summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--examples.mk14
-rw-r--r--examples/vp8_multi_resolution_encoder.c8
-rw-r--r--examples/vp9_spatial_svc_encoder.c32
-rw-r--r--examples/vpx_temporal_svc_encoder.c29
-rwxr-xr-xtest/tools_common.sh2
-rwxr-xr-xtest/vpx_temporal_svc_encoder.sh7
-rw-r--r--tools_common.c90
-rw-r--r--tools_common.h8
-rw-r--r--vp9/encoder/vp9_encoder.c101
-rw-r--r--vp9/encoder/vp9_pickmode.c12
-rw-r--r--vpxdec.c8
-rw-r--r--vpxenc.c85
12 files changed, 262 insertions, 134 deletions
diff --git a/examples.mk b/examples.mk
index a1d4eb68c..e0da4caa2 100644
--- a/examples.mk
+++ b/examples.mk
@@ -72,6 +72,7 @@ vpxdec.SRCS += vpx_ports/vpx_timer.h
vpxdec.SRCS += vpx/vpx_integer.h
vpxdec.SRCS += args.c args.h
vpxdec.SRCS += ivfdec.c ivfdec.h
+vpxdec.SRCS += y4minput.c y4minput.h
vpxdec.SRCS += tools_common.c tools_common.h
vpxdec.SRCS += y4menc.c y4menc.h
ifeq ($(CONFIG_LIBYUV),yes)
@@ -113,6 +114,7 @@ vpxenc.DESCRIPTION = Full featured encoder
EXAMPLES-$(CONFIG_VP9_ENCODER) += vp9_spatial_svc_encoder.c
vp9_spatial_svc_encoder.SRCS += args.c args.h
vp9_spatial_svc_encoder.SRCS += ivfenc.c ivfenc.h
+vp9_spatial_svc_encoder.SRCS += y4minput.c y4minput.h
vp9_spatial_svc_encoder.SRCS += tools_common.c tools_common.h
vp9_spatial_svc_encoder.SRCS += video_common.h
vp9_spatial_svc_encoder.SRCS += video_writer.h video_writer.c
@@ -129,6 +131,7 @@ endif
EXAMPLES-$(CONFIG_ENCODERS) += vpx_temporal_svc_encoder.c
vpx_temporal_svc_encoder.SRCS += ivfenc.c ivfenc.h
+vpx_temporal_svc_encoder.SRCS += y4minput.c y4minput.h
vpx_temporal_svc_encoder.SRCS += tools_common.c tools_common.h
vpx_temporal_svc_encoder.SRCS += video_common.h
vpx_temporal_svc_encoder.SRCS += video_writer.h video_writer.c
@@ -138,6 +141,7 @@ vpx_temporal_svc_encoder.DESCRIPTION = Temporal SVC Encoder
EXAMPLES-$(CONFIG_DECODERS) += simple_decoder.c
simple_decoder.GUID = D3BBF1E9-2427-450D-BBFF-B2843C1D44CC
simple_decoder.SRCS += ivfdec.h ivfdec.c
+simple_decoder.SRCS += y4minput.c y4minput.h
simple_decoder.SRCS += tools_common.h tools_common.c
simple_decoder.SRCS += video_common.h
simple_decoder.SRCS += video_reader.h video_reader.c
@@ -147,6 +151,7 @@ simple_decoder.SRCS += vpx_ports/msvc.h
simple_decoder.DESCRIPTION = Simplified decoder loop
EXAMPLES-$(CONFIG_DECODERS) += postproc.c
postproc.SRCS += ivfdec.h ivfdec.c
+postproc.SRCS += y4minput.c y4minput.h
postproc.SRCS += tools_common.h tools_common.c
postproc.SRCS += video_common.h
postproc.SRCS += video_reader.h video_reader.c
@@ -158,6 +163,7 @@ postproc.DESCRIPTION = Decoder postprocessor control
EXAMPLES-$(CONFIG_DECODERS) += decode_to_md5.c
decode_to_md5.SRCS += md5_utils.h md5_utils.c
decode_to_md5.SRCS += ivfdec.h ivfdec.c
+decode_to_md5.SRCS += y4minput.c y4minput.h
decode_to_md5.SRCS += tools_common.h tools_common.c
decode_to_md5.SRCS += video_common.h
decode_to_md5.SRCS += video_reader.h video_reader.c
@@ -168,6 +174,7 @@ decode_to_md5.GUID = 59120B9B-2735-4BFE-B022-146CA340FE42
decode_to_md5.DESCRIPTION = Frame by frame MD5 checksum
EXAMPLES-$(CONFIG_ENCODERS) += simple_encoder.c
simple_encoder.SRCS += ivfenc.h ivfenc.c
+simple_encoder.SRCS += y4minput.c y4minput.h
simple_encoder.SRCS += tools_common.h tools_common.c
simple_encoder.SRCS += video_common.h
simple_encoder.SRCS += video_writer.h video_writer.c
@@ -176,6 +183,7 @@ simple_encoder.GUID = 4607D299-8A71-4D2C-9B1D-071899B6FBFD
simple_encoder.DESCRIPTION = Simplified encoder loop
EXAMPLES-$(CONFIG_VP9_ENCODER) += vp9_lossless_encoder.c
vp9_lossless_encoder.SRCS += ivfenc.h ivfenc.c
+vp9_lossless_encoder.SRCS += y4minput.c y4minput.h
vp9_lossless_encoder.SRCS += tools_common.h tools_common.c
vp9_lossless_encoder.SRCS += video_common.h
vp9_lossless_encoder.SRCS += video_writer.h video_writer.c
@@ -184,6 +192,7 @@ vp9_lossless_encoder.GUID = B63C7C88-5348-46DC-A5A6-CC151EF93366
vp9_lossless_encoder.DESCRIPTION = Simplified lossless VP9 encoder
