libvpx: Pull from upstream

source/libvpx/examples/vpx_temporal_scalable_patterns.c

 /*
  *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 23 matching lines @@
 
 // For rate control encoding stats.
 struct RateControlMetrics {
   // Number of input frames per layer.
   int layer_input_frames[VPX_TS_MAX_LAYERS];
   // Total (cumulative) number of encoded frames per layer.
   int layer_tot_enc_frames[VPX_TS_MAX_LAYERS];
   // Number of encoded non-key frames per layer.
   int layer_enc_frames[VPX_TS_MAX_LAYERS];
   // Framerate per layer layer (cumulative).
-  float layer_framerate[VPX_TS_MAX_LAYERS];
+  double layer_framerate[VPX_TS_MAX_LAYERS];
   // Target average frame size per layer (per-frame-bandwidth per layer).
-  float layer_pfb[VPX_TS_MAX_LAYERS];
+  double layer_pfb[VPX_TS_MAX_LAYERS];
   // Actual average frame size per layer.
-  float layer_avg_frame_size[VPX_TS_MAX_LAYERS];
+  double layer_avg_frame_size[VPX_TS_MAX_LAYERS];
   // Average rate mismatch per layer (|target - actual| / target).
-  float layer_avg_rate_mismatch[VPX_TS_MAX_LAYERS];
+  double layer_avg_rate_mismatch[VPX_TS_MAX_LAYERS];
   // Actual encoding bitrate per layer (cumulative).
-  float layer_encoding_bitrate[VPX_TS_MAX_LAYERS];
+  double layer_encoding_bitrate[VPX_TS_MAX_LAYERS];
 };
 
+// Note: these rate control metrics assume only 1 key frame in the
+// sequence (i.e., first frame only). So for temporal pattern# 7
+// (which has key frame for every frame on base layer), the metrics
+// computation will be off/wrong.
+// TODO(marpan): Update these metrics to account for multiple key frames
+// in the stream.
 static void set_rate_control_metrics(struct RateControlMetrics *rc,
                                      vpx_codec_enc_cfg_t *cfg) {
-  int i = 0;
+  unsigned int i = 0;
   // Set the layer (cumulative) framerate and the target layer (non-cumulative)
   // per-frame-bandwidth, for the rate control encoding stats below.
-  float framerate = cfg->g_timebase.den / cfg->g_timebase.num;
+  const double framerate = cfg->g_timebase.den / cfg->g_timebase.num;
   rc->layer_framerate[0] = framerate / cfg->ts_rate_decimator[0];
   rc->layer_pfb[0] = 1000.0 * cfg->ts_target_bitrate[0] /
       rc->layer_framerate[0];
   for (i = 0; i < cfg->ts_number_layers; ++i) {
     if (i > 0) {
       rc->layer_framerate[i] = framerate / cfg->ts_rate_decimator[i];
       rc->layer_pfb[i] = 1000.0 *
           (cfg->ts_target_bitrate[i] - cfg->ts_target_bitrate[i - 1]) /
           (rc->layer_framerate[i] - rc->layer_framerate[i - 1]);
     }
     rc->layer_input_frames[i] = 0;
     rc->layer_enc_frames[i] = 0;
     rc->layer_tot_enc_frames[i] = 0;
     rc->layer_encoding_bitrate[i] = 0.0;
     rc->layer_avg_frame_size[i] = 0.0;
     rc->layer_avg_rate_mismatch[i] = 0.0;
   }
 }
 
 static void printout_rate_control_summary(struct RateControlMetrics *rc,
                                           vpx_codec_enc_cfg_t *cfg,
                                           int frame_cnt) {
-  int i = 0;
-  int check_num_frames = 0;
+  unsigned int i = 0;
+  int tot_num_frames = 0;
   printf("Total number of processed frames: %d\n\n", frame_cnt -1);
   printf("Rate control layer stats for %d layer(s):\n\n",
       cfg->ts_number_layers);
   for (i = 0; i < cfg->ts_number_layers; ++i) {
     const int num_dropped = (i > 0) ?
         (rc->layer_input_frames[i] - rc->layer_enc_frames[i]) :
         (rc->layer_input_frames[i] - rc->layer_enc_frames[i] - 1);
+    tot_num_frames += rc->layer_input_frames[i];
     rc->layer_encoding_bitrate[i] = 0.001 * rc->layer_framerate[i] *
-        rc->layer_encoding_bitrate[i] / rc->layer_tot_enc_frames[i];
+        rc->layer_encoding_bitrate[i] / tot_num_frames;
     rc->layer_avg_frame_size[i] = rc->layer_avg_frame_size[i] /
         rc->layer_enc_frames[i];
     rc->layer_avg_rate_mismatch[i] = 100.0 * rc->layer_avg_rate_mismatch[i] /
         rc->layer_enc_frames[i];
     printf("For layer#: %d \n", i);
     printf("Bitrate (target vs actual): %d %f \n", cfg->ts_target_bitrate[i],
            rc->layer_encoding_bitrate[i]);
     printf("Average frame size (target vs actual): %f %f \n", rc->layer_pfb[i],
            rc->layer_avg_frame_size[i]);
     printf("Average rate_mismatch: %f \n", rc->layer_avg_rate_mismatch[i]);
     printf("Number of input frames, encoded (non-key) frames, "
         "and perc dropped frames: %d %d %f \n", rc->layer_input_frames[i],
         rc->layer_enc_frames[i],
         100.0 * num_dropped / rc->layer_input_frames[i]);
-    check_num_frames += rc->layer_input_frames[i];
     printf("\n");
   }
-  if ((frame_cnt - 1) != check_num_frames)
+  if ((frame_cnt - 1) != tot_num_frames)
     die("Error: Number of input frames not equal to output! \n");
 }
 
