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 #if HAVE_AS_DN_DIRECTIVE
  27 extern void rgbx_to_nv12_neon_32(const uint8_t *src, uint8_t *y, uint8_t *chroma,
  28                 int width, int height,
  29                 int y_stride, int c_stride, int src_stride,
  30                 int32_t coeff_tbl[9]);
  31
  32 extern void rgbx_to_nv12_neon_16(const uint8_t *src, uint8_t *y, uint8_t *chroma,
  33                 int width, int height,
  34                 int y_stride, int c_stride, int src_stride,
  35                 int32_t coeff_tbl[9]);
  36
  37 static int rgbx_to_nv12_neon_32_wrapper(SwsInternal *context, const uint8_t *const src[],
  38                         const int srcStride[], int srcSliceY, int srcSliceH,
  39                         uint8_t *const dst[], const int dstStride[]) {
  40
  41     rgbx_to_nv12_neon_32(src[0] + srcSliceY * srcStride[0],
  42             dst[0] + srcSliceY * dstStride[0],
  43             dst[1] + (srcSliceY / 2) * dstStride[1],
  44             context->opts.src_w, srcSliceH,
  45             dstStride[0], dstStride[1], srcStride[0],
  46             context->input_rgb2yuv_table);
  47
  48     return 0;
  49 }
  50
  51 static int rgbx_to_nv12_neon_16_wrapper(SwsInternal *context, const uint8_t *const src[],
  52                         const int srcStride[], int srcSliceY, int srcSliceH,
  53                         uint8_t *const dst[], int dstStride[]) {
  54
  55     rgbx_to_nv12_neon_16(src[0] + srcSliceY * srcStride[0],
  56             dst[0] + srcSliceY * dstStride[0],
  57             dst[1] + (srcSliceY / 2) * dstStride[1],
  58             context->opts.src_w, srcSliceH,
  59             dstStride[0], dstStride[1], srcStride[0],
  60             context->input_rgb2yuv_table);
  61
  62     return 0;
  63 }
  64
  65 #define YUV_TO_RGB_TABLE                                                                    \
  66         c->yuv2rgb_v2r_coeff,                                                               \
  67         c->yuv2rgb_u2g_coeff,                                                               \
  68         c->yuv2rgb_v2g_coeff,                                                               \
  69         c->yuv2rgb_u2b_coeff,                                                               \
  70
  71 #define DECLARE_FF_YUVX_TO_RGBX_FUNCS(ifmt, ofmt)                                           \
  72 int ff_##ifmt##_to_##ofmt##_neon(int w, int h,                                              \
  73                                  uint8_t *dst, int linesize,                                \
  74                                  const uint8_t *srcY, int linesizeY,                        \
  75                                  const uint8_t *srcU, int linesizeU,                        \
  76                                  const uint8_t *srcV, int linesizeV,                        \
  77                                  const int16_t *table,                                      \
  78                                  int y_offset,                                              \
  79                                  int y_coeff);                                              \
  80                                                                                             \
  81 static int ifmt##_to_##ofmt##_neon_wrapper(SwsInternal *c, const uint8_t *const src[],      \
  82                                            const int srcStride[], int srcSliceY,            \
  83                                            int srcSliceH, uint8_t *const dst[],             \
  84                                            const int dstStride[]) {                         \
  85     const int16_t yuv2rgb_table[] = { YUV_TO_RGB_TABLE };                                   \
  86                                                                                             \
  87     ff_##ifmt##_to_##ofmt##_neon(c->opts.src_w, srcSliceH,                                  \
  88                                  dst[0] + srcSliceY * dstStride[0], dstStride[0],           \
  89                                  src[0], srcStride[0],                                      \
  90                                  src[1], srcStride[1],                                      \
  91                                  src[2], srcStride[2],                                      \
  92                                  yuv2rgb_table,                                             \
  93                                  c->yuv2rgb_y_offset >> 6,                                  \
  94                                  c->yuv2rgb_y_coeff);                                       \
  95                                                                                             \
  96     return 0;                                                                               \
  97 }                                                                                           \
  98
  99 #define DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuvx)                                             \
 100 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, argb)                                                   \
 101 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, rgba)                                                   \
 102 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, abgr)                                                   \
 103 DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, bgra)                                                   \
 104
 105 DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuv420p)