EXAMPLES-$(CONFIG_ENCODERS) += twopass_encoder.c
twopass_encoder.SRCS += ivfenc.h ivfenc.c
+twopass_encoder.SRCS += y4minput.c y4minput.h
twopass_encoder.SRCS += tools_common.h tools_common.c
twopass_encoder.SRCS += video_common.h
twopass_encoder.SRCS += video_writer.h video_writer.c
@@ -192,6 +201,7 @@ twopass_encoder.GUID = 73494FA6-4AF9-4763-8FBB-265C92402FD8
twopass_encoder.DESCRIPTION = Two-pass encoder loop
EXAMPLES-$(CONFIG_DECODERS) += decode_with_drops.c
decode_with_drops.SRCS += ivfdec.h ivfdec.c
+decode_with_drops.SRCS += y4minput.c y4minput.h
decode_with_drops.SRCS += tools_common.h tools_common.c
decode_with_drops.SRCS += video_common.h
decode_with_drops.SRCS += video_reader.h video_reader.c
@@ -202,6 +212,7 @@ decode_with_drops.GUID = CE5C53C4-8DDA-438A-86ED-0DDD3CDB8D26
decode_with_drops.DESCRIPTION = Drops frames while decoding
EXAMPLES-$(CONFIG_ENCODERS) += set_maps.c
set_maps.SRCS += ivfenc.h ivfenc.c
+set_maps.SRCS += y4minput.c y4minput.h
set_maps.SRCS += tools_common.h tools_common.c
set_maps.SRCS += video_common.h
set_maps.SRCS += video_writer.h video_writer.c
@@ -210,6 +221,7 @@ set_maps.GUID = ECB2D24D-98B8-4015-A465-A4AF3DCC145F
set_maps.DESCRIPTION = Set active and ROI maps
EXAMPLES-$(CONFIG_VP8_ENCODER) += vp8cx_set_ref.c
vp8cx_set_ref.SRCS += ivfenc.h ivfenc.c
+vp8cx_set_ref.SRCS += y4minput.c y4minput.h
vp8cx_set_ref.SRCS += tools_common.h tools_common.c
vp8cx_set_ref.SRCS += video_common.h
vp8cx_set_ref.SRCS += video_writer.h video_writer.c
@@ -221,6 +233,7 @@ ifeq ($(CONFIG_VP9_ENCODER),yes)
ifeq ($(CONFIG_DECODERS),yes)
EXAMPLES-yes += vp9cx_set_ref.c
vp9cx_set_ref.SRCS += ivfenc.h ivfenc.c
+vp9cx_set_ref.SRCS += y4minput.c y4minput.h
vp9cx_set_ref.SRCS += tools_common.h tools_common.c
vp9cx_set_ref.SRCS += video_common.h
vp9cx_set_ref.SRCS += video_writer.h video_writer.c
@@ -233,6 +246,7 @@ ifeq ($(CONFIG_MULTI_RES_ENCODING),yes)
ifeq ($(CONFIG_LIBYUV),yes)
EXAMPLES-$(CONFIG_VP8_ENCODER) += vp8_multi_resolution_encoder.c
vp8_multi_resolution_encoder.SRCS += ivfenc.h ivfenc.c
+vp8_multi_resolution_encoder.SRCS += y4minput.c y4minput.h
vp8_multi_resolution_encoder.SRCS += tools_common.h tools_common.c
vp8_multi_resolution_encoder.SRCS += video_writer.h video_writer.c
vp8_multi_resolution_encoder.SRCS += vpx_ports/msvc.h
diff --git a/examples/vp8_multi_resolution_encoder.c b/examples/vp8_multi_resolution_encoder.c
index b14b1ff39..e72f8a019 100644
--- a/examples/vp8_multi_resolution_encoder.c
+++ b/examples/vp8_multi_resolution_encoder.c
@@ -61,7 +61,7 @@ void usage_exit(void) { exit(EXIT_FAILURE); }
int (*read_frame_p)(FILE *f, vpx_image_t *img);
-static int read_frame(FILE *f, vpx_image_t *img) {
+static int mulres_read_frame(FILE *f, vpx_image_t *img) {
size_t nbytes, to_read;
int res = 1;
@@ -75,7 +75,7 @@ static int read_frame(FILE *f, vpx_image_t *img) {
return res;
}
-static int read_frame_by_row(FILE *f, vpx_image_t *img) {
+static int mulres_read_frame_by_row(FILE *f, vpx_image_t *img) {
size_t nbytes, to_read;
int res = 1;
int plane;
@@ -471,9 +471,9 @@ int main(int argc, char **argv) {
die("Failed to allocate image", cfg[i].g_w, cfg[i].g_h);
if (raw[0].stride[VPX_PLANE_Y] == (int)raw[0].d_w)
- read_frame_p = read_frame;
+ read_frame_p = mulres_read_frame;
else
- read_frame_p = read_frame_by_row;
+ read_frame_p = mulres_read_frame_by_row;
for (i = 0; i < NUM_ENCODERS; i++)
if (outfile[i]) write_ivf_file_header(outfile[i], &cfg[i], 0);
diff --git a/examples/vp9_spatial_svc_encoder.c b/examples/vp9_spatial_svc_encoder.c
index f8093e1bf..92b310684 100644
--- a/examples/vp9_spatial_svc_encoder.c
+++ b/examples/vp9_spatial_svc_encoder.c
@@ -30,6 +30,8 @@
#include "vpx/vpx_encoder.h"
#include "../vpxstats.h"
#include "vp9/encoder/vp9_encoder.h"
+#include "./y4minput.h"
+
#define OUTPUT_RC_STATS 1
static const arg_def_t skip_frames_arg =
@@ -161,7 +163,6 @@ static const int32_t default_speed = -1; // -1 means use library default.
