OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2016 Google Inc. | 2 * Copyright 2016 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 SkSwizzler_opts_DEFINED | 8 #ifndef SkSwizzler_opts_DEFINED |
9 #define SkSwizzler_opts_DEFINED | 9 #define SkSwizzler_opts_DEFINED |
10 | 10 |
(...skipping 385 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
396 static void grayA_to_RGBA(uint32_t dst[], const void* src, int count) { | 396 static void grayA_to_RGBA(uint32_t dst[], const void* src, int count) { |
397 expand_grayA<false>(dst, src, count); | 397 expand_grayA<false>(dst, src, count); |
398 } | 398 } |
399 | 399 |
400 static void grayA_to_rgbA(uint32_t dst[], const void* src, int count) { | 400 static void grayA_to_rgbA(uint32_t dst[], const void* src, int count) { |
401 expand_grayA<true>(dst, src, count); | 401 expand_grayA<true>(dst, src, count); |
402 } | 402 } |
403 | 403 |
404 #elif SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3 | 404 #elif SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3 |
405 | 405 |
| 406 // Scale a byte by another. |
| 407 // Inputs are stored in 16-bit lanes, but are not larger than 8-bits. |
| 408 static __m128i scale(__m128i x, __m128i y) { |
| 409 const __m128i _128 = _mm_set1_epi16(128); |
| 410 const __m128i _257 = _mm_set1_epi16(257); |
| 411 |
| 412 // (x+127)/255 == ((x+128)*257)>>16 for 0 <= x <= 255*255. |
| 413 return _mm_mulhi_epu16(_mm_add_epi16(_mm_mullo_epi16(x, y), _128), _257); |
| 414 } |
| 415 |
406 template <bool kSwapRB> | 416 template <bool kSwapRB> |
407 static void premul_should_swapRB(uint32_t* dst, const void* vsrc, int count) { | 417 static void premul_should_swapRB(uint32_t* dst, const void* vsrc, int count) { |
408 auto src = (const uint32_t*)vsrc; | 418 auto src = (const uint32_t*)vsrc; |
409 | 419 |
410 auto premul8 = [](__m128i* lo, __m128i* hi) { | 420 auto premul8 = [](__m128i* lo, __m128i* hi) { |
411 const __m128i zeros = _mm_setzero_si128(); | 421 const __m128i zeros = _mm_setzero_si128(); |
412 const __m128i _128 = _mm_set1_epi16(128); | |
413 const __m128i _257 = _mm_set1_epi16(257); | |
414 __m128i planar; | 422 __m128i planar; |
415 if (kSwapRB) { | 423 if (kSwapRB) { |
416 planar = _mm_setr_epi8(2,6,10,14, 1,5,9,13, 0,4,8,12, 3,7,11,15); | 424 planar = _mm_setr_epi8(2,6,10,14, 1,5,9,13, 0,4,8,12, 3,7,11,15); |
417 } else { | 425 } else { |
418 planar = _mm_setr_epi8(0,4,8,12, 1,5,9,13, 2,6,10,14, 3,7,11,15); | 426 planar = _mm_setr_epi8(0,4,8,12, 1,5,9,13, 2,6,10,14, 3,7,11,15); |
419 } | 427 } |
420 | 428 |
421 // Swizzle the pixels to 8-bit planar. | 429 // Swizzle the pixels to 8-bit planar. |
422 *lo = _mm_shuffle_epi8(*lo, planar); // rrrrgggg bb
bbaaaa | 430 *lo = _mm_shuffle_epi8(*lo, planar); // rrrrgggg bb
bbaaaa |
