blob: 7d4d55bb618d2ae3939c07cff4860c6d8654eb03
1 | /* |
2 | * Intel MediaSDK QSV based HEVC encoder |
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 | |
22 | #include <stdint.h> |
23 | #include <sys/types.h> |
24 | |
25 | #include <mfx/mfxvideo.h> |
26 | |
27 | #include "libavutil/common.h" |
28 | #include "libavutil/opt.h" |
29 | |
30 | #include "avcodec.h" |
31 | #include "bytestream.h" |
32 | #include "get_bits.h" |
33 | #include "hevc.h" |
34 | #include "hevcdec.h" |
35 | #include "h2645_parse.h" |
36 | #include "internal.h" |
37 | #include "qsv.h" |
38 | #include "qsv_internal.h" |
39 | #include "qsvenc.h" |
40 | |
41 | enum LoadPlugin { |
42 | LOAD_PLUGIN_NONE, |
43 | LOAD_PLUGIN_HEVC_SW, |
44 | LOAD_PLUGIN_HEVC_HW, |
45 | }; |
46 | |
47 | typedef struct QSVHEVCEncContext { |
48 | AVClass *class; |
49 | QSVEncContext qsv; |
50 | int load_plugin; |
51 | } QSVHEVCEncContext; |
52 | |
53 | static int generate_fake_vps(QSVEncContext *q, AVCodecContext *avctx) |
54 | { |
55 | GetByteContext gbc; |
56 | PutByteContext pbc; |
57 | |
58 | GetBitContext gb; |
59 | H2645NAL sps_nal = { NULL }; |
60 | HEVCSPS sps = { 0 }; |
61 | HEVCVPS vps = { 0 }; |
62 | uint8_t vps_buf[128], vps_rbsp_buf[128]; |
63 | uint8_t *new_extradata; |
64 | unsigned int sps_id; |
65 | int ret, i, type, vps_size; |
66 | |
67 | if (!avctx->extradata_size) { |
68 | av_log(avctx, AV_LOG_ERROR, "No extradata returned from libmfx\n"); |
69 | return AVERROR_UNKNOWN; |
70 | } |
71 | |
72 | /* parse the SPS */ |
73 | ret = ff_h2645_extract_rbsp(avctx->extradata + 4, avctx->extradata_size - 4, &sps_nal, 1); |
74 | if (ret < 0) { |
75 | av_log(avctx, AV_LOG_ERROR, "Error unescaping the SPS buffer\n"); |
76 | return ret; |
77 | } |
78 | |
79 | ret = init_get_bits8(&gb, sps_nal.data, sps_nal.size); |
80 | if (ret < 0) { |
81 | av_freep(&sps_nal.rbsp_buffer); |
82 | return ret; |
83 | } |
84 | |
85 | get_bits(&gb, 1); |
86 | type = get_bits(&gb, 6); |
87 | if (type != HEVC_NAL_SPS) { |
88 | av_log(avctx, AV_LOG_ERROR, "Unexpected NAL type in the extradata: %d\n", |
89 | type); |
90 | av_freep(&sps_nal.rbsp_buffer); |
91 | return AVERROR_INVALIDDATA; |
92 | } |
93 | get_bits(&gb, 9); |
94 | |
95 | ret = ff_hevc_parse_sps(&sps, &gb, &sps_id, 0, NULL, avctx); |
96 | av_freep(&sps_nal.rbsp_buffer); |
97 | if (ret < 0) { |
98 | av_log(avctx, AV_LOG_ERROR, "Error parsing the SPS\n"); |
99 | return ret; |
100 | } |
101 | |
102 | /* generate the VPS */ |
103 | vps.vps_max_layers = 1; |
104 | vps.vps_max_sub_layers = sps.max_sub_layers; |
105 | memcpy(&vps.ptl, &sps.ptl, sizeof(vps.ptl)); |
106 | vps.vps_sub_layer_ordering_info_present_flag = 1; |
107 | for (i = 0; i < HEVC_MAX_SUB_LAYERS; i++) { |
108 | vps.vps_max_dec_pic_buffering[i] = sps.temporal_layer[i].max_dec_pic_buffering; |
109 | vps.vps_num_reorder_pics[i] = sps.temporal_layer[i].num_reorder_pics; |
110 | vps.vps_max_latency_increase[i] = sps.temporal_layer[i].max_latency_increase; |
111 | } |
112 | |
113 | vps.vps_num_layer_sets = 1; |
114 | vps.vps_timing_info_present_flag = sps.vui.vui_timing_info_present_flag; |
115 | vps.vps_num_units_in_tick = sps.vui.vui_num_units_in_tick; |
