Merge "vp9: fix crash in inline loopfilter w/corrupt file"
diff --git a/examples.mk b/examples.mk
index 537d8ff..bd38c41 100644
--- a/examples.mk
+++ b/examples.mk
@@ -186,7 +186,9 @@
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 += tools_common.h tools_common.c
+vp8_multi_resolution_encoder.SRCS += video_writer.h video_writer.c
vp8_multi_resolution_encoder.SRCS += $(LIBYUV_SRCS)
vp8_multi_resolution_encoder.GUID = 04f8738e-63c8-423b-90fa-7c2703a374de
vp8_multi_resolution_encoder.DESCRIPTION = VP8 Multiple-resolution Encoding
diff --git a/examples/vp8_multi_resolution_encoder.c b/examples/vp8_multi_resolution_encoder.c
index d41e442..783c9d3 100644
--- a/examples/vp8_multi_resolution_encoder.c
+++ b/examples/vp8_multi_resolution_encoder.c
@@ -8,446 +8,293 @@
* be found in the AUTHORS file in the root of the source tree.
*/
-/*
- * This is an example demonstrating multi-resolution encoding in VP8.
- * High-resolution input video is down-sampled to lower-resolutions. The
- * encoder then encodes the video and outputs multiple bitstreams with
- * different resolutions.
- */
+
+// This is an example demonstrating multi-resolution encoding in VP8.
+// High-resolution input video is down-sampled to lower-resolutions. The
+// encoder then encodes the video and outputs multiple bitstreams with
+// different resolutions.
+//
+// Configure with --enable-multi-res-encoding flag to enable this example.
+
#include <stdio.h>
#include <stdlib.h>
-#include <stdarg.h>
#include <string.h>
-#include <math.h>
-#define VPX_CODEC_DISABLE_COMPAT 1
-#include "vpx/vpx_encoder.h"
-#include "vpx/vp8cx.h"
-#include "vpx_ports/mem_ops.h"
-#include "./tools_common.h"
-#define interface (vpx_codec_vp8_cx())
-#define fourcc 0x30385056
-void usage_exit() {
- exit(EXIT_FAILURE);
-}
-
-/*
- * The input video frame is downsampled several times to generate a multi-level
- * hierarchical structure. NUM_ENCODERS is defined as the number of encoding
- * levels required. For example, if the size of input video is 1280x720,
- * NUM_ENCODERS is 3, and down-sampling factor is 2, the encoder outputs 3
- * bitstreams with resolution of 1280x720(level 0), 640x360(level 1), and
- * 320x180(level 2) respectively.
- */
-#define NUM_ENCODERS 3
-
-/* This example uses the scaler function in libyuv. */
#include "third_party/libyuv/include/libyuv/basic_types.h"
#include "third_party/libyuv/include/libyuv/scale.h"
#include "third_party/libyuv/include/libyuv/cpu_id.h"
-int (*read_frame_p)(FILE *f, vpx_image_t *img);
+#define VPX_CODEC_DISABLE_COMPAT 1
+#include "vpx/vpx_encoder.h"
+#include "vpx/vp8cx.h"
-static int read_frame(FILE *f, vpx_image_t *img) {
- size_t nbytes, to_read;
- int res = 1;
+#include "./tools_common.h"
+#include "./video_writer.h"
- 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 const char *exec_name;
+
+void usage_exit() {
+ fprintf(stderr,
+ "Usage: %s <width> <height> <infile> <outfile(s)> <output psnr?>\n",
+ exec_name);
+ exit(EXIT_FAILURE);
}
-static int read_frame_by_row(FILE *f, vpx_image_t *img) {
- size_t nbytes, to_read;
- int res = 1;
- int plane;
+int main(int argc, char *argv[]) {
+ // The input video frame is downsampled several times to generate a
+ // multi-level hierarchical structure. kNumEncoders is defined as the number
+ // of encoding levels required. For example, if the size of input video is
+ // 1280x720, kNumEncoders is 3, and down-sampling factor is 2, the encoder
+ // outputs 3 bitstreams with resolution of 1280x720(level 0),
+ // 640x360(level 1), and 320x180(level 2) respectively.
+ static const int kNumEncoders = 3;
- for (plane = 0; plane < 3; plane++)
+ int frame_cnt = 0;
+ FILE *infile = NULL;
+ VpxVideoWriter *writers[kNumEncoders];
+ vpx_codec_ctx_t codec[kNumEncoders];
+ vpx_codec_enc_cfg_t cfg[kNumEncoders];
+ vpx_image_t raw[kNumEncoders];
+ const VpxInterface *const encoder = get_vpx_encoder_by_name("vp8");
+ // Currently, only realtime mode is supported in multi-resolution encoding.
+ const int arg_deadline = VPX_DL_REALTIME;
+ int i;
+ int width = 0;
+ int height = 0;
+ int frame_avail = 0;
+ int got_data = 0;
+
+ // Set show_psnr to 1/0 to show/not show PSNR. Choose show_psnr=0 if you
+ // don't need to know PSNR, which will skip PSNR calculation and save
+ // encoding time.
+ int show_psnr = 0;
+ uint64_t psnr_sse_total[kNumEncoders] = {0};
+ uint64_t psnr_samples_total[kNumEncoders] = {0};
+ double psnr_totals[kNumEncoders][4] = {{0, 0}};
+ int psnr_count[kNumEncoders] = {0};
+
+ // Set the required target bitrates for each resolution level.
+ // If target bitrate for highest-resolution level is set to 0,
+ // (i.e. target_bitrate[0]=0), we skip encoding at that level.
