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
27extern 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
32extern 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
37static 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
51static 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
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)                                           \
72int 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                                                                                            \
81static int ifmt##_to_##ofmt##_neon_wrapper(SwsContext *c, const uint8_t *src[],             \
82                                           int srcStride[], int srcSliceY, int srcSliceH,   \
83                                           uint8_t *dst[], int dstStride[]) {               \
84    const int16_t yuv2rgb_table[] = { YUV_TO_RGB_TABLE };                                   \
85                                                                                            \
86    ff_##ifmt##_to_##ofmt##_neon(c->srcW, srcSliceH,                                        \
87                                 dst[0] + srcSliceY * dstStride[0], dstStride[0],           \
88                                 src[0], srcStride[0],                                      \
89                                 src[1], srcStride[1],                                      \
90                                 src[2], srcStride[2],                                      \
91                                 yuv2rgb_table,                                             \
92                                 c->yuv2rgb_y_offset >> 6,                                  \
93                                 c->yuv2rgb_y_coeff);                                       \
94                                                                                            \
95    return 0;                                                                               \
96}                                                                                           \
97
98#define DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuvx)                                             \
99DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, argb)                                                   \
100DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, rgba)                                                   \
101DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, abgr)                                                   \
102DECLARE_FF_YUVX_TO_RGBX_FUNCS(yuvx, bgra)                                                   \
103
104DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuv420p)
105DECLARE_FF_YUVX_TO_ALL_RGBX_FUNCS(yuv422p)
106
107#define DECLARE_FF_NVX_TO_RGBX_FUNCS(ifmt, ofmt)                                            \
108int ff_##ifmt##_to_##ofmt##_neon(int w, int h,                                              \
109                                 uint8_t *dst, int linesize,                                \
110                                 const uint8_t *srcY, int linesizeY,                        \
111                                 const uint8_t *srcC, int linesizeC,                        \
112                                 const int16_t *table,                                      \
113                                 int y_offset,                                              \
114                                 int y_coeff);                                              \
115                                                                                            \
116static int ifmt##_to_##ofmt##_neon_wrapper(SwsContext *c, const uint8_t *src[],             \
117                                           int srcStride[], int srcSliceY, int srcSliceH,   \
118                                           uint8_t *dst[], int dstStride[]) {               \
119    const int16_t yuv2rgb_table[] = { YUV_TO_RGB_TABLE };                                   \
120                                                                                            \
121    ff_##ifmt##_to_##ofmt##_neon(c->srcW, srcSliceH,                                        \
122                                 dst[0] + srcSliceY * dstStride[0], dstStride[0],           \
123                                 src[0], srcStride[0], src[1], srcStride[1],                \
124                                 yuv2rgb_table,                                             \
125                                 c->yuv2rgb_y_offset >> 6,                                  \
126                                 c->yuv2rgb_y_coeff);                                       \
127                                                                                            \
128    return 0;                                                                               \
129}                                                                                           \
130
131#define DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nvx)                                               \
132DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, argb)                                                     \
133DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, rgba)                                                     \
134DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, abgr)                                                     \
135DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, bgra)                                                     \
136
137DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv12)
138DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv21)
139
140/* We need a 16 pixel width alignment. This constraint can easily be removed
141 * for input reading but for the output which is 4-bytes per pixel (RGBA) the
142 * assembly might be writing as much as 4*15=60 extra bytes at the end of the
143 * line, which won't fit the 32-bytes buffer alignment. */
144#define SET_FF_NVX_TO_RGBX_FUNC(ifmt, IFMT, ofmt, OFMT, accurate_rnd) do {                  \
145    if (c->srcFormat == AV_PIX_FMT_##IFMT                                                   \
146        && c->dstFormat == AV_PIX_FMT_##OFMT                                                \
147        && !(c->srcH & 1)                                                                   \
148        && !(c->srcW & 15)                                                                  \
149        && !accurate_rnd) {                                                                 \
150        c->convert_unscaled = ifmt##_to_##ofmt##_neon_wrapper;                              \
151    }                                                                                       \
152} while (0)
153
154#define SET_FF_NVX_TO_ALL_RGBX_FUNC(nvx, NVX, accurate_rnd) do {                            \
155    SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, argb, ARGB, accurate_rnd);                            \
156    SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, rgba, RGBA, accurate_rnd);                            \
157    SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, abgr, ABGR, accurate_rnd);                            \
158    SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, bgra, BGRA, accurate_rnd);                            \
159} while (0)
160
161static void get_unscaled_swscale_neon(SwsContext *c) {
162    int accurate_rnd = c->flags & SWS_ACCURATE_RND;
163    if (c->srcFormat == AV_PIX_FMT_RGBA
164            && c->dstFormat == AV_PIX_FMT_NV12
165            && (c->srcW >= 16)) {
166        c->convert_unscaled = accurate_rnd ? rgbx_to_nv12_neon_32_wrapper
167                                           : rgbx_to_nv12_neon_16_wrapper;
168    }
169
170    SET_FF_NVX_TO_ALL_RGBX_FUNC(nv12, NV12, accurate_rnd);
171    SET_FF_NVX_TO_ALL_RGBX_FUNC(nv21, NV21, accurate_rnd);
172    SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv420p, YUV420P, accurate_rnd);
173    SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv422p, YUV422P, accurate_rnd);
174}
175
176void ff_get_unscaled_swscale_arm(SwsContext *c)
177{
178    int cpu_flags = av_get_cpu_flags();
179    if (have_neon(cpu_flags))
180        get_unscaled_swscale_neon(c);
181}
182#else
183void ff_get_unscaled_swscale_arm(SwsContext *c)
184{
185}
186#endif
187