116 | vps.vps_time_scale = sps.vui.vui_time_scale; |
117 | vps.vps_poc_proportional_to_timing_flag = sps.vui.vui_poc_proportional_to_timing_flag; |
118 | vps.vps_num_ticks_poc_diff_one = sps.vui.vui_num_ticks_poc_diff_one_minus1 + 1; |
119 | |
120 | /* generate the encoded RBSP form of the VPS */ |
121 | ret = ff_hevc_encode_nal_vps(&vps, sps.vps_id, vps_rbsp_buf, sizeof(vps_rbsp_buf)); |
122 | if (ret < 0) { |
123 | av_log(avctx, AV_LOG_ERROR, "Error writing the VPS\n"); |
124 | return ret; |
125 | } |
126 | |
127 | /* escape and add the startcode */ |
128 | bytestream2_init(&gbc, vps_rbsp_buf, ret); |
129 | bytestream2_init_writer(&pbc, vps_buf, sizeof(vps_buf)); |
130 | |
131 | bytestream2_put_be32(&pbc, 1); // startcode |
132 | bytestream2_put_byte(&pbc, HEVC_NAL_VPS << 1); // NAL |
133 | bytestream2_put_byte(&pbc, 1); // header |
134 | |
135 | while (bytestream2_get_bytes_left(&gbc)) { |
136 | uint32_t b = bytestream2_peek_be24(&gbc); |
137 | if (b <= 3) { |
138 | bytestream2_put_be24(&pbc, 3); |
139 | bytestream2_skip(&gbc, 2); |
140 | } else |
141 | bytestream2_put_byte(&pbc, bytestream2_get_byte(&gbc)); |
142 | } |
143 | |
144 | vps_size = bytestream2_tell_p(&pbc); |
145 | new_extradata = av_mallocz(vps_size + avctx->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE); |
146 | if (!new_extradata) |
147 | return AVERROR(ENOMEM); |
148 | memcpy(new_extradata, vps_buf, vps_size); |
149 | memcpy(new_extradata + vps_size, avctx->extradata, avctx->extradata_size); |
150 | |
151 | av_freep(&avctx->extradata); |
152 | avctx->extradata = new_extradata; |
153 | avctx->extradata_size += vps_size; |
154 | |
155 | return 0; |
156 | } |
157 | |
158 | static av_cold int qsv_enc_init(AVCodecContext *avctx) |
159 | { |
160 | QSVHEVCEncContext *q = avctx->priv_data; |
161 | int ret; |
162 | |
163 | if (q->load_plugin != LOAD_PLUGIN_NONE) { |
164 | static const char *uid_hevcenc_sw = "2fca99749fdb49aeb121a5b63ef568f7"; |
165 | static const char *uid_hevcenc_hw = "6fadc791a0c2eb479ab6dcd5ea9da347"; |
166 | |
167 | if (q->qsv.load_plugins[0]) { |
168 | av_log(avctx, AV_LOG_WARNING, |
169 | "load_plugins is not empty, but load_plugin is not set to 'none'." |
170 | "The load_plugin value will be ignored.\n"); |
171 | } else { |
172 | av_freep(&q->qsv.load_plugins); |
173 | |
174 | if (q->load_plugin == LOAD_PLUGIN_HEVC_SW) |
175 | q->qsv.load_plugins = av_strdup(uid_hevcenc_sw); |
176 | else |
177 | q->qsv.load_plugins = av_strdup(uid_hevcenc_hw); |
178 | |
179 | if (!q->qsv.load_plugins) |
180 | return AVERROR(ENOMEM); |
181 | } |
182 | } |
183 | |
184 | ret = ff_qsv_enc_init(avctx, &q->qsv); |
185 | if (ret < 0) |
186 | return ret; |
187 | |
188 | ret = generate_fake_vps(&q->qsv, avctx); |
189 | if (ret < 0) { |
190 | ff_qsv_enc_close(avctx, &q->qsv); |
191 | return ret; |
192 | } |
193 | |
194 | return 0; |
195 | } |
196 | |
197 | static int qsv_enc_frame(AVCodecContext *avctx, AVPacket *pkt, |
198 | const AVFrame *frame, int *got_packet) |
199 | { |
200 | QSVHEVCEncContext *q = avctx->priv_data; |
201 | |
202 | return ff_qsv_encode(avctx, &q->qsv, pkt, frame, got_packet); |
