static gboolean
gst_av1_enc_set_format (GstVideoEncoder * encoder, GstVideoCodecState * state)
{
GstVideoCodecState *output_state;
GstAV1Enc *av1enc = GST_AV1_ENC_CAST (encoder);
GstAV1EncClass *av1enc_class = GST_AV1_ENC_GET_CLASS (av1enc);
output_state =
gst_video_encoder_set_output_state (encoder,
gst_pad_get_pad_template_caps (GST_VIDEO_ENCODER_SRC_PAD (encoder)),
state);
gst_video_codec_state_unref (output_state);
if (av1enc->input_state) {
gst_video_codec_state_unref (av1enc->input_state);
}
av1enc->input_state = gst_video_codec_state_ref (state);
g_mutex_lock (&av1enc->encoder_lock);
if (aom_codec_enc_config_default (av1enc_class->codec_algo, &av1enc->aom_cfg,
0)) {
gst_av1_codec_error (&av1enc->encoder,
"Failed to get default codec config.");
return FALSE;
}
GST_DEBUG_OBJECT (av1enc, "Got default encoder config");
gst_av1_enc_debug_encoder_cfg (&av1enc->aom_cfg);
gst_av1_enc_set_latency (av1enc);
av1enc->aom_cfg.g_w = av1enc->input_state->info.width;
av1enc->aom_cfg.g_h = av1enc->input_state->info.height;
av1enc->aom_cfg.g_timebase.num = av1enc->input_state->info.fps_d;
av1enc->aom_cfg.g_timebase.den = av1enc->input_state->info.fps_n;
/* FIXME : Make configuration properties */
av1enc->aom_cfg.rc_target_bitrate = 3000;
av1enc->aom_cfg.g_error_resilient = AOM_ERROR_RESILIENT_DEFAULT;
GST_DEBUG_OBJECT (av1enc, "Calling encoder init with config:");
gst_av1_enc_debug_encoder_cfg (&av1enc->aom_cfg);
if (aom_codec_enc_init (&av1enc->encoder, av1enc_class->codec_algo,
&av1enc->aom_cfg, 0)) {
gst_av1_codec_error (&av1enc->encoder, "Failed to initialize encoder");
return FALSE;
}
av1enc->encoder_inited = TRUE;
GST_AV1_ENC_APPLY_CODEC_CONTROL (av1enc, AOME_SET_CPUUSED, av1enc->cpu_used);
g_mutex_unlock (&av1enc->encoder_lock);
return TRUE;
}
static void pass1(aom_image_t *raw, FILE *infile, const char *outfile_name,
const AvxInterface *encoder, aom_codec_enc_cfg_t *cfg,
int lf_width, int lf_height, int lf_blocksize, int flags,
aom_image_t *raw_shift) {
AvxVideoInfo info = { encoder->fourcc,
cfg->g_w,
cfg->g_h,
{ cfg->g_timebase.num, cfg->g_timebase.den },
0 };
AvxVideoWriter *writer = NULL;
aom_codec_ctx_t codec;
int frame_count = 0;
int image_size_bytes = aom_img_size_bytes(raw);
int bu, bv;
int u_blocks, v_blocks;
aom_image_t *frame_to_encode;
aom_image_t reference_images[MAX_EXTERNAL_REFERENCES];
int reference_image_num = 0;
int i;
writer = aom_video_writer_open(outfile_name, kContainerIVF, &info);
if (!writer) die("Failed to open %s for writing", outfile_name);
if (aom_codec_enc_init(&codec, encoder->codec_interface(), cfg, flags))
die_codec(&codec, "Failed to initialize encoder");
if (aom_codec_control(&codec, AOME_SET_ENABLEAUTOALTREF, 0))
die_codec(&codec, "Failed to turn off auto altref");
if (aom_codec_control(&codec, AV1E_SET_FRAME_PARALLEL_DECODING, 0))
die_codec(&codec, "Failed to set frame parallel decoding");
// Note: The superblock is a sequence parameter and has to be the same for 1
// sequence. In lightfield application, must choose the superblock size(either
// 64x64 or 128x128) before the encoding starts. Otherwise, the default is
// AOM_SUPERBLOCK_SIZE_DYNAMIC, and the superblock size will be set to 64x64
// internally.
