audio: rewrite filtering glue code

Use the new filtering code for audio too.

1 files changed, 717 insertions, 0 deletions

diff --git a/filters/f_swresample.c b/filters/f_swresample.c
new file mode 100644
index 0000000000..48bd08d847
--- /dev/null
+++ b/filters/f_swresample.c
@@ -0,0 +1,717 @@
+/*
+ * This file is part of mpv.
+ *
+ * mpv is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * mpv is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with mpv.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <libavutil/opt.h>
+#include <libavutil/common.h>
+#include <libavutil/samplefmt.h>
+#include <libavutil/channel_layout.h>
+#include <libavutil/mathematics.h>
+
+#include "config.h"
+
+#include "audio/aframe.h"
+#include "audio/fmt-conversion.h"
+#include "audio/format.h"
+#include "common/common.h"
+#include "common/av_common.h"
+#include "common/msg.h"
+#include "options/m_config.h"
+#include "options/m_option.h"
+
+#include "f_swresample.h"
+#include "filter_internal.h"
+
+#define HAVE_LIBSWRESAMPLE (!HAVE_LIBAV)
+#define HAVE_LIBAVRESAMPLE HAVE_LIBAV
+
+#if HAVE_LIBAVRESAMPLE
+#include <libavresample/avresample.h>
+#elif HAVE_LIBSWRESAMPLE
+#include <libswresample/swresample.h>
+#define AVAudioResampleContext SwrContext
+#define avresample_alloc_context swr_alloc
+#define avresample_open swr_init
+#define avresample_close(x) do { } while(0)
+#define avresample_free swr_free
+#define avresample_available(x) 0
+#define avresample_convert(ctx, out, out_planesize, out_samples, in, in_planesize, in_samples) \
+    swr_convert(ctx, out, out_samples, (const uint8_t**)(in), in_samples)
+#define avresample_set_channel_mapping swr_set_channel_mapping
+#define avresample_set_compensation swr_set_compensation
+#else
+#error "config.h broken or no resampler found"
+#endif
+
+struct priv {
+    struct mp_log *log;
+    bool is_resampling;
+    struct AVAudioResampleContext *avrctx;
+    struct mp_aframe *avrctx_fmt; // output format of avrctx
+    struct mp_aframe *pool_fmt; // format used to allocate frames for avrctx output
+    struct mp_aframe *pre_out_fmt; // format before final conversion
+    struct AVAudioResampleContext *avrctx_out; // for output channel reordering
+    struct mp_resample_opts *opts; // opts requested by the user
+    // At least libswresample keeps a pointer around for this:
+    int reorder_in[MP_NUM_CHANNELS];
+    int reorder_out[MP_NUM_CHANNELS];
+    struct mp_aframe_pool *reorder_buffer;
+    struct mp_aframe_pool *out_pool;
+
+    int in_rate_user; // user input sample rate
+    int in_rate;      // actual rate (used by lavr), adjusted for playback speed
+    int in_format;
+    struct mp_chmap in_channels;
+    int out_rate;
+    int out_format;
+    struct mp_chmap out_channels;
+
+    double current_pts;
+
+    double cmd_speed;
+    double speed;
+
+    struct mp_swresample public;
+};
+
+#define OPT_BASE_STRUCT struct mp_resample_opts
+const struct m_sub_options resample_conf = {
+    .opts = (const m_option_t[]) {
+        OPT_INTRANGE("audio-resample-filter-size", filter_size, 0, 0, 32),
+        OPT_INTRANGE("audio-resample-phase-shift", phase_shift, 0, 0, 30),
+        OPT_FLAG("audio-resample-linear", linear, 0),
+        OPT_DOUBLE("audio-resample-cutoff", cutoff, M_OPT_RANGE,
+                   .min = 0, .max = 1),