static const uint32_t default_threads = 0; // zero means use library default.
typedef struct {
- const char *input_filename;
const char *output_filename;
uint32_t frames_to_code;
uint32_t frames_to_skip;
@@ -393,10 +394,16 @@ static void parse_command_line(int argc, const char **argv_,
if (argv[0] == NULL || argv[1] == 0) {
usage_exit();
}
- app_input->input_filename = argv[0];
+ app_input->input_ctx.filename = argv[0];
app_input->output_filename = argv[1];
free(argv);
+ open_input_file(&app_input->input_ctx);
+ if (app_input->input_ctx.file_type == FILE_TYPE_Y4M) {
+ enc_cfg->g_w = app_input->input_ctx.width;
+ enc_cfg->g_h = app_input->input_ctx.height;
+ }
+
if (enc_cfg->g_w < 16 || enc_cfg->g_w % 2 || enc_cfg->g_h < 16 ||
enc_cfg->g_h % 2)
die("Invalid resolution: %d x %d\n", enc_cfg->g_w, enc_cfg->g_h);
@@ -752,7 +759,6 @@ int main(int argc, const char **argv) {
vpx_codec_err_t res;
int pts = 0; /* PTS starts at 0 */
int frame_duration = 1; /* 1 timebase tick per frame */
- FILE *infile = NULL;
int end_of_stream = 0;
int frames_received = 0;
#if OUTPUT_RC_STATS
@@ -773,6 +779,13 @@ int main(int argc, const char **argv) {
memset(&layer_id, 0, sizeof(vpx_svc_layer_id_t));
memset(&rc, 0, sizeof(struct RateControlStats));
exec_name = argv[0];
+
+ /* Setup default input stream settings */
+ app_input.input_ctx.framerate.numerator = 30;
+ app_input.input_ctx.framerate.denominator = 1;
+ app_input.input_ctx.only_i420 = 1;
+ app_input.input_ctx.bit_depth = 0;
+
parse_command_line(argc, argv, &app_input, &svc_ctx, &enc_cfg);
// Allocate image buffer
@@ -789,9 +802,6 @@ int main(int argc, const char **argv) {
}
#endif // CONFIG_VP9_HIGHBITDEPTH
- if (!(infile = fopen(app_input.input_filename, "rb")))
- die("Failed to open %s for reading\n", app_input.input_filename);
-
// Initialize codec
if (vpx_svc_init(&svc_ctx, &codec, vpx_codec_vp9_cx(), &enc_cfg) !=
VPX_CODEC_OK)
@@ -835,7 +845,8 @@ int main(int argc, const char **argv) {
#endif
// skip initial frames
- for (i = 0; i < app_input.frames_to_skip; ++i) vpx_img_read(&raw, infile);
+ for (i = 0; i < app_input.frames_to_skip; ++i)
+ read_frame(&app_input.input_ctx, &raw);
if (svc_ctx.speed != -1)
vpx_codec_control(&codec, VP8E_SET_CPUUSED, svc_ctx.speed);
@@ -875,7 +886,8 @@ int main(int argc, const char **argv) {
// layers, with SL0 only has TL0, and SL1 has both TL0 and TL1. This example
// uses the extended API.
int example_pattern = 0;
- if (frame_cnt >= app_input.frames_to_code || !vpx_img_read(&raw, infile)) {
+ if (frame_cnt >= app_input.frames_to_code ||
+ !read_frame(&app_input.input_ctx, &raw)) {
// We need one extra vpx_svc_encode call at end of stream to flush
// encoder and get remaining data
end_of_stream = 1;
@@ -1091,7 +1103,9 @@ int main(int argc, const char **argv) {
}
printf("Processed %d frames\n", frame_cnt);
- fclose(infile);
+
+ close_input_file(&app_input.input_ctx);
+
#if OUTPUT_RC_STATS
if (svc_ctx.output_rc_stat) {
printout_rate_control_summary(&rc, &enc_cfg, frame_cnt);
diff --git a/examples/vpx_temporal_svc_encoder.c b/examples/vpx_temporal_svc_encoder.c
index f49ef7b1d..aa2213a5b 100644
--- a/examples/vpx_temporal_svc_encoder.c
+++ b/examples/vpx_temporal_svc_encoder.c
@@ -19,6 +19,7 @@
#include <string.h>
#include "./vpx_config.h"
+#include "./y4minput.h"
#include "../vpx_ports/vpx_timer.h"
#include "vpx/vp8cx.h"
#include "vpx/vpx_encoder.h"
@@ -594,7 +595,7 @@ int main(int argc, char **argv) {
#endif
vpx_svc_layer_id_t layer_id;
const VpxInterface *encoder = NULL;
- FILE *infile = NULL;
+ struct VpxInputContext input_ctx;
struct RateControlMetrics rc;
int64_t cx_time = 0;
const int min_args_base = 13;
@@ -611,6 +612,13 @@ int main(int argc, char **argv) {
zero(rc.layer_target_bitrate);
memset(&layer_id, 0, sizeof(vpx_svc_layer_id_t));
+ memset(&input_ctx, 0, sizeof(input_ctx));
+ /* Setup default input stream settings */
+ input_ctx.framerate.numerator = 30;
+ input_ctx.framerate.denominator = 1;
+ input_ctx.only_i420 = 1;
+ input_ctx.bit_depth = 0;
+
exec_name = argv[0];
// Check usage and arguments.
