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

   1 /*
   2  * Copyright (C) 2013 Xiaolei Yu <dreifachstein@gmail.com>
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * FFmpeg is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with FFmpeg; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include "config.h"
  22 #include "libswscale/swscale.h"
  23 #include "libswscale/swscale_internal.h"
  24 #include "libavutil/arm/cpu.h"
  25
  26 extern void rgbx_to_nv12_neon_32(const uint8_t *src, uint8_t *y, uint8_t *chroma,
  27                 int width, int height,
  28                 int y_stride, int c_stride, int src_stride,
  29                 int32_t coeff_tbl[9]);
  30
  31 extern void rgbx_to_nv12_neon_16(const uint8_t *src, uint8_t *y, uint8_t *chroma,
  32                 int width, int height,
  33                 int y_stride, int c_stride, int src_stride,
  34                 int32_t coeff_tbl[9]);
  35
  36 static int rgbx_to_nv12_neon_32_wrapper(SwsContext *context, const uint8_t *src[],
  37                         int srcStride[], int srcSliceY, int srcSliceH,
  38                         uint8_t *dst[], int dstStride[]) {
  39
  40     rgbx_to_nv12_neon_32(src[0] + srcSliceY * srcStride[0],
  41             dst[0] + srcSliceY * dstStride[0],
  42             dst[1] + (srcSliceY / 2) * dstStride[1],
  43             context->srcW, srcSliceH,
  44             dstStride[0], dstStride[1], srcStride[0],
  45             context->input_rgb2yuv_table);
  46
  47     return 0;
  48 }
  49
  50 static int rgbx_to_nv12_neon_16_wrapper(SwsContext *context, const uint8_t *src[],
  51                         int srcStride[], int srcSliceY, int srcSliceH,
  52                         uint8_t *dst[], int dstStride[]) {
  53
  54     rgbx_to_nv12_neon_16(src[0] + srcSliceY * srcStride[0],
  55             dst[0] + srcSliceY * dstStride[0],
  56             dst[1] + (srcSliceY / 2) * dstStride[1],
  57             context->srcW, srcSliceH,
  58             dstStride[0], dstStride[1], srcStride[0],
  59             context->input_rgb2yuv_table);
  60
  61     return 0;
  62 }
  63
  64 #define YUV_TO_RGB_TABLE(precision)                                                         \
  65         c->yuv2rgb_v2r_coeff / ((precision) == 16 ? 1 << 7 : 1),                            \
  66         c->yuv2rgb_u2g_coeff / ((precision) == 16 ? 1 << 7 : 1),                            \
  67         c->yuv2rgb_v2g_coeff / ((precision) == 16 ? 1 << 7 : 1),                            \
  68         c->yuv2rgb_u2b_coeff / ((precision) == 16 ? 1 << 7 : 1),                            \
  69
  70 #define DECLARE_FF_YUVX_TO_RGBX_FUNCS(ifmt, ofmt, precision)                                \
  71 int ff_##ifmt##_to_##ofmt##_neon_##precision(int w, int h,                                  \
  72                                  uint8_t *dst, int linesize,                                \
  73                                  const uint8_t *srcY, int linesizeY,                        \
  74                                  const uint8_t *srcU, int linesizeU,                        \
  75                                  const uint8_t *srcV, int linesizeV,                        \
  76                                  const int16_t *table,                                      \
  77                                  int y_offset,                                              \
  78                                  int y_coeff);                                              \
  79                                                                                             \
  80 static int ifmt##_to_##ofmt##_neon_wrapper_##precision(SwsContext *c, const uint8_t *src[], \
  81                                            int srcStride[], int srcSliceY, int srcSliceH,   \
  82                                            uint8_t *dst[], int dstStride[]) {               \
  83     const int16_t yuv2rgb_table[] = { YUV_TO_RGB_TABLE(precision) };                        \
  84                                                                                             \
  85     ff_##ifmt##_to_##ofmt##_neon_##precision(c->srcW, srcSliceH,                            \
  86                                  dst[0] + srcSliceY * dstStride[0], dstStride[0],           \
  87                                  src[0], srcStride[0],                                      \
  88                                  src[1], srcStride[1],                                      \
  89                                  src[2], srcStride[2],                                      \
  90                                  yuv2rgb_table,                                             \
  91                                  c->yuv2rgb_y_offset >> 9,                                  \
  92                                  c->yuv2rgb_y_coeff / ((precision) == 16 ? 1 << 7 : 1));    \
  93                                                                                             \
  94     return 0;                                                                               \
  95 }                                                                                           \
  96
  97 #define DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuvx, precision)                                  \
  98 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, argb, precision)                                        \
  99 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, rgba, precision)                                        \
 100 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, abgr, precision)                                        \
 101 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, bgra, precision)                                        \
 102
 103 #define DECLARE_FF_YUVX_TO_ALL_RGBX_ALL_PRECISION_FUNCS(yuvx)                               \
 104 DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuvx, 16)                                                 \
