1 /*
2  * V210 decoder
3  *
4  * Copyright (C) 2009 Michael Niedermayer <michaelni@gmx.at>
5  * Copyright (c) 2009 Baptiste Coudurier <baptiste dot coudurier at gmail dot com>
6  *
7  * This file is part of FFmpeg.
8  *
9  * FFmpeg is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU Lesser General Public
11  * License as published by the Free Software Foundation; either
12  * version 2.1 of the License, or (at your option) any later version.
13  *
14  * FFmpeg is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * Lesser General Public License for more details.
18  *
19  * You should have received a copy of the GNU Lesser General Public
20  * License along with FFmpeg; if not, write to the Free Software
21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22  */
23 
24 #include "avcodec.h"
25 #include "codec_internal.h"
26 #include "v210dec.h"
27 #include "v210dec_init.h"
28 #include "libavutil/bswap.h"
29 #include "libavutil/imgutils.h"
30 #include "libavutil/internal.h"
31 #include "libavutil/intreadwrite.h"
32 #include "thread.h"
33 
34 typedef struct ThreadData {
35     AVFrame *frame;
36     uint8_t *buf;
37     int stride;
38 } ThreadData;
39 
decode_init(AVCodecContext *avctx)40 static av_cold int decode_init(AVCodecContext *avctx)
41 {
42     V210DecContext *s = avctx->priv_data;
43 
44     avctx->pix_fmt             = AV_PIX_FMT_YUV422P10;
45     avctx->bits_per_raw_sample = 10;
46 
47     s->thread_count  = av_clip(avctx->thread_count, 1, avctx->height/4);
48     s->aligned_input = 0;
49     ff_v210dec_init(s);
50 
51     return 0;
52 }
53 
decode_row(const uint32_t *src, uint16_t *y, uint16_t *u, uint16_t *v, const int width, void (*unpack_frame)(const uint32_t *src, uint16_t *y, uint16_t *u, uint16_t *v, int width))54 static void decode_row(const uint32_t *src, uint16_t *y, uint16_t *u, uint16_t *v, const int width,
55                        void (*unpack_frame)(const uint32_t *src, uint16_t *y, uint16_t *u, uint16_t *v, int width))
56 {
57     uint32_t val;
58     int w = (FFMAX(0, width - 12) / 12) * 12;
59 
60     unpack_frame(src, y, u, v, w);
61 
62     y += w;
63     u += w >> 1;
64     v += w >> 1;
65     src += (w << 1) / 3;
66 
67     while (w < width - 5) {
68         READ_PIXELS(u, y, v);
69         READ_PIXELS(y, u, y);
70         READ_PIXELS(v, y, u);
71         READ_PIXELS(y, v, y);
72         w += 6;
73     }
74 
75     if (w++ < width) {
76         READ_PIXELS(u, y, v);
77 
78         if (w++ < width) {
79             val  = av_le2ne32(*src++);
80             *y++ =  val & 0x3FF;
81 
82             if (w++ < width) {
83                 *u++ = (val >> 10) & 0x3FF;
84                 *y++ = (val >> 20) & 0x3FF;
85                 val  = av_le2ne32(*src++);
86                 *v++ =  val & 0x3FF;
87 
88                 if (w++ < width) {
89                     *y++ = (val >> 10) & 0x3FF;
90 
91                     if (w++ < width) {
92                         *u++ = (val >> 20) & 0x3FF;
93                         val  = av_le2ne32(*src++);
94                         *y++ =  val & 0x3FF;
95                         *v++ = (val >> 10) & 0x3FF;
96 
97                         if (w++ < width)
98                             *y++ = (val >> 20) & 0x3FF;
99                     }
100                 }
101             }
102         }
103     }
104 }
105 
v210_decode_slice(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)106 static int v210_decode_slice(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
107 {
108     V210DecContext *s = avctx->priv_data;
109     ThreadData *td = arg;
110     AVFrame *frame = td->frame;
111     int stride = td->stride;
112     int slice_start = (avctx->height *  jobnr) / s->thread_count;
113     int slice_end = (avctx->height * (jobnr+1)) / s->thread_count;
114     uint8_t *psrc = td->buf + stride * slice_start;
115     int16_t *py = (uint16_t*)frame->data[0] + slice_start * frame->linesize[0] / 2;
116     int16_t *pu = (uint16_t*)frame->data[1] + slice_start * frame->linesize[1] / 2;
117     int16_t *pv = (uint16_t*)frame->data[2] + slice_start * frame->linesize[2] / 2;
118 
119     for (int h = slice_start; h < slice_end; h++) {
120         decode_row((const uint32_t *)psrc, py, pu, pv, avctx->width, s->unpack_frame);
121         psrc += stride;
122         py += frame->linesize[0] / 2;
123         pu += frame->linesize[1] / 2;
124         pv += frame->linesize[2] / 2;
125     }
126 
127     return 0;
128 }
129 
v210_stride(int width, int align)130 static int v210_stride(int width, int align) {
131     int aligned_width = ((width + align - 1) / align) * align;
132     return aligned_width * 8 / 3;
133 }
134 
decode_frame(AVCodecContext *avctx, AVFrame *pic, int *got_frame, AVPacket *avpkt)135 static int decode_frame(AVCodecContext *avctx, AVFrame *pic,