if (argc < min_args) {
@@ -754,9 +762,18 @@ int main(int argc, char **argv) {
// Set to layer_target_bitrate for highest layer (total bitrate).
cfg.rc_target_bitrate = rc.layer_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]);
+ input_ctx.filename = argv[1];
+ open_input_file(&input_ctx);
+
+ if (input_ctx.file_type == FILE_TYPE_Y4M) {
+ if (input_ctx.width != cfg.g_w || input_ctx.height != cfg.g_h) {
+ die("Incorrect width or height: %d x %d", cfg.g_w, cfg.g_h);
+ }
+ if (input_ctx.framerate.numerator != cfg.g_timebase.den ||
+ input_ctx.framerate.denominator != cfg.g_timebase.num) {
+ die("Incorrect framerate: numerator %d denominator %d",
+ cfg.g_timebase.num, cfg.g_timebase.den);
+ }
}
framerate = cfg.g_timebase.den / cfg.g_timebase.num;
@@ -865,7 +882,7 @@ int main(int argc, char **argv) {
}
flags = layer_flags[frame_cnt % flag_periodicity];
if (layering_mode == 0) flags = 0;
- frame_avail = vpx_img_read(&raw, infile);
+ frame_avail = read_frame(&input_ctx, &raw);
if (frame_avail) ++rc.layer_input_frames[layer_id.temporal_layer_id];
vpx_usec_timer_start(&timer);
if (vpx_codec_encode(&codec, frame_avail ? &raw : NULL, pts, 1, flags,
@@ -933,7 +950,7 @@ int main(int argc, char **argv) {
++frame_cnt;
pts += frame_duration;
}
- fclose(infile);
+ close_input_file(&input_ctx);
printout_rate_control_summary(&rc, &cfg, frame_cnt);
printf("\n");
printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f \n",
diff --git a/test/tools_common.sh b/test/tools_common.sh
index 7caad9f06..844a12534 100755
--- a/test/tools_common.sh
+++ b/test/tools_common.sh
@@ -412,6 +412,8 @@ YUV_RAW_INPUT_HEIGHT=288
Y4M_NOSQ_PAR_INPUT="${LIBVPX_TEST_DATA_PATH}/park_joy_90p_8_420_a10-1.y4m"
Y4M_720P_INPUT="${LIBVPX_TEST_DATA_PATH}/niklas_1280_720_30.y4m"
+Y4M_720P_INPUT_WIDTH=1280
+Y4M_720P_INPUT_HEIGHT=720
# Setup a trap function to clean up after tests complete.
trap cleanup EXIT
diff --git a/test/vpx_temporal_svc_encoder.sh b/test/vpx_temporal_svc_encoder.sh
index fd1045e7f..5e5bac8fa 100755
--- a/test/vpx_temporal_svc_encoder.sh
+++ b/test/vpx_temporal_svc_encoder.sh
@@ -38,6 +38,7 @@ vpx_tsvc_encoder() {
local output_file="${VPX_TEST_OUTPUT_DIR}/${output_file_base}"
local timebase_num="1"
local timebase_den="1000"
+ local timebase_den_y4m="30"
local speed="6"
local frame_drop_thresh="30"
local max_threads="4"
@@ -58,6 +59,12 @@ vpx_tsvc_encoder() {
"${YUV_RAW_INPUT_HEIGHT}" "${timebase_num}" "${timebase_den}" \
"${speed}" "${frame_drop_thresh}" "${error_resilient}" "${threads}" \
"$@" ${devnull}
+ # Test for y4m input.
