WebM Codec SDK
twopass_encoder
1 /*
2  * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3  *
4  * Use of this source code is governed by a BSD-style license
5  * that can be found in the LICENSE file in the root of the source
6  * tree. An additional intellectual property rights grant can be found
7  * in the file PATENTS. All contributing project authors may
8  * be found in the AUTHORS file in the root of the source tree.
9  */
10 
11 // Two Pass Encoder
12 // ================
13 //
14 // This is an example of a two pass encoder loop. It takes an input file in
15 // YV12 format, passes it through the encoder twice, and writes the compressed
16 // frames to disk in IVF format. It builds upon the simple_encoder example.
17 //
18 // Twopass Variables
19 // -----------------
20 // Twopass mode needs to track the current pass number and the buffer of
21 // statistics packets.
22 //
23 // Updating The Configuration
24 // ---------------------------------
25 // In two pass mode, the configuration has to be updated on each pass. The
26 // statistics buffer is passed on the last pass.
27 //
28 // Encoding A Frame
29 // ----------------
30 // Encoding a frame in two pass mode is identical to the simple encoder
31 // example. To increase the quality while sacrificing encoding speed,
32 // VPX_DL_BEST_QUALITY can be used in place of VPX_DL_GOOD_QUALITY.
33 //
34 // Processing Statistics Packets
35 // -----------------------------
36 // Each packet of type `VPX_CODEC_CX_FRAME_PKT` contains the encoded data
37 // for this frame. We write a IVF frame header, followed by the raw data.
38 //
39 //
40 // Pass Progress Reporting
41 // -----------------------------
42 // It's sometimes helpful to see when each pass completes.
43 //
44 //
45 // Clean-up
46 // -----------------------------
47 // Destruction of the encoder instance must be done on each pass. The
48 // raw image should be destroyed at the end as usual.
49 
50 #include <stdio.h>
51 #include <stdlib.h>
52 #include <string.h>
53 
54 #include "vpx/vpx_encoder.h"
55 
56 #include "../tools_common.h"
57 #include "../video_writer.h"
58 
59 static const char *exec_name;
60 
61 void usage_exit(void) {
62  fprintf(stderr,
63  "Usage: %s <codec> <width> <height> <infile> <outfile> "
64  "<frame limit>\n",
65  exec_name);
66  exit(EXIT_FAILURE);
67 }
68 
69 static int get_frame_stats(vpx_codec_ctx_t *ctx, const vpx_image_t *img,
70  vpx_codec_pts_t pts, unsigned int duration,
71  vpx_enc_frame_flags_t flags, unsigned int deadline,
72  vpx_fixed_buf_t *stats) {
73  int got_pkts = 0;
74  vpx_codec_iter_t iter = NULL;
75  const vpx_codec_cx_pkt_t *pkt = NULL;
76  const vpx_codec_err_t res =
77  vpx_codec_encode(ctx, img, pts, duration, flags, deadline);
78  if (res != VPX_CODEC_OK) die_codec(ctx, "Failed to get frame stats.");
79 
80  while ((pkt = vpx_codec_get_cx_data(ctx, &iter)) != NULL) {
81  got_pkts = 1;
82 
83  if (pkt->kind == VPX_CODEC_STATS_PKT) {
84  const uint8_t *const pkt_buf = pkt->data.twopass_stats.buf;
85  const size_t pkt_size = pkt->data.twopass_stats.sz;
