| 1 | /* |
| 2 | * Copyright (c) 2011 Nicolas George <nicolas.george@normalesup.org> |
| 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 |
| 14 | * GNU 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 | * @file |
| 23 | * Audio merging filter |
| 24 | */ |
| 25 | |
| 26 | #include "libavutil/avstring.h" |
| 27 | #include "libavutil/bprint.h" |
| 28 | #include "libavutil/channel_layout.h" |
| 29 | #include "libavutil/opt.h" |
| 30 | #include "avfilter.h" |
| 31 | #include "audio.h" |
| 32 | #include "bufferqueue.h" |
| 33 | #include "internal.h" |
| 34 | |
| 35 | #define SWR_CH_MAX 32 |
| 36 | |
| 37 | typedef struct { |
| 38 | const AVClass *class; |
| 39 | int nb_inputs; |
| 40 | int route[SWR_CH_MAX]; /**< channels routing, see copy_samples */ |
| 41 | int bps; |
| 42 | struct amerge_input { |
| 43 | struct FFBufQueue queue; |
| 44 | int nb_ch; /**< number of channels for the input */ |
| 45 | int nb_samples; |
| 46 | int pos; |
| 47 | } *in; |
| 48 | } AMergeContext; |
| 49 | |
| 50 | #define OFFSET(x) offsetof(AMergeContext, x) |
| 51 | #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM |
| 52 | |
| 53 | static const AVOption amerge_options[] = { |
| 54 | { "inputs", "specify the number of inputs", OFFSET(nb_inputs), |
| 55 | AV_OPT_TYPE_INT, { .i64 = 2 }, 2, SWR_CH_MAX, FLAGS }, |
| 56 | { NULL } |
| 57 | }; |
| 58 | |
| 59 | AVFILTER_DEFINE_CLASS(amerge); |
| 60 | |
| 61 | static av_cold void uninit(AVFilterContext *ctx) |
| 62 | { |
| 63 | AMergeContext *am = ctx->priv; |
| 64 | int i; |
| 65 | |
| 66 | for (i = 0; i < am->nb_inputs; i++) { |
| 67 | if (am->in) |
| 68 | ff_bufqueue_discard_all(&am->in[i].queue); |
| 69 | if (ctx->input_pads) |
| 70 | av_freep(&ctx->input_pads[i].name); |
| 71 | } |
| 72 | av_freep(&am->in); |
| 73 | } |
| 74 | |
| 75 | static int query_formats(AVFilterContext *ctx) |
| 76 | { |
| 77 | AMergeContext *am = ctx->priv; |
| 78 | int64_t inlayout[SWR_CH_MAX], outlayout = 0; |
| 79 | AVFilterFormats *formats; |
| 80 | AVFilterChannelLayouts *layouts; |
| 81 | int i, overlap = 0, nb_ch = 0; |
| 82 | |
| 83 | for (i = 0; i < am->nb_inputs; i++) { |
| 84 | if (!ctx->inputs[i]->in_channel_layouts || |
| 85 | !ctx->inputs[i]->in_channel_layouts->nb_channel_layouts) { |
| 86 | av_log(ctx, AV_LOG_WARNING, |
| 87 | "No channel layout for input %d\n", i + 1); |
| 88 | return AVERROR(EAGAIN); |
| 89 | } |
| 90 | inlayout[i] = ctx->inputs[i]->in_channel_layouts->channel_layouts[0]; |
| 91 | if (ctx->inputs[i]->in_channel_layouts->nb_channel_layouts > 1) { |
| 92 | char buf[256]; |
| 93 | av_get_channel_layout_string(buf, sizeof(buf), 0, inlayout[i]); |
| 94 | av_log(ctx, AV_LOG_INFO, "Using \"%s\" for input %d\n", buf, i + 1); |
| 95 | } |
| 96 | am->in[i].nb_ch = av_get_channel_layout_nb_channels(inlayout[i]); |
| 97 | if (outlayout & inlayout[i]) |
| 98 | overlap++; |
| 99 | outlayout |= inlayout[i]; |
| 100 | nb_ch += am->in[i].nb_ch; |
| 101 | } |
| 102 | if (nb_ch > SWR_CH_MAX) { |
| 103 | av_log(ctx, AV_LOG_ERROR, "Too many channels (max %d)\n", SWR_CH_MAX); |
| 104 | return AVERROR(EINVAL); |
| 105 | } |
