]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/libdav1d.c
avutil/frame: Remove deprecated AVFrame.pkt_pts field
[ffmpeg] / libavcodec / libdav1d.c
index 8c8584f4e804ef54978d39ec61d7cca9ea5eb9fd..90c531fb91811bf0130f2f3f66bb22c0d2161c37 100644 (file)
 #include <dav1d/dav1d.h>
 
 #include "libavutil/avassert.h"
+#include "libavutil/film_grain_params.h"
 #include "libavutil/mastering_display_metadata.h"
 #include "libavutil/imgutils.h"
 #include "libavutil/opt.h"
 
+#include "atsc_a53.h"
 #include "avcodec.h"
+#include "bytestream.h"
 #include "decode.h"
 #include "internal.h"
 
@@ -38,7 +41,10 @@ typedef struct Libdav1dContext {
 
     Dav1dData data;
     int tile_threads;
+    int frame_threads;
     int apply_grain;
+    int operating_point;
+    int all_layers;
 } Libdav1dContext;
 
 static const enum AVPixelFormat pix_fmt[][3] = {
@@ -48,6 +54,10 @@ static const enum AVPixelFormat pix_fmt[][3] = {
     [DAV1D_PIXEL_LAYOUT_I444] = { AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV444P12 },
 };
 
+static const enum AVPixelFormat pix_fmt_rgb[3] = {
+    AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRP12,
+};
+
 static void libdav1d_log_callback(void *opaque, const char *fmt, va_list vl)
 {
     AVCodecContext *c = opaque;
@@ -59,12 +69,11 @@ static int libdav1d_picture_allocator(Dav1dPicture *p, void *cookie)
 {
     Libdav1dContext *dav1d = cookie;
     enum AVPixelFormat format = pix_fmt[p->p.layout][p->seq_hdr->hbd];
-    int ret, linesize[4], h = FFALIGN(p->p.h, 128);
+    int ret, linesize[4], h = FFALIGN(p->p.h, 128), w = FFALIGN(p->p.w, 128);
     uint8_t *aligned_ptr, *data[4];
     AVBufferRef *buf;
 
-    ret = av_image_fill_arrays(data, linesize, NULL, format, FFALIGN(p->p.w, 128),
-                               h, DAV1D_PICTURE_ALIGNMENT);
+    ret = av_image_get_buffer_size(format, w, h, DAV1D_PICTURE_ALIGNMENT);
     if (ret < 0)
         return ret;
 
@@ -72,8 +81,10 @@ static int libdav1d_picture_allocator(Dav1dPicture *p, void *cookie)
         av_buffer_pool_uninit(&dav1d->pool);
         // Use twice the amount of required padding bytes for aligned_ptr below.
         dav1d->pool = av_buffer_pool_init(ret + DAV1D_PICTURE_ALIGNMENT * 2, NULL);
-        if (!dav1d->pool)
+        if (!dav1d->pool) {
+            dav1d->pool_size = 0;
             return AVERROR(ENOMEM);
+        }
         dav1d->pool_size = ret;
     }
     buf = av_buffer_pool_get(dav1d->pool);
@@ -85,7 +96,8 @@ static int libdav1d_picture_allocator(Dav1dPicture *p, void *cookie)
     // Use the extra DAV1D_PICTURE_ALIGNMENT padding bytes in the buffer to align it
     // if required.
     aligned_ptr = (uint8_t *)FFALIGN((uintptr_t)buf->data, DAV1D_PICTURE_ALIGNMENT);
-    ret = av_image_fill_pointers(data, format, h, aligned_ptr, linesize);
+    ret = av_image_fill_arrays(data, linesize, aligned_ptr, format, w, h,
+                               DAV1D_PICTURE_ALIGNMENT);
     if (ret < 0) {
         av_buffer_unref(&buf);
         return ret;
@@ -112,6 +124,7 @@ static av_cold int libdav1d_init(AVCodecContext *c)
 {
     Libdav1dContext *dav1d = c->priv_data;
     Dav1dSettings s;
+    int threads = (c->thread_count ? c->thread_count : av_cpu_count()) * 3 / 2;
     int res;
 
