git.sesse.net Git - ffmpeg/blob - libswscale/aarch64/swscale_unscaled.c

   1 /*
   2  * This file is part of FFmpeg.
   3  *
   4  * FFmpeg is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU Lesser General Public
   6  * License as published by the Free Software Foundation; either
   7  * version 2.1 of the License, or (at your option) any later version.
   8  *
   9  * FFmpeg is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12  * Lesser General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU Lesser General Public
  15  * License along with FFmpeg; if not, write to the Free Software
  16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  17  */
  18
  19 #include "config.h"
  20 #include "libswscale/swscale.h"
  21 #include "libswscale/swscale_internal.h"
  22 #include "libavutil/aarch64/cpu.h"
  23
  24 #define YUV_TO_RGB_TABLE                                                                    \
  25         c->yuv2rgb_v2r_coeff,                                                               \
  26         c->yuv2rgb_u2g_coeff,                                                               \
  27         c->yuv2rgb_v2g_coeff,                                                               \
  28         c->yuv2rgb_u2b_coeff,                                                               \
  29
  30 #define DECLARE_FF_YUVX_TO_RGBX_FUNCS(ifmt, ofmt)                                           \
  31 int ff_##ifmt##_to_##ofmt##_neon(int w, int h,                                              \
  32                                  uint8_t *dst, int linesize,                                \
  33                                  const uint8_t *srcY, int linesizeY,                        \
  34                                  const uint8_t *srcU, int linesizeU,                        \
  35                                  const uint8_t *srcV, int linesizeV,                        \
  36                                  const int16_t *table,                                      \
  37                                  int y_offset,                                              \
  38                                  int y_coeff);                                              \
  39                                                                                             \
  40 static int ifmt##_to_##ofmt##_neon_wrapper(SwsContext *c, const uint8_t *src[],             \
  41                                            int srcStride[], int srcSliceY, int srcSliceH,   \
  42                                            uint8_t *dst[], int dstStride[]) {               \
  43     const int16_t yuv2rgb_table[] = { YUV_TO_RGB_TABLE };                                   \
  44                                                                                             \
  45     ff_##ifmt##_to_##ofmt##_neon(c->srcW, srcSliceH,                                        \
  46                                  dst[0] + srcSliceY * dstStride[0], dstStride[0],           \
  47                                  src[0], srcStride[0],                                      \
  48                                  src[1], srcStride[1],                                      \
  49                                  src[2], srcStride[2],                                      \
  50                                  yuv2rgb_table,                                             \
  51                                  c->yuv2rgb_y_offset >> 6,                                  \
  52                                  c->yuv2rgb_y_coeff);                                       \
  53     return 0;                                                                               \
  54 }                                                                                           \
  55
  56 #define DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuvx)                                             \
  57 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, argb)                                                   \
  58 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, rgba)                                                   \
  59 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, abgr)                                                   \
  60 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, bgra)                                                   \
  61
  62 DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuv420p)
  63 DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuv422p)
  64
  65 #define DECLARE_FF_NVX_TO_RGBX_FUNCS(ifmt, ofmt)                                            \
  66 int ff_##ifmt##_to_##ofmt##_neon(int w, int h,                                              \
  67                                  uint8_t *dst, int linesize,                                \
  68                                  const uint8_t *srcY, int linesizeY,                        \
  69                                  const uint8_t *srcC, int linesizeC,                        \
  70                                  const int16_t *table,                                      \
  71                                  int y_offset,                                              \
  72                                  int y_coeff);                                              \
  73                                                                                             \
  74 static int ifmt##_to_##ofmt##_neon_wrapper(SwsContext *c, const uint8_t *src[],             \
  75                                            int srcStride[], int srcSliceY, int srcSliceH,   \
  76                                            uint8_t *dst[], int dstStride[]) {               \
  77     const int16_t yuv2rgb_table[] = { YUV_TO_RGB_TABLE };                                   \
  78                                                                                             \
  79     ff_##ifmt##_to_##ofmt##_neon(c->srcW, srcSliceH,                                        \
  80                                  dst[0] + srcSliceY * dstStride[0], dstStride[0],           \
  81                                  src[0], srcStride[0], src[1], srcStride[1],                \
  82                                  yuv2rgb_table,                                             \
  83                                  c->yuv2rgb_y_offset >> 6,                                  \
  84                                  c->yuv2rgb_y_coeff);                                       \
  85                                                                                             \
  86     return 0;                                                                               \
  87 }                                                                                           \
  88
  89 #define DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nvx)                                               \
  90 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, argb)                                                     \
  91 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, rgba)                                                     \
  92 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, abgr)                                                     \
  93 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, bgra)                                                     \
  94
  95 DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv12)
  96 DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv21)
  97
  98 /* We need a 16 pixel width alignment. This constraint can easily be removed
  99  * for input reading but for the output which is 4-bytes per pixel (RGBA) the
 100  * assembly might be writing as much as 4*15=60 extra bytes at the end of the
 101  * line, which won't fit the 32-bytes buffer alignment. */
 102 #define SET_FF_NVX_TO_RGBX_FUNC(ifmt, IFMT, ofmt, OFMT, accurate_rnd) do {                  \
 103     if (c->srcFormat == AV_PIX_FMT_##IFMT                                                   \
 104         && c->dstFormat == AV_PIX_FMT_##OFMT                                                \
 105         && !(c->srcH & 1)                                                                   \
 106         && !(c->srcW & 15)                                                                  \
 107         && !accurate_rnd)                                                                   \
 108         c->swscale = ifmt##_to_##ofmt##_neon_wrapper;                                       \
 109 } while (0)
 110
 111 #define SET_FF_NVX_TO_ALL_RGBX_FUNC(nvx, NVX, accurate_rnd) do {                            \
 112     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, argb, ARGB, accurate_rnd);                            \
 113     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, rgba, RGBA, accurate_rnd);                            \
 114     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, abgr, ABGR, accurate_rnd);                            \
 115     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, bgra, BGRA, accurate_rnd);                            \
 116 } while (0)
 117
 118 static void get_unscaled_swscale_neon(SwsContext *c) {
 119     int accurate_rnd = c->flags & SWS_ACCURATE_RND;
 120
 121     SET_FF_NVX_TO_ALL_RGBX_FUNC(nv12, NV12, accurate_rnd);
 122     SET_FF_NVX_TO_ALL_RGBX_FUNC(nv21, NV21, accurate_rnd);
 123     SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv420p, YUV420P, accurate_rnd);
 124     SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv422p, YUV422P, accurate_rnd);
 125 }
 126
 127 void ff_get_unscaled_swscale_aarch64(SwsContext *c)
 128 {
 129     int cpu_flags = av_get_cpu_flags();
 130     if (have_neon(cpu_flags))
 131         get_unscaled_swscale_neon(c);
 132 }