| 106 | if (overlap) { |
| 107 | av_log(ctx, AV_LOG_WARNING, |
| 108 | "Input channel layouts overlap: " |
| 109 | "output layout will be determined by the number of distinct input channels\n"); |
| 110 | for (i = 0; i < nb_ch; i++) |
| 111 | am->route[i] = i; |
| 112 | outlayout = av_get_default_channel_layout(nb_ch); |
| 113 | if (!outlayout) |
| 114 | outlayout = ((int64_t)1 << nb_ch) - 1; |
| 115 | } else { |
| 116 | int *route[SWR_CH_MAX]; |
| 117 | int c, out_ch_number = 0; |
| 118 | |
| 119 | route[0] = am->route; |
| 120 | for (i = 1; i < am->nb_inputs; i++) |
| 121 | route[i] = route[i - 1] + am->in[i - 1].nb_ch; |
| 122 | for (c = 0; c < 64; c++) |
| 123 | for (i = 0; i < am->nb_inputs; i++) |
| 124 | if ((inlayout[i] >> c) & 1) |
| 125 | *(route[i]++) = out_ch_number++; |
| 126 | } |
| 127 | formats = ff_make_format_list(ff_packed_sample_fmts_array); |
| 128 | ff_set_common_formats(ctx, formats); |
| 129 | for (i = 0; i < am->nb_inputs; i++) { |
| 130 | layouts = NULL; |
| 131 | ff_add_channel_layout(&layouts, inlayout[i]); |
| 132 | ff_channel_layouts_ref(layouts, &ctx->inputs[i]->out_channel_layouts); |
| 133 | } |
| 134 | layouts = NULL; |
| 135 | ff_add_channel_layout(&layouts, outlayout); |
| 136 | ff_channel_layouts_ref(layouts, &ctx->outputs[0]->in_channel_layouts); |
| 137 | ff_set_common_samplerates(ctx, ff_all_samplerates()); |
| 138 | return 0; |
| 139 | } |
| 140 | |
| 141 | static int config_output(AVFilterLink *outlink) |
| 142 | { |
| 143 | AVFilterContext *ctx = outlink->src; |
| 144 | AMergeContext *am = ctx->priv; |
| 145 | AVBPrint bp; |
| 146 | int i; |
| 147 | |
| 148 | for (i = 1; i < am->nb_inputs; i++) { |
| 149 | if (ctx->inputs[i]->sample_rate != ctx->inputs[0]->sample_rate) { |
| 150 | av_log(ctx, AV_LOG_ERROR, |
| 151 | "Inputs must have the same sample rate " |
| 152 | "%d for in%d vs %d\n", |
| 153 | ctx->inputs[i]->sample_rate, i, ctx->inputs[0]->sample_rate); |
| 154 | return AVERROR(EINVAL); |
| 155 | } |
| 156 | } |
| 157 | am->bps = av_get_bytes_per_sample(ctx->outputs[0]->format); |
| 158 | outlink->sample_rate = ctx->inputs[0]->sample_rate; |
| 159 | outlink->time_base = ctx->inputs[0]->time_base; |
| 160 | |
| 161 | av_bprint_init(&bp, 0, 1); |
| 162 | for (i = 0; i < am->nb_inputs; i++) { |
| 163 | av_bprintf(&bp, "%sin%d:", i ? " + " : "", i); |
| 164 | av_bprint_channel_layout(&bp, -1, ctx->inputs[i]->channel_layout); |
| 165 | } |
| 166 | av_bprintf(&bp, " -> out:"); |
| 167 | av_bprint_channel_layout(&bp, -1, ctx->outputs[0]->channel_layout); |
| 168 | av_log(ctx, AV_LOG_VERBOSE, "%s\n", bp.str); |
| 169 | |
| 170 | return 0; |
| 171 | } |
| 172 | |
| 173 | static int request_frame(AVFilterLink *outlink) |
| 174 | { |
| 175 | AVFilterContext *ctx = outlink->src; |
| 176 | AMergeContext *am = ctx->priv; |
| 177 | int i, ret; |
| 178 | |
| 179 | for (i = 0; i < am->nb_inputs; i++) |
| 180 | if (!am->in[i].nb_samples) |
| 181 | if ((ret = ff_request_frame(ctx->inputs[i])) < 0) |
| 182 | return ret; |
| 183 | return 0; |
| 184 | } |
| 185 | |
| 186 | /** |
