external/skia/source/skia_opts_internal.hxx

   1 /*
   2  * Use of this source code is governed by a BSD-style license that can be
   3  * found in the LICENSE file.
   4  */
   5
   6 #ifndef SKIA_OPTS_INTERNAL_H
   7 #define SKIA_OPTS_INTERNAL_H
   8
   9 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
  10     #include <immintrin.h>
  11 #endif
  12
  13 namespace SK_OPTS_NS {
  14
  15 static void RGB1_to_RGB_portable(uint8_t dst[], const uint32_t* src, int count) {
  16     for (int i = 0; i < count; i++) {
  17         dst[0] = src[i] >> 0;
  18         dst[1] = src[i] >> 8;
  19         dst[2] = src[i] >> 16;
  20         dst += 3;
  21     }
  22 }
  23 static void RGB1_to_R_portable(uint8_t dst[], const uint32_t* src, int count) {
  24     for (int i = 0; i < count; i++) {
  25         dst[i] = src[i] & 0xFF;
  26     }
  27 }
  28
  29 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
  30 inline void RGB1_to_RGB(uint8_t dst[], const uint32_t* src, int count) {
  31         const uint8_t X = 0xFF; // Used a placeholder.  The value of X is irrelevant.
  32         __m128i pack = _mm_setr_epi8(0,1,2, 4,5,6, 8,9,10, 12,13,14, X,X,X,X);
  33
  34 // Storing 4 pixels should store 12 bytes, but here it stores 16, so test count >= 6
  35 // in order to not overrun the output buffer.
  36         while (count >= 6) {
  37             __m128i rgba = _mm_loadu_si128((const __m128i*) src);
  38
  39             __m128i rgb = _mm_shuffle_epi8(rgba, pack);
  40
  41             // Store 4 pixels.
  42             _mm_storeu_si128((__m128i*) dst, rgb);
  43
  44             src += 4;
  45             dst += 4*3;
  46             count -= 4;
  47         }
  48         RGB1_to_RGB_portable(dst, src, count);
  49 }
  50
  51 inline void RGB1_to_R(uint8_t dst[], const uint32_t* src, int count) {
  52         const uint8_t X = 0xFF; // Used a placeholder.  The value of X is irrelevant.
  53         __m128i pack = _mm_setr_epi8(0,4,8,12, X,X,X,X,X,X,X,X,X,X,X,X);
  54
  55         while (count >= 4) {
  56             __m128i rgba = _mm_loadu_si128((const __m128i*) src);
  57
  58             __m128i rgb = _mm_shuffle_epi8(rgba, pack);
  59
  60             // Store 4 pixels.
  61             *((uint32_t*)dst) = _mm_cvtsi128_si32(rgb);
  62
  63             src += 4;
  64             dst += 4;
  65             count -= 4;
  66         }
  67         RGB1_to_R_portable(dst, src, count);
  68 }
  69
  70 #else
  71 inline void RGB1_to_RGB(uint8_t dst[], const uint32_t* src, int count) {
  72     RGB1_to_RGB_portable(dst, src, count);
  73 }
  74 inline void RGB1_to_R(uint8_t dst[], const uint32_t* src, int count) {
  75     RGB1_to_R_portable(dst, src, count);
  76 }
  77 #endif
  78
  79 } // namespace
  80
  81 #endif