Merge remote branch 'qatar/master'
[ffmpeg.git] / libavcodec / libtheoraenc.c
index c154cbd..e57fda2 100644 (file)
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
-/*!
- * \file libtheoraenc.c
- * \brief Theora encoder using libtheora.
- * \author Paul Richards <paul.richards@gmail.com>
+/**
+ * @file
+ * @brief Theora encoder using libtheora.
+ * @author Paul Richards <paul.richards@gmail.com>
  *
  * A lot of this is copy / paste from other output codecs in
  * libavcodec or pure guesswork (or both).
  * and o_ prefixes on variables which are libogg types.
  */
 
-/* FFmpeg includes */
+/* Libav includes */
 #include "libavutil/intreadwrite.h"
 #include "libavutil/log.h"
+#include "libavutil/base64.h"
 #include "avcodec.h"
 
 /* libtheora includes */
-#include <theora/theora.h>
+#include <theora/theoraenc.h>
 
 typedef struct TheoraContext {
-    theora_state t_state;
+    th_enc_ctx *t_state;
+    uint8_t    *stats;
+    int         stats_size;
+    int         stats_offset;
+    int         uv_hshift;
+    int         uv_vshift;
+    int         keyframe_mask;
 } TheoraContext;
 
-/*!
-    Concatenates an ogg_packet into the extradata.
-*/
+/** Concatenate an ogg_packet into the extradata. */
 static int concatenate_packet(unsigned int* offset,
                               AVCodecContext* avc_context,
                               const ogg_packet* packet)
@@ -78,22 +83,89 @@ static int concatenate_packet(unsigned int* offset,
     return 0;
 }
 
+static int get_stats(AVCodecContext *avctx, int eos)
+{
+#ifdef TH_ENCCTL_2PASS_OUT
+    TheoraContext *h = avctx->priv_data;
+    uint8_t *buf;
+    int bytes;
+
+    bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
+    if (bytes < 0) {
+        av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
+        return -1;
+    }
+    if (!eos) {
+        h->stats = av_fast_realloc(h->stats, &h->stats_size,
+                                   h->stats_offset + bytes);
+        memcpy(h->stats + h->stats_offset, buf, bytes);
+        h->stats_offset += bytes;
+    } else {
+        int b64_size = AV_BASE64_SIZE(h->stats_offset);
+        // libtheora generates a summary header at the end
+        memcpy(h->stats, buf, bytes);
+        avctx->stats_out = av_malloc(b64_size);
+        av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
+    }
+    return 0;
+#else
+    av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
+    return -1;
+#endif
+}
+
+// libtheora won't read the entire buffer we give it at once, so we have to
+// repeatedly submit it...
+static int submit_stats(AVCodecContext *avctx)
+{
+#ifdef TH_ENCCTL_2PASS_IN
+    TheoraContext *h = avctx->priv_data;
+    int bytes;
+    if (!h->stats) {
+        if (!avctx->stats_in) {
+            av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
+            return -1;
+        }
+        h->stats_size = strlen(avctx->stats_in) * 3/4;
+        h->stats      = av_malloc(h->stats_size);
+        h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
+    }
+    while (h->stats_size - h->stats_offset > 0) {
+        bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
+                              h->stats + h->stats_offset,
+                              h->stats_size - h->stats_offset);
+        if (bytes < 0) {
+            av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
+            return -1;
+        }
+        if (!bytes)
+            return 0;
+        h->stats_offset += bytes;
+    }
+    return 0;
+#else
+    av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
+    return -1;
+#endif
+}
+
 static av_cold int encode_init(AVCodecContext* avc_context)
 {
-    theora_info t_info;
-    theora_comment t_comment;
+    th_info t_info;
+    th_comment t_comment;
     ogg_packet o_packet;
     unsigned int offset;
     TheoraContext *h = avc_context->priv_data;
+    uint32_t gop_size = avc_context->gop_size;
 