| 187 | * Copy samples from several input streams to one output stream. |
| 188 | * @param nb_inputs number of inputs |
| 189 | * @param in inputs; used only for the nb_ch field; |
| 190 | * @param route routing values; |
| 191 | * input channel i goes to output channel route[i]; |
| 192 | * i < in[0].nb_ch are the channels from the first output; |
| 193 | * i >= in[0].nb_ch are the channels from the second output |
| 194 | * @param ins pointer to the samples of each inputs, in packed format; |
| 195 | * will be left at the end of the copied samples |
| 196 | * @param outs pointer to the samples of the output, in packet format; |
| 197 | * must point to a buffer big enough; |
| 198 | * will be left at the end of the copied samples |
| 199 | * @param ns number of samples to copy |
| 200 | * @param bps bytes per sample |
| 201 | */ |
| 202 | static inline void copy_samples(int nb_inputs, struct amerge_input in[], |
| 203 | int *route, uint8_t *ins[], |
| 204 | uint8_t **outs, int ns, int bps) |
| 205 | { |
| 206 | int *route_cur; |
| 207 | int i, c, nb_ch = 0; |
| 208 | |
| 209 | for (i = 0; i < nb_inputs; i++) |
| 210 | nb_ch += in[i].nb_ch; |
| 211 | while (ns--) { |
| 212 | route_cur = route; |
| 213 | for (i = 0; i < nb_inputs; i++) { |
| 214 | for (c = 0; c < in[i].nb_ch; c++) { |
| 215 | memcpy((*outs) + bps * *(route_cur++), ins[i], bps); |
| 216 | ins[i] += bps; |
| 217 | } |
| 218 | } |
| 219 | *outs += nb_ch * bps; |
| 220 | } |
| 221 | } |
| 222 | |
| 223 | static int filter_frame(AVFilterLink *inlink, AVFrame *insamples) |
| 224 | { |
| 225 | AVFilterContext *ctx = inlink->dst; |
| 226 | AMergeContext *am = ctx->priv; |
| 227 | AVFilterLink *const outlink = ctx->outputs[0]; |
| 228 | int input_number; |
| 229 | int nb_samples, ns, i; |
| 230 | AVFrame *outbuf, *inbuf[SWR_CH_MAX]; |
| 231 | uint8_t *ins[SWR_CH_MAX], *outs; |
| 232 | |
| 233 | for (input_number = 0; input_number < am->nb_inputs; input_number++) |
| 234 | if (inlink == ctx->inputs[input_number]) |
| 235 | break; |
| 236 | av_assert1(input_number < am->nb_inputs); |
| 237 | if (ff_bufqueue_is_full(&am->in[input_number].queue)) { |
| 238 | av_frame_free(&insamples); |
| 239 | return AVERROR(ENOMEM); |
| 240 | } |
| 241 | ff_bufqueue_add(ctx, &am->in[input_number].queue, av_frame_clone(insamples)); |
| 242 | am->in[input_number].nb_samples += insamples->nb_samples; |
| 243 | av_frame_free(&insamples); |
| 244 | nb_samples = am->in[0].nb_samples; |
| 245 | for (i = 1; i < am->nb_inputs; i++) |
| 246 | nb_samples = FFMIN(nb_samples, am->in[i].nb_samples); |
| 247 | if (!nb_samples) |
| 248 | return 0; |
| 249 | |
| 250 | outbuf = ff_get_audio_buffer(ctx->outputs[0], nb_samples); |
| 251 | if (!outbuf) |
| 252 | return AVERROR(ENOMEM); |
| 253 | outs = outbuf->data[0]; |
| 254 | for (i = 0; i < am->nb_inputs; i++) { |
| 255 | inbuf[i] = ff_bufqueue_peek(&am->in[i].queue, 0); |
| 256 | ins[i] = inbuf[i]->data[0] + |
| 257 | am->in[i].pos * am->in[i].nb_ch * am->bps; |
| 258 | } |
| 259 | av_frame_copy_props(outbuf, inbuf[0]); |
| 260 | outbuf->pts = inbuf[0]->pts == AV_NOPTS_VALUE ? AV_NOPTS_VALUE : |