+ unsigned int target_bitrate[kNumEncoders] = {1000, 500, 100};
+
+ // Enter the frame rate of the input video.
+ const int framerate = 30;
+ // Set down-sampling factor for each resolution level.
+ // dsf[0] controls down sampling from level 0 to level 1;
+ // dsf[1] controls down sampling from level 1 to level 2;
+ // dsf[2] is not used.
+ vpx_rational_t dsf[kNumEncoders] = {{2, 1}, {2, 1}, {1, 1}};
+
+ exec_name = argv[0];
+
+ if (!encoder)
+ die("Unsupported codec.");
+
+ // exe_name, input width, input height, input file,
+ // output file 1, output file 2, output file 3, psnr on/off
+ if (argc != (5 + kNumEncoders))
+ die("Invalid number of input options.");
+
+ printf("Using %s\n", vpx_codec_iface_name(encoder->codec_interface()));
+
+ width = strtol(argv[1], NULL, 0);
+ height = strtol(argv[2], NULL, 0);
+
+ if (width < 16 || width % 2 || height < 16 || height % 2)
+ die("Invalid resolution: %ldx%ld", width, height);
+
+ // Open input video file for encoding
+ if (!(infile = fopen(argv[3], "rb")))
+ die("Failed to open %s for reading", argv[3]);
+
+ show_psnr = strtol(argv[kNumEncoders + 4], NULL, 0);
+
+ // Populate default encoder configuration
+ for (i = 0; i < kNumEncoders; ++i) {
+ vpx_codec_err_t res =
+ vpx_codec_enc_config_default(encoder->codec_interface(), &cfg[i], 0);
+ if (res != VPX_CODEC_OK) {
+ printf("Failed to get config: %s\n", vpx_codec_err_to_string(res));
+ return EXIT_FAILURE;
+ }
+ }
+
+ // Update the default configuration according to needs of the application.
+ // Highest-resolution encoder settings
+ cfg[0].g_w = width;
+ cfg[0].g_h = height;
+ cfg[0].g_threads = 1;
+ cfg[0].rc_dropframe_thresh = 30;
+ cfg[0].rc_end_usage = VPX_CBR;
+ cfg[0].rc_resize_allowed = 0;
+ cfg[0].rc_min_quantizer = 4;
+ cfg[0].rc_max_quantizer = 56;
+ cfg[0].rc_undershoot_pct = 98;
+ cfg[0].rc_overshoot_pct = 100;
+ cfg[0].rc_buf_initial_sz = 500;
+ cfg[0].rc_buf_optimal_sz = 600;
+ cfg[0].rc_buf_sz = 1000;
+ cfg[0].g_error_resilient = 1;
+ cfg[0].g_lag_in_frames = 0;
+ cfg[0].kf_mode = VPX_KF_AUTO; // VPX_KF_DISABLED
+ cfg[0].kf_min_dist = 3000;
+ cfg[0].kf_max_dist = 3000;
+ cfg[0].rc_target_bitrate = target_bitrate[0];
+ cfg[0].g_timebase.num = 1;
+ cfg[0].g_timebase.den = framerate;
+
+ // Other-resolution encoder settings
+ for (i = 1; i < kNumEncoders; ++i) {
+ cfg[i] = cfg[0];
+ cfg[i].g_threads = 1;
+ cfg[i].rc_target_bitrate = target_bitrate[i];
+
+ // Note: Width & height of other-resolution encoders are calculated
+ // from the highest-resolution encoder's size and the corresponding
+ // down_sampling_factor.
{
- 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;
+ unsigned int iw = cfg[i - 1].g_w * dsf[i - 1].den + dsf[i - 1].num - 1;
+ unsigned int ih = cfg[i - 1].g_h * dsf[i - 1].den + dsf[i - 1].num - 1;
+ cfg[i].g_w = iw / dsf[i - 1].num;
+ cfg[i].g_h = ih / dsf[i - 1].num;
+ }
- /* Determine the correct plane based on the image format. The for-loop
- * always counts in Y,U,V order, but this may not match the order of
- * the data on disk.
- */
- switch (plane)
- {
- case 1:
- ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
+ // Make width & height to be multiplier of 2.
+ if ((cfg[i].g_w) % 2)
+ cfg[i].g_w++;
+
+ if ((cfg[i].g_h) % 2)
+ cfg[i].g_h++;
+ }
+
+ // Open output file for each encoder to output bitstreams
+ for (i = 0; i < kNumEncoders; ++i) {
+ VpxVideoInfo info = {
+ encoder->fourcc,
+ cfg[i].g_w,
+ cfg[i].g_h,
+ {cfg[i].g_timebase.num, cfg[i].g_timebase.den}
+ };
+
+ if (!(writers[i] = vpx_video_writer_open(argv[i+4], kContainerIVF, &info)))
+ die("Failed to open %s for writing", argv[i+4]);
+ }
+
+ // Allocate image for each encoder
+ for (i = 0; i < kNumEncoders; ++i)
+ if (!vpx_img_alloc(&raw[i], VPX_IMG_FMT_I420, cfg[i].g_w, cfg[i].g_h, 32))
+ die("Failed to allocate image", cfg[i].g_w, cfg[i].g_h);
+
+ // Initialize multi-encoder
+ if (vpx_codec_enc_init_multi(&codec[0], encoder->codec_interface(), &cfg[0],
+ kNumEncoders,
+ show_psnr ? VPX_CODEC_USE_PSNR : 0, &dsf[0]))
+ die_codec(&codec[0], "Failed to initialize encoder");
+
+ // The extra encoding configuration parameters can be set as follows.