if (aom_codec_control(&codec, AV1E_SET_SUPERBLOCK_SIZE,
AOM_SUPERBLOCK_SIZE_64X64))
die_codec(&codec, "Failed to set SB size");
u_blocks = (lf_width + lf_blocksize - 1) / lf_blocksize;
v_blocks = (lf_height + lf_blocksize - 1) / lf_blocksize;
reference_image_num = u_blocks * v_blocks;
aom_img_fmt_t ref_fmt = AOM_IMG_FMT_I420;
if (!CONFIG_LOWBITDEPTH) ref_fmt |= AOM_IMG_FMT_HIGHBITDEPTH;
// Allocate memory with the border so that it can be used as a reference.
for (i = 0; i < reference_image_num; i++) {
if (!aom_img_alloc_with_border(&reference_images[i], ref_fmt, cfg->g_w,
cfg->g_h, 32, 8, AOM_BORDER_IN_PIXELS)) {
die("Failed to allocate image.");
}
}
printf("\n Second pass: "******"Encoding Reference Images\n");
for (bv = 0; bv < v_blocks; ++bv) {
for (bu = 0; bu < u_blocks; ++bu) {
const int block_u_min = bu * lf_blocksize;
const int block_v_min = bv * lf_blocksize;
int block_u_end = (bu + 1) * lf_blocksize;
int block_v_end = (bv + 1) * lf_blocksize;
int u_block_size, v_block_size;
int block_ref_u, block_ref_v;
block_u_end = block_u_end < lf_width ? block_u_end : lf_width;
block_v_end = block_v_end < lf_height ? block_v_end : lf_height;
u_block_size = block_u_end - block_u_min;
v_block_size = block_v_end - block_v_min;
block_ref_u = block_u_min + u_block_size / 2;
block_ref_v = block_v_min + v_block_size / 2;
printf("A%d, ", (block_ref_u + block_ref_v * lf_width));
fseek(infile, (block_ref_u + block_ref_v * lf_width) * image_size_bytes,
SEEK_SET);
aom_img_read(raw, infile);
get_raw_image(&frame_to_encode, raw, raw_shift);
// Reference frames may be encoded without tiles.
++frame_count;
printf("Encoding reference image %d of %d\n", bv * u_blocks + bu,
u_blocks * v_blocks);
encode_frame(&codec, frame_to_encode, frame_count, 1,
AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 |
AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF |
AOM_EFLAG_NO_REF_BWD | AOM_EFLAG_NO_REF_ARF2 |
AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF |
AOM_EFLAG_NO_UPD_ARF | AOM_EFLAG_NO_UPD_ENTROPY,
writer);
if (aom_codec_control(&codec, AV1_COPY_NEW_FRAME_IMAGE,
&reference_images[frame_count - 1]))
die_codec(&codec, "Failed to copy decoder reference frame");
}
}
cfg->large_scale_tile = 1;
// Fixed q encoding for camera frames.
cfg->rc_end_usage = AOM_Q;
if (aom_codec_enc_config_set(&codec, cfg))
die_codec(&codec, "Failed to configure encoder");
// The fixed q value used in encoding.
if (aom_codec_control(&codec, AOME_SET_CQ_LEVEL, 36))
die_codec(&codec, "Failed to set cq level");
if (aom_codec_control(&codec, AV1E_SET_FRAME_PARALLEL_DECODING, 1))
die_codec(&codec, "Failed to set frame parallel decoding");
if (aom_codec_control(&codec, AV1E_SET_SINGLE_TILE_DECODING, 1))
die_codec(&codec, "Failed to turn on single tile decoding");
// Set tile_columns and tile_rows to MAX values, which guarantees the tile
// size of 64 x 64 pixels(i.e. 1 SB) for <= 4k resolution.
