Revert "Merge "SVC improvements""
This reverts commit a82001b1cfd688601bbda9b08b3d20e9b46b14d4, reversing
changes made to f6d870f7ae6f968bdbc342af80c8f71fa98f2ac2.
This commit breaks windows builds and needs some work to fix those and
some additional comments.
Change-Id: Ic0b0228e36704b127e5e399ce59db26182cfffe7
diff --git a/vp9_spatial_scalable_encoder.c b/vp9_spatial_scalable_encoder.c
index 847d9c4..8bb582f 100644
--- a/vp9_spatial_scalable_encoder.c
+++ b/vp9_spatial_scalable_encoder.c
@@ -23,22 +23,24 @@
#define VPX_CODEC_DISABLE_COMPAT 1
#include "vpx/vpx_encoder.h"
#include "vpx/vp8cx.h"
-#include "vpx/svc_context.h"
-
#define interface (vpx_codec_vp9_cx())
#define fourcc 0x30395056
#define IVF_FILE_HDR_SZ (32)
#define IVF_FRAME_HDR_SZ (12)
+#define NUM_BUFFERS 8
char *input_filename;
char *output_filename;
unsigned int number_frames_to_code = 60 * 60;
unsigned int number_frames_to_skip = 0;
-unsigned int gop_size = 100;
+unsigned int number_spatial_layers = 5;
+unsigned int key_period = 100;
-char *scaling_factor;
-char *quantizer;
-SVC_ENCODING_MODE encoding_mode = INTER_LAYER_PREDICTION_IP;
+typedef enum ENCODING_MODE {
+ INTER_LAYER_PREDICTION_I,
+ INTER_LAYER_PREDICTION_IP,
+ USE_GOLDEN_FRAME
+} ENCODING_MODE;
static void mem_put_le16(char *mem, unsigned int val) {
mem[0] = val;
@@ -55,10 +57,7 @@
static void usage(char *program_name) {
printf(
"Usage: %s [-f frames] [-s skip_frames] [-w width] [-h height] \n\t"
- "[-n rate_num] [-d rate_den] [-b bitrate] [-l layers] [-g gop_size] \n\t"
- "[-z dummy_frame (default 1) \n\t"
- "[-q quantizer (lowest to highest)] \n\t"
- "[-r 1/16th scale factor (lowest to highest layer)] "
+ "[-n rate_num] [-d rate_den] [-b bitrate] [-l layers] "
"<input_filename> <output_filename>\n",
basename(program_name));
exit(EXIT_FAILURE);
@@ -84,98 +83,78 @@
static int read_frame(FILE *f, vpx_image_t *img) {
size_t nbytes, to_read;
int res = 1;
- int plane;
- for (plane = 0; plane < 3; plane++) {
- unsigned char *ptr;
- int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
- int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
- int r;
-
- switch (plane) {
- case 1:
- ptr = img->planes[VPX_PLANE_U];
- break;
- case 2:
- ptr = img->planes[VPX_PLANE_V];
- break;
- default:
- ptr = img->planes[plane];
- }
- for (r = 0; r < h; r++) {
- to_read = w;
-
- nbytes = fread(ptr, 1, to_read, f);
- if (nbytes != to_read) {
- res = 0;
- if (nbytes > 0)
- printf("Warning: Read partial frame. Check your width & height!\n");
- break;
- }
- ptr += img->stride[plane];
- }
- if (!res) break;
+ to_read = img->w * img->h * 3 / 2;
+ nbytes = fread(img->planes[0], 1, to_read, f);
+ if (nbytes != to_read) {
+ res = 0;
+ if (nbytes > 0)
+ printf("Warning: Read partial frame. Check your width & height!\n");
}
return res;
}
-static int create_dummy_frame(vpx_image_t *img) {
- size_t buf_size;
- buf_size = img->w * img->h * 3 / 2;
- memset(img->planes[0], 129, buf_size);
+static int read_dummy_frame(vpx_image_t *img) {
+ size_t to_read;
+
+ to_read = img->w * img->h * 3 / 2;
+ memset(img->planes[0], 129, to_read);
return 1;
}
-static void write_ivf_file_header(FILE *outfile, unsigned int width,
- unsigned int height, int timebase_num,
- int timebase_den, int frame_cnt) {
+static void write_ivf_file_header(FILE *outfile, const vpx_codec_enc_cfg_t *cfg,
+ int frame_cnt) {
char header[32];
+ if (cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS) return;
header[0] = 'D';
header[1] = 'K';
header[2] = 'I';
header[3] = 'F';
- mem_put_le16(header + 4, 0); /* version */
- mem_put_le16(header + 6, 32); /* headersize */