136                         int *got_frame, AVPacket *avpkt)
137 {
138     V210DecContext *s = avctx->priv_data;
139     ThreadData td;
140     int ret, stride, aligned_input;
141     const uint8_t *psrc = avpkt->data;
142 
143     if (s->custom_stride )
144         stride = s->custom_stride > 0 ? s->custom_stride : 0;
145     else {
146         stride = v210_stride(avctx->width, 48);
147         if (avpkt->size < stride * avctx->height) {
148             int align;
149             for (align = 24; align >= 6; align >>= 1) {
150                 int small_stride = v210_stride(avctx->width, align);
151                 if (avpkt->size == small_stride * avctx->height) {
152                     stride = small_stride;
153                     if (!s->stride_warning_shown)
154                         av_log(avctx, AV_LOG_WARNING, "Broken v210 with too small padding (%d byte) detected\n", align * 8 / 3);
155                     s->stride_warning_shown = 1;
156                     break;
157                 }
158             }
159             if (align < 6 && avctx->codec_tag == MKTAG('b', 'x', 'y', '2'))
160                 stride = 0;
161         }
162     }
163 
164     if (stride == 0 && ((avctx->width & 1) || (int64_t)avctx->width * avctx->height > INT_MAX / 6)) {
165         av_log(avctx, AV_LOG_ERROR, "Strideless v210 is not supported for size %dx%d\n", avctx->width, avctx->height);
166         return AVERROR_INVALIDDATA;
167     }
168 
169     if (stride  > 0 && avpkt->size < (int64_t)stride * avctx->height ||
170         stride == 0 && avpkt->size < v210_stride(avctx->width * avctx->height, 6)) {
171         av_log(avctx, AV_LOG_ERROR, "packet too small\n");
172         return AVERROR_INVALIDDATA;
173     }
174     if (   avctx->codec_tag == MKTAG('C', '2', '1', '0')
175         && avpkt->size > 64
176         && AV_RN32(psrc) == AV_RN32("INFO")
177         && avpkt->size - 64 >= stride * avctx->height)
178         psrc += 64;
179 
180     aligned_input = !((uintptr_t)psrc & 0x1f) && !(stride & 0x1f);
181     if (aligned_input != s->aligned_input) {
182         s->aligned_input = aligned_input;
183         ff_v210dec_init(s);
184     }
185 
186     if ((ret = ff_thread_get_buffer(avctx, pic, 0)) < 0)
187         return ret;
188 
189     pic->pict_type = AV_PICTURE_TYPE_I;
190     pic->key_frame = 1;
191 
192     if (stride) {
193         td.stride = stride;
194         td.buf = (uint8_t*)psrc;
195         td.frame = pic;
196         avctx->execute2(avctx, v210_decode_slice, &td, NULL, s->thread_count);
197     } else {
198         uint8_t *pointers[4];
199         int linesizes[4];
200         int ret = av_image_alloc(pointers, linesizes, avctx->width, avctx->height, avctx->pix_fmt, 1);
201         if (ret < 0)
202             return ret;
203         decode_row((const uint32_t *)psrc, (uint16_t *)pointers[0], (uint16_t *)pointers[1], (uint16_t *)pointers[2], avctx->width * avctx->height, s->unpack_frame);
204         av_image_copy(pic->data, pic->linesize, (const uint8_t **)pointers, linesizes, avctx->pix_fmt, avctx->width, avctx->height);
205         av_freep(&pointers[0]);
206     }
207 
208     if (avctx->field_order > AV_FIELD_PROGRESSIVE) {
209         /* we have interlaced material flagged in container */
210         pic->interlaced_frame = 1;
211         if (avctx->field_order == AV_FIELD_TT || avctx->field_order == AV_FIELD_TB)
212             pic->top_field_first = 1;
213     }
214 
215     *got_frame      = 1;
216 
217     return avpkt->size;
218 }
219 
220 #define V210DEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM
221 static const AVOption v210dec_options[] = {
222     {"custom_stride", "Custom V210 stride", offsetof(V210DecContext, custom_stride), AV_OPT_TYPE_INT,
223      {.i64 = 0}, -1, INT_MAX, V210DEC_FLAGS},
224     {NULL}
225 };
226 
227 static const AVClass v210dec_class = {
228     .class_name = "V210 Decoder",
229     .item_name  = av_default_item_name,
230     .option     = v210dec_options,
231     .version    = LIBAVUTIL_VERSION_INT,
232 };
233 
234 const FFCodec ff_v210_decoder = {
235     .p.name         = "v210",
236     .p.long_name    = NULL_IF_CONFIG_SMALL("Uncompressed 4:2:2 10-bit"),
237     .p.type         = AVMEDIA_TYPE_VIDEO,
238     .p.id           = AV_CODEC_ID_V210,
239     .priv_data_size = sizeof(V210DecContext),
240     .init           = decode_init,
241     FF_CODEC_DECODE_CB(decode_frame),
242     .p.capabilities = AV_CODEC_CAP_DR1 |
243                       AV_CODEC_CAP_SLICE_THREADS |
244                       AV_CODEC_CAP_FRAME_THREADS,
245     .p.priv_class   = &v210dec_class,
246     .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE,
247 };
248