     /* Set up the theora_info struct */
-    theora_info_init(&t_info);
-    t_info.width        = FFALIGN(avc_context->width,  16);
-    t_info.height       = FFALIGN(avc_context->height, 16);
-    t_info.frame_width  = avc_context->width;
-    t_info.frame_height = avc_context->height;
-    t_info.offset_x     = 0;
-    t_info.offset_y     = avc_context->height & 0xf;
+    th_info_init(&t_info);
+    t_info.frame_width  = FFALIGN(avc_context->width,  16);
+    t_info.frame_height = FFALIGN(avc_context->height, 16);
+    t_info.pic_width    = avc_context->width;
+    t_info.pic_height   = avc_context->height;
+    t_info.pic_x        = 0;
+    t_info.pic_y        = 0;
     /* Swap numerator and denominator as time_base in AVCodecContext gives the
      * time period between frames, but theora_info needs the framerate.  */
     t_info.fps_numerator   = avc_context->time_base.den;
@@ -105,19 +177,25 @@ static av_cold int encode_init(AVCodecContext* avc_context)
         t_info.aspect_numerator   = 1;
         t_info.aspect_denominator = 1;
     }
-    t_info.colorspace               = OC_CS_UNSPECIFIED;
-    t_info.pixelformat              = OC_PF_420;
-    t_info.keyframe_frequency       = avc_context->gop_size;
-    t_info.keyframe_frequency_force = avc_context->gop_size;
-    t_info.keyframe_mindistance     = avc_context->keyint_min;
-
-    t_info.quick_p                      = 1;
-    t_info.dropframes_p                 = 0;
-    t_info.keyframe_auto_p              = 1;
-    t_info.keyframe_data_target_bitrate = t_info.target_bitrate * 1.5;
-    t_info.keyframe_auto_threshold      = 80;
-    t_info.noise_sensitivity            = 1;
-    t_info.sharpness                    = 0;
+
+    if (avc_context->color_primaries == AVCOL_PRI_BT470M)
+        t_info.colorspace = TH_CS_ITU_REC_470M;
+    else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
+        t_info.colorspace = TH_CS_ITU_REC_470BG;
+    else
+        t_info.colorspace = TH_CS_UNSPECIFIED;
+
+    if (avc_context->pix_fmt == PIX_FMT_YUV420P)
+        t_info.pixel_fmt = TH_PF_420;
+    else if (avc_context->pix_fmt == PIX_FMT_YUV422P)
+        t_info.pixel_fmt = TH_PF_422;
+    else if (avc_context->pix_fmt == PIX_FMT_YUV444P)
+        t_info.pixel_fmt = TH_PF_444;
+    else {
+        av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
+        return -1;
+    }
+    avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
 
     if (avc_context->flags & CODEC_FLAG_QSCALE) {
         /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
@@ -133,13 +211,30 @@ static av_cold int encode_init(AVCodecContext* avc_context)
     }
 
     /* Now initialise libtheora */
-    if (theora_encode_init(&(h->t_state), &t_info)) {
+    h->t_state = th_encode_alloc(&t_info);
+    if (!h->t_state) {
         av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
         return -1;
     }
 
+    h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
     /* Clear up theora_info struct */
-    theora_info_clear(&t_info);
+    th_info_clear(&t_info);
+
+    if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
+                      &gop_size, sizeof(gop_size))) {
+        av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
+        return -1;
+    }
+
+    // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
+    if (avc_context->flags & CODEC_FLAG_PASS1) {
+        if (get_stats(avc_context, 0))
+            return -1;
+    } else if (avc_context->flags & CODEC_FLAG_PASS2) {
+        if (submit_stats(avc_context))
+            return -1;
+    }
 
     /*
         Output first header packet consisting of theora
@@ -150,26 +245,14 @@ static av_cold int encode_init(AVCodecContext* avc_context)
     */
     offset = 0;
 
-    /* Header */
-    theora_encode_header(&(h->t_state), &o_packet);
-    if (concatenate_packet(&offset, avc_context, &o_packet))
-        return -1;
+    /* Headers */
+    th_comment_init(&t_comment);
 
-    /* Comment */
-    theora_comment_init(&t_comment);
-    theora_encode_comment(&t_comment, &o_packet);
-    if (concatenate_packet(&offset, avc_context, &o_packet))
-        return -1;
-    /* Clear up theora_comment struct before we reset the packet */
-    theora_comment_clear(&t_comment);
-    /* And despite documentation to the contrary, theora_comment_clear
-     * does not release the packet */
-    ogg_packet_clear(&o_packet);
-
-    /* Tables */
-    theora_encode_tables(&(h->t_state), &o_packet);
-    if (concatenate_packet(&offset, avc_context, &o_packet))
-        return -1;
+    while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
+        if (concatenate_packet(&offset, avc_context, &o_packet))
+            return -1;
+
+    th_comment_clear(&t_comment);
 
     /* Set up the output AVFrame */
     avc_context->coded_frame= avcodec_alloc_frame();
@@ -180,40 +263,42 @@ static av_cold int encode_init(AVCodecContext* avc_context)
 static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf,
                         int buf_size, void *data)
 {
-    yuv_buffer t_yuv_buffer;
+    th_ycbcr_buffer t_yuv_buffer;
     TheoraContext *h = avc_context->priv_data;
     AVFrame *frame = data;
     ogg_packet o_packet;
-    int result;
-
-    assert(avc_context->pix_fmt == PIX_FMT_YUV420P);
+    int result, i;
+
+    // EOS, finish and get 1st pass stats if applicable
+    if (!frame) {
+        th_encode_packetout(h->t_state, 1, &o_packet);
+        if (avc_context->flags & CODEC_FLAG_PASS1)
+            if (get_stats(avc_context, 1))
+                return -1;
+        return 0;
+    }
 