     av_log(c, AV_LOG_INFO, "libdav1d %s\n", dav1d_version());
@@ -122,9 +135,24 @@ static av_cold int libdav1d_init(AVCodecContext *c)
     s.allocator.cookie = dav1d;
     s.allocator.alloc_picture_callback = libdav1d_picture_allocator;
     s.allocator.release_picture_callback = libdav1d_picture_release;
-    s.n_tile_threads = dav1d->tile_threads;
-    s.apply_grain = dav1d->apply_grain;
-    s.n_frame_threads = FFMIN(c->thread_count ? c->thread_count : av_cpu_count(), DAV1D_MAX_FRAME_THREADS);
+    s.frame_size_limit = c->max_pixels;
+    if (dav1d->apply_grain >= 0)
+        s.apply_grain = dav1d->apply_grain;
+    else if (c->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN)
+        s.apply_grain = 0;
+
+    s.all_layers = dav1d->all_layers;
+    if (dav1d->operating_point >= 0)
+        s.operating_point = dav1d->operating_point;
+
+    s.n_tile_threads = dav1d->tile_threads
+                     ? dav1d->tile_threads
+                     : FFMIN(floor(sqrt(threads)), DAV1D_MAX_TILE_THREADS);
+    s.n_frame_threads = dav1d->frame_threads
+                      ? dav1d->frame_threads
+                      : FFMIN(ceil(threads / s.n_tile_threads), DAV1D_MAX_FRAME_THREADS);
+    av_log(c, AV_LOG_DEBUG, "Using %d frame threads, %d tile threads\n",
+           s.n_frame_threads, s.n_tile_threads);
 
     res = dav1d_open(&dav1d->c, &s);
     if (res < 0)
@@ -147,18 +175,16 @@ static void libdav1d_data_free(const uint8_t *data, void *opaque) {
     av_buffer_unref(&buf);
 }
 
-static void libdav1d_frame_free(void *opaque, uint8_t *data) {
-    Dav1dPicture *p = opaque;
-
-    dav1d_picture_unref(p);
-    av_free(p);
+static void libdav1d_user_data_free(const uint8_t *data, void *opaque) {
+    av_assert0(data == opaque);
+    av_free(opaque);
 }
 
 static int libdav1d_receive_frame(AVCodecContext *c, AVFrame *frame)
 {
     Libdav1dContext *dav1d = c->priv_data;
     Dav1dData *data = &dav1d->data;
-    Dav1dPicture *p;
+    Dav1dPicture pic = { 0 }, *p = &pic;
     int res;
 
     if (!data->sz) {
@@ -181,6 +207,23 @@ static int libdav1d_receive_frame(AVCodecContext *c, AVFrame *frame)
 
             pkt.buf = NULL;
             av_packet_unref(&pkt);
+
+            if (c->reordered_opaque != AV_NOPTS_VALUE) {
+                uint8_t *reordered_opaque = av_malloc(sizeof(c->reordered_opaque));
+                if (!reordered_opaque) {
+                    dav1d_data_unref(data);
+                    return AVERROR(ENOMEM);
+                }
+
+                memcpy(reordered_opaque, &c->reordered_opaque, sizeof(c->reordered_opaque));
+                res = dav1d_data_wrap_user_data(data, reordered_opaque,
+                                                libdav1d_user_data_free, reordered_opaque);
+                if (res < 0) {
+                    av_free(reordered_opaque);
+                    dav1d_data_unref(data);
+                    return res;
+                }
+            }
         }
     }
 
@@ -192,10 +235,6 @@ static int libdav1d_receive_frame(AVCodecContext *c, AVFrame *frame)
             return res;
     }
 
-    p = av_mallocz(sizeof(*p));
-    if (!p)
-        return AVERROR(ENOMEM);
-
     res = dav1d_get_picture(dav1d->c, p);
     if (res < 0) {
         if (res == AVERROR(EINVAL))
@@ -203,17 +242,15 @@ static int libdav1d_receive_frame(AVCodecContext *c, AVFrame *frame)
         else if (res == AVERROR(EAGAIN) && c->internal->draining)
             res = AVERROR_EOF;
 
-        av_free(p);
         return res;
     }
 
-    av_assert0(p->data[0] != NULL);
+    av_assert0(p->data[0] && p->allocator_data);
 
-    frame->buf[0] = av_buffer_create(NULL, 0, libdav1d_frame_free,
-                                     p, AV_BUFFER_FLAG_READONLY);
+    // This requires the custom allocator above
+    frame->buf[0] = av_buffer_ref(p->allocator_data);
     if (!frame->buf[0]) {
         dav1d_picture_unref(p);
-        av_free(p);
         return AVERROR(ENOMEM);
     }
 
@@ -225,7 +262,8 @@ static int libdav1d_receive_frame(AVCodecContext *c, AVFrame *frame)
     frame->linesize[2] = p->stride[1];
 
     c->profile = p->seq_hdr->profile;
-    frame->format = c->pix_fmt = pix_fmt[p->p.layout][p->seq_hdr->hbd];
+    c->level = ((p->seq_hdr->operating_points[0].major_level - 2) << 2)
+               | p->seq_hdr->operating_points[0].minor_level;
     frame->width = p->p.w;
     frame->height = p->p.h;
     if (c->width != p->p.w || c->height != p->p.h) {
@@ -234,6 +272,13 @@ static int libdav1d_receive_frame(AVCodecContext *c, AVFrame *frame)
             goto fail;
     }
 