if (aom_codec_control(&codec, AV1E_SET_TILE_COLUMNS, 6))
die_codec(&codec, "Failed to set tile width");
if (aom_codec_control(&codec, AV1E_SET_TILE_ROWS, 6))
die_codec(&codec, "Failed to set tile height");
for (bv = 0; bv < v_blocks; ++bv) {
for (bu = 0; bu < u_blocks; ++bu) {
const int block_u_min = bu * lf_blocksize;
const int block_v_min = bv * lf_blocksize;
int block_u_end = (bu + 1) * lf_blocksize;
int block_v_end = (bv + 1) * lf_blocksize;
int u, v;
block_u_end = block_u_end < lf_width ? block_u_end : lf_width;
block_v_end = block_v_end < lf_height ? block_v_end : lf_height;
for (v = block_v_min; v < block_v_end; ++v) {
for (u = block_u_min; u < block_u_end; ++u) {
av1_ref_frame_t ref;
ref.idx = 0;
ref.use_external_ref = 1;
ref.img = reference_images[bv * u_blocks + bu];
if (aom_codec_control(&codec, AV1_SET_REFERENCE, &ref))
die_codec(&codec, "Failed to set reference frame");
printf("C%d, ", (u + v * lf_width));
fseek(infile, (u + v * lf_width) * image_size_bytes, SEEK_SET);
aom_img_read(raw, infile);
get_raw_image(&frame_to_encode, raw, raw_shift);
++frame_count;
printf("Encoding image %d of %d\n",
frame_count - (u_blocks * v_blocks), lf_width * lf_height);
encode_frame(&codec, frame_to_encode, frame_count, 1,
AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 |
AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF |
AOM_EFLAG_NO_REF_BWD | AOM_EFLAG_NO_REF_ARF2 |
AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF |
AOM_EFLAG_NO_UPD_ARF | AOM_EFLAG_NO_UPD_ENTROPY,
writer);
}
}
}
}
// Flush encoder.
// No ARF, this should not be needed.
while (encode_frame(&codec, NULL, -1, 1, 0, writer)) {
}
for (i = 0; i < reference_image_num; i++) aom_img_free(&reference_images[i]);
if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec.");
aom_video_writer_close(writer);
printf("\nSecond pass complete. Processed %d frames.\n", frame_count);
}
static aom_fixed_buf_t pass0(aom_image_t *raw, FILE *infile,
const AvxInterface *encoder,
const aom_codec_enc_cfg_t *cfg, int lf_width,
int lf_height, int lf_blocksize, int flags,
aom_image_t *raw_shift) {
aom_codec_ctx_t codec;
int frame_count = 0;
int image_size_bytes = aom_img_size_bytes(raw);
int u_blocks, v_blocks;
int bu, bv;
aom_fixed_buf_t stats = { NULL, 0 };
aom_image_t *frame_to_encode;
if (aom_codec_enc_init(&codec, encoder->codec_interface(), cfg, flags))
die_codec(&codec, "Failed to initialize encoder");
if (aom_codec_control(&codec, AOME_SET_ENABLEAUTOALTREF, 0))
die_codec(&codec, "Failed to turn off auto altref");
if (aom_codec_control(&codec, AV1E_SET_FRAME_PARALLEL_DECODING, 0))
die_codec(&codec, "Failed to set frame parallel decoding");
// How many reference images we need to encode.
u_blocks = (lf_width + lf_blocksize - 1) / lf_blocksize;
v_blocks = (lf_height + lf_blocksize - 1) / lf_blocksize;
printf("\n First pass: "******"A%d, ", (block_ref_u + block_ref_v * lf_width));
fseek(infile, (block_ref_u + block_ref_v * lf_width) * image_size_bytes,
SEEK_SET);
aom_img_read(raw, infile);
get_raw_image(&frame_to_encode, raw, raw_shift);
// Reference frames can be encoded encoded without tiles.
++frame_count;
get_frame_stats(&codec, frame_to_encode, frame_count, 1,
AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 |
AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF |
AOM_EFLAG_NO_REF_BWD | AOM_EFLAG_NO_REF_ARF2 |
AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF |
AOM_EFLAG_NO_UPD_ARF,
&stats);
}
}
if (aom_codec_control(&codec, AV1E_SET_FRAME_PARALLEL_DECODING, 1))
die_codec(&codec, "Failed to set frame parallel decoding");
for (bv = 0; bv < v_blocks; ++bv) {
for (bu = 0; bu < u_blocks; ++bu) {
const int block_u_min = bu * lf_blocksize;
const int block_v_min = bv * lf_blocksize;
int block_u_end = (bu + 1) * lf_blocksize;
int block_v_end = (bv + 1) * lf_blocksize;
int u, v;
block_u_end = block_u_end < lf_width ? block_u_end : lf_width;
block_v_end = block_v_end < lf_height ? block_v_end : lf_height;
for (v = block_v_min; v < block_v_end; ++v) {
for (u = block_u_min; u < block_u_end; ++u) {
printf("C%d, ", (u + v * lf_width));
fseek(infile, (u + v * lf_width) * image_size_bytes, SEEK_SET);
aom_img_read(raw, infile);
get_raw_image(&frame_to_encode, raw, raw_shift);
++frame_count;
get_frame_stats(&codec, frame_to_encode, frame_count, 1,
AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 |
AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF |
AOM_EFLAG_NO_REF_BWD | AOM_EFLAG_NO_REF_ARF2 |
AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF |
AOM_EFLAG_NO_UPD_ARF | AOM_EFLAG_NO_UPD_ENTROPY,
&stats);
}
}
}
}
// Flush encoder.