+ for (i = 0; i < kNumEncoders; i++) {
+ // Set encoding speed
+ if (vpx_codec_control(&codec[i], VP8E_SET_CPUUSED, -6))
+ die_codec(&codec[i], "Failed to set cpu_used");
+
+ // Set static threshold.
+ if (vpx_codec_control(&codec[i], VP8E_SET_STATIC_THRESHOLD, 1))
+ die_codec(&codec[i], "Failed to set static threshold");
+
+ // Set NOISE_SENSITIVITY to do TEMPORAL_DENOISING
+ // Enable denoising for the highest-resolution encoder.
+ if (vpx_codec_control(&codec[0], VP8E_SET_NOISE_SENSITIVITY, i == 0))
+ die_codec(&codec[0], "Failed to set noise_sensitivity");
+ }
+
+ frame_avail = 1;
+ got_data = 0;
+
+ while (frame_avail || got_data) {
+ vpx_codec_iter_t iter[kNumEncoders] = {NULL};
+ const vpx_codec_cx_pkt_t *pkt[kNumEncoders];
+
+ frame_avail = vpx_img_read(&raw[0], infile);
+
+ if (frame_avail) {
+ for (i = 1; i < kNumEncoders; ++i) {
+ vpx_image_t *const prev = &raw[i - 1];
+
+ // Scale the image down a number of times by downsampling factor
+ // FilterMode 1 or 2 give better psnr than FilterMode 0.
+ I420Scale(prev->planes[VPX_PLANE_Y], prev->stride[VPX_PLANE_Y],
+ prev->planes[VPX_PLANE_U], prev->stride[VPX_PLANE_U],
+ prev->planes[VPX_PLANE_V], prev->stride[VPX_PLANE_V],
+ prev->d_w, prev->d_h,
+ raw[i].planes[VPX_PLANE_Y], raw[i].stride[VPX_PLANE_Y],
+ raw[i].planes[VPX_PLANE_U], raw[i].stride[VPX_PLANE_U],
+ raw[i].planes[VPX_PLANE_V], raw[i].stride[VPX_PLANE_V],
+ raw[i].d_w, raw[i].d_h, 1);
+ }
+ }
+
+ // Encode frame.
+ if (vpx_codec_encode(&codec[0], frame_avail? &raw[0] : NULL,
+ frame_cnt, 1, 0, arg_deadline)) {
+ die_codec(&codec[0], "Failed to encode frame");
+ }
+
+ for (i = kNumEncoders - 1; i >= 0; i--) {
+ got_data = 0;
+
+ while ((pkt[i] = vpx_codec_get_cx_data(&codec[i], &iter[i]))) {
+ got_data = 1;
+ switch (pkt[i]->kind) {
+ case VPX_CODEC_CX_FRAME_PKT:
+ vpx_video_writer_write_frame(writers[i], pkt[i]->data.frame.buf,
+ pkt[i]->data.frame.sz, frame_cnt - 1);
+ break;
+ case VPX_CODEC_PSNR_PKT:
+ if (show_psnr) {
+ int j;
+ psnr_sse_total[i] += pkt[i]->data.psnr.sse[0];
+ psnr_samples_total[i] += pkt[i]->data.psnr.samples[0];
+ for (j = 0; j < 4; j++)
+ psnr_totals[i][j] += pkt[i]->data.psnr.psnr[j];
+ psnr_count[i]++;
+ }
break;
- case 2:
- ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
+ default:
break;
- default:
- ptr = img->planes[plane];
}
+ printf(pkt[i]->kind == VPX_CODEC_CX_FRAME_PKT &&
+ (pkt[i]->data.frame.flags & VPX_FRAME_IS_KEY)? "K":".");
+ fflush(stdout);
+ }
+ }
+ frame_cnt++;
+ }
+ printf("\n");
- for (r = 0; r < h; r++)
- {
- to_read = w;
+ fclose(infile);
- 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;
- }
+ printf("Processed %d frames.\n", frame_cnt - 1);
+ for (i = 0; i < kNumEncoders; ++i) {
+ // Calculate PSNR and print it out
+ if (show_psnr && psnr_count[i] > 0) {
+ int j;
+ double ovpsnr = sse_to_psnr(psnr_samples_total[i], 255.0,
+ psnr_sse_total[i]);
- ptr += img->stride[plane];
- }
- if (!res)
- break;
+ fprintf(stderr, "\n ENC%d PSNR (Overall/Avg/Y/U/V)", i);
+ fprintf(stderr, " %.3lf", ovpsnr);
+ for (j = 0; j < 4; j++)
+ fprintf(stderr, " %.3lf", psnr_totals[i][j]/psnr_count[i]);
}
- return res;
-}
+ if (vpx_codec_destroy(&codec[i]))
+ die_codec(&codec[i], "Failed to destroy codec");
-static void write_ivf_file_header(FILE *outfile,
- const vpx_codec_enc_cfg_t *cfg,
- int frame_cnt) {
- char header[32];
+ vpx_img_free(&raw[i]);
+ vpx_video_writer_close(writers[i]);
+ }
+ printf("\n");
- 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, 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,
- const vpx_codec_cx_pkt_t *pkt)
-{
- char header[12];
- 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);
-}
-
-int main(int argc, char **argv)
-{
- FILE *infile, *outfile[NUM_ENCODERS];
- vpx_codec_ctx_t codec[NUM_ENCODERS];
- vpx_codec_enc_cfg_t cfg[NUM_ENCODERS];
- vpx_codec_pts_t frame_cnt = 0;
- vpx_image_t raw[NUM_ENCODERS];
- vpx_codec_err_t res[NUM_ENCODERS];
-
- int i;
- long width;
- long height;
- int frame_avail;
- int got_data;
- int flags = 0;
-
- /*Currently, only realtime mode is supported in multi-resolution encoding.*/
- int arg_deadline = VPX_DL_REALTIME;
-
- /* Set show_psnr to 1/0 to show/not show PSNR. Choose show_psnr=0 if you
- don't need to know PSNR, which will skip PSNR calculation and save
- encoding time. */
- int show_psnr = 0;
- uint64_t psnr_sse_total[NUM_ENCODERS] = {0};
- uint64_t psnr_samples_total[NUM_ENCODERS] = {0};
- double psnr_totals[NUM_ENCODERS][4] = {{0,0}};
- int psnr_count[NUM_ENCODERS] = {0};
-
- /* Set the required target bitrates for each resolution level.