 // Temporal scaling parameters:
 // NOTE: The 3 prediction frames cannot be used interchangeably due to
 // differences in the way they are handled throughout the code. The
 // frames should be allocated to layers in the order LAST, GF, ARF.
 // Other combinations work, but may produce slightly inferior results.
 static void set_temporal_layer_pattern(int layering_mode,
                                        vpx_codec_enc_cfg_t *cfg,
@@ skipping 303 matching lines @@
   vpx_codec_ctx_t codec;
   vpx_codec_enc_cfg_t cfg;
   int frame_cnt = 0;
   vpx_image_t raw;
   vpx_codec_err_t res;
   unsigned int width;
   unsigned int height;
   int frame_avail;
   int got_data;
   int flags = 0;
-  int i;
+  unsigned int i;
   int pts = 0;  // PTS starts at 0.
   int frame_duration = 1;  // 1 timebase tick per frame.
   int layering_mode = 0;
   int layer_flags[VPX_TS_MAX_PERIODICITY] = {0};
   int flag_periodicity = 1;
   int max_intra_size_pct;
   vpx_svc_layer_id_t layer_id = {0, 0};
   const VpxInterface *encoder = NULL;
   FILE *infile = NULL;
   struct RateControlMetrics rc;
@@ skipping 39 matching lines @@
   }
 
   // Update the default configuration with our settings.
   cfg.g_w = width;
   cfg.g_h = height;
 
   // Timebase format e.g. 30fps: numerator=1, demoninator = 30.
   cfg.g_timebase.num = strtol(argv[6], NULL, 0);
   cfg.g_timebase.den = strtol(argv[7], NULL, 0);
 
-  for (i = 10; i < 10 + mode_to_num_layers[layering_mode]; ++i) {
+  for (i = 10; (int)i < 10 + mode_to_num_layers[layering_mode]; ++i) {
     cfg.ts_target_bitrate[i - 10] = strtol(argv[i], NULL, 0);
   }
 
   // Real time parameters.
   cfg.rc_dropframe_thresh = strtol(argv[8], NULL, 0);
   cfg.rc_end_usage = VPX_CBR;
   cfg.rc_resize_allowed = 0;
   cfg.rc_min_quantizer = 2;
   cfg.rc_max_quantizer = 56;
   cfg.rc_undershoot_pct = 50;
   cfg.rc_overshoot_pct = 50;
   cfg.rc_buf_initial_sz = 500;
   cfg.rc_buf_optimal_sz = 600;
   cfg.rc_buf_sz = 1000;
 
   // Enable error resilient mode.
   cfg.g_error_resilient = 1;
   cfg.g_lag_in_frames   = 0;
   cfg.kf_mode = VPX_KF_DISABLED;
 
   // Disable automatic keyframe placement.
   cfg.kf_min_dist = cfg.kf_max_dist = 3000;
 
-  // Default setting for bitrate: used in special case of 1 layer (case 0).
-  cfg.rc_target_bitrate = cfg.ts_target_bitrate[0];
-
   set_temporal_layer_pattern(layering_mode,
                              &cfg,
                              layer_flags,
                              &flag_periodicity);
 
   set_rate_control_metrics(&rc, &cfg);
 
+  // Target bandwidth for the whole stream.
+  // Set to ts_target_bitrate for highest layer (total bitrate).
+  cfg.rc_target_bitrate = cfg.ts_target_bitrate[cfg.ts_number_layers - 1];
+
   // Open input file.
   if (!(infile = fopen(argv[1], "rb"))) {
     die("Failed to open %s for reading", argv[1]);
   }
 
   // Open an output file for each stream.
   for (i = 0; i < cfg.ts_number_layers; ++i) {
     char file_name[PATH_MAX];
     VpxVideoInfo info;
     info.codec_fourcc = encoder->fourcc;
@@ skipping 18 matching lines @@
   vpx_codec_control(&codec, VP8E_SET_NOISE_SENSITIVITY, 1);
   if (strncmp(encoder->name, "vp9", 3) == 0) {
     vpx_codec_control(&codec, VP8E_SET_CPUUSED, 3);
     vpx_codec_control(&codec, VP8E_SET_NOISE_SENSITIVITY, 0);
     if (vpx_codec_control(&codec, VP9E_SET_SVC, 1)) {
       die_codec(&codec, "Failed to set SVC");
     }
   }
   vpx_codec_control(&codec, VP8E_SET_STATIC_THRESHOLD, 1);
   vpx_codec_control(&codec, VP8E_SET_TOKEN_PARTITIONS, 1);
+  // This controls the maximum target size of the key frame.
+  // For generating smaller key frames, use a smaller max_intra_size_pct
+  // value, like 100 or 200.
   max_intra_size_pct = (int) (((double)cfg.rc_buf_optimal_sz * 0.5)
       * ((double) cfg.g_timebase.den / cfg.g_timebase.num) / 10.0);
   vpx_codec_control(&codec, VP8E_SET_MAX_INTRA_BITRATE_PCT, max_intra_size_pct);
 
   frame_avail = 1;
   while (frame_avail || got_data) {
     vpx_codec_iter_t iter = NULL;
     const vpx_codec_cx_pkt_t *pkt;
     // Update the temporal layer_id. No spatial layers in this test.
     layer_id.spatial_layer_id = 0;
@@ skipping 48 matching lines @@
 
   if (vpx_codec_destroy(&codec))
     die_codec(&codec, "Failed to destroy codec");
 
   // Try to rewrite the output file headers with the actual frame count.
   for (i = 0; i < cfg.ts_number_layers; ++i)
     vpx_video_writer_close(outfile[i]);
 
   return EXIT_SUCCESS;
 }