 106 DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuv422p)
 107
 108 #define DECLARE_FF_NVX_TO_RGBX_FUNCS(ifmt, ofmt)                                            \
 109 int ff_##ifmt##_to_##ofmt##_neon(int w, int h,                                              \
 110                                  uint8_t *dst, int linesize,                                \
 111                                  const uint8_t *srcY, int linesizeY,                        \
 112                                  const uint8_t *srcC, int linesizeC,                        \
 113                                  const int16_t *table,                                      \
 114                                  int y_offset,                                              \
 115                                  int y_coeff);                                              \
 116                                                                                             \
 117 static int ifmt##_to_##ofmt##_neon_wrapper(SwsInternal *c, const uint8_t *const src[],      \
 118                                            const int srcStride[], int srcSliceY,            \
 119                                            int srcSliceH, uint8_t *const dst[],             \
 120                                            const int dstStride[]) {                         \
 121     const int16_t yuv2rgb_table[] = { YUV_TO_RGB_TABLE };                                   \
 122                                                                                             \
 123     ff_##ifmt##_to_##ofmt##_neon(c->opts.src_w, 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 >> 6,                                  \
 128                                  c->yuv2rgb_y_coeff);                                       \
 129                                                                                             \
 130     return 0;                                                                               \
 131 }                                                                                           \
 132
 133 #define DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nvx)                                               \
 134 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, argb)                                                     \
 135 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, rgba)                                                     \
 136 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, abgr)                                                     \
 137 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, bgra)                                                     \
 138
 139 DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv12)
 140 DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv21)
 141
 142 /* We need a 16 pixel width alignment. This constraint can easily be removed
 143  * for input reading but for the output which is 4-bytes per pixel (RGBA) the
 144  * assembly might be writing as much as 4*15=60 extra bytes at the end of the
 145  * line, which won't fit the 32-bytes buffer alignment. */
 146 #define SET_FF_NVX_TO_RGBX_FUNC(ifmt, IFMT, ofmt, OFMT, accurate_rnd) do {                  \
 147     if (c->opts.src_format == AV_PIX_FMT_##IFMT                                             \
 148         && c->opts.dst_format == AV_PIX_FMT_##OFMT                                          \
 149         && !(c->opts.src_h & 1)                                                             \
 150         && !(c->opts.src_w & 15)                                                            \
 151         && !accurate_rnd) {                                                                 \
 152         c->convert_unscaled = ifmt##_to_##ofmt##_neon_wrapper;                              \
 153     }                                                                                       \
 154 } while (0)
 155
 156 #define SET_FF_NVX_TO_ALL_RGBX_FUNC(nvx, NVX, accurate_rnd) do {                            \
 157     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, argb, ARGB, accurate_rnd);                            \
 158     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, rgba, RGBA, accurate_rnd);                            \
 159     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, abgr, ABGR, accurate_rnd);                            \
 160     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, bgra, BGRA, accurate_rnd);                            \
 161 } while (0)
 162
 163 static void get_unscaled_swscale_neon(SwsInternal *c) {
 164     int accurate_rnd = c->opts.flags & SWS_ACCURATE_RND;
 165     if (c->opts.src_format == AV_PIX_FMT_RGBA
 166             && c->opts.dst_format == AV_PIX_FMT_NV12
 167             && (c->opts.src_w >= 16)) {
 168         c->convert_unscaled = accurate_rnd ? rgbx_to_nv12_neon_32_wrapper
 169                                            : rgbx_to_nv12_neon_16_wrapper;
 170     }
 171
 172     SET_FF_NVX_TO_ALL_RGBX_FUNC(nv12, NV12, accurate_rnd);
 173     SET_FF_NVX_TO_ALL_RGBX_FUNC(nv21, NV21, accurate_rnd);
 174     SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv420p, YUV420P, accurate_rnd);
 175     SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv422p, YUV422P, accurate_rnd);
 176 }
 177
 178 void ff_get_unscaled_swscale_arm(SwsInternal *c)
 179 {
 180     int cpu_flags = av_get_cpu_flags();
 181     if (have_neon(cpu_flags))
 182         get_unscaled_swscale_neon(c);
 183 }
 184 #else
 185 void ff_get_unscaled_swscale_arm(SwsInternal *c)
 186 {
 187 }
 188 #endif