- * If target bitrate for highest-resolution level is set to 0,
- * (i.e. target_bitrate[0]=0), we skip encoding at that level.
- */
- unsigned int target_bitrate[NUM_ENCODERS]={1000, 500, 100};
- /* Enter the frame rate of the input video */
- int framerate = 30;
- /* Set down-sampling factor for each resolution level.
- dsf[0] controls down sampling from level 0 to level 1;
- dsf[1] controls down sampling from level 1 to level 2;
- dsf[2] is not used. */
- vpx_rational_t dsf[NUM_ENCODERS] = {{2, 1}, {2, 1}, {1, 1}};
-
- if(argc!= (5+NUM_ENCODERS))
- die("Usage: %s <width> <height> <infile> <outfile(s)> <output psnr?>\n",
- argv[0]);
-
- printf("Using %s\n",vpx_codec_iface_name(interface));
-
- width = strtol(argv[1], NULL, 0);
- height = strtol(argv[2], NULL, 0);
-
- if(width < 16 || width%2 || height <16 || height%2)
- die("Invalid resolution: %ldx%ld", width, height);
-
- /* Open input video file for encoding */
- if(!(infile = fopen(argv[3], "rb")))
- die("Failed to open %s for reading", argv[3]);
-
- /* Open output file for each encoder to output bitstreams */
- for (i=0; i< NUM_ENCODERS; i++)
- {
- if(!target_bitrate[i])
- {
- outfile[i] = NULL;
- continue;
- }
-
- if(!(outfile[i] = fopen(argv[i+4], "wb")))
- die("Failed to open %s for writing", argv[i+4]);
- }
-
- show_psnr = strtol(argv[NUM_ENCODERS + 4], NULL, 0);
-
- /* Populate default encoder configuration */
- for (i=0; i< NUM_ENCODERS; i++)
- {
- res[i] = vpx_codec_enc_config_default(interface, &cfg[i], 0);
- if(res[i]) {
- printf("Failed to get config: %s\n", vpx_codec_err_to_string(res[i]));
- return EXIT_FAILURE;
- }
- }
-
- /*
- * Update the default configuration according to needs of the application.
- */
- /* Highest-resolution encoder settings */
- cfg[0].g_w = width;
- cfg[0].g_h = height;
- cfg[0].g_threads = 1; /* number of threads used */
- cfg[0].rc_dropframe_thresh = 30;
- cfg[0].rc_end_usage = VPX_CBR;
- cfg[0].rc_resize_allowed = 0;
- cfg[0].rc_min_quantizer = 4;
- cfg[0].rc_max_quantizer = 56;
- cfg[0].rc_undershoot_pct = 98;
- cfg[0].rc_overshoot_pct = 100;
- cfg[0].rc_buf_initial_sz = 500;
- cfg[0].rc_buf_optimal_sz = 600;
- cfg[0].rc_buf_sz = 1000;
- cfg[0].g_error_resilient = 1; /* Enable error resilient mode */
- cfg[0].g_lag_in_frames = 0;
-
- /* Disable automatic keyframe placement */
- /* Note: These 3 settings are copied to all levels. But, except the lowest
- * resolution level, all other levels are set to VPX_KF_DISABLED internally.
- */
- //cfg[0].kf_mode = VPX_KF_DISABLED;
- cfg[0].kf_mode = VPX_KF_AUTO;
- cfg[0].kf_min_dist = 3000;
- cfg[0].kf_max_dist = 3000;
-
- cfg[0].rc_target_bitrate = target_bitrate[0]; /* Set target bitrate */
- cfg[0].g_timebase.num = 1; /* Set fps */
- cfg[0].g_timebase.den = framerate;
-
- /* Other-resolution encoder settings */
- for (i=1; i< NUM_ENCODERS; i++)
- {
- memcpy(&cfg[i], &cfg[0], sizeof(vpx_codec_enc_cfg_t));
-
- cfg[i].g_threads = 1; /* number of threads used */
- cfg[i].rc_target_bitrate = target_bitrate[i];
-
- /* Note: Width & height of other-resolution encoders are calculated
- * from the highest-resolution encoder's size and the corresponding
- * down_sampling_factor.
- */
- {
- unsigned int iw = cfg[i-1].g_w*dsf[i-1].den + dsf[i-1].num - 1;
- unsigned int ih = cfg[i-1].g_h*dsf[i-1].den + dsf[i-1].num - 1;
- cfg[i].g_w = iw/dsf[i-1].num;
- cfg[i].g_h = ih/dsf[i-1].num;
- }
-
- /* Make width & height to be multiplier of 2. */
- // Should support odd size ???