423 *hi = _mm_shuffle_epi8(*hi, planar); // RRRRGGGG BB
BBAAAA | 431 *hi = _mm_shuffle_epi8(*hi, planar); // RRRRGGGG BB
BBAAAA |
424 __m128i rg = _mm_unpacklo_epi32(*lo, *hi), // rrrrRRRR gg
ggGGGG | 432 __m128i rg = _mm_unpacklo_epi32(*lo, *hi), // rrrrRRRR gg
ggGGGG |
425 ba = _mm_unpackhi_epi32(*lo, *hi); // bbbbBBBB aa
aaAAAA | 433 ba = _mm_unpackhi_epi32(*lo, *hi); // bbbbBBBB aa
aaAAAA |
426 | 434 |
427 // Unpack to 16-bit planar. | 435 // Unpack to 16-bit planar. |
428 __m128i r = _mm_unpacklo_epi8(rg, zeros), // r_r_r_r_ R_
R_R_R_ | 436 __m128i r = _mm_unpacklo_epi8(rg, zeros), // r_r_r_r_ R_
R_R_R_ |
429 g = _mm_unpackhi_epi8(rg, zeros), // g_g_g_g_ G_
G_G_G_ | 437 g = _mm_unpackhi_epi8(rg, zeros), // g_g_g_g_ G_
G_G_G_ |
430 b = _mm_unpacklo_epi8(ba, zeros), // b_b_b_b_ B_
B_B_B_ | 438 b = _mm_unpacklo_epi8(ba, zeros), // b_b_b_b_ B_
B_B_B_ |
431 a = _mm_unpackhi_epi8(ba, zeros); // a_a_a_a_ A_
A_A_A_ | 439 a = _mm_unpackhi_epi8(ba, zeros); // a_a_a_a_ A_
A_A_A_ |
432 | 440 |
433 // Premultiply! (x+127)/255 == ((x+128)*257)>>16 for 0 <= x <= 255*255. | 441 // Premultiply! |
434 r = _mm_mulhi_epu16(_mm_add_epi16(_mm_mullo_epi16(r, a), _128), _257); | 442 r = scale(r, a); |
435 g = _mm_mulhi_epu16(_mm_add_epi16(_mm_mullo_epi16(g, a), _128), _257); | 443 g = scale(g, a); |
436 b = _mm_mulhi_epu16(_mm_add_epi16(_mm_mullo_epi16(b, a), _128), _257); | 444 b = scale(b, a); |
437 | 445 |
438 // Repack into interlaced pixels. | 446 // Repack into interlaced pixels. |
439 rg = _mm_or_si128(r, _mm_slli_epi16(g, 8)); // rgrgrgrg RG
RGRGRG | 447 rg = _mm_or_si128(r, _mm_slli_epi16(g, 8)); // rgrgrgrg RG
RGRGRG |
440 ba = _mm_or_si128(b, _mm_slli_epi16(a, 8)); // babababa BA
BABABA | 448 ba = _mm_or_si128(b, _mm_slli_epi16(a, 8)); // babababa BA
BABABA |
441 *lo = _mm_unpacklo_epi16(rg, ba); // rgbargba rg
bargba | 449 *lo = _mm_unpacklo_epi16(rg, ba); // rgbargba rg
bargba |
442 *hi = _mm_unpackhi_epi16(rg, ba); // RGBARGBA RG
BARGBA | 450 *hi = _mm_unpackhi_epi16(rg, ba); // RGBARGBA RG
BARGBA |
443 }; | 451 }; |
444 | 452 |
445 while (count >= 8) { | 453 while (count >= 8) { |
446 __m128i lo = _mm_loadu_si128((const __m128i*) (src + 0)), | 454 __m128i lo = _mm_loadu_si128((const __m128i*) (src + 0)), |
(...skipping 118 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
565 _mm_storeu_si128((__m128i*) (dst + 12), ggga3); | 573 _mm_storeu_si128((__m128i*) (dst + 12), ggga3); |
566 | 574 |
567 src += 16; | 575 src += 16; |
568 dst += 16; | 576 dst += 16; |
569 count -= 16; | 577 count -= 16; |
570 } | 578 } |
571 | 579 |
572 gray_to_RGB1_portable(dst, src, count); | 580 gray_to_RGB1_portable(dst, src, count); |
573 } | 581 } |
574 | 582 |
575 static void grayA_to_RGBA(uint32_t dst[], const void* src, int count) { | 583 static void grayA_to_RGBA(uint32_t dst[], const void* vsrc, int count) { |