- mem_put_le32(header + 8, fourcc); /* headersize */
- mem_put_le16(header + 12, width); /* width */
- mem_put_le16(header + 14, height); /* height */
- mem_put_le32(header + 16, timebase_den); /* rate */
- mem_put_le32(header + 20, timebase_num); /* scale */
- mem_put_le32(header + 24, frame_cnt); /* length */
- mem_put_le32(header + 28, 0); /* unused */
+ mem_put_le16(header + 4, 0); /* version */
+ mem_put_le16(header + 6, 32); /* headersize */
+ mem_put_le32(header + 8, fourcc); /* headersize */
+ mem_put_le16(header + 12, cfg->g_w); /* width */
+ mem_put_le16(header + 14, cfg->g_h); /* height */
+ mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
+ mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
+ mem_put_le32(header + 24, frame_cnt); /* length */
+ mem_put_le32(header + 28, 0); /* unused */
(void)fwrite(header, 1, 32, outfile);
}
-static void write_ivf_frame_header(FILE *outfile, vpx_codec_pts_t pts,
- size_t sz) {
+static void write_ivf_frame_header(FILE *outfile,
+ const vpx_codec_cx_pkt_t *pkt) {
char header[12];
- mem_put_le32(header, sz);
+ vpx_codec_pts_t pts;
+
+ if (pkt->kind != VPX_CODEC_CX_FRAME_PKT) return;
+
+ pts = pkt->data.frame.pts;
+ mem_put_le32(header, pkt->data.frame.sz);
mem_put_le32(header + 4, pts & 0xFFFFFFFF);
mem_put_le32(header + 8, pts >> 32);
(void)fwrite(header, 1, 12, outfile);
}
-static void parse_command_line(int argc, char **argv, SvcContext *svc_ctx,
- vpx_codec_enc_cfg_t *enc_cfg) {
+static void check_parameters() {
+ if (number_spatial_layers > 5) die("Cannot support more than 5 layers");
+}
+
+static void parse_command_line(int argc, char **argv,
+ vpx_codec_enc_cfg_t *cfg) {
unsigned int width = 1920;
unsigned int height = 1080;
unsigned int timebase_num = 1;
unsigned int timebase_den = 60;
unsigned int bitrate = 1000;
- unsigned int number_spatial_layers = 5;
- int use_dummy_frame = 1;
-
int c;
vpx_codec_err_t res;
- int r = 0;
- int q = 0;
opterr = 0;
- while ((c = getopt(argc, argv, "f:w:h:n:d:b:s:l:g:r:q:z:")) != -1)
- switch (c) {
+ while ((c = getopt(argc, argv, "f:w:h:n:d:b:s:l:p:")) != -1) switch (c) {
case 'f':
number_frames_to_code = atoi(optarg);
break;
@@ -200,17 +179,8 @@
case 'l':
number_spatial_layers = atoi(optarg);
break;
- case 'g':
- gop_size = atoi(optarg);
- break;
- case 'r':
- scaling_factor = optarg;
- break;
- case 'q':
- quantizer = optarg;
- break;
- case 'z':
- use_dummy_frame = atoi(optarg);
+ case 'p':
+ key_period = atoi(optarg);
break;
case '?':
usage(argv[0]);
@@ -221,28 +191,14 @@
usage(argv[0]);
}
- if ((r != 0 && q != 0 && r != q) || (r != number_spatial_layers && r != 0)) {
- usage(argv[0]);
- }
-
input_filename = argv[optind];
output_filename = argv[optind + 1];
if (width < 16 || width % 2 || height < 16 || height % 2)
die("Invalid resolution: %d x %d", width, height);
- // initialize SvcContext
- svc_ctx->log_level = SVC_LOG_DEBUG;
- svc_ctx->spatial_layers = number_spatial_layers;
- svc_ctx->encoding_mode = encoding_mode;
- svc_ctx->gop_size = gop_size;
- svc_ctx->quantizer_values = quantizer;
- svc_ctx->scale_factors = scaling_factor;
- // when using a dummy frame, that frame is only encoded to be full size
- svc_ctx->first_frame_full_size = use_dummy_frame;
-
/* Populate encoder configuration */
- res = vpx_codec_enc_config_default(interface, enc_cfg, 0);
+ res = vpx_codec_enc_config_default(interface, cfg, 0);
if (res) {
die("Failed to get config: %s\n", vpx_codec_err_to_string(res));
}
@@ -250,52 +206,200 @@
"Codec %s\nframes: %d, skip: %d, layers: %d\n"
"width %d, height: %d, \n"
"num: %d, den: %d, bitrate: %d, \n"