- if((cfg[i].g_w)%2)cfg[i].g_w++;
- if((cfg[i].g_h)%2)cfg[i].g_h++;
- }
-
- /* Allocate image for each encoder */
- for (i=0; i< NUM_ENCODERS; i++)
- if(!vpx_img_alloc(&raw[i], VPX_IMG_FMT_I420, cfg[i].g_w, cfg[i].g_h, 32))
- die("Failed to allocate image", cfg[i].g_w, cfg[i].g_h);
-
- if (raw[0].stride[VPX_PLANE_Y] == raw[0].d_w)
- read_frame_p = read_frame;
- else
- read_frame_p = read_frame_by_row;
-
- for (i=0; i< NUM_ENCODERS; i++)
- if(outfile[i])
- write_ivf_file_header(outfile[i], &cfg[i], 0);
-
- /* Initialize multi-encoder */
- if(vpx_codec_enc_init_multi(&codec[0], interface, &cfg[0], NUM_ENCODERS,
- (show_psnr ? VPX_CODEC_USE_PSNR : 0), &dsf[0]))
- die_codec(&codec[0], "Failed to initialize encoder");
-
- /* The extra encoding configuration parameters can be set as follows. */
- /* Set encoding speed */
- for ( i=0; i<NUM_ENCODERS; i++)
- {
- int speed = -6;
- if(vpx_codec_control(&codec[i], VP8E_SET_CPUUSED, speed))
- die_codec(&codec[i], "Failed to set cpu_used");
- }
-
- /* Set static threshold. */
- for ( i=0; i<NUM_ENCODERS; i++)
- {
- unsigned int static_thresh = 1;
- if(vpx_codec_control(&codec[i], VP8E_SET_STATIC_THRESHOLD, static_thresh))
- die_codec(&codec[i], "Failed to set static threshold");
- }
-
- /* Set NOISE_SENSITIVITY to do TEMPORAL_DENOISING */
- /* Enable denoising for the highest-resolution encoder. */
- if(vpx_codec_control(&codec[0], VP8E_SET_NOISE_SENSITIVITY, 1))
- die_codec(&codec[0], "Failed to set noise_sensitivity");
- for ( i=1; i< NUM_ENCODERS; i++)
- {
- if(vpx_codec_control(&codec[i], VP8E_SET_NOISE_SENSITIVITY, 0))
- die_codec(&codec[i], "Failed to set noise_sensitivity");
- }
-
-
- frame_avail = 1;
- got_data = 0;
-
- while(frame_avail || got_data)
- {
- vpx_codec_iter_t iter[NUM_ENCODERS]={NULL};
- const vpx_codec_cx_pkt_t *pkt[NUM_ENCODERS];
-
- flags = 0;
- frame_avail = read_frame_p(infile, &raw[0]);
-
- if(frame_avail)
- {
- for ( i=1; i<NUM_ENCODERS; i++)
- {
- /*Scale the image down a number of times by downsampling factor*/
- /* FilterMode 1 or 2 give better psnr than FilterMode 0. */
- I420Scale(raw[i-1].planes[VPX_PLANE_Y], raw[i-1].stride[VPX_PLANE_Y],
- raw[i-1].planes[VPX_PLANE_U], raw[i-1].stride[VPX_PLANE_U],
- raw[i-1].planes[VPX_PLANE_V], raw[i-1].stride[VPX_PLANE_V],
- raw[i-1].d_w, raw[i-1].d_h,
- raw[i].planes[VPX_PLANE_Y], raw[i].stride[VPX_PLANE_Y],
- raw[i].planes[VPX_PLANE_U], raw[i].stride[VPX_PLANE_U],
- raw[i].planes[VPX_PLANE_V], raw[i].stride[VPX_PLANE_V],
- raw[i].d_w, raw[i].d_h, 1);
- }
- }
-
- /* Encode each frame at multi-levels */
- if(vpx_codec_encode(&codec[0], frame_avail? &raw[0] : NULL,
- frame_cnt, 1, flags, arg_deadline))
- die_codec(&codec[0], "Failed to encode frame");
-
- for (i=NUM_ENCODERS-1; i>=0 ; i--)
- {
- got_data = 0;
-
- while( (pkt[i] = vpx_codec_get_cx_data(&codec[i], &iter[i])) )
- {
- got_data = 1;
- switch(pkt[i]->kind) {
- case VPX_CODEC_CX_FRAME_PKT:
- write_ivf_frame_header(outfile[i], pkt[i]);
- (void) fwrite(pkt[i]->data.frame.buf, 1,
- pkt[i]->data.frame.sz, outfile[i]);
- break;
- case VPX_CODEC_PSNR_PKT:
- if (show_psnr)
- {
- int j;
-
- psnr_sse_total[i] += pkt[i]->data.psnr.sse[0];
- psnr_samples_total[i] += pkt[i]->data.psnr.samples[0];
- for (j = 0; j < 4; j++)
- {
- //fprintf(stderr, "%.3lf ", pkt[i]->data.psnr.psnr[j]);
- psnr_totals[i][j] += pkt[i]->data.psnr.psnr[j];
- }
- psnr_count[i]++;
- }
-
- break;
- default:
- break;
- }
- printf(pkt[i]->kind == VPX_CODEC_CX_FRAME_PKT
- && (pkt[i]->data.frame.flags & VPX_FRAME_IS_KEY)? "K":".");
- fflush(stdout);
- }
- }
- frame_cnt++;
- }
- printf("\n");
-
- fclose(infile);
-
- printf("Processed %ld frames.\n",(long int)frame_cnt-1);
- for (i=0; i< NUM_ENCODERS; i++)
- {
- /* Calculate PSNR and print it out */
- if ( (show_psnr) && (psnr_count[i]>0) )
- {
- int j;
- double ovpsnr = sse_to_psnr(psnr_samples_total[i], 255.0,
- psnr_sse_total[i]);
-
- fprintf(stderr, "\n ENC%d PSNR (Overall/Avg/Y/U/V)", i);
-
- fprintf(stderr, " %.3lf", ovpsnr);
- for (j = 0; j < 4; j++)
- {
- fprintf(stderr, " %.3lf", psnr_totals[i][j]/psnr_count[i]);
- }
- }
-
- if(vpx_codec_destroy(&codec[i]))
- die_codec(&codec[i], "Failed to destroy codec");
-
- vpx_img_free(&raw[i]);
-
- if(!outfile[i])
- continue;
-
- /* Try to rewrite the file header with the actual frame count */
- if(!fseek(outfile[i], 0, SEEK_SET))
- write_ivf_file_header(outfile[i], &cfg[i], frame_cnt-1);
- fclose(outfile[i]);
- }
- printf("\n");
-
- return EXIT_SUCCESS;
+ return EXIT_SUCCESS;
}
diff --git a/test/datarate_test.cc b/test/datarate_test.cc
index e8cbfbe..a3d730a 100644
--- a/test/datarate_test.cc
+++ b/test/datarate_test.cc
@@ -126,7 +126,36 @@
int denoiser_on_;
};
-TEST_P(DatarateTestLarge, BasicBufferModelDenoiserOff) {
+// Check basic datarate targeting, for a single bitrate, but loop over the
+// various denoiser settings.