86  stats->buf = realloc(stats->buf, stats->sz + pkt_size);
87  if (!stats->buf) die("Failed to reallocate stats buffer.");
88  memcpy((uint8_t *)stats->buf + stats->sz, pkt_buf, pkt_size);
89  stats->sz += pkt_size;
90  }
91  }
92 
93  return got_pkts;
94 }
95 
96 static int encode_frame(vpx_codec_ctx_t *ctx, const vpx_image_t *img,
97  vpx_codec_pts_t pts, unsigned int duration,
98  vpx_enc_frame_flags_t flags, unsigned int deadline,
99  VpxVideoWriter *writer) {
100  int got_pkts = 0;
101  vpx_codec_iter_t iter = NULL;
102  const vpx_codec_cx_pkt_t *pkt = NULL;
103  const vpx_codec_err_t res =
104  vpx_codec_encode(ctx, img, pts, duration, flags, deadline);
105  if (res != VPX_CODEC_OK) die_codec(ctx, "Failed to encode frame.");
106 
107  while ((pkt = vpx_codec_get_cx_data(ctx, &iter)) != NULL) {
108  got_pkts = 1;
109  if (pkt->kind == VPX_CODEC_CX_FRAME_PKT) {
110  const int keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY) != 0;
111 
112  if (!vpx_video_writer_write_frame(writer, pkt->data.frame.buf,
113  pkt->data.frame.sz,
114  pkt->data.frame.pts))
115  die_codec(ctx, "Failed to write compressed frame.");
116  printf(keyframe ? "K" : ".");
117  fflush(stdout);
118  }
119  }
120 
121  return got_pkts;
122 }
123 
124 static vpx_fixed_buf_t pass0(vpx_image_t *raw, FILE *infile,
125  const VpxInterface *encoder,
126  const vpx_codec_enc_cfg_t *cfg, int max_frames) {
127  vpx_codec_ctx_t codec;
128  int frame_count = 0;
129  vpx_fixed_buf_t stats = { NULL, 0 };
130 
131  if (vpx_codec_enc_init(&codec, encoder->codec_interface(), cfg, 0))
132  die("Failed to initialize encoder");
133 
134  // Calculate frame statistics.
135  while (vpx_img_read(raw, infile)) {
136  ++frame_count;
137  get_frame_stats(&codec, raw, frame_count, 1, 0, VPX_DL_GOOD_QUALITY,
138  &stats);
139  if (max_frames > 0 && frame_count >= max_frames) break;
140  }
141 
142  // Flush encoder.
143  while (get_frame_stats(&codec, NULL, frame_count, 1, 0, VPX_DL_GOOD_QUALITY,
144  &stats)) {
145  }
146 
147  printf("Pass 0 complete. Processed %d frames.\n", frame_count);
148  if (vpx_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec.");
149 
150  return stats;
151 }
152 
153 static void pass1(vpx_image_t *raw, FILE *infile, const char *outfile_name,
154  const VpxInterface *encoder, const vpx_codec_enc_cfg_t *cfg,
155  int max_frames) {
156  VpxVideoInfo info = { encoder->fourcc,
157  cfg->g_w,
158  cfg->g_h,
159  { cfg->g_timebase.num, cfg->g_timebase.den } };
160  VpxVideoWriter *writer = NULL;
161  vpx_codec_ctx_t codec;
162  int frame_count = 0;
163 
164  writer = vpx_video_writer_open(outfile_name, kContainerIVF, &info);
165  if (!writer) die("Failed to open %s for writing", outfile_name);
166 
167  if (vpx_codec_enc_init(&codec, encoder->codec_interface(), cfg, 0))
168  die("Failed to initialize encoder");
169 
170  // Encode frames.
171  while (vpx_img_read(raw, infile)) {
172  ++frame_count;
173  encode_frame(&codec, raw, frame_count, 1, 0, VPX_DL_GOOD_QUALITY, writer);
174 
175  if (max_frames > 0 && frame_count >= max_frames) break;
176  }
177 
178  // Flush encoder.
