |
7 // the rights to use, copy, modify, merge, publish, distribute, sublicense, | 7 // the rights to use, copy, modify, merge, publish, distribute, sublicense, |
8 // and/or sell copies of the Software, and to permit persons to whom the Softwar
e | 8 // and/or sell copies of the Software, and to permit persons to whom the Softwar
e |
9 // is furnished to do so, subject to the following conditions: | 9 // is furnished to do so, subject to the following conditions: |
10 // | 10 // |
(...skipping 254 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
265 *((__m64 *)&output[0]) = _mm_cvtps_pi32(result); | 265 *((__m64 *)&output[0]) = _mm_cvtps_pi32(result); |
266 result = _mm_movehl_ps(result, result); | 266 result = _mm_movehl_ps(result, result); |
267 *((__m64 *)&output[2]) = _mm_cvtps_pi32(result); | 267 *((__m64 *)&output[2]) = _mm_cvtps_pi32(result); |
268 | 268 |
269 dest[0] = otdata_r[output[0]]; | 269 dest[0] = otdata_r[output[0]]; |
270 dest[1] = otdata_g[output[1]]; | 270 dest[1] = otdata_g[output[1]]; |
271 dest[2] = otdata_b[output[2]]; | 271 dest[2] = otdata_b[output[2]]; |
272 | 272 |
273 _mm_empty(); | 273 _mm_empty(); |
274 } | 274 } |
| 275 |
| 276 void qcms_transform_data_bgra_out_lut_sse1(qcms_transform *transform, |
| 277 unsigned char *src, |
| 278 unsigned char *dest, |
| 279 size_t length) |
| 280 { |
| 281 unsigned int i; |
| 282 float (*mat)[4] = transform->matrix; |
| 283 char input_back[32]; |
| 284 /* Ensure we have a buffer that's 16 byte aligned regardless of the original |
| 285 * stack alignment. We can't use __attribute__((aligned(16))) or __declspec(
align(32)) |
| 286 * because they don't work on stack variables. gcc 4.4 does do the right thi
ng |
| 287 * on x86 but that's too new for us right now. For more info: gcc bug #16660
*/ |
| 288 float const * input = (float*)(((uintptr_t)&input_back[16]) & ~0xf); |
| 289 /* share input and output locations to save having to keep the |
| 290 * locations in separate registers */ |
| 291 uint32_t const * output = (uint32_t*)input; |
| 292 |
| 293 /* deref *transform now to avoid it in loop */ |
| 294 const float *igtbl_r = transform->input_gamma_table_r; |
| 295 const float *igtbl_g = transform->input_gamma_table_g; |
| 296 const float *igtbl_b = transform->input_gamma_table_b; |
| 297 |
| 298 /* deref *transform now to avoid it in loop */ |
| 299 const uint8_t *otdata_r = &transform->output_table_r->data[0]; |
| 300 const uint8_t *otdata_g = &transform->output_table_g->data[0]; |
| 301 const uint8_t *otdata_b = &transform->output_table_b->data[0]; |
| 302 |
| 303 /* input matrix values never change */ |
| 304 const __m128 mat0 = _mm_load_ps(mat[0]); |
| 305 const __m128 mat1 = _mm_load_ps(mat[1]); |
| 306 const __m128 mat2 = _mm_load_ps(mat[2]); |
| 307 |
| 308 /* these values don't change, either */ |
| 309 const __m128 max = _mm_load_ps(clampMaxValueX4); |
| 310 const __m128 min = _mm_setzero_ps(); |