- "gop size: %d, use_dummy_frame: %d \n",
+ "key period: %d \n",
vpx_codec_iface_name(interface), number_frames_to_code,
number_frames_to_skip, number_spatial_layers, width, height, timebase_num,
- timebase_den, bitrate, gop_size, use_dummy_frame);
+ timebase_den, bitrate, key_period);
- enc_cfg->rc_target_bitrate = bitrate;
- enc_cfg->g_w = width;
- enc_cfg->g_h = height;
- enc_cfg->g_timebase.num = timebase_num;
- enc_cfg->g_timebase.den = timebase_den;
+ // Do minimal check at the application level. Encoder parameters will be
+ // checked internally
+ check_parameters();
+
+ cfg->rc_target_bitrate = bitrate;
+ cfg->g_w = width;
+ cfg->g_h = height;
+ cfg->g_timebase.num = timebase_num;
+ cfg->g_timebase.den = timebase_den;
+ cfg->ss_number_layers = number_spatial_layers;
+}
+
+static void set_default_configuration(vpx_codec_enc_cfg_t *cfg) {
+ /* Real time parameters */
+ cfg->rc_dropframe_thresh = 0;
+ cfg->rc_end_usage = VPX_CBR;
+ cfg->rc_resize_allowed = 0;
+ cfg->rc_min_quantizer = 33;
+ cfg->rc_max_quantizer = 33;
+ cfg->rc_undershoot_pct = 100;
+ cfg->rc_overshoot_pct = 15;
+ 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;
+
+ /* Disable automatic keyframe placement */
+ cfg->kf_mode = VPX_KF_DISABLED;
+ cfg->kf_min_dist = cfg->kf_max_dist = 3000;
+}
+
+static void initialize_codec(vpx_codec_ctx_t *codec, vpx_codec_enc_cfg_t *cfg) {
+ int max_intra_size_pct;
+
+ /* Initialize codec */
+ if (vpx_codec_enc_init(codec, interface, cfg, VPX_CODEC_USE_PSNR))
+ die_codec(codec, "Failed to initialize encoder");
+
+ vpx_codec_control(codec, VP9E_SET_SVC, 1);
+ /* Cap CPU & first I-frame size */
+ vpx_codec_control(codec, VP8E_SET_CPUUSED, 1);
+ vpx_codec_control(codec, VP8E_SET_STATIC_THRESHOLD, 1);
+ vpx_codec_control(codec, VP8E_SET_NOISE_SENSITIVITY, 1);
+ vpx_codec_control(codec, VP8E_SET_TOKEN_PARTITIONS, 1);
+
+ max_intra_size_pct =
+ (int)(((double)cfg->rc_buf_optimal_sz * 0.5) *
+ ((double)cfg->g_timebase.den / cfg->g_timebase.num) / 10.0);
+ /* printf ("max_intra_size_pct=%d\n", max_intra_size_pct); */
+
+ vpx_codec_control(codec, VP8E_SET_MAX_INTRA_BITRATE_PCT, max_intra_size_pct);
+}
+
+static int calculate_layer(int frame_cnt, int number_spatial_layers) {
+ if (frame_cnt == 0)
+ return 0;
+ else
+ return (frame_cnt + number_spatial_layers - 1) % number_spatial_layers;
+}
+
+static void switch_to_layer(int layer, unsigned int initial_width,
+ unsigned int initial_height,
+ vpx_codec_ctx_t *codec) {
+ // Set layer size
+ int scaling_factor_num[MAX_LAYERS] = {2, 1, 4, 2, 1};
+ int scaling_factor_den[MAX_LAYERS] = {9, 3, 9, 3, 1};
+
+ int quantizer[MAX_LAYERS] = {60, 53, 39, 33, 27};
+
+ unsigned int current_width;
+ unsigned int current_height;
+
+ current_width = initial_width *
+ scaling_factor_num[layer + 5 - number_spatial_layers] /
+ scaling_factor_den[layer + 5 - number_spatial_layers];
+ current_height = initial_height *
+ scaling_factor_num[layer + 5 - number_spatial_layers] /
+ scaling_factor_den[layer + 5 - number_spatial_layers];
+
+ current_width += current_width % 2;
+ current_height += current_height % 2;
+
+ vpx_codec_control(codec, VP9E_SET_WIDTH, ¤t_width);
+ vpx_codec_control(codec, VP9E_SET_HEIGHT, ¤t_height);
+
+ // Set layer context
+ vpx_codec_control(codec, VP9E_SET_LAYER, &layer);
+ vpx_codec_control(codec, VP9E_SET_MAX_Q,
+ quantizer[layer + 5 - number_spatial_layers]);
+ vpx_codec_control(codec, VP9E_SET_MIN_Q,
+ quantizer[layer + 5 - number_spatial_layers]);
+}
+
+static int get_flag(int is_I_frame_in_layer, int layer, ENCODING_MODE mode) {
+ // First layer