179  while (encode_frame(&codec, NULL, -1, 1, 0, VPX_DL_GOOD_QUALITY, writer)) {
180  }
181 
182  printf("\n");
183 
184  if (vpx_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec.");
185 
186  vpx_video_writer_close(writer);
187 
188  printf("Pass 1 complete. Processed %d frames.\n", frame_count);
189 }
190 
191 int main(int argc, char **argv) {
192  FILE *infile = NULL;
193  int w, h;
194  vpx_codec_ctx_t codec;
196  vpx_image_t raw;
197  vpx_codec_err_t res;
198  vpx_fixed_buf_t stats;
199 
200  const VpxInterface *encoder = NULL;
201  const int fps = 30; // TODO(dkovalev) add command line argument
202  const int bitrate = 200; // kbit/s TODO(dkovalev) add command line argument
203  const char *const codec_arg = argv[1];
204  const char *const width_arg = argv[2];
205  const char *const height_arg = argv[3];
206  const char *const infile_arg = argv[4];
207  const char *const outfile_arg = argv[5];
208  int max_frames = 0;
209  exec_name = argv[0];
210 
211  if (argc != 7) die("Invalid number of arguments.");
212 
213  max_frames = (int)strtol(argv[6], NULL, 0);
214 
215  encoder = get_vpx_encoder_by_name(codec_arg);
216  if (!encoder) die("Unsupported codec.");
217 
218  w = (int)strtol(width_arg, NULL, 0);
219  h = (int)strtol(height_arg, NULL, 0);
220 
221  if (w <= 0 || h <= 0 || (w % 2) != 0 || (h % 2) != 0)
222  die("Invalid frame size: %dx%d", w, h);
223 
224  if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, w, h, 1))
225  die("Failed to allocate image (%dx%d)", w, h);
226 
227  printf("Using %s\n", vpx_codec_iface_name(encoder->codec_interface()));
228 
229  // Configuration
230  res = vpx_codec_enc_config_default(encoder->codec_interface(), &cfg, 0);
231  if (res) die_codec(&codec, "Failed to get default codec config.");
232 
233  cfg.g_w = w;
234  cfg.g_h = h;
235  cfg.g_timebase.num = 1;
236  cfg.g_timebase.den = fps;
237  cfg.rc_target_bitrate = bitrate;
238 
239  if (!(infile = fopen(infile_arg, "rb")))
240  die("Failed to open %s for reading", infile_arg);
241 
242  // Pass 0
244  stats = pass0(&raw, infile, encoder, &cfg, max_frames);
245 
246  // Pass 1
247  rewind(infile);
248  cfg.g_pass = VPX_RC_LAST_PASS;
249  cfg.rc_twopass_stats_in = stats;
250  pass1(&raw, infile, outfile_arg, encoder, &cfg, max_frames);
251  free(stats.buf);
252 
253  vpx_img_free(&raw);
254  fclose(infile);
255 
256  return EXIT_SUCCESS;
257 }
vpx_codec_ctx
Codec context structure.
Definition: vpx_codec.h:200
vpx_fixed_buf
Generic fixed size buffer structure.
Definition: vpx_encoder.h:98
vpx_codec_cx_pkt::kind
enum vpx_codec_cx_pkt_kind kind
Definition: vpx_encoder.h:162
vpx_img_free
void vpx_img_free(vpx_image_t *img)
Close an image descriptor.
vpx_codec_cx_pkt::data
union vpx_codec_cx_pkt::@1 data
vpx_codec_enc_cfg::rc_twopass_stats_in
vpx_fixed_buf_t rc_twopass_stats_in
Two-pass stats buffer.
Definition: vpx_encoder.h:449
vpx_codec_enc_cfg
Encoder configuration structure.
Definition: vpx_encoder.h:270
vpx_codec_iface_name
const char * vpx_codec_iface_name(vpx_codec_iface_t *iface)
Return the name for a given interface.
vpx_enc_frame_flags_t
long vpx_enc_frame_flags_t
Encoded Frame Flags.
Definition: vpx_encoder.h:261
vpx_img_alloc
vpx_image_t * vpx_img_alloc(vpx_image_t *img, vpx_img_fmt_t fmt, unsigned int d_w, unsigned int d_h, unsigned int align)
Open a descriptor, allocating storage for the underlying image.
vpx_rational::num
int num
Definition: vpx_encoder.h:221
VPX_CODEC_OK
@ VPX_CODEC_OK
Operation completed without error.