+        OPT_FLAG("audio-normalize-downmix", normalize, 0),
+        OPT_KEYVALUELIST("audio-swresample-o", avopts, 0),
+        {0}
+    },
+    .size = sizeof(struct mp_resample_opts),
+    .defaults = &(const struct mp_resample_opts)MP_RESAMPLE_OPTS_DEF,
+    .change_flags = UPDATE_AUDIO,
+};
+
+#if HAVE_LIBAVRESAMPLE
+static double get_delay(struct priv *p)
+{
+    return avresample_get_delay(p->avrctx) / (double)p->in_rate +
+           avresample_available(p->avrctx) / (double)p->out_rate;
+}
+static int get_out_samples(struct priv *p, int in_samples)
+{
+    return avresample_get_out_samples(p->avrctx, in_samples);
+}
+#else
+static double get_delay(struct priv *p)
+{
+    int64_t base = p->in_rate * (int64_t)p->out_rate;
+    return swr_get_delay(p->avrctx, base) / (double)base;
+}
+static int get_out_samples(struct priv *p, int in_samples)
+{
+    return swr_get_out_samples(p->avrctx, in_samples);
+}
+#endif
+
+static void close_lavrr(struct priv *p)
+{
+    if (p->avrctx)
+        avresample_close(p->avrctx);
+    avresample_free(&p->avrctx);
+    if (p->avrctx_out)
+        avresample_close(p->avrctx_out);
+    avresample_free(&p->avrctx_out);
+
+    TA_FREEP(&p->pre_out_fmt);
+    TA_FREEP(&p->avrctx_fmt);
+    TA_FREEP(&p->pool_fmt);
+}
+
+static int rate_from_speed(int rate, double speed)
+{
+    return lrint(rate * speed);
+}
+
+static struct mp_chmap fudge_pairs[][2] = {
+    {MP_CHMAP2(BL,  BR),  MP_CHMAP2(SL,  SR)},
+    {MP_CHMAP2(SL,  SR),  MP_CHMAP2(BL,  BR)},
+    {MP_CHMAP2(SDL, SDR), MP_CHMAP2(SL,  SR)},
+    {MP_CHMAP2(SL,  SR),  MP_CHMAP2(SDL, SDR)},
+};
+
+// Modify out_layout and return the new value. The intention is reducing the
+// loss libswresample's rematrixing will cause by exchanging similar, but
+// strictly speaking incompatible channel pairs. For example, 7.1 should be
+// changed to 7.1(wide) without dropping the SL/SR channels. (We still leave
+// it to libswresample to create the remix matrix.)
+static uint64_t fudge_layout_conversion(struct priv *p,
+                                        uint64_t in, uint64_t out)
+{
+    for (int n = 0; n < MP_ARRAY_SIZE(fudge_pairs); n++) {
+        uint64_t a = mp_chmap_to_lavc(&fudge_pairs[n][0]);
+        uint64_t b = mp_chmap_to_lavc(&fudge_pairs[n][1]);
+        if ((in & a) == a && (in & b) == 0 &&
+            (out & a) == 0 && (out & b) == b)
+        {
+            out = (out & ~b) | a;
+
+            MP_VERBOSE(p, "Fudge: %s -> %s\n",
+                       mp_chmap_to_str(&fudge_pairs[n][0]),
+                       mp_chmap_to_str(&fudge_pairs[n][1]));
+        }
+    }
+    return out;
+}
+
+// mp_chmap_get_reorder() performs:
+//  to->speaker[n] = from->speaker[src[n]]
+// but libavresample does:
+//  to->speaker[dst[n]] = from->speaker[n]
+static void transpose_order(int *map, int num)
+{
+    int nmap[MP_NUM_CHANNELS] = {0};
+    for (int n = 0; n < num; n++) {
+        for (int i = 0; i < num; i++) {
+            if (map[n] == i)
+                nmap[i] = n;
+        }
+    }
+    memcpy(map, nmap, sizeof(nmap));
+}
+
+static bool configure_lavrr(struct priv *p, bool verbose)
+{
+    close_lavrr(p);
+
+    p->in_rate = rate_from_speed(p->in_rate_user, p->speed);
+
+    MP_VERBOSE(p, "%dHz %s %s -> %dHz %s %s\n",
+               p->in_rate, mp_chmap_to_str(&p->in_channels),