+TEST_P(DatarateTestLarge, DenoiserLevels) {
+ cfg_.rc_buf_initial_sz = 500;
+ cfg_.rc_dropframe_thresh = 1;
+ cfg_.rc_max_quantizer = 56;
+ cfg_.rc_end_usage = VPX_CBR;
+ ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+ 30, 1, 0, 140);
+ for (int j = 1; j < 5; ++j) {
+ // Run over the denoiser levels.
+ // For the temporal denoiser (#if CONFIG_TEMPORAL_DENOISING) the level j
+ // refers to the 4 denoiser modes: denoiserYonly, denoiserOnYUV,
+ // denoiserOnAggressive, and denoiserOnAdaptive.
+ // For the spatial denoiser (if !CONFIG_TEMPORAL_DENOISING), the level j
+ // refers to the blur thresholds: 20, 40, 60 80.
+ // The j = 0 case (denoiser off) is covered in the tests below.
+ denoiser_on_ = j;
+ cfg_.rc_target_bitrate = 300;
+ ResetModel();
+ ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+ ASSERT_GE(cfg_.rc_target_bitrate, effective_datarate_ * 0.95)
+ << " The datarate for the file exceeds the target!";
+
+ ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.3)
+ << " The datarate for the file missed the target!";
+ }
+}
+
+TEST_P(DatarateTestLarge, BasicBufferModel) {
denoiser_on_ = 0;
cfg_.rc_buf_initial_sz = 500;
cfg_.rc_dropframe_thresh = 1;
@@ -158,40 +187,8 @@
}
}
-TEST_P(DatarateTestLarge, BasicBufferModel) {
- denoiser_on_ = 1;
- cfg_.rc_buf_initial_sz = 500;
- cfg_.rc_dropframe_thresh = 1;
- cfg_.rc_max_quantizer = 56;
- cfg_.rc_end_usage = VPX_CBR;
- // 2 pass cbr datarate control has a bug hidden by the small # of
- // frames selected in this encode. The problem is that even if the buffer is
- // negative we produce a keyframe on a cutscene. Ignoring datarate
- // constraints
- // TODO(jimbankoski): ( Fix when issue
- // http://code.google.com/p/webm/issues/detail?id=495 is addressed. )
- ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
- 30, 1, 0, 140);
-
- // There is an issue for low bitrates in real-time mode, where the
- // effective_datarate slightly overshoots the target bitrate.
- // This is same the issue as noted about (#495).
- // TODO(jimbankoski/marpan): Update test to run for lower bitrates (< 100),
- // when the issue is resolved.
- for (int i = 100; i < 800; i += 200) {
- cfg_.rc_target_bitrate = i;
- ResetModel();
- ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
- ASSERT_GE(cfg_.rc_target_bitrate, effective_datarate_ * 0.95)
- << " The datarate for the file exceeds the target!";
-
- ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.3)
- << " The datarate for the file missed the target!";
- }
-}
-
TEST_P(DatarateTestLarge, ChangingDropFrameThresh) {
- denoiser_on_ = 1;
+ denoiser_on_ = 0;
cfg_.rc_buf_initial_sz = 500;
cfg_.rc_max_quantizer = 36;
cfg_.rc_end_usage = VPX_CBR;
diff --git a/test/vp8_multi_resolution_encoder.sh b/test/vp8_multi_resolution_encoder.sh
new file mode 100755
index 0000000..a8b7fe7
--- /dev/null
+++ b/test/vp8_multi_resolution_encoder.sh
@@ -0,0 +1,75 @@
+#!/bin/sh
+##
+## Copyright (c) 2014 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.
+##
+## This file tests the libvpx vp8_multi_resolution_encoder example. To add new
+## tests to this file, do the following:
+## 1. Write a shell function (this is your test).
+## 2. Add the function to vp8_mre_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: $YUV_RAW_INPUT is required.
+vp8_multi_resolution_encoder_verify_environment() {
+ if [ "$(vpx_config_option_enabled CONFIG_MULTI_RES_ENCODING)" = "yes" ]; then
+ if [ ! -e "${YUV_RAW_INPUT}" ]; then
+ elog "Libvpx test data must exist in LIBVPX_TEST_DATA_PATH."