 105
 106 DECLARE_FF_YUVX_TO_ALL_RGBX_ALL_PRECISION_FUNCS(yuv420p)
 107 DECLARE_FF_YUVX_TO_ALL_RGBX_ALL_PRECISION_FUNCS(yuv422p)
 108
 109 #define DECLARE_FF_NVX_TO_RGBX_FUNCS(ifmt, ofmt, precision)                                 \
 110 int ff_##ifmt##_to_##ofmt##_neon_##precision(int w, int h,                                  \
 111                                  uint8_t *dst, int linesize,                                \
 112                                  const uint8_t *srcY, int linesizeY,                        \
 113                                  const uint8_t *srcC, int linesizeC,                        \
 114                                  const int16_t *table,                                      \
 115                                  int y_offset,                                              \
 116                                  int y_coeff);                                              \
 117                                                                                             \
 118 static int ifmt##_to_##ofmt##_neon_wrapper_##precision(SwsContext *c, const uint8_t *src[], \
 119                                            int srcStride[], int srcSliceY, int srcSliceH,   \
 120                                            uint8_t *dst[], int dstStride[]) {               \
 121     const int16_t yuv2rgb_table[] = { YUV_TO_RGB_TABLE(precision) };                        \
 122                                                                                             \
 123     ff_##ifmt##_to_##ofmt##_neon_##precision(c->srcW, srcSliceH,                            \
 124                                  dst[0] + srcSliceY * dstStride[0], dstStride[0],           \
 125                                  src[0], srcStride[0], src[1], srcStride[1],                \
 126                                  yuv2rgb_table,                                             \
 127                                  c->yuv2rgb_y_offset >> 9,                                  \
 128                                  c->yuv2rgb_y_coeff / ((precision) == 16 ? 1 << 7 : 1));    \
 129                                                                                             \
 130     return 0;                                                                               \
 131 }                                                                                           \
 132
 133 #define DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nvx, precision)                                    \
 134 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, argb, precision)                                          \
 135 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, rgba, precision)                                          \
 136 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, abgr, precision)                                          \
 137 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, bgra, precision)                                          \
 138
 139 #define DECLARE_FF_NVX_TO_ALL_RGBX_ALL_PRECISION_FUNCS(nvx)                                 \
 140 DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nvx, 16)                                                   \
 141
 142 DECLARE_FF_NVX_TO_ALL_RGBX_ALL_PRECISION_FUNCS(nv12)
 143 DECLARE_FF_NVX_TO_ALL_RGBX_ALL_PRECISION_FUNCS(nv21)
 144
 145 /* We need a 16 pixel width alignment. This constraint can easily be removed
 146  * for input reading but for the output which is 4-bytes per pixel (RGBA) the
 147  * assembly might be writing as much as 4*15=60 extra bytes at the end of the
 148  * line, which won't fit the 32-bytes buffer alignment. */
 149 #define SET_FF_NVX_TO_RGBX_FUNC(ifmt, IFMT, ofmt, OFMT, accurate_rnd) do {                  \
 150     if (c->srcFormat == AV_PIX_FMT_##IFMT                                                   \
 151         && c->dstFormat == AV_PIX_FMT_##OFMT                                                \
 152         && !(c->srcH & 1)                                                                   \
 153         && !(c->srcW & 15)                                                                  \
 154         && !accurate_rnd) {                                                                 \
 155         c->swscale = ifmt##_to_##ofmt##_neon_wrapper_16;                                    \
 156     }                                                                                       \
 157 } while (0)
 158
 159 #define SET_FF_NVX_TO_ALL_RGBX_FUNC(nvx, NVX, accurate_rnd) do {                            \
 160     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, argb, ARGB, accurate_rnd);                            \
 161     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, rgba, RGBA, accurate_rnd);                            \
 162     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, abgr, ABGR, accurate_rnd);                            \
 163     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, bgra, BGRA, accurate_rnd);                            \
 164 } while (0)
 165
 166 static void get_unscaled_swscale_neon(SwsContext *c) {
 167     int accurate_rnd = c->flags & SWS_ACCURATE_RND;
 168     if (c->srcFormat == AV_PIX_FMT_RGBA
 169             && c->dstFormat == AV_PIX_FMT_NV12
 170             && (c->srcW >= 16)) {
 171         c->swscale = accurate_rnd ? rgbx_to_nv12_neon_32_wrapper
 172                         : rgbx_to_nv12_neon_16_wrapper;
 173     }
 174
 175     SET_FF_NVX_TO_ALL_RGBX_FUNC(nv12, NV12, accurate_rnd);
 176     SET_FF_NVX_TO_ALL_RGBX_FUNC(nv21, NV21, accurate_rnd);
 177     SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv420p, YUV420P, accurate_rnd);
 178     SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv422p, YUV422P, accurate_rnd);
 179 }
 180
 181 void ff_get_unscaled_swscale_arm(SwsContext *c)
 182 {
 183     int cpu_flags = av_get_cpu_flags();
 184     if (have_neon(cpu_flags))
 185         get_unscaled_swscale_neon(c);
 186 }