+               af_fmt_to_str(p->in_format),
+               p->out_rate, mp_chmap_to_str(&p->out_channels),
+               af_fmt_to_str(p->out_format));
+
+    p->avrctx = avresample_alloc_context();
+    p->avrctx_out = avresample_alloc_context();
+    if (!p->avrctx || !p->avrctx_out)
+        goto error;
+
+    enum AVSampleFormat in_samplefmt = af_to_avformat(p->in_format);
+    enum AVSampleFormat out_samplefmt = af_to_avformat(p->out_format);
+    enum AVSampleFormat out_samplefmtp = av_get_planar_sample_fmt(out_samplefmt);
+
+    if (in_samplefmt == AV_SAMPLE_FMT_NONE ||
+        out_samplefmt == AV_SAMPLE_FMT_NONE ||
+        out_samplefmtp == AV_SAMPLE_FMT_NONE)
+    {
+        MP_ERR(p, "unsupported conversion: %s -> %s\n",
+               af_fmt_to_str(p->in_format), af_fmt_to_str(p->out_format));
+        goto error;
+    }
+
+    av_opt_set_int(p->avrctx, "filter_size",        p->opts->filter_size, 0);
+    av_opt_set_int(p->avrctx, "phase_shift",        p->opts->phase_shift, 0);
+    av_opt_set_int(p->avrctx, "linear_interp",      p->opts->linear, 0);
+
+    double cutoff = p->opts->cutoff;
+    if (cutoff <= 0.0)
+        cutoff = MPMAX(1.0 - 6.5 / (p->opts->filter_size + 8), 0.80);
+    av_opt_set_double(p->avrctx, "cutoff",          cutoff, 0);
+
+    int normalize = p->opts->normalize;
+#if HAVE_LIBSWRESAMPLE
+    av_opt_set_double(p->avrctx, "rematrix_maxval", normalize ? 1 : 1000, 0);
+#else
+    av_opt_set_int(p->avrctx, "normalize_mix_level", !!normalize, 0);
+#endif
+
+    if (mp_set_avopts(p->log, p->avrctx, p->opts->avopts) < 0)
+        goto error;
+
+    struct mp_chmap map_in = p->in_channels;
+    struct mp_chmap map_out = p->out_channels;
+
+    // Try not to do any remixing if at least one is "unknown". Some corner
+    // cases also benefit from disabling all channel handling logic if the
+    // src/dst layouts are the same (like fl-fr-na -> fl-fr-na).
+    if (mp_chmap_is_unknown(&map_in) || mp_chmap_is_unknown(&map_out) ||
+        mp_chmap_equals(&map_in, &map_out))
+    {
+        mp_chmap_set_unknown(&map_in, map_in.num);
+        mp_chmap_set_unknown(&map_out, map_out.num);
+    }
+
+    // unchecked: don't take any channel reordering into account
+    uint64_t in_ch_layout = mp_chmap_to_lavc_unchecked(&map_in);
+    uint64_t out_ch_layout = mp_chmap_to_lavc_unchecked(&map_out);
+
+    struct mp_chmap in_lavc, out_lavc;
+    mp_chmap_from_lavc(&in_lavc, in_ch_layout);
+    mp_chmap_from_lavc(&out_lavc, out_ch_layout);
+
+    if (verbose && !mp_chmap_equals(&in_lavc, &out_lavc)) {
+        MP_VERBOSE(p, "Remix: %s -> %s\n", mp_chmap_to_str(&in_lavc),
+                                            mp_chmap_to_str(&out_lavc));
+    }
+
+    if (in_lavc.num != map_in.num) {
+        // For handling NA channels, we would have to add a planarization step.
+        MP_FATAL(p, "Unsupported input channel layout %s.\n",
+                 mp_chmap_to_str(&map_in));
+        goto error;
+    }
+
+    mp_chmap_get_reorder(p->reorder_in, &map_in, &in_lavc);
+    transpose_order(p->reorder_in, map_in.num);
+
+    if (mp_chmap_equals(&out_lavc, &map_out)) {
+        // No intermediate step required - output new format directly.
+        out_samplefmtp = out_samplefmt;
+    } else {
+        // Verify that we really just reorder and/or insert NA channels.