+ eval "${VPX_TEST_PREFIX}" "${encoder}" "${Y4M_720P_INPUT}" \
+ "${output_file}" "${codec}" "${Y4M_720P_INPUT_WIDTH}" \
+ "${Y4M_720P_INPUT_HEIGHT}" "${timebase_num}" "${timebase_den_y4m}" \
+ "${speed}" "${frame_drop_thresh}" "${error_resilient}" "${threads}" \
+ "$@" ${devnull}
else
eval "${VPX_TEST_PREFIX}" "${encoder}" "${YUV_RAW_INPUT}" \
"${output_file}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
diff --git a/tools_common.c b/tools_common.c
index b1c35e815..2cd8a0704 100644
--- a/tools_common.c
+++ b/tools_common.c
@@ -46,6 +46,14 @@
va_end(ap); \
} while (0)
+#if CONFIG_ENCODERS
+/* Swallow warnings about unused results of fread/fwrite */
+static size_t wrap_fread(void *ptr, size_t size, size_t nmemb, FILE *stream) {
+ return fread(ptr, size, nmemb, stream);
+}
+#define fread wrap_fread
+#endif
+
FILE *set_binary_mode(FILE *stream) {
(void)stream;
#if defined(_WIN32) || defined(__OS2__)
@@ -264,6 +272,88 @@ double sse_to_psnr(double samples, double peak, double sse) {
}
}
+#if CONFIG_ENCODERS
+int read_frame(struct VpxInputContext *input_ctx, vpx_image_t *img) {
+ FILE *f = input_ctx->file;
+ y4m_input *y4m = &input_ctx->y4m;
+ int shortread = 0;
+
+ if (input_ctx->file_type == FILE_TYPE_Y4M) {
+ if (y4m_input_fetch_frame(y4m, f, img) < 1) return 0;
+ } else {
+ shortread = read_yuv_frame(input_ctx, img);
+ }
+
+ return !shortread;
+}
+
+int file_is_y4m(const char detect[4]) {
+ if (memcmp(detect, "YUV4", 4) == 0) {
+ return 1;
+ }
+ return 0;
+}
+
+int fourcc_is_ivf(const char detect[4]) {
+ if (memcmp(detect, "DKIF", 4) == 0) {
+ return 1;
+ }
+ return 0;
+}
+
+void open_input_file(struct VpxInputContext *input) {
+ /* Parse certain options from the input file, if possible */
+ input->file = strcmp(input->filename, "-") ? fopen(input->filename, "rb")
+ : set_binary_mode(stdin);
+
+ if (!input->file) fatal("Failed to open input file");
+
+ if (!fseeko(input->file, 0, SEEK_END)) {
+ /* Input file is seekable. Figure out how long it is, so we can get
+ * progress info.
+ */
+ input->length = ftello(input->file);
+ rewind(input->file);
+ }
+
+ /* Default to 1:1 pixel aspect ratio. */
+ input->pixel_aspect_ratio.numerator = 1;
+ input->pixel_aspect_ratio.denominator = 1;
+
+ /* For RAW input sources, these bytes will applied on the first frame
+ * in read_frame().
+ */
+ input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
+ input->detect.position = 0;
+
+ if (input->detect.buf_read == 4 && file_is_y4m(input->detect.buf)) {
+ if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4,
+ input->only_i420) >= 0) {
+ input->file_type = FILE_TYPE_Y4M;
+ input->width = input->y4m.pic_w;
+ input->height = input->y4m.pic_h;
+ input->pixel_aspect_ratio.numerator = input->y4m.par_n;
+ input->pixel_aspect_ratio.denominator = input->y4m.par_d;
+ input->framerate.numerator = input->y4m.fps_n;
+ input->framerate.denominator = input->y4m.fps_d;
+ input->fmt = input->y4m.vpx_fmt;
+ input->bit_depth = input->y4m.bit_depth;
+ } else {
+ fatal("Unsupported Y4M stream.");
+ }
+ } else if (input->detect.buf_read == 4 && fourcc_is_ivf(input->detect.buf)) {
+ fatal("IVF is not supported as input.");
+ } else {
+ input->file_type = FILE_TYPE_RAW;
+ }
+}
+
+void close_input_file(struct VpxInputContext *input) {
+ fclose(input->file);
+ if (input->file_type == FILE_TYPE_Y4M) y4m_input_close(&input->y4m);
+}
+#endif
+
// TODO(debargha): Consolidate the functions below into a separate file.
#if CONFIG_VP9_HIGHBITDEPTH
static void highbd_img_upshift(vpx_image_t *dst, vpx_image_t *src,
diff --git a/tools_common.h b/tools_common.h
index bb169351b..28695e4f1 100644
--- a/tools_common.h
+++ b/tools_common.h
@@ -152,6 +152,14 @@ int vpx_img_read(vpx_image_t *img, FILE *file);
double sse_to_psnr(double samples, double peak, double mse);
+#if CONFIG_ENCODERS
+int read_frame(struct VpxInputContext *input_ctx, vpx_image_t *img);
+int file_is_y4m(const char detect[4]);
+int fourcc_is_ivf(const char detect[4]);
+void open_input_file(struct VpxInputContext *input);