+ switch (mode) {
+ case INTER_LAYER_PREDICTION_I:
+ if (is_I_frame_in_layer && layer == 0) return VPX_EFLAG_FORCE_KF;
+ if (layer == 0)
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
+ VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF;
+ else if (is_I_frame_in_layer)
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
+ VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_LAST;
+ else
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
+ VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF;
+ break;
+
+ case INTER_LAYER_PREDICTION_IP:
+ if (is_I_frame_in_layer && layer == 0) return VPX_EFLAG_FORCE_KF;
+ if (layer == 0)
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
+ VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF;
+ else if (is_I_frame_in_layer)
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
+ VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_LAST;
+ else
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_REF_GF;
+ break;
+
+ case USE_GOLDEN_FRAME:
+ if (is_I_frame_in_layer && layer == 0) return VPX_EFLAG_FORCE_KF;
+ if (2 * number_spatial_layers - NUM_BUFFERS <= layer) {
+ if (layer == 0)
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
+ VP8_EFLAG_NO_REF_ARF;
+ else if (is_I_frame_in_layer)
+ return VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_REF_GF |
+ VP8_EFLAG_NO_REF_LAST;
+ else
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
+ } else {
+ if (layer == 0)
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
+ VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF;
+ else if (is_I_frame_in_layer)
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
+ VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_LAST;
+ else
+ return VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
+ VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF;
+ }
+ break;
+ default:
+ return VPX_EFLAG_FORCE_KF;
+ }
}
int main(int argc, char **argv) {
- FILE *infile, *outfile;
+ FILE *infile, *outfile[MAX_LAYERS];
vpx_codec_ctx_t codec;
- vpx_codec_enc_cfg_t enc_cfg;
- SvcContext svc_ctx;
- int i;
+ vpx_codec_enc_cfg_t cfg;
int frame_cnt = 0;
vpx_image_t raw;
+ int frame_avail = 1;
+ int got_data = 0;
+ int i;
+ int frames_in_layer[MAX_LAYERS] = {0};
clock_t before;
clock_t after;
- vpx_codec_err_t res;
int pts = 0; /* PTS starts at 0 */
int frame_duration = 1; /* 1 timebase tick per frame */
- memset(&svc_ctx, 0, sizeof(svc_ctx));
- svc_ctx.log_print = 1;
- parse_command_line(argc, argv, &svc_ctx, &enc_cfg);
+ parse_command_line(argc, argv, &cfg);
// Allocate image buffer
- if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, enc_cfg.g_w, enc_cfg.g_h, 32))
- die("Failed to allocate image", enc_cfg.g_w, enc_cfg.g_h);
+ if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, cfg.g_w, cfg.g_h, 32))
+ die("Failed to allocate image", cfg.g_w, cfg.g_h);
+ set_default_configuration(&cfg);
+
+ /* Open input file */
if (!(infile = fopen(input_filename, "rb")))
die("Failed to open %s for reading", argv[1]);
- if (!(outfile = fopen(output_filename, "wb")))
- die("Failed to open %s for writing", output_filename);
+ /* Open output file */
+ for (i = 0; i < number_spatial_layers; i++) {
+ char file_name[512];
+ snprintf(file_name, sizeof(file_name), "%s_%d.ivf", output_filename, i);
+ if (!(outfile[i] = fopen(file_name, "wb")))
+ die("Failed to open %s for writing", file_name);
+ write_ivf_file_header(outfile[i], &cfg, 0);
+ }
- // Initialize codec
- if (vpx_svc_init(&svc_ctx, &codec, interface, &enc_cfg) != VPX_CODEC_OK)
- die("Failed to initialize encoder");