| 311 const __m128 scale = _mm_load_ps(floatScaleX4); |
| 312 |
| 313 /* working variables */ |
| 314 __m128 vec_r, vec_g, vec_b, result; |
| 315 unsigned char alpha; |
| 316 |
| 317 /* CYA */ |
| 318 if (!length) |
| 319 return; |
| 320 |
| 321 /* one pixel is handled outside of the loop */ |
| 322 length--; |
| 323 |
| 324 /* setup for transforming 1st pixel */ |
| 325 vec_b = _mm_load_ss(&igtbl_b[src[0]]); |
| 326 vec_g = _mm_load_ss(&igtbl_g[src[1]]); |
| 327 vec_r = _mm_load_ss(&igtbl_r[src[2]]); |
| 328 alpha = src[3]; |
| 329 src += 4; |
| 330 |
| 331 /* transform all but final pixel */ |
| 332 |
| 333 for (i=0; i<length; i++) |
| 334 { |
| 335 /* position values from gamma tables */ |
| 336 vec_r = _mm_shuffle_ps(vec_r, vec_r, 0); |
| 337 vec_g = _mm_shuffle_ps(vec_g, vec_g, 0); |
| 338 vec_b = _mm_shuffle_ps(vec_b, vec_b, 0); |
| 339 |
| 340 /* gamma * matrix */ |
| 341 vec_r = _mm_mul_ps(vec_r, mat0); |
| 342 vec_g = _mm_mul_ps(vec_g, mat1); |
| 343 vec_b = _mm_mul_ps(vec_b, mat2); |
| 344 |
| 345 /* store alpha for this pixel; load alpha for next */ |
| 346 dest[3] = alpha; |
| 347 alpha = src[3]; |
| 348 |
| 349 /* crunch, crunch, crunch */ |
| 350 vec_r = _mm_add_ps(vec_r, _mm_add_ps(vec_g, vec_b)); |
| 351 vec_r = _mm_max_ps(min, vec_r); |
| 352 vec_r = _mm_min_ps(max, vec_r); |
| 353 result = _mm_mul_ps(vec_r, scale); |
| 354 |
| 355 /* store calc'd output tables indices */ |
| 356 *((__m64 *)&output[0]) = _mm_cvtps_pi32(result); |
| 357 result = _mm_movehl_ps(result, result); |
| 358 *((__m64 *)&output[2]) = _mm_cvtps_pi32(result); |
| 359 |
| 360 /* load gamma values for next loop while store completes */ |
| 361 vec_b = _mm_load_ss(&igtbl_b[src[0]]); |
| 362 vec_g = _mm_load_ss(&igtbl_g[src[1]]); |
| 363 vec_r = _mm_load_ss(&igtbl_r[src[2]]); |
| 364 src += 4; |
| 365 |
| 366 /* use calc'd indices to output RGB values */ |
| 367 dest[0] = otdata_b[output[2]]; |
| 368 dest[1] = otdata_g[output[1]]; |
| 369 dest[2] = otdata_r[output[0]]; |
| 370 dest += 4; |
| 371 } |
| 372 |
| 373 /* handle final (maybe only) pixel */ |
| 374 |
| 375 vec_r = _mm_shuffle_ps(vec_r, vec_r, 0); |
| 376 vec_g = _mm_shuffle_ps(vec_g, vec_g, 0); |
| 377 vec_b = _mm_shuffle_ps(vec_b, vec_b, 0); |
| 378 |
| 379 vec_r = _mm_mul_ps(vec_r, mat0); |
| 380 vec_g = _mm_mul_ps(vec_g, mat1); |
| 381 vec_b = _mm_mul_ps(vec_b, mat2); |
| 382 |
| 383 dest[3] = alpha; |
| 384 |
| 385 vec_r = _mm_add_ps(vec_r, _mm_add_ps(vec_g, vec_b)); |
| 386 vec_r = _mm_max_ps(min, vec_r); |
| 387 vec_r = _mm_min_ps(max, vec_r); |
| 388 result = _mm_mul_ps(vec_r, scale); |
| 389 |
| 390 *((__m64 *)&output[0]) = _mm_cvtps_pi32(result); |
| 391 result = _mm_movehl_ps(result, result); |
| 392 *((__m64 *)&output[2]) = _mm_cvtps_pi32(result); |
| 393 |
| 394 dest[0] = otdata_b[output[2]]; |
| 395 dest[1] = otdata_g[output[1]]; |
| 396 dest[2] = otdata_r[output[0]]; |
| 397 |
| 398 _mm_empty(); |
| 399 } |
OLD | NEW |