+void close_input_file(struct VpxInputContext *input);
+#endif
+
#if CONFIG_VP9_HIGHBITDEPTH
void vpx_img_upshift(vpx_image_t *dst, vpx_image_t *src, int input_shift);
void vpx_img_downshift(vpx_image_t *dst, vpx_image_t *src, int down_shift);
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index eaeb3d96e..b8c86ea43 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -2583,6 +2583,7 @@ void vp9_remove_compressor(VP9_COMP *cpi) {
for (sqr_bsize = 0; sqr_bsize < SQUARE_BLOCK_SIZES; ++sqr_bsize) {
vpx_free(cpi->tpl_stats[frame].pyramid_mv_arr[rf_idx][sqr_bsize]);
}
+ vpx_free(cpi->tpl_stats[frame].mv_mode_arr[rf_idx]);
}
#endif
vpx_free(cpi->tpl_stats[frame].tpl_stats_ptr);
@@ -6014,22 +6015,27 @@ static void mode_estimation(VP9_COMP *cpi, MACROBLOCK *x, MACROBLOCKD *xd,
}
#if CONFIG_NON_GREEDY_MV
-static void get_block_src_pred_buf(MACROBLOCKD *xd, GF_PICTURE *gf_picture,
- int frame_idx, int rf_idx, int mi_row,
- int mi_col, struct buf_2d *src,
- struct buf_2d *pre) {
+static int get_block_src_pred_buf(MACROBLOCKD *xd, GF_PICTURE *gf_picture,
+ int frame_idx, int rf_idx, int mi_row,
+ int mi_col, struct buf_2d *src,
+ struct buf_2d *pre) {
const int mb_y_offset =
mi_row * MI_SIZE * xd->cur_buf->y_stride + mi_col * MI_SIZE;
YV12_BUFFER_CONFIG *ref_frame = NULL;
int ref_frame_idx = gf_picture[frame_idx].ref_frame[rf_idx];
if (ref_frame_idx != -1) {
ref_frame = gf_picture[ref_frame_idx].frame;
+ src->buf = xd->cur_buf->y_buffer + mb_y_offset;
+ src->stride = xd->cur_buf->y_stride;
+ pre->buf = ref_frame->y_buffer + mb_y_offset;
+ pre->stride = ref_frame->y_stride;
+ assert(src->stride == pre->stride);
+ return 1;
+ } else {
+ printf("invalid ref_frame_idx");
+ assert(ref_frame_idx != -1);
+ return 0;
}
- src->buf = xd->cur_buf->y_buffer + mb_y_offset;
- src->stride = xd->cur_buf->y_stride;
- pre->buf = ref_frame->y_buffer + mb_y_offset;
- pre->stride = ref_frame->y_stride;
- assert(src->stride == pre->stride);
}
#define kMvPreCheckLines 5
@@ -6135,18 +6141,60 @@ static double get_mv_dist(int mv_mode, VP9_COMP *cpi, MACROBLOCKD *xd,
*mv = get_mv_from_mv_mode(mv_mode, cpi, tpl_frame, rf_idx, bsize, mi_row,
mi_col);
full_mv = get_full_mv(&mv->as_mv);
- get_block_src_pred_buf(xd, gf_picture, frame_idx, rf_idx, mi_row, mi_col,
- &src, &pre);
- // TODO(angiebird): Consider subpixel when computing the sse.
- cpi->fn_ptr[bsize].vf(src.buf, src.stride, get_buf_from_mv(&pre, &full_mv),
- pre.stride, &sse);
- return (double)sse;
+ if (get_block_src_pred_buf(xd, gf_picture, frame_idx, rf_idx, mi_row, mi_col,
+ &src, &pre)) {
+ // TODO(angiebird): Consider subpixel when computing the sse.
+ cpi->fn_ptr[bsize].vf(src.buf, src.stride, get_buf_from_mv(&pre, &full_mv),
+ pre.stride, &sse);
+ return (double)sse;
+ } else {
+ assert(0);
+ return 0;
+ }
}
-static double get_mv_cost(int mv_mode) {
- // TODO(angiebird): Implement this function.
- (void)mv_mode;
- return 0;
+static int get_mv_mode_cost(int mv_mode) {
+ // TODO(angiebird): The probabilities are roughly inferred from
+ // default_inter_mode_probs. Check if there is a better way to set the
+ // probabilities.
+ const int zero_mv_prob = 9;
+ const int new_mv_prob = 77;
+ const int ref_mv_prob = 170;
+ assert(zero_mv_prob + new_mv_prob + ref_mv_prob == 256);
+ switch (mv_mode) {
+ case ZERO_MV_MODE: return vp9_prob_cost[zero_mv_prob]; break;
+ case NEW_MV_MODE: return vp9_prob_cost[new_mv_prob]; break;
+ case NEAREST_MV_MODE: return vp9_prob_cost[ref_mv_prob]; break;
+ case NEAR_MV_MODE: return vp9_prob_cost[ref_mv_prob]; break;
+ default: assert(0); return -1;
+ }
+}
+
+static INLINE double get_mv_diff_cost(MV *new_mv, MV *ref_mv) {
+ double mv_diff_cost = log2(1 + abs(new_mv->row - ref_mv->row)) +
+ log2(1 + abs(new_mv->col - ref_mv->col));
+ mv_diff_cost *= (1 << VP9_PROB_COST_SHIFT);
+ return mv_diff_cost;
+}
+static double get_mv_cost(int mv_mode, VP9_COMP *cpi, TplDepFrame *tpl_frame,
+ int rf_idx, BLOCK_SIZE bsize, int mi_row,
+ int mi_col) {
+ double mv_cost = get_mv_mode_cost(mv_mode);