+        struct mp_chmap withna = out_lavc;
+        mp_chmap_fill_na(&withna, map_out.num);
+        if (withna.num != map_out.num)
+            goto error;
+    }
+    mp_chmap_get_reorder(p->reorder_out, &out_lavc, &map_out);
+
+    p->pre_out_fmt = mp_aframe_create();
+    mp_aframe_set_rate(p->pre_out_fmt, p->out_rate);
+    mp_aframe_set_chmap(p->pre_out_fmt, &p->out_channels);
+    mp_aframe_set_format(p->pre_out_fmt, p->out_format);
+
+    p->avrctx_fmt = mp_aframe_create();
+    mp_aframe_config_copy(p->avrctx_fmt, p->pre_out_fmt);
+    mp_aframe_set_chmap(p->avrctx_fmt, &out_lavc);
+    mp_aframe_set_format(p->avrctx_fmt, af_from_avformat(out_samplefmtp));
+
+    // If there are NA channels, the final output will have more channels than
+    // the avrctx output. Also, avrctx will output planar (out_samplefmtp was
+    // not overwritten). Allocate the output frame with more channels, so the
+    // NA channels can be trivially added.
+    p->pool_fmt = mp_aframe_create();
+    mp_aframe_config_copy(p->pool_fmt, p->avrctx_fmt);
+    if (map_out.num > out_lavc.num)
+        mp_aframe_set_chmap(p->pool_fmt, &map_out);
+
+    out_ch_layout = fudge_layout_conversion(p, in_ch_layout, out_ch_layout);
+
+    // Real conversion; output is input to avrctx_out.
+    av_opt_set_int(p->avrctx, "in_channel_layout",  in_ch_layout, 0);
+    av_opt_set_int(p->avrctx, "out_channel_layout", out_ch_layout, 0);
+    av_opt_set_int(p->avrctx, "in_sample_rate",     p->in_rate, 0);
+    av_opt_set_int(p->avrctx, "out_sample_rate",    p->out_rate, 0);
+    av_opt_set_int(p->avrctx, "in_sample_fmt",      in_samplefmt, 0);
+    av_opt_set_int(p->avrctx, "out_sample_fmt",     out_samplefmtp, 0);
+
+    // Just needs the correct number of channels for deplanarization.
+    struct mp_chmap fake_chmap;
+    mp_chmap_set_unknown(&fake_chmap, map_out.num);
+    uint64_t fake_out_ch_layout = mp_chmap_to_lavc_unchecked(&fake_chmap);
+    if (!fake_out_ch_layout)
+        goto error;
+    av_opt_set_int(p->avrctx_out, "in_channel_layout",  fake_out_ch_layout, 0);
+    av_opt_set_int(p->avrctx_out, "out_channel_layout", fake_out_ch_layout, 0);
+
+    av_opt_set_int(p->avrctx_out, "in_sample_fmt",      out_samplefmtp, 0);
+    av_opt_set_int(p->avrctx_out, "out_sample_fmt",     out_samplefmt, 0);
+    av_opt_set_int(p->avrctx_out, "in_sample_rate",     p->out_rate, 0);
+    av_opt_set_int(p->avrctx_out, "out_sample_rate",    p->out_rate, 0);
+
+    // API has weird requirements, quoting avresample.h:
+    //  * This function can only be called when the allocated context is not open.
+    //  * Also, the input channel layout must have already been set.
+    avresample_set_channel_mapping(p->avrctx, p->reorder_in);
+
+    p->is_resampling = false;
+
+    if (avresample_open(p->avrctx) < 0 || avresample_open(p->avrctx_out) < 0) {
+        MP_ERR(p, "Cannot open Libavresample context.\n");
+        goto error;
+    }
+    return true;
+
+error:
+    close_lavrr(p);
+    mp_filter_internal_mark_failed(p->public.f);
+    MP_FATAL(p, "libswresample failed to initialize.\n");
+    return false;
+}
+
+static void reset(struct mp_filter *f)
+{
+    struct priv *p = f->priv;
+
+    p->current_pts = MP_NOPTS_VALUE;
+
+    if (!p->avrctx)
+        return;
+#if HAVE_LIBSWRESAMPLE
+    swr_close(p->avrctx);
+    if (swr_init(p->avrctx) < 0)
+        close_lavrr(p);
+#else
+    while (avresample_read(p->avrctx, NULL, 1000) > 0) {}
+#endif
+}
+
+static void extra_output_conversion(struct mp_aframe *mpa)
+{
+    int format = af_fmt_from_planar(mp_aframe_get_format(mpa));
+    int num_planes = mp_aframe_get_planes(mpa);
+    uint8_t **planes = mp_aframe_get_data_rw(mpa);
+    if (!planes)
+        return;
+    for (int p = 0; p < num_planes; p++) {
+        void *ptr = planes[p];
+        int total = mp_aframe_get_total_plane_samples(mpa);
+        if (format == AF_FORMAT_FLOAT) {
+            for (int s = 0; s < total; s++)
+                ((float *)ptr)[s] = av_clipf(((float *)ptr)[s], -1.0f, 1.0f);
+        } else if (format == AF_FORMAT_DOUBLE) {
+            for (int s = 0; s < total; s++)
+                ((double *)ptr)[s] = MPCLAMP(((double *)ptr)[s], -1.0, 1.0);
+        }
+    }
+}
+
+// This relies on the tricky way mpa was allocated.