Definition: vpx_codec.h:95
vpx_codec_get_cx_data
const vpx_codec_cx_pkt_t * vpx_codec_get_cx_data(vpx_codec_ctx_t *ctx, vpx_codec_iter_t *iter)
Encoded data iterator.
vpx_codec_cx_pkt
Encoder output packet.
Definition: vpx_encoder.h:161
vpx_codec_iter_t
const typedef void * vpx_codec_iter_t
Iterator.
Definition: vpx_codec.h:190
vpx_codec_enc_cfg::g_w
unsigned int g_w
Width of the frame.
Definition: vpx_encoder.h:306
VPX_CODEC_CX_FRAME_PKT
@ VPX_CODEC_CX_FRAME_PKT
Definition: vpx_encoder.h:149
vpx_codec_enc_init
#define vpx_codec_enc_init(ctx, iface, cfg, flags)
Convenience macro for vpx_codec_enc_init_ver()
Definition: vpx_encoder.h:889
VPX_CODEC_STATS_PKT
@ VPX_CODEC_STATS_PKT
Definition: vpx_encoder.h:150
vpx_codec_pts_t
int64_t vpx_codec_pts_t
Time Stamp Type.
Definition: vpx_encoder.h:108
VPX_RC_LAST_PASS
@ VPX_RC_LAST_PASS
Definition: vpx_encoder.h:229
vpx_codec_enc_cfg::g_timebase
struct vpx_rational g_timebase
Stream timebase units.
Definition: vpx_encoder.h:345
VPX_IMG_FMT_I420
@ VPX_IMG_FMT_I420
Definition: vpx_image.h:42
vpx_image
Image Descriptor.
Definition: vpx_image.h:72
vpx_codec_enc_cfg::g_pass
enum vpx_enc_pass g_pass
Multi-pass Encoding Mode.
Definition: vpx_encoder.h:360
vpx_codec_enc_cfg::g_h
unsigned int g_h
Height of the frame.
Definition: vpx_encoder.h:315
vpx_codec_err_t
vpx_codec_err_t
Algorithm return codes.
Definition: vpx_codec.h:93
VPX_RC_FIRST_PASS
@ VPX_RC_FIRST_PASS
Definition: vpx_encoder.h:228
VPX_FRAME_IS_KEY
#define VPX_FRAME_IS_KEY
Definition: vpx_encoder.h:118
vpx_fixed_buf::sz
size_t sz
Definition: vpx_encoder.h:100
vpx_codec_enc_cfg::rc_target_bitrate
unsigned int rc_target_bitrate
Target data rate.
Definition: vpx_encoder.h:462
vpx_fixed_buf::buf
void * buf
Definition: vpx_encoder.h:99
vpx_codec_cx_pkt::twopass_stats
vpx_fixed_buf_t twopass_stats
Definition: vpx_encoder.h:184
vpx_codec_encode
vpx_codec_err_t vpx_codec_encode(vpx_codec_ctx_t *ctx, const vpx_image_t *img, vpx_codec_pts_t pts, unsigned long duration, vpx_enc_frame_flags_t flags, unsigned long deadline)
Encode a frame.
vpx_codec_cx_pkt::frame
struct vpx_codec_cx_pkt::@1::@2 frame
vpx_rational::den
int den
Definition: vpx_encoder.h:222
vpx_codec_enc_config_default
vpx_codec_err_t vpx_codec_enc_config_default(vpx_codec_iface_t *iface, vpx_codec_enc_cfg_t *cfg, unsigned int usage)
Get a default configuration.
vpx_codec_destroy
vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx)
Destroy a codec instance.
vpx_encoder.h
Describes the encoder algorithm interface to applications.
VPX_DL_GOOD_QUALITY
#define VPX_DL_GOOD_QUALITY
deadline parameter analogous to VPx GOOD QUALITY mode.
Definition: vpx_encoder.h:980