+ if (mv_mode == NEW_MV_MODE) {
+ MV new_mv = get_mv_from_mv_mode(mv_mode, cpi, tpl_frame, rf_idx, bsize,
+ mi_row, mi_col)
+ .as_mv;
+ MV nearest_mv = get_mv_from_mv_mode(NEAREST_MV_MODE, cpi, tpl_frame, rf_idx,
+ bsize, mi_row, mi_col)
+ .as_mv;
+ MV near_mv = get_mv_from_mv_mode(NEAR_MV_MODE, cpi, tpl_frame, rf_idx,
+ bsize, mi_row, mi_col)
+ .as_mv;
+ double nearest_cost = get_mv_diff_cost(&new_mv, &nearest_mv);
+ double near_cost = get_mv_diff_cost(&new_mv, &near_mv);
+ mv_cost += nearest_cost < near_cost ? nearest_cost : near_cost;
+ }
+ return mv_cost;
}
static double rd_cost(int rdmult, int rddiv, double rate, double dist) {
@@ -6160,7 +6208,9 @@ static double eval_mv_mode(int mv_mode, VP9_COMP *cpi, MACROBLOCK *x,
MACROBLOCKD *xd = &x->e_mbd;
double mv_dist = get_mv_dist(mv_mode, cpi, xd, gf_picture, frame_idx,
tpl_frame, rf_idx, bsize, mi_row, mi_col, mv);
- double mv_cost = get_mv_cost(mv_mode);
+ double mv_cost =
+ get_mv_cost(mv_mode, cpi, tpl_frame, rf_idx, bsize, mi_row, mi_col);
+
return rd_cost(x->rdmult, x->rddiv, mv_cost, mv_dist);
}
@@ -6218,7 +6268,7 @@ static void predict_mv_mode(VP9_COMP *cpi, MACROBLOCK *x,
// no new mv
// diagnal scan order
tmp_idx = 0;
- for (idx = 0; idx < kMvPreCheckSize; ++idx) {
+ for (idx = 0; idx < kMvPreCheckLines; ++idx) {
int r;
for (r = 0; r <= idx; ++r) {
int c = idx - r;
@@ -6245,7 +6295,7 @@ static void predict_mv_mode(VP9_COMP *cpi, MACROBLOCK *x,
&select_mv_arr[mi_row * stride + mi_col]);
// We start from idx = 1 because idx = 0 is evaluated as NEW_MV_MODE
// beforehand.
- for (idx = 1; idx < kMvPreCheckSize; ++idx) {
+ for (idx = 1; idx < kMvPreCheckLines; ++idx) {
int r;
for (r = 0; r <= idx; ++r) {
int c = idx - r;
@@ -6266,7 +6316,7 @@ static void predict_mv_mode(VP9_COMP *cpi, MACROBLOCK *x,
tmp_idx = 0;
if (no_new_mv_rd < new_mv_rd) {
*rd = no_new_mv_rd;
- for (idx = 0; idx < kMvPreCheckSize; ++idx) {
+ for (idx = 0; idx < kMvPreCheckLines; ++idx) {
int r;
for (r = 0; r <= idx; ++r) {
int c = idx - r;
@@ -6783,6 +6833,11 @@ static void init_tpl_buffer(VP9_COMP *cpi) {
sizeof(
*cpi->tpl_stats[frame].pyramid_mv_arr[rf_idx][sqr_bsize])));
}
+ vpx_free(cpi->tpl_stats[frame].mv_mode_arr[rf_idx]);
+ CHECK_MEM_ERROR(
+ cm, cpi->tpl_stats[frame].mv_mode_arr[rf_idx],
+ vpx_calloc(mi_rows * mi_cols * 4,
+ sizeof(*cpi->tpl_stats[frame].mv_mode_arr[rf_idx])));
}
#endif
vpx_free(cpi->tpl_stats[frame].tpl_stats_ptr);
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 0fdc61649..8cd1e6e31 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -1943,8 +1943,9 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data,
// For SVC with quality layers, when QP of lower layer is lower
// than current layer: force check of GF-ZEROMV before early exit
// due to skip flag.
- if (svc->spatial_layer_id > 0 && usable_ref_frame == GOLDEN_FRAME &&
- no_scaling && cm->base_qindex > svc->lower_layer_qindex + 10)
+ if (svc->spatial_layer_id > 0 && no_scaling &&
+ (cpi->ref_frame_flags & flag_list[GOLDEN_FRAME]) &&
+ cm->base_qindex > svc->lower_layer_qindex + 10)
force_test_gf_zeromv = 1;
for (idx = 0; idx < num_inter_modes + comp_modes; ++idx) {
@@ -2407,6 +2408,8 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data,
// Perform intra prediction search, if the best SAD is above a certain
// threshold.
if (best_rdc.rdcost == INT64_MAX ||
+ (cpi->oxcf.content == VP9E_CONTENT_SCREEN && x->source_variance == 0 &&
+ !x->zero_temp_sad_source) ||
(scene_change_detected && perform_intra_pred) ||
((!force_skip_low_temp_var || bsize < BLOCK_32X32 ||
x->content_state_sb == kVeryHighSad) &&
@@ -2449,8 +2452,11 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data,
const PREDICTION_MODE this_mode = intra_mode_list[i];
THR_MODES mode_index = mode_idx[INTRA_FRAME][mode_offset(this_mode)];
int mode_rd_thresh = rd_threshes[mode_index];
+ // For spatially flat blocks, under short_circuit_flat_blocks flag:
+ // only check DC mode for stationary blocks, otherwise also check
+ // H and V mode.