+    av_reduce(&frame->sample_aspect_ratio.num,
+              &frame->sample_aspect_ratio.den,
+              frame->height * (int64_t)p->frame_hdr->render_width,
+              frame->width  * (int64_t)p->frame_hdr->render_height,
+              INT_MAX);
+    ff_set_sar(c, frame->sample_aspect_ratio);
+
     switch (p->seq_hdr->chr) {
     case DAV1D_CHR_VERTICAL:
         frame->chroma_location = c->chroma_sample_location = AVCHROMA_LOC_LEFT;
@@ -247,13 +292,28 @@ static int libdav1d_receive_frame(AVCodecContext *c, AVFrame *frame)
     frame->color_trc = c->color_trc = (enum AVColorTransferCharacteristic) p->seq_hdr->trc;
     frame->color_range = c->color_range = p->seq_hdr->color_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
 
+    if (p->p.layout == DAV1D_PIXEL_LAYOUT_I444 &&
+        p->seq_hdr->mtrx == DAV1D_MC_IDENTITY &&
+        p->seq_hdr->pri  == DAV1D_COLOR_PRI_BT709 &&
+        p->seq_hdr->trc  == DAV1D_TRC_SRGB)
+        frame->format = c->pix_fmt = pix_fmt_rgb[p->seq_hdr->hbd];
+    else
+        frame->format = c->pix_fmt = pix_fmt[p->p.layout][p->seq_hdr->hbd];
+
+    if (p->m.user_data.data)
+        memcpy(&frame->reordered_opaque, p->m.user_data.data, sizeof(frame->reordered_opaque));
+    else
+        frame->reordered_opaque = AV_NOPTS_VALUE;
+
+    if (p->seq_hdr->num_units_in_tick && p->seq_hdr->time_scale) {
+        av_reduce(&c->framerate.den, &c->framerate.num,
+                  p->seq_hdr->num_units_in_tick, p->seq_hdr->time_scale, INT_MAX);
+        if (p->seq_hdr->equal_picture_interval)
+            c->ticks_per_frame = p->seq_hdr->num_ticks_per_picture;
+    }
+
     // match timestamps and packet size
-    frame->pts = frame->best_effort_timestamp = p->m.timestamp;
-#if FF_API_PKT_PTS
-FF_DISABLE_DEPRECATION_WARNINGS
-    frame->pkt_pts = p->m.timestamp;
-FF_ENABLE_DEPRECATION_WARNINGS
-#endif
+    frame->pts = p->m.timestamp;
     frame->pkt_dts = p->m.timestamp;
     frame->pkt_pos = p->m.offset;
     frame->pkt_size = p->m.size;
@@ -305,9 +365,76 @@ FF_ENABLE_DEPRECATION_WARNINGS
         light->MaxCLL = p->content_light->max_content_light_level;
         light->MaxFALL = p->content_light->max_frame_average_light_level;
     }
+    if (p->itut_t35) {
+        GetByteContext gb;
+        unsigned int user_identifier;
+
+        bytestream2_init(&gb, p->itut_t35->payload, p->itut_t35->payload_size);
+        bytestream2_skip(&gb, 1); // terminal provider code
+        bytestream2_skip(&gb, 1); // terminal provider oriented code
+        user_identifier = bytestream2_get_be32(&gb);
+        switch (user_identifier) {
+        case MKBETAG('G', 'A', '9', '4'): { // closed captions
+            AVBufferRef *buf = NULL;
+
+            res = ff_parse_a53_cc(&buf, gb.buffer, bytestream2_get_bytes_left(&gb));
+            if (res < 0)
+                goto fail;
+            if (!res)
+                break;
+
+            if (!av_frame_new_side_data_from_buf(frame, AV_FRAME_DATA_A53_CC, buf))
+                av_buffer_unref(&buf);
+
+            c->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
+            break;
+        }
+        default: // ignore unsupported identifiers
+            break;
+        }
+    }
+    if (p->frame_hdr->film_grain.present && (!dav1d->apply_grain ||
+        (c->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN))) {
+        AVFilmGrainParams *fgp = av_film_grain_params_create_side_data(frame);
+        if (!fgp) {
+            res = AVERROR(ENOMEM);
+            goto fail;
+        }
+
+        fgp->type = AV_FILM_GRAIN_PARAMS_AV1;
+        fgp->seed = p->frame_hdr->film_grain.data.seed;
+        fgp->codec.aom.num_y_points = p->frame_hdr->film_grain.data.num_y_points;
+        fgp->codec.aom.chroma_scaling_from_luma = p->frame_hdr->film_grain.data.chroma_scaling_from_luma;
+        fgp->codec.aom.scaling_shift = p->frame_hdr->film_grain.data.scaling_shift;
+        fgp->codec.aom.ar_coeff_lag = p->frame_hdr->film_grain.data.ar_coeff_lag;
+        fgp->codec.aom.ar_coeff_shift = p->frame_hdr->film_grain.data.ar_coeff_shift;
+        fgp->codec.aom.grain_scale_shift = p->frame_hdr->film_grain.data.grain_scale_shift;
+        fgp->codec.aom.overlap_flag = p->frame_hdr->film_grain.data.overlap_flag;
+        fgp->codec.aom.limit_output_range = p->frame_hdr->film_grain.data.clip_to_restricted_range;
+
+        memcpy(&fgp->codec.aom.y_points, &p->frame_hdr->film_grain.data.y_points,
+               sizeof(fgp->codec.aom.y_points));
+        memcpy(&fgp->codec.aom.num_uv_points, &p->frame_hdr->film_grain.data.num_uv_points,
+               sizeof(fgp->codec.aom.num_uv_points));
+        memcpy(&fgp->codec.aom.uv_points, &p->frame_hdr->film_grain.data.uv_points,
+               sizeof(fgp->codec.aom.uv_points));
+        memcpy(&fgp->codec.aom.ar_coeffs_y, &p->frame_hdr->film_grain.data.ar_coeffs_y,
+               sizeof(fgp->codec.aom.ar_coeffs_y));
+        memcpy(&fgp->codec.aom.ar_coeffs_uv[0], &p->frame_hdr->film_grain.data.ar_coeffs_uv[0],
+               sizeof(fgp->codec.aom.ar_coeffs_uv[0]));
+        memcpy(&fgp->codec.aom.ar_coeffs_uv[1], &p->frame_hdr->film_grain.data.ar_coeffs_uv[1],
+               sizeof(fgp->codec.aom.ar_coeffs_uv[1]));
+        memcpy(&fgp->codec.aom.uv_mult, &p->frame_hdr->film_grain.data.uv_mult,
+               sizeof(fgp->codec.aom.uv_mult));
+        memcpy(&fgp->codec.aom.uv_mult_luma, &p->frame_hdr->film_grain.data.uv_luma_mult,
+               sizeof(fgp->codec.aom.uv_mult_luma));
+        memcpy(&fgp->codec.aom.uv_offset, &p->frame_hdr->film_grain.data.uv_offset,
+               sizeof(fgp->codec.aom.uv_offset));
+    }
 
