OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2012 The Android Open Source Project | 2 * Copyright 2012 The Android Open Source Project |
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 #include <emmintrin.h> | 8 #include <emmintrin.h> |
9 #include "SkBitmapProcState_opts_SSE2.h" | 9 #include "SkBitmapProcState_opts_SSE2.h" |
10 #include "SkBlitRow_opts_SSE2.h" | 10 #include "SkBlitRow_opts_SSE2.h" |
(...skipping 423 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
434 __m128i *d = reinterpret_cast<__m128i*>(dst); | 434 __m128i *d = reinterpret_cast<__m128i*>(dst); |
435 __m128i rb_mask = _mm_set1_epi32(0x00FF00FF); | 435 __m128i rb_mask = _mm_set1_epi32(0x00FF00FF); |
436 __m128i c_256 = _mm_set1_epi16(256); | 436 __m128i c_256 = _mm_set1_epi16(256); |
437 __m128i c_1 = _mm_set1_epi16(1); | 437 __m128i c_1 = _mm_set1_epi16(1); |
438 __m128i src_pixel = _mm_set1_epi32(color); | 438 __m128i src_pixel = _mm_set1_epi32(color); |
439 while (count >= 4) { | 439 while (count >= 4) { |
440 // Load 4 pixels each of src and dest. | 440 // Load 4 pixels each of src and dest. |
441 __m128i dst_pixel = _mm_load_si128(d); | 441 __m128i dst_pixel = _mm_load_si128(d); |
442 | 442 |
443 //set the aphla value | 443 //set the aphla value |
444 __m128i src_scale_wide = _mm_set_epi8(0, *(mask+3),\ | 444 __m128i src_scale_wide = _mm_cvtsi32_si128(*reinterpret_cast<con
st uint32_t*>(mask)); |
445 0, *(mask+3),0, \ | 445 src_scale_wide = _mm_unpacklo_epi8(src_scale_wide, |
446 *(mask+2),0, *(mask+2),\ | 446 _mm_setzero_si128()); |
447 0,*(mask+1), 0,*(mask+1),\ | 447 src_scale_wide = _mm_unpacklo_epi16(src_scale_wide, src_scale_wi
de); |
448 0, *mask,0,*mask); | |
449 | 448 |
450 //call SkAlpha255To256() | 449 //call SkAlpha255To256() |
451 src_scale_wide = _mm_add_epi16(src_scale_wide, c_1); | 450 src_scale_wide = _mm_add_epi16(src_scale_wide, c_1); |
452 | 451 |
453 // Get red and blue pixels into lower byte of each word. | 452 // Get red and blue pixels into lower byte of each word. |
454 __m128i dst_rb = _mm_and_si128(rb_mask, dst_pixel); | 453 __m128i dst_rb = _mm_and_si128(rb_mask, dst_pixel); |
455 __m128i src_rb = _mm_and_si128(rb_mask, src_pixel); | 454 __m128i src_rb = _mm_and_si128(rb_mask, src_pixel); |
456 | 455 |
457 // Get alpha and green into lower byte of each word. | 456 // Get alpha and green into lower byte of each word. |
458 __m128i dst_ag = _mm_srli_epi16(dst_pixel, 8); | 457 __m128i dst_ag = _mm_srli_epi16(dst_pixel, 8); |
(...skipping 893 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
1352 uint32_t dst_expanded = SkExpand_rgb_16(*dst); | 1351 uint32_t dst_expanded = SkExpand_rgb_16(*dst); |
1353 dst_expanded = dst_expanded * (SkAlpha255To256(255 - a) >> 3); | 1352 dst_expanded = dst_expanded * (SkAlpha255To256(255 - a) >> 3); |
1354 // now src and dst expanded are in g:11 r:10 x:1 b:10 | 1353 // now src and dst expanded are in g:11 r:10 x:1 b:10 |
1355 *dst = SkCompact_rgb_16((src_expanded + dst_expanded) >> 5); | 1354 *dst = SkCompact_rgb_16((src_expanded + dst_expanded) >> 5); |
1356 } | 1355 } |
1357 dst += 1; | 1356 dst += 1; |
1358 DITHER_INC_X(x); | 1357 DITHER_INC_X(x); |
1359 } while (--count != 0); | 1358 } while (--count != 0); |
1360 } | 1359 } |
1361 } | 1360 } |
OLD | NEW |