OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2014 The LibYuv Project Authors. All rights reserved. | 2 * Copyright 2014 The LibYuv Project Authors. All rights reserved. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license | 4 * Use of this source code is governed by a BSD-style license |
5 * that can be found in the LICENSE file in the root of the source | 5 * that can be found in the LICENSE file in the root of the source |
6 * tree. An additional intellectual property rights grant can be found | 6 * tree. An additional intellectual property rights grant can be found |
7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
9 */ | 9 */ |
10 | 10 |
(...skipping 432 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
443 : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20", | 443 : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20", |
444 "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30" | 444 "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30" |
445 ); | 445 ); |
446 } | 446 } |
447 #endif // HAS_I422TOARGB4444ROW_NEON | 447 #endif // HAS_I422TOARGB4444ROW_NEON |
448 | 448 |
449 #ifdef HAS_I400TOARGBROW_NEON | 449 #ifdef HAS_I400TOARGBROW_NEON |
450 void I400ToARGBRow_NEON(const uint8* src_y, | 450 void I400ToARGBRow_NEON(const uint8* src_y, |
451 uint8* dst_argb, | 451 uint8* dst_argb, |
452 int width) { | 452 int width) { |
453 int64 width64 = (int64)(width); | |
454 asm volatile ( | 453 asm volatile ( |
455 YUVTORGB_SETUP | 454 YUVTORGB_SETUP |
456 "movi v23.8b, #255 \n" | 455 "movi v23.8b, #255 \n" |
457 "1: \n" | 456 "1: \n" |
458 READYUV400 | 457 READYUV400 |
459 YUVTORGB(v22, v21, v20) | 458 YUVTORGB(v22, v21, v20) |
460 "subs %w2, %w2, #8 \n" | 459 "subs %w2, %w2, #8 \n" |
461 MEMACCESS(1) | 460 MEMACCESS(1) |
462 "st4 {v20.8b,v21.8b,v22.8b,v23.8b}, [%1], #32 \n" | 461 "st4 {v20.8b,v21.8b,v22.8b,v23.8b}, [%1], #32 \n" |
463 "b.gt 1b \n" | 462 "b.gt 1b \n" |
464 : "+r"(src_y), // %0 | 463 : "+r"(src_y), // %0 |
465 "+r"(dst_argb), // %1 | 464 "+r"(dst_argb), // %1 |
466 "+r"(width64) // %2 | 465 "+r"(width) // %2 |
467 : [kUVToRB]"r"(&kYuvI601Constants.kUVToRB), | 466 : [kUVToRB]"r"(&kYuvI601Constants.kUVToRB), |
468 [kUVToG]"r"(&kYuvI601Constants.kUVToG), | 467 [kUVToG]"r"(&kYuvI601Constants.kUVToG), |
469 [kUVBiasBGR]"r"(&kYuvI601Constants.kUVBiasBGR), | 468 [kUVBiasBGR]"r"(&kYuvI601Constants.kUVBiasBGR), |
470 [kYToRgb]"r"(&kYuvI601Constants.kYToRgb) | 469 [kYToRgb]"r"(&kYuvI601Constants.kYToRgb) |
471 : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20", | 470 : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20", |
472 "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30" | 471 "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30" |
473 ); | 472 ); |
474 } | 473 } |
475 #endif // HAS_I400TOARGBROW_NEON | 474 #endif // HAS_I400TOARGBROW_NEON |
476 | 475 |
(...skipping 920 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
1397 : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7" | 1396 : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7" |
1398 ); | 1397 ); |
1399 } | 1398 } |
1400 #endif // HAS_ARGBTOYROW_NEON | 1399 #endif // HAS_ARGBTOYROW_NEON |
1401 | 1400 |
1402 #ifdef HAS_ARGBEXTRACTALPHAROW_NEON | 1401 #ifdef HAS_ARGBEXTRACTALPHAROW_NEON |
1403 void ARGBExtractAlphaRow_NEON(const uint8* src_argb, uint8* dst_a, int width) { | 1402 void ARGBExtractAlphaRow_NEON(const uint8* src_argb, uint8* dst_a, int width) { |
1404 asm volatile ( | 1403 asm volatile ( |
1405 "1: \n" | 1404 "1: \n" |
1406 MEMACCESS(0) | 1405 MEMACCESS(0) |
1407 "ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n" // load row 8 pixels | 1406 "ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n" // load row 16 pix
els |
1408 "subs %w2, %w2, #8 \n" // 8 processed per loop | 1407 "subs %w2, %w2, #16 \n" // 16 processed per loop |
1409 MEMACCESS(1) | 1408 MEMACCESS(1) |
1410 "st1 {v3.8b}, [%1], #8 \n" // store 8 A's. | 1409 "st1 {v3.16b}, [%1], #16 \n" // store 16 A's. |
1411 "b.gt 1b \n" | 1410 "b.gt 1b \n" |
1412 : "+r"(src_argb), // %0 | 1411 : "+r"(src_argb), // %0 |
1413 "+r"(dst_a), // %1 | 1412 "+r"(dst_a), // %1 |
1414 "+r"(width) // %2 | 1413 "+r"(width) // %2 |
1415 : | 1414 : |
1416 : "cc", "memory", "v0", "v1", "v2", "v3" // Clobber List | 1415 : "cc", "memory", "v0", "v1", "v2", "v3" // Clobber List |
1417 ); | 1416 ); |
1418 } | 1417 } |
1419 #endif // HAS_ARGBEXTRACTALPHAROW_NEON | 1418 #endif // HAS_ARGBEXTRACTALPHAROW_NEON |
1420 | 1419 |
(...skipping 1550 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
2971 : "cc", "memory", "v0", "v1", "v2", "v3" // Clobber List | 2970 : "cc", "memory", "v0", "v1", "v2", "v3" // Clobber List |
2972 ); | 2971 ); |
2973 } | 2972 } |
2974 #endif // HAS_SOBELYROW_NEON | 2973 #endif // HAS_SOBELYROW_NEON |
2975 #endif // !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__) | 2974 #endif // !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__) |
2976 | 2975 |
2977 #ifdef __cplusplus | 2976 #ifdef __cplusplus |
2978 } // extern "C" | 2977 } // extern "C" |
2979 } // namespace libyuv | 2978 } // namespace libyuv |
2980 #endif | 2979 #endif |
OLD | NEW |