| 261 | inbuf[0]->pts + |
| 262 | av_rescale_q(am->in[0].pos, |
| 263 | av_make_q(1, ctx->inputs[0]->sample_rate), |
| 264 | ctx->outputs[0]->time_base); |
| 265 | |
| 266 | outbuf->nb_samples = nb_samples; |
| 267 | outbuf->channel_layout = outlink->channel_layout; |
| 268 | av_frame_set_channels(outbuf, outlink->channels); |
| 269 | |
| 270 | while (nb_samples) { |
| 271 | ns = nb_samples; |
| 272 | for (i = 0; i < am->nb_inputs; i++) |
| 273 | ns = FFMIN(ns, inbuf[i]->nb_samples - am->in[i].pos); |
| 274 | /* Unroll the most common sample formats: speed +~350% for the loop, |
| 275 | +~13% overall (including two common decoders) */ |
| 276 | switch (am->bps) { |
| 277 | case 1: |
| 278 | copy_samples(am->nb_inputs, am->in, am->route, ins, &outs, ns, 1); |
| 279 | break; |
| 280 | case 2: |
| 281 | copy_samples(am->nb_inputs, am->in, am->route, ins, &outs, ns, 2); |
| 282 | break; |
| 283 | case 4: |
| 284 | copy_samples(am->nb_inputs, am->in, am->route, ins, &outs, ns, 4); |
| 285 | break; |
| 286 | default: |
| 287 | copy_samples(am->nb_inputs, am->in, am->route, ins, &outs, ns, am->bps); |
| 288 | break; |
| 289 | } |
| 290 | |
| 291 | nb_samples -= ns; |
| 292 | for (i = 0; i < am->nb_inputs; i++) { |
| 293 | am->in[i].nb_samples -= ns; |
| 294 | am->in[i].pos += ns; |
| 295 | if (am->in[i].pos == inbuf[i]->nb_samples) { |
| 296 | am->in[i].pos = 0; |
| 297 | av_frame_free(&inbuf[i]); |
| 298 | ff_bufqueue_get(&am->in[i].queue); |
| 299 | inbuf[i] = ff_bufqueue_peek(&am->in[i].queue, 0); |
| 300 | ins[i] = inbuf[i] ? inbuf[i]->data[0] : NULL; |
| 301 | } |
| 302 | } |
| 303 | } |
| 304 | return ff_filter_frame(ctx->outputs[0], outbuf); |
| 305 | } |
| 306 | |
| 307 | static av_cold int init(AVFilterContext *ctx) |
| 308 | { |
| 309 | AMergeContext *am = ctx->priv; |
| 310 | int i; |
| 311 | |
| 312 | am->in = av_calloc(am->nb_inputs, sizeof(*am->in)); |
| 313 | if (!am->in) |
| 314 | return AVERROR(ENOMEM); |
| 315 | for (i = 0; i < am->nb_inputs; i++) { |
| 316 | char *name = av_asprintf("in%d", i); |
| 317 | AVFilterPad pad = { |
| 318 | .name = name, |
| 319 | .type = AVMEDIA_TYPE_AUDIO, |
| 320 | .filter_frame = filter_frame, |
| 321 | }; |
| 322 | if (!name) |
| 323 | return AVERROR(ENOMEM); |
| 324 | ff_insert_inpad(ctx, i, &pad); |
| 325 | } |
| 326 | return 0; |
| 327 | } |
| 328 | |
| 329 | static const AVFilterPad amerge_outputs[] = { |
| 330 | { |
| 331 | .name = "default", |
| 332 | .type = AVMEDIA_TYPE_AUDIO, |
| 333 | .config_props = config_output, |
| 334 | .request_frame = request_frame, |
| 335 | }, |
| 336 | { NULL } |
| 337 | }; |
| 338 | |
| 339 | AVFilter ff_af_amerge = { |
| 340 | .name = "amerge", |
| 341 | .description = NULL_IF_CONFIG_SMALL("Merge two or more audio streams into " |
| 342 | "a single multi-channel stream."), |
| 343 | .priv_size = sizeof(AMergeContext), |
| 344 | .init = init, |
| 345 | .uninit = uninit, |
| 346 | .query_formats = query_formats, |
| 347 | .inputs = NULL, |
| 348 | .outputs = amerge_outputs, |
| 349 | .priv_class = &amerge_class, |
| 350 | .flags = AVFILTER_FLAG_DYNAMIC_INPUTS, |
| 351 | }; |