swscale/arm: add ff_nv{12,21}_to_{argb,rgba,abgr,bgra}_neon
[ffmpeg.git] / 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 0
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(SwsContext *context, const uint8_t *src[],
38                         int srcStride[], int srcSliceY, int srcSliceH,
39                         uint8_t *dst[], 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->srcW, 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(SwsContext *context, const uint8_t *src[],
52                         int srcStride[], int srcSliceY, int srcSliceH,
53                         uint8_t *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->srcW, srcSliceH,
59             dstStride[0], dstStride[1], srcStride[0],
60             context->input_rgb2yuv_table);
61
62     return 0;
63 }
64 #endif
65
66 #define DECLARE_FF_NVX_TO_RGBX_FUNCS(ifmt, ofmt)                                            \
67 int ff_##ifmt##_to_##ofmt##_neon(int w, int h,                                              \
68                                  uint8_t *dst, int linesize,                                \
69                                  const uint8_t *srcY, int linesizeY,                        \
70                                  const uint8_t *srcC, int linesizeC,                        \
71                                  const int16_t *table,                                      \
72                                  int y_offset,                                              \
73                                  int y_coeff);                                              \
74                                                                                             \
75 static int ifmt##_to_##ofmt##_neon_wrapper(SwsContext *c, const uint8_t *src[],             \
76                                            int srcStride[], int srcSliceY, int srcSliceH,   \
77                                            uint8_t *dst[], int dstStride[]) {               \
78     const int16_t yuv2rgb_table[] = {                                                       \
79         c->yuv2rgb_v2r_coeff,                                                               \
80         c->yuv2rgb_u2g_coeff,                                                               \
81         c->yuv2rgb_v2g_coeff,                                                               \
82         c->yuv2rgb_u2b_coeff,                                                               \
83     };                                                                                      \
84                                                                                             \
85     ff_##ifmt##_to_##ofmt##_neon(c->srcW, srcSliceH,                                        \
86                                  dst[0] +  srcSliceY      * dstStride[0], dstStride[0],     \
87                                  src[0] +  srcSliceY      * srcStride[0], srcStride[0],     \
88                                  src[1] + (srcSliceY / 2) * srcStride[1], srcStride[1],     \
89                                  yuv2rgb_table,                                             \
90                                  c->yuv2rgb_y_offset >> 9,                                  \
91                                  c->yuv2rgb_y_coeff);                                       \
92                                                                                             \
93     return 0;                                                                               \
94 }
95
96 #define DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nvx)                                               \
97 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, argb)                                                     \
98 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, rgba)                                                     \
99 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, abgr)                                                     \
100 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, bgra)                                                     \
101
102 DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv12)
103 DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv21)
104
105 /* We need a 16 pixel width alignment. This constraint can easily be removed
106  * for input reading but for the output which is 4-bytes per pixel (RGBA) the
107  * assembly might be writing as much as 4*15=60 extra bytes at the end of the
108  * line, which won't fit the 32-bytes buffer alignment. */
109 #define SET_FF_NVX_TO_RGBX_FUNC(ifmt, IFMT, ofmt, OFMT) do {                                \
110     if (c->srcFormat == AV_PIX_FMT_##IFMT                                                   \
111         && c->dstFormat == AV_PIX_FMT_##OFMT                                                \
112         && !(c->srcH & 1)                                                                   \
113         && !(c->srcW & 15)) {                                                               \
114         c->swscale = ifmt##_to_##ofmt##_neon_wrapper;                                       \
115     }                                                                                       \
116 } while (0)
117
118 #define SET_FF_NVX_TO_ALL_RGBX_FUNC(nvx, NVX) do {                                          \
119     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, argb, ARGB);                                          \
120     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, rgba, RGBA);                                          \
121     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, abgr, ABGR);                                          \
122     SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, bgra, BGRA);                                          \
123 } while (0)
124
125 static void get_unscaled_swscale_neon(SwsContext *c) {
126 #if 0
127     int accurate_rnd = c->flags & SWS_ACCURATE_RND;
128     if (c->srcFormat == AV_PIX_FMT_RGBA
129             && c->dstFormat == AV_PIX_FMT_NV12
130             && (c->srcW >= 16)) {
131         c->swscale = accurate_rnd ? rgbx_to_nv12_neon_32_wrapper
132                         : rgbx_to_nv12_neon_16_wrapper;
133     }
134 #endif
135
136     SET_FF_NVX_TO_ALL_RGBX_FUNC(nv12, NV12);
137     SET_FF_NVX_TO_ALL_RGBX_FUNC(nv21, NV21);
138 }
139
140 void ff_get_unscaled_swscale_arm(SwsContext *c)
141 {
142     int cpu_flags = av_get_cpu_flags();
143     if (have_neon(cpu_flags))
144         get_unscaled_swscale_neon(c);
145 }