+ return 1
+ fi
+ local readonly app="vp8_multi_resolution_encoder"
+ if [ -z "$(vpx_tool_path "${app}")" ]; then
+ elog "${app} not found. It must exist in LIBVPX_BIN_PATH or its parent."
+ return 1
+ fi
+ fi
+}
+
+# Runs vp8_multi_resolution_encoder. Simply forwards all arguments to
+# vp8_multi_resolution_encoder after building path to the executable.
+vp8_mre() {
+ local readonly encoder="$(vpx_tool_path vp8_multi_resolution_encoder)"
+ if [ ! -x "${encoder}" ]; then
+ elog "${encoder} does not exist or is not executable."
+ return 1
+ fi
+
+ eval "${VPX_TEST_PREFIX}" "${encoder}" "$@" ${devnull}
+}
+
+vp8_multi_resolution_encoder_three_formats() {
+ local readonly output_files="${VPX_TEST_OUTPUT_DIR}/vp8_mre_0.ivf
+ ${VPX_TEST_OUTPUT_DIR}/vp8_mre_1.ivf
+ ${VPX_TEST_OUTPUT_DIR}/vp8_mre_2.ivf"
+
+ if [ "$(vpx_config_option_enabled CONFIG_MULTI_RES_ENCODING)" = "yes" ]; then
+ if [ "$(vp8_encode_available)" = "yes" ]; then
+ # Param order:
+ # Input width
+ # Input height
+ # Input file path
+ # Output file names
+ # Output PSNR
+ vp8_mre "${YUV_RAW_INPUT_WIDTH}" \
+ "${YUV_RAW_INPUT_HEIGHT}" \
+ "${YUV_RAW_INPUT}" \
+ ${output_files} \
+ 0
+
+ for output_file in ${output_files}; do
+ if [ ! -e "${output_file}" ]; then
+ elog "Missing output file: ${output_file}"
+ return 1
+ fi
+ done
+ fi
+ fi
+}
+
+vp8_mre_tests="vp8_multi_resolution_encoder_three_formats"
+run_tests vp8_multi_resolution_encoder_verify_environment "${vp8_mre_tests}"
diff --git a/vp8/vp8_cx_iface.c b/vp8/vp8_cx_iface.c
index 1654cd8..5be645c 100644
--- a/vp8/vp8_cx_iface.c
+++ b/vp8/vp8_cx_iface.c
@@ -620,7 +620,6 @@
{
vpx_codec_err_t res = VPX_CODEC_OK;
struct vpx_codec_alg_priv *priv;
- struct VP8_COMP *optr;
vp8_rtcd();
@@ -633,9 +632,8 @@
return VPX_CODEC_MEM_ERROR;
}
- ctx->priv = &priv->base;
- ctx->priv->sz = sizeof(*ctx->priv);
- ctx->priv->alg_priv = priv;
+ ctx->priv = (vpx_codec_priv_t *)priv;
+ ctx->priv->sz = sizeof(struct vpx_codec_alg_priv);
ctx->priv->init_flags = ctx->init_flags;
if (ctx->config.enc)
@@ -643,11 +641,10 @@
/* Update the reference to the config structure to an
* internal copy.
*/
- ctx->priv->alg_priv->cfg = *ctx->config.enc;
- ctx->config.enc = &ctx->priv->alg_priv->cfg;
+ priv->cfg = *ctx->config.enc;
+ ctx->config.enc = &priv->cfg;
}
-
priv->vp8_cfg = default_extracfg;
priv->vp8_cfg.pkt_list = &priv->pkt_list.head;
@@ -671,17 +668,10 @@
if (!res)
{
- set_vp8e_config(&ctx->priv->alg_priv->oxcf,
- ctx->priv->alg_priv->cfg,
- ctx->priv->alg_priv->vp8_cfg,
- mr_cfg);
-
- optr = vp8_create_compressor(&ctx->priv->alg_priv->oxcf);
-
- if (!optr)
+ set_vp8e_config(&priv->oxcf, priv->cfg, priv->vp8_cfg, mr_cfg);
+ priv->cpi = vp8_create_compressor(&priv->oxcf);
+ if (!priv->cpi)
res = VPX_CODEC_MEM_ERROR;
- else
- ctx->priv->alg_priv->cpi = optr;
}
}
diff --git a/vp8/vp8_dx_iface.c b/vp8/vp8_dx_iface.c
index 05f32ba..2820320 100644
--- a/vp8/vp8_dx_iface.c
+++ b/vp8/vp8_dx_iface.c
@@ -80,29 +80,32 @@
static void vp8_init_ctx(vpx_codec_ctx_t *ctx)
{
- ctx->priv =
- (vpx_codec_priv_t *)vpx_memalign(8, sizeof(vpx_codec_alg_priv_t));
- vpx_memset(ctx->priv, 0, sizeof(vpx_codec_alg_priv_t));
+ vpx_codec_alg_priv_t *priv =
+ (vpx_codec_alg_priv_t *)vpx_memalign(8, sizeof(*priv));
+ vpx_memset(priv, 0, sizeof(*priv));
+
+ ctx->priv = (vpx_codec_priv_t *)priv;
ctx->priv->sz = sizeof(*ctx->priv);
- ctx->priv->alg_priv = (vpx_codec_alg_priv_t *)ctx->priv;
- ctx->priv->alg_priv->si.sz = sizeof(ctx->priv->alg_priv->si);
- ctx->priv->alg_priv->decrypt_cb = NULL;
- ctx->priv->alg_priv->decrypt_state = NULL;
- ctx->priv->alg_priv->flushed = 0;
ctx->priv->init_flags = ctx->init_flags;
+ priv->si.sz = sizeof(priv->si);
+ priv->decrypt_cb = NULL;
+ priv->decrypt_state = NULL;
+ priv->flushed = 0;
+
if (ctx->config.dec)
{
/* Update the reference to the config structure to an internal copy. */
- ctx->priv->alg_priv->cfg = *ctx->config.dec;
- ctx->config.dec = &ctx->priv->alg_priv->cfg;
+ priv->cfg = *ctx->config.dec;
+ ctx->config.dec = &priv->cfg;
}
}
static vpx_codec_err_t vp8_init(vpx_codec_ctx_t *ctx,