+static bool reorder_planes(struct mp_aframe *mpa, int *reorder,
+                           struct mp_chmap *newmap)
+{
+    if (!mp_aframe_set_chmap(mpa, newmap))
+        return false;
+
+    int num_planes = newmap->num;
+    uint8_t **planes = mp_aframe_get_data_rw(mpa);
+    uint8_t *old_planes[MP_NUM_CHANNELS];
+    assert(num_planes <= MP_NUM_CHANNELS);
+    for (int n = 0; n < num_planes; n++)
+        old_planes[n] = planes[n];
+
+    int next_na = 0;
+    for (int n = 0; n < num_planes; n++)
+        next_na += newmap->speaker[n] != MP_SPEAKER_ID_NA;
+
+    for (int n = 0; n < num_planes; n++) {
+        int src = reorder[n];
+        assert(src >= -1 && src < num_planes);
+        if (src >= 0) {
+            planes[n] = old_planes[src];
+        } else {
+            assert(next_na < num_planes);
+            planes[n] = old_planes[next_na++];
+            // The NA planes were never written by avrctx, so clear them.
+            af_fill_silence(planes[n],
+                            mp_aframe_get_sstride(mpa) * mp_aframe_get_size(mpa),
+                            mp_aframe_get_format(mpa));
+        }
+    }
+
+    return true;
+}
+
+static int resample_frame(struct AVAudioResampleContext *r,
+                          struct mp_aframe *out, struct mp_aframe *in)
+{
+    // Be aware that the channel layout and count can be different for in and
+    // out frames. In some situations the caller will fix up the frames before
+    // or after conversion. The sample rates can also be different.
+    AVFrame *av_i = in ? mp_aframe_get_raw_avframe(in) : NULL;
+    AVFrame *av_o = out ? mp_aframe_get_raw_avframe(out) : NULL;
+    return avresample_convert(r,
+        av_o ? av_o->extended_data : NULL,
+        av_o ? av_o->linesize[0] : 0,
+        av_o ? av_o->nb_samples : 0,
+        av_i ? av_i->extended_data : NULL,
+        av_i ? av_i->linesize[0] : 0,
+        av_i ? av_i->nb_samples : 0);
+}
+
+static struct mp_frame filter_resample_output(struct priv *p,
+                                              struct mp_aframe *in)
+{
+    struct mp_aframe *out = NULL;
+
+    if (!p->avrctx)
+        goto error;
+
+    int samples = get_out_samples(p, in ? mp_aframe_get_size(in) : 0);
+    out = mp_aframe_create();
+    mp_aframe_config_copy(out, p->pool_fmt);
+    if (mp_aframe_pool_allocate(p->out_pool, out, samples) < 0)
+        goto error;
+
+    int out_samples = 0;
+    if (samples) {
+        out_samples = resample_frame(p->avrctx, out, in);
+        if (out_samples < 0 || out_samples > samples)
+            goto error;
+        mp_aframe_set_size(out, out_samples);
+    }
+
+    struct mp_chmap out_chmap;
+    if (!mp_aframe_get_chmap(p->pool_fmt, &out_chmap))
+        goto error;
+    if (!reorder_planes(out, p->reorder_out, &out_chmap))
+        goto error;
+
+    if (!mp_aframe_config_equals(out, p->pre_out_fmt)) {
+        struct mp_aframe *new = mp_aframe_create();
+        mp_aframe_config_copy(new, p->pre_out_fmt);
+        if (mp_aframe_pool_allocate(p->reorder_buffer, new, out_samples) < 0) {
+            talloc_free(new);