// No ARF, this should not be needed.
while (get_frame_stats(&codec, NULL, frame_count, 1, 0, &stats)) {
}
if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec.");
printf("\nFirst pass complete. Processed %d frames.\n", frame_count);
return stats;
}
// TODO(tomfinegan): Improve command line parsing and add args for bitrate/fps.
int main(int argc, char **argv) {
FILE *infile = NULL;
aom_codec_ctx_t codec;
aom_codec_enc_cfg_t cfg;
int frame_count = 0;
aom_image_t raw;
aom_codec_err_t res;
AvxVideoInfo info;
AvxVideoWriter *writer = NULL;
const AvxInterface *encoder = NULL;
const int fps = 30;
const int bitrate = 200;
int keyframe_interval = 0;
int max_frames = 0;
int frames_encoded = 0;
const char *codec_arg = NULL;
const char *width_arg = NULL;
const char *height_arg = NULL;
const char *infile_arg = NULL;
const char *outfile_arg = NULL;
const char *keyframe_interval_arg = NULL;
exec_name = argv[0];
// Clear explicitly, as simply assigning "{ 0 }" generates
// "missing-field-initializers" warning in some compilers.
memset(&info, 0, sizeof(info));
if (argc != 9) die("Invalid number of arguments");
codec_arg = argv[1];
width_arg = argv[2];
height_arg = argv[3];
infile_arg = argv[4];
outfile_arg = argv[5];
keyframe_interval_arg = argv[6];
max_frames = (int)strtol(argv[8], NULL, 0);
encoder = get_aom_encoder_by_name(codec_arg);
if (!encoder) die("Unsupported codec.");
info.codec_fourcc = encoder->fourcc;
info.frame_width = (int)strtol(width_arg, NULL, 0);
info.frame_height = (int)strtol(height_arg, NULL, 0);
info.time_base.numerator = 1;
info.time_base.denominator = fps;
if (info.frame_width <= 0 || info.frame_height <= 0 ||
(info.frame_width % 2) != 0 || (info.frame_height % 2) != 0) {
die("Invalid frame size: %dx%d", info.frame_width, info.frame_height);
}
if (!aom_img_alloc(&raw, AOM_IMG_FMT_I420, info.frame_width,
info.frame_height, 1)) {
die("Failed to allocate image.");
}
keyframe_interval = (int)strtol(keyframe_interval_arg, NULL, 0);
if (keyframe_interval < 0) die("Invalid keyframe interval value.");
printf("Using %s\n", aom_codec_iface_name(encoder->codec_interface()));
res = aom_codec_enc_config_default(encoder->codec_interface(), &cfg, 0);
if (res) die_codec(&codec, "Failed to get default codec config.");
cfg.g_w = info.frame_width;
cfg.g_h = info.frame_height;
cfg.g_timebase.num = info.time_base.numerator;
cfg.g_timebase.den = info.time_base.denominator;
cfg.rc_target_bitrate = bitrate;
cfg.g_error_resilient = (aom_codec_er_flags_t)strtoul(argv[7], NULL, 0);
writer = aom_video_writer_open(outfile_arg, kContainerIVF, &info);
if (!writer) die("Failed to open %s for writing.", outfile_arg);
if (!(infile = fopen(infile_arg, "rb")))
die("Failed to open %s for reading.", infile_arg);
if (aom_codec_enc_init(&codec, encoder->codec_interface(), &cfg, 0))
die_codec(&codec, "Failed to initialize encoder");
// Encode frames.
while (aom_img_read(&raw, infile)) {
int flags = 0;
if (keyframe_interval > 0 && frame_count % keyframe_interval == 0)
flags |= AOM_EFLAG_FORCE_KF;
encode_frame(&codec, &raw, frame_count++, flags, writer);
frames_encoded++;
if (max_frames > 0 && frames_encoded >= max_frames) break;
}
// Flush encoder.
while (encode_frame(&codec, NULL, -1, 0, writer)) continue;
printf("\n");
fclose(infile);
printf("Processed %d frames.\n", frame_count);
aom_img_free(&raw);
if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec.");
aom_video_writer_close(writer);
return EXIT_SUCCESS;
}