     res = 0;
 fail:
+    dav1d_picture_unref(p);
     if (res < 0)
         av_frame_unref(frame);
     return res;
@@ -327,8 +454,11 @@ static av_cold int libdav1d_close(AVCodecContext *c)
 #define OFFSET(x) offsetof(Libdav1dContext, x)
 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
 static const AVOption libdav1d_options[] = {
-    { "tilethreads", "Tile threads", OFFSET(tile_threads), AV_OPT_TYPE_INT, { .i64 = 1 }, 1, DAV1D_MAX_TILE_THREADS, VD },
-    { "filmgrain", "Apply Film Grain", OFFSET(apply_grain), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, VD },
+    { "tilethreads", "Tile threads", OFFSET(tile_threads), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, DAV1D_MAX_TILE_THREADS, VD },
+    { "framethreads", "Frame threads", OFFSET(frame_threads), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, DAV1D_MAX_FRAME_THREADS, VD },
+    { "filmgrain", "Apply Film Grain", OFFSET(apply_grain), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VD | AV_OPT_FLAG_DEPRECATED },
+    { "oppoint",  "Select an operating point of the scalable bitstream", OFFSET(operating_point), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 31, VD },
+    { "alllayers", "Output all spatial layers", OFFSET(all_layers), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VD },
     { NULL }
 };
 
@@ -349,8 +479,9 @@ AVCodec ff_libdav1d_decoder = {
     .close          = libdav1d_close,
     .flush          = libdav1d_flush,
     .receive_frame  = libdav1d_receive_frame,
-    .capabilities   = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_AUTO_THREADS,
-    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_SETS_PKT_DTS,
+    .capabilities   = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_OTHER_THREADS,
+    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_SETS_PKT_DTS |
+                      FF_CODEC_CAP_AUTO_THREADS,
     .priv_class     = &libdav1d_class,
     .wrapper_name   = "libdav1d",
 };