+            goto error;
+        }
+        int got = 0;
+        if (out_samples)
+            got = resample_frame(p->avrctx_out, new, out);
+        talloc_free(out);
+        out = new;
+        if (got != out_samples)
+            goto error;
+    }
+
+    extra_output_conversion(out);
+
+    if (in) {
+        mp_aframe_copy_attributes(out, in);
+        p->current_pts = mp_aframe_end_pts(in);
+    }
+
+    if (out_samples) {
+        if (p->current_pts != MP_NOPTS_VALUE) {
+            double delay = get_delay(p) * mp_aframe_get_speed(out) +
+                           mp_aframe_duration(out);
+            mp_aframe_set_pts(out, p->current_pts - delay);
+            mp_aframe_mul_speed(out, p->speed);
+        }
+    } else {
+        TA_FREEP(&out);
+    }
+
+    return out ? MAKE_FRAME(MP_FRAME_AUDIO, out) : MP_NO_FRAME;
+error:
+    talloc_free(out);
+    MP_ERR(p, "Error on resampling.\n");
+    mp_filter_internal_mark_failed(p->public.f);
+    return MP_NO_FRAME;
+}
+
+static void process(struct mp_filter *f)
+{
+    struct priv *p = f->priv;
+
+    if (!mp_pin_can_transfer_data(f->ppins[1], f->ppins[0]))
+        return;
+
+    p->speed = p->cmd_speed * p->public.speed;
+
+    struct mp_frame frame = mp_pin_out_read(f->ppins[0]);
+
+    struct mp_aframe *input = NULL;
+    if (frame.type == MP_FRAME_AUDIO) {
+        input = frame.data;
+    } else if (frame.type != MP_FRAME_EOF) {
+        MP_ERR(p, "Unsupported frame type.\n");
+        mp_frame_unref(&frame);
+        mp_filter_internal_mark_failed(f);
+        return;
+    }
+
+    if (!input && !p->avrctx) {
+        // Obviously no draining needed.
+        mp_pin_in_write(f->ppins[1], MP_EOF_FRAME);
+        return;
+    }
+
+    if (input) {
+        struct mp_swresample *s = &p->public;
+
+        int in_rate = mp_aframe_get_rate(input);
+        int in_format = mp_aframe_get_format(input);
+        struct mp_chmap in_channels = {0};
+        mp_aframe_get_chmap(input, &in_channels);
+
+        if (!in_rate || !in_format || !in_channels.num) {
+            MP_ERR(p, "Frame with invalid format unsupported\n");
+            mp_frame_unref(&frame);
+            mp_filter_internal_mark_failed(f);
+            return;
+        }
+
+        int out_rate = s->out_rate ? s->out_rate : in_rate;
+        int out_format = s->out_format ? s->out_format : in_format;
+        struct mp_chmap out_channels =
+            s->out_channels.num ? s->out_channels : in_channels;
+
+        if (p->in_rate_user != in_rate ||
+            p->in_format != in_format ||
+            !mp_chmap_equals(&p->in_channels, &in_channels) ||
+            p->out_rate != out_rate ||
+            p->out_format != out_format ||
+            !mp_chmap_equals(&p->out_channels, &out_channels) ||
+            !p->avrctx)
+        {
+            if (p->avrctx) {
+                // drain remaining audio
+                struct mp_frame out = filter_resample_output(p, NULL);
+                if (out.type) {
+                    mp_pin_in_write(f->ppins[1], out);
+                    // continue filtering next time.