| 584 const uint8_t* src = (const uint8_t*) vsrc; |
| 585 while (count >= 8) { |
| 586 __m128i ga = _mm_loadu_si128((const __m128i*) src); |
| 587 |
| 588 __m128i gg = _mm_or_si128(_mm_and_si128(ga, _mm_set1_epi16(0x00FF)), |
| 589 _mm_slli_epi16(ga, 8)); |
| 590 |
| 591 __m128i ggga_lo = _mm_unpacklo_epi16(gg, ga); |
| 592 __m128i ggga_hi = _mm_unpackhi_epi16(gg, ga); |
| 593 |
| 594 _mm_storeu_si128((__m128i*) (dst + 0), ggga_lo); |
| 595 _mm_storeu_si128((__m128i*) (dst + 4), ggga_hi); |
| 596 |
| 597 src += 8*2; |
| 598 dst += 8; |
| 599 count -= 8; |
| 600 } |
| 601 |
576 grayA_to_RGBA_portable(dst, src, count); | 602 grayA_to_RGBA_portable(dst, src, count); |
577 } | 603 } |
578 | 604 |
579 static void grayA_to_rgbA(uint32_t dst[], const void* src, int count) { | 605 static void grayA_to_rgbA(uint32_t dst[], const void* vsrc, int count) { |
| 606 const uint8_t* src = (const uint8_t*) vsrc; |
| 607 while (count >= 8) { |
| 608 __m128i grayA = _mm_loadu_si128((const __m128i*) src); |
| 609 |
| 610 __m128i g0 = _mm_and_si128(grayA, _mm_set1_epi16(0x00FF)); |
| 611 __m128i a0 = _mm_srli_epi16(grayA, 8); |
| 612 |
| 613 // Premultiply |
| 614 g0 = scale(g0, a0); |
| 615 |
| 616 __m128i gg = _mm_or_si128(g0, _mm_slli_epi16(g0, 8)); |
| 617 __m128i ga = _mm_or_si128(g0, _mm_slli_epi16(a0, 8)); |
| 618 |
| 619 |
| 620 __m128i ggga_lo = _mm_unpacklo_epi16(gg, ga); |
| 621 __m128i ggga_hi = _mm_unpackhi_epi16(gg, ga); |
| 622 |
| 623 _mm_storeu_si128((__m128i*) (dst + 0), ggga_lo); |
| 624 _mm_storeu_si128((__m128i*) (dst + 4), ggga_hi); |
| 625 |
| 626 src += 8*2; |
| 627 dst += 8; |
| 628 count -= 8; |
| 629 } |
| 630 |
580 grayA_to_rgbA_portable(dst, src, count); | 631 grayA_to_rgbA_portable(dst, src, count); |
581 } | 632 } |
582 | 633 |
583 #else | 634 #else |
584 | 635 |
585 static void RGBA_to_rgbA(uint32_t* dst, const void* src, int count) { | 636 static void RGBA_to_rgbA(uint32_t* dst, const void* src, int count) { |
586 RGBA_to_rgbA_portable(dst, src, count); | 637 RGBA_to_rgbA_portable(dst, src, count); |
587 } | 638 } |
588 | 639 |
589 static void RGBA_to_bgrA(uint32_t* dst, const void* src, int count) { | 640 static void RGBA_to_bgrA(uint32_t* dst, const void* src, int count) { |
(...skipping 22 matching lines...) Expand all Loading... |
612 | 663 |
613 static void grayA_to_rgbA(uint32_t dst[], const void* src, int count) { | 664 static void grayA_to_rgbA(uint32_t dst[], const void* src, int count) { |
614 grayA_to_rgbA_portable(dst, src, count); | 665 grayA_to_rgbA_portable(dst, src, count); |
615 } | 666 } |
616 | 667 |
617 #endif | 668 #endif |
618 | 669 |
619 } | 670 } |
620 | 671 |
621 #endif // SkSwizzler_opts_DEFINED | 672 #endif // SkSwizzler_opts_DEFINED |
OLD | NEW |