-
- write_ivf_file_header(outfile, enc_cfg.g_w, enc_cfg.g_h,
- enc_cfg.g_timebase.num, enc_cfg.g_timebase.den, 0);
+ initialize_codec(&codec, &cfg);
// skip initial frames
for (i = 0; i < number_frames_to_skip; i++) {
@@ -303,47 +407,81 @@
}
before = clock();
- // Encode frames
- while (frame_cnt <= number_frames_to_code) {
- if (frame_cnt == 0 && svc_ctx.first_frame_full_size) {
- create_dummy_frame(&raw);
- } else {
- if (!read_frame(infile, &raw)) break;
+ // Encoding frames
+ while ((frame_avail || got_data) &&
+ frame_cnt <= number_frames_to_code * number_spatial_layers) {
+ int flags = 0;
+ vpx_codec_iter_t iter = NULL;
+ const vpx_codec_cx_pkt_t *pkt;
+
+ int layer = calculate_layer(frame_cnt, number_spatial_layers);
+ int is_I_frame_in_layer =
+ (((frame_cnt - 1) / number_spatial_layers % key_period) == 0);
+ int is_dummy = (frame_cnt == 0);
+
+ if (is_dummy) { // Dummy frame
+ flags = VPX_EFLAG_FORCE_KF;
+ frame_avail = read_dummy_frame(&raw);
+
+ } else { // Regular frame
+ // Read a new frame only at the base layer
+ if (layer == 0) frame_avail = read_frame(infile, &raw);
+ switch_to_layer(layer, cfg.g_w, cfg.g_h, &codec);
+ flags = get_flag(is_I_frame_in_layer, layer, INTER_LAYER_PREDICTION_I);
}
- res = vpx_svc_encode(&svc_ctx, &codec, &raw, pts, frame_duration,
- VPX_DL_REALTIME);
- printf("%s", svc_get_message(&svc_ctx));
- if (res != VPX_CODEC_OK) {
+
+ // Actual Encoding
+ if (vpx_codec_encode(&codec, frame_avail ? &raw : NULL, pts, 1, flags,
+ VPX_DL_REALTIME))
die_codec(&codec, "Failed to encode frame");
- }
- if (svc_get_frame_size(&svc_ctx) > 0) {
- write_ivf_frame_header(outfile, pts, svc_get_frame_size(&svc_ctx));
- (void)fwrite(svc_get_buffer(&svc_ctx), 1, svc_get_frame_size(&svc_ctx),
- outfile);
+
+ got_data = 0;
+ // Process data / Get PSNR statistics
+ while ((pkt = vpx_codec_get_cx_data(&codec, &iter))) {
+ got_data = 1;
+ switch (pkt->kind) {
+ case VPX_CODEC_CX_FRAME_PKT:
+ for (i = layer; i < number_spatial_layers; i++) {
+ write_ivf_frame_header(outfile[i], pkt);
+ (void)fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz,
+ outfile[i]);
+ frames_in_layer[i]++;
+ }
+ break;
+ case VPX_CODEC_PSNR_PKT:
+ if (frame_cnt != 0)
+ printf(
+ "Processed Frame %d, layer %d, PSNR(Total/Y/U/V): "
+ "%2.3f %2.3f %2.3f %2.3f \n",
+ (frame_cnt - 1) / number_spatial_layers + 1, layer,
+ pkt->data.psnr.psnr[0], pkt->data.psnr.psnr[1],
+ pkt->data.psnr.psnr[2], pkt->data.psnr.psnr[3]);
+ break;
+ default:
+ break;
+ }
}
frame_cnt++;
+ // TODO(ivan): Modify ts later if(!layer)
pts += frame_duration;
- } // end encode frames loop
+ }
+ // end while
after = clock();
printf("Processed %d frames in different resolutions in %ld ms.\n",
- frame_cnt - svc_ctx.first_frame_full_size,
- (int)(after - before) / (CLOCKS_PER_SEC / 1000));
+ frame_cnt - 1, (int)(after - before) / (CLOCKS_PER_SEC / 1000));
fclose(infile);
+
if (vpx_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec");
- // rewrite the output file headers with the actual frame count
- if (!fseek(outfile, 0, SEEK_SET)) {
- write_ivf_file_header(outfile, enc_cfg.g_w, enc_cfg.g_h,
- enc_cfg.g_timebase.num, enc_cfg.g_timebase.den,
- frame_cnt);
+ /* Try to rewrite the output file headers with the actual frame count */
+ for (i = 0; i < number_spatial_layers; i++) {
+ if (!fseek(outfile[i], 0, SEEK_SET)) {
+ write_ivf_file_header(outfile[i], &cfg, frames_in_layer[i]);
+ }
+ fclose(outfile[i]);
}
- fclose(outfile);
-
- // display average size, psnr
- svc_dump_statistics(&svc_ctx);
- printf("%s", svc_get_message(&svc_ctx));
return EXIT_SUCCESS;
}