+                    mp_pin_out_unread(f->ppins[0], frame);
+                    input = NULL;
+                }
+            }
+
+            MP_VERBOSE(p, "format change, reinitializing resampler\n");
+
+            p->in_rate_user = in_rate;
+            p->in_format = in_format;
+            p->in_channels = in_channels;
+            p->out_rate = out_rate;
+            p->out_format = out_format;
+            p->out_channels = out_channels;
+
+            if (!configure_lavrr(p, true)) {
+                talloc_free(input);
+                return;
+            }
+
+            if (!input) {
+                // continue filtering next time
+                mp_filter_internal_mark_progress(f);
+                return;
+            }
+        }
+    }
+
+    int new_rate = rate_from_speed(p->in_rate_user, p->speed);
+    if (p->avrctx && !(!p->is_resampling && new_rate == p->in_rate)) {
+        AVRational r =
+            av_d2q(p->speed * p->in_rate_user / p->in_rate, INT_MAX / 2);
+        // Essentially, swr/avresample_set_compensation() does 2 things:
+        // - adjust output sample rate by sample_delta/compensation_distance
+        // - reset the adjustment after compensation_distance output samples
+        // Increase the compensation_distance to avoid undesired reset
+        // semantics - we want to keep the ratio for the whole frame we're
+        // feeding it, until the next filter() call.
+        int mult = INT_MAX / 2 / MPMAX(MPMAX(abs(r.num), abs(r.den)), 1);
+        r = (AVRational){ r.num * mult, r.den * mult };
+        if (avresample_set_compensation(p->avrctx, r.den - r.num, r.den) >= 0) {
+            new_rate = p->in_rate;
+            p->is_resampling = true;
+        }
+    }
+
+    bool need_reinit = fabs(new_rate / (double)p->in_rate - 1) > 0.01;
+    if (need_reinit && new_rate != p->in_rate) {
+        // Before reconfiguring, drain the audio that is still buffered
+        // in the resampler.
+        struct mp_frame out = filter_resample_output(p, NULL);
+        bool need_drain = !!out.type;
+        if (need_drain) {
+            mp_pin_in_write(f->ppins[1], out);
+            // Drain; continue filtering next time.
+            mp_pin_out_unread(f->ppins[0], frame);
+        }
+        // Reinitialize resampler.
+        configure_lavrr(p, false);
+        if (need_drain) {
+            mp_filter_internal_mark_progress(f);
+            return;
+        }
+    }
+
+    struct mp_frame out = filter_resample_output(p, input);
+
+    if (input && out.type) {
+        mp_pin_in_write(f->ppins[1], out);
+        mp_pin_out_request_data(f->ppins[0]);
+    } else if (!input && out.type) {
+        mp_pin_in_write(f->ppins[1], out);
+        mp_pin_out_repeat_eof(f->ppins[0]);
+    } else if (!input) {
+        mp_pin_in_write(f->ppins[1], MP_EOF_FRAME);
+    }
+
+    talloc_free(input);
+}
+
+double mp_swresample_get_delay(struct mp_swresample *s)
+{
+    struct priv *p = s->f->priv;
+
+    return get_delay(p);
+}
+
+static bool command(struct mp_filter *f, struct mp_filter_command *cmd)
+{
+    struct priv *p = f->priv;
+
+    if (cmd->type == MP_FILTER_COMMAND_SET_SPEED_RESAMPLE) {
+        p->cmd_speed = cmd->speed;
+        return true;
+    }
+
+    return false;
+}
+
+static void destroy(struct mp_filter *f)
+{
+    struct priv *p = f->priv;
+
+    close_lavrr(p);
+}
+
+static const struct mp_filter_info swresample_filter = {
+    .name = "swresample",
+    .priv_size = sizeof(struct priv),
+    .process = process,
+    .command = command,
+    .reset = reset,
+    .destroy = destroy,
+};
+
+struct mp_swresample *mp_swresample_create(struct mp_filter *parent,
+                                           struct mp_resample_opts *opts)
+{
+    struct mp_filter *f = mp_filter_create(parent, &swresample_filter);
+    if (!f)
+        return NULL;
+
+    mp_filter_add_pin(f, MP_PIN_IN, "in");
+    mp_filter_add_pin(f, MP_PIN_OUT, "out");
+
+    struct priv *p = f->priv;
+    p->public.f = f;
+    p->public.speed = 1.0;
+    p->cmd_speed = 1.0;
+    p->log = f->log;
+
+    if (opts) {
+        p->opts = talloc_dup(p, opts);
+        p->opts->avopts = mp_dup_str_array(p, p->opts->avopts);
+    } else {
+        p->opts = mp_get_config_group(p, f->global, &resample_conf);
+    }
+
+    p->reorder_buffer = mp_aframe_pool_create(p);
+    p->out_pool = mp_aframe_pool_create(p);
+
+    return &p->public;
+}