     /* Copy planes to the theora yuv_buffer */
-    if (frame->linesize[1] != frame->linesize[2]) {
-        av_log(avc_context, AV_LOG_ERROR, "U and V stride differ\n");
-        return -1;
+    for (i = 0; i < 3; i++) {
+        t_yuv_buffer[i].width  = FFALIGN(avc_context->width,  16) >> (i && h->uv_hshift);
+        t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
+        t_yuv_buffer[i].stride = frame->linesize[i];
+        t_yuv_buffer[i].data   = frame->data[i];
     }
 
-    t_yuv_buffer.y_width   = FFALIGN(avc_context->width, 16);
-    t_yuv_buffer.y_height  = FFALIGN(avc_context->height, 16);
-    t_yuv_buffer.y_stride  = frame->linesize[0];
-    t_yuv_buffer.uv_width  = t_yuv_buffer.y_width / 2;
-    t_yuv_buffer.uv_height = t_yuv_buffer.y_height / 2;
-    t_yuv_buffer.uv_stride = frame->linesize[1];
-
-    t_yuv_buffer.y = frame->data[0];
-    t_yuv_buffer.u = frame->data[1];
-    t_yuv_buffer.v = frame->data[2];
+    if (avc_context->flags & CODEC_FLAG_PASS2)
+        if (submit_stats(avc_context))
+            return -1;
 
     /* Now call into theora_encode_YUVin */
-    result = theora_encode_YUVin(&(h->t_state), &t_yuv_buffer);
+    result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
     if (result) {
         const char* message;
         switch (result) {
         case -1:
             message = "differing frame sizes";
             break;
-        case OC_EINVAL:
+        case TH_EINVAL:
             message = "encoder is not ready or is finished";
             break;
         default:
@@ -224,8 +309,12 @@ static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf,
         return -1;
     }
 
+    if (avc_context->flags & CODEC_FLAG_PASS1)
+        if (get_stats(avc_context, 0))
+            return -1;
+
     /* Pick up returned ogg_packet */
-    result = theora_encode_packetout(&(h->t_state), 0, &o_packet);
+    result = th_encode_packetout(h->t_state, 0, &o_packet);
     switch (result) {
     case 0:
         /* No packet is ready */
@@ -245,52 +334,38 @@ static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf,
     }
     memcpy(outbuf, o_packet.packet, o_packet.bytes);
 
-    // HACK: does not take codec delay into account (neither does the decoder though)
+    // HACK: assumes no encoder delay, this is true until libtheora becomes
+    // multithreaded (which will be disabled unless explictly requested)
     avc_context->coded_frame->pts = frame->pts;
+    avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask);
 
     return o_packet.bytes;
 }
 
 static av_cold int encode_close(AVCodecContext* avc_context)
 {
-    ogg_packet o_packet;
     TheoraContext *h = avc_context->priv_data;
-    int result;
-    const char* message;
 
-    result = theora_encode_packetout(&(h->t_state), 1, &o_packet);
-    theora_clear(&(h->t_state));
+    th_encode_free(h->t_state);
+    av_freep(&h->stats);
     av_freep(&avc_context->coded_frame);
+    av_freep(&avc_context->stats_out);
     av_freep(&avc_context->extradata);
     avc_context->extradata_size = 0;
 
-    switch (result) {
-    case 0:  /* No packet is ready */
-    case -1: /* Encoding finished */
-        return 0;
-    case 1:
-        /* We have a packet */
-        message = "gave us a packet";
-        break;
-    default:
-        message = "unknown reason";
-        break;
-    }
-    av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed (%s) [%d]\n", message, result);
-    return -1;
+    return 0;
 }
 
-static const enum PixelFormat supported_pixel_formats[] = { PIX_FMT_YUV420P, PIX_FMT_NONE };
-
-/*! AVCodec struct exposed to libavcodec */
-AVCodec libtheora_encoder = {
+/** AVCodec struct exposed to libavcodec */
+AVCodec ff_libtheora_encoder = {
     .name = "libtheora",
-    .type = CODEC_TYPE_VIDEO,
+    .type = AVMEDIA_TYPE_VIDEO,
     .id = CODEC_ID_THEORA,
     .priv_data_size = sizeof(TheoraContext),
     .init = encode_init,
     .close = encode_close,
     .encode = encode_frame,
-    .pix_fmts = supported_pixel_formats,
+    .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
+    .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE},
     .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),
 };