203 | } |
204 | |
205 | static av_cold int qsv_enc_close(AVCodecContext *avctx) |
206 | { |
207 | QSVHEVCEncContext *q = avctx->priv_data; |
208 | |
209 | return ff_qsv_enc_close(avctx, &q->qsv); |
210 | } |
211 | |
212 | #define OFFSET(x) offsetof(QSVHEVCEncContext, x) |
213 | #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM |
214 | static const AVOption options[] = { |
215 | QSV_COMMON_OPTS |
216 | |
217 | { "load_plugin", "A user plugin to load in an internal session", OFFSET(load_plugin), AV_OPT_TYPE_INT, { .i64 = LOAD_PLUGIN_HEVC_SW }, LOAD_PLUGIN_NONE, LOAD_PLUGIN_HEVC_HW, VE, "load_plugin" }, |
218 | { "none", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = LOAD_PLUGIN_NONE }, 0, 0, VE, "load_plugin" }, |
219 | { "hevc_sw", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = LOAD_PLUGIN_HEVC_SW }, 0, 0, VE, "load_plugin" }, |
220 | { "hevc_hw", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = LOAD_PLUGIN_HEVC_HW }, 0, 0, VE, "load_plugin" }, |
221 | |
222 | { "load_plugins", "A :-separate list of hexadecimal plugin UIDs to load in an internal session", |
223 | OFFSET(qsv.load_plugins), AV_OPT_TYPE_STRING, { .str = "" }, 0, 0, VE }, |
224 | |
225 | { "profile", NULL, OFFSET(qsv.profile), AV_OPT_TYPE_INT, { .i64 = MFX_PROFILE_UNKNOWN }, 0, INT_MAX, VE, "profile" }, |
226 | { "unknown", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_PROFILE_UNKNOWN }, INT_MIN, INT_MAX, VE, "profile" }, |
227 | { "main", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_PROFILE_HEVC_MAIN }, INT_MIN, INT_MAX, VE, "profile" }, |
228 | { "main10", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_PROFILE_HEVC_MAIN10 }, INT_MIN, INT_MAX, VE, "profile" }, |
229 | { "mainsp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_PROFILE_HEVC_MAINSP }, INT_MIN, INT_MAX, VE, "profile" }, |
230 | |
231 | { NULL }, |
232 | }; |
233 | |
234 | static const AVClass class = { |
235 | .class_name = "hevc_qsv encoder", |
236 | .item_name = av_default_item_name, |
237 | .option = options, |
238 | .version = LIBAVUTIL_VERSION_INT, |
239 | }; |
240 | |
241 | static const AVCodecDefault qsv_enc_defaults[] = { |
242 | { "b", "1M" }, |
243 | { "refs", "0" }, |
244 | // same as the x264 default |
245 | { "g", "248" }, |
246 | { "bf", "8" }, |
247 | |
248 | { "flags", "+cgop" }, |
249 | #if FF_API_PRIVATE_OPT |
250 | { "b_strategy", "-1" }, |
251 | #endif |
252 | { NULL }, |
253 | }; |
254 | |
255 | AVCodec ff_hevc_qsv_encoder = { |
256 | .name = "hevc_qsv", |
257 | .long_name = NULL_IF_CONFIG_SMALL("HEVC (Intel Quick Sync Video acceleration)"), |
258 | .priv_data_size = sizeof(QSVHEVCEncContext), |
259 | .type = AVMEDIA_TYPE_VIDEO, |
260 | .id = AV_CODEC_ID_HEVC, |
261 | .init = qsv_enc_init, |
262 | .encode2 = qsv_enc_frame, |
263 | .close = qsv_enc_close, |
264 | .capabilities = AV_CODEC_CAP_DELAY, |
265 | .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_NV12, |
266 | AV_PIX_FMT_P010, |
267 | AV_PIX_FMT_QSV, |
268 | AV_PIX_FMT_NONE }, |
269 | .priv_class = &class, |
270 | .defaults = qsv_enc_defaults, |
271 | .caps_internal = FF_CODEC_CAP_INIT_CLEANUP, |
272 | }; |
273 |