if (sf->short_circuit_flat_blocks && x->source_variance == 0 &&
- this_mode != DC_PRED) {
+ ((x->zero_temp_sad_source && this_mode != DC_PRED) || i > 2)) {
continue;
}
diff --git a/vpxdec.c b/vpxdec.c
index 7f544d4bc..c60eb5c30 100644
--- a/vpxdec.c
+++ b/vpxdec.c
@@ -265,8 +265,8 @@ static int raw_read_frame(FILE *infile, uint8_t **buffer, size_t *bytes_read,
return 1;
}
-static int read_frame(struct VpxDecInputContext *input, uint8_t **buf,
- size_t *bytes_in_buffer, size_t *buffer_size) {
+static int dec_read_frame(struct VpxDecInputContext *input, uint8_t **buf,
+ size_t *bytes_in_buffer, size_t *buffer_size) {
switch (input->vpx_input_ctx->file_type) {
#if CONFIG_WEBM_IO
case FILE_TYPE_WEBM:
@@ -806,7 +806,7 @@ static int main_loop(int argc, const char **argv_) {
if (arg_skip) fprintf(stderr, "Skipping first %d frames.\n", arg_skip);
while (arg_skip) {
- if (read_frame(&input, &buf, &bytes_in_buffer, &buffer_size)) break;
+ if (dec_read_frame(&input, &buf, &bytes_in_buffer, &buffer_size)) break;
arg_skip--;
}
@@ -837,7 +837,7 @@ static int main_loop(int argc, const char **argv_) {
frame_avail = 0;
if (!stop_after || frame_in < stop_after) {
- if (!read_frame(&input, &buf, &bytes_in_buffer, &buffer_size)) {
+ if (!dec_read_frame(&input, &buf, &bytes_in_buffer, &buffer_size)) {
frame_avail = 1;
frame_in++;
diff --git a/vpxenc.c b/vpxenc.c
index b7841522d..a6d3d41d7 100644
--- a/vpxenc.c
+++ b/vpxenc.c
@@ -50,12 +50,6 @@
#endif
#include "./y4minput.h"
-/* Swallow warnings about unused results of fread/fwrite */
-static size_t wrap_fread(void *ptr, size_t size, size_t nmemb, FILE *stream) {
- return fread(ptr, size, nmemb, stream);
-}
-#define fread wrap_fread
-
static size_t wrap_fwrite(const void *ptr, size_t size, size_t nmemb,
FILE *stream) {
return fwrite(ptr, size, nmemb, stream);
@@ -95,34 +89,6 @@ static void warn_or_exit_on_error(vpx_codec_ctx_t *ctx, int fatal,
va_end(ap);
}
-static int read_frame(struct VpxInputContext *input_ctx, vpx_image_t *img) {
- FILE *f = input_ctx->file;
- y4m_input *y4m = &input_ctx->y4m;
- int shortread = 0;
-
- if (input_ctx->file_type == FILE_TYPE_Y4M) {
- if (y4m_input_fetch_frame(y4m, f, img) < 1) return 0;
- } else {
- shortread = read_yuv_frame(input_ctx, img);
- }
-
- return !shortread;
-}
-
-static int file_is_y4m(const char detect[4]) {
- if (memcmp(detect, "YUV4", 4) == 0) {
- return 1;
- }
- return 0;
-}
-
-static int fourcc_is_ivf(const char detect[4]) {
- if (memcmp(detect, "DKIF", 4) == 0) {
- return 1;
- }
- return 0;
-}
-
static const arg_def_t help =
ARG_DEF(NULL, "help", 0, "Show usage options and exit");
static const arg_def_t debugmode =
@@ -1020,57 +986,6 @@ static void parse_global_config(struct VpxEncoderConfig *global, char **argv) {
}
}
-static void open_input_file(struct VpxInputContext *input) {
- /* Parse certain options from the input file, if possible */
- input->file = strcmp(input->filename, "-") ? fopen(input->filename, "rb")
- : set_binary_mode(stdin);
-
- if (!input->file) fatal("Failed to open input file");
-
- if (!fseeko(input->file, 0, SEEK_END)) {
- /* Input file is seekable. Figure out how long it is, so we can get
- * progress info.
- */
- input->length = ftello(input->file);
- rewind(input->file);
- }
-
- /* Default to 1:1 pixel aspect ratio. */
- input->pixel_aspect_ratio.numerator = 1;
- input->pixel_aspect_ratio.denominator = 1;
-
- /* For RAW input sources, these bytes will applied on the first frame
- * in read_frame().
- */
- input->detect.buf_read = fread(input->detect.buf, 1, 4, input->file);
- input->detect.position = 0;
-
- if (input->detect.buf_read == 4 && file_is_y4m(input->detect.buf)) {
- if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4,
- input->only_i420) >= 0) {
- input->file_type = FILE_TYPE_Y4M;
- input->width = input->y4m.pic_w;
- input->height = input->y4m.pic_h;
- input->pixel_aspect_ratio.numerator = input->y4m.par_n;
- input->pixel_aspect_ratio.denominator = input->y4m.par_d;
- input->framerate.numerator = input->y4m.fps_n;
- input->framerate.denominator = input->y4m.fps_d;
- input->fmt = input->y4m.vpx_fmt;
- input->bit_depth = input->y4m.bit_depth;
- } else
- fatal("Unsupported Y4M stream.");
- } else if (input->detect.buf_read == 4 && fourcc_is_ivf(input->detect.buf)) {
- fatal("IVF is not supported as input.");
- } else {
- input->file_type = FILE_TYPE_RAW;
- }
-}
-
-static void close_input_file(struct VpxInputContext *input) {
- fclose(input->file);
- if (input->file_type == FILE_TYPE_Y4M) y4m_input_close(&input->y4m);
-}
-
static struct stream_state *new_stream(struct VpxEncoderConfig *global,
struct stream_state *prev) {
struct stream_state *stream;