vpx_codec_priv_enc_mr_cfg_t *data)
{
- vpx_codec_err_t res = VPX_CODEC_OK;
+ vpx_codec_err_t res = VPX_CODEC_OK;
+ vpx_codec_alg_priv_t *priv = NULL;
(void) data;
vp8_rtcd();
@@ -114,29 +117,30 @@
if (!ctx->priv)
{
vp8_init_ctx(ctx);
+ priv = (vpx_codec_alg_priv_t *)ctx->priv;
/* initialize number of fragments to zero */
- ctx->priv->alg_priv->fragments.count = 0;
+ priv->fragments.count = 0;
/* is input fragments enabled? */
- ctx->priv->alg_priv->fragments.enabled =
- (ctx->priv->alg_priv->base.init_flags &
- VPX_CODEC_USE_INPUT_FRAGMENTS);
+ priv->fragments.enabled =
+ (priv->base.init_flags & VPX_CODEC_USE_INPUT_FRAGMENTS);
/*post processing level initialized to do nothing */
}
+ else
+ {
+ priv = (vpx_codec_alg_priv_t *)ctx->priv;
+ }
- ctx->priv->alg_priv->yv12_frame_buffers.use_frame_threads =
- (ctx->priv->alg_priv->base.init_flags &
- VPX_CODEC_USE_FRAME_THREADING);
+ priv->yv12_frame_buffers.use_frame_threads =
+ (ctx->priv->init_flags & VPX_CODEC_USE_FRAME_THREADING);
/* for now, disable frame threading */
- ctx->priv->alg_priv->yv12_frame_buffers.use_frame_threads = 0;
+ priv->yv12_frame_buffers.use_frame_threads = 0;
- if(ctx->priv->alg_priv->yv12_frame_buffers.use_frame_threads &&
- (( ctx->priv->alg_priv->base.init_flags &
- VPX_CODEC_USE_ERROR_CONCEALMENT)
- || ( ctx->priv->alg_priv->base.init_flags &
- VPX_CODEC_USE_INPUT_FRAGMENTS) ) )
+ if (priv->yv12_frame_buffers.use_frame_threads &&
+ ((ctx->priv->init_flags & VPX_CODEC_USE_ERROR_CONCEALMENT) ||
+ (ctx->priv->init_flags & VPX_CODEC_USE_INPUT_FRAGMENTS)))
{
/* row-based threading, error concealment, and input fragments will
* not be supported when using frame-based threading */
diff --git a/vp9/encoder/vp9_context_tree.h b/vp9/encoder/vp9_context_tree.h
index d60e6c3..0cbb244 100644
--- a/vp9/encoder/vp9_context_tree.h
+++ b/vp9/encoder/vp9_context_tree.h
@@ -34,6 +34,9 @@
int num_4x4_blk;
int skip;
int skip_txfm[MAX_MB_PLANE];
+ // For current partition, only if all Y, U, and V transform blocks'
+ // coefficients are quantized to 0, skippable is set to 0.
+ int skippable;
int best_mode_index;
int hybrid_pred_diff;
int comp_pred_diff;
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 950a6c8..07134dc 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -727,6 +727,7 @@
p[i].eobs = ctx->eobs_pbuf[i][0];
}
ctx->is_coded = 0;
+ ctx->skippable = 0;
x->skip_recode = 0;
// Set to zero to make sure we do not use the previous encoded frame stats
@@ -2158,8 +2159,8 @@
sum_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_dist);
if (sum_rd < best_rd) {
- int64_t stop_thresh = 4096;
- int64_t stop_thresh_rd;
+ int64_t dist_breakout_thr = cpi->sf.partition_search_breakout_dist_thr;
+ int rate_breakout_thr = cpi->sf.partition_search_breakout_rate_thr;
best_rate = this_rate;
best_dist = this_dist;
@@ -2167,14 +2168,18 @@
if (bsize >= BLOCK_8X8)
pc_tree->partitioning = PARTITION_NONE;
- // Adjust threshold according to partition size.
- stop_thresh >>= 8 - (b_width_log2(bsize) +
+ // Adjust dist breakout threshold according to the partition size.
+ dist_breakout_thr >>= 8 - (b_width_log2(bsize) +
b_height_log2(bsize));
- stop_thresh_rd = RDCOST(x->rdmult, x->rddiv, 0, stop_thresh);
- // If obtained distortion is very small, choose current partition
- // and stop splitting.
- if (!x->e_mbd.lossless && best_rd < stop_thresh_rd) {
+ // If all y, u, v transform blocks in this partition are skippable, and
+ // the dist & rate are within the thresholds, the partition search is
+ // terminated for current branch of the partition search tree.
+ // The dist & rate thresholds are set to 0 at speed 0 to disable the
+ // early termination at that speed.
+ if (!x->e_mbd.lossless &&
+ (ctx->skippable && best_dist < dist_breakout_thr &&
+ best_rate < rate_breakout_thr)) {
do_split = 0;
do_rect = 0;
}
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 8464882..88b67cf 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -1445,40 +1445,6 @@
vp9_extend_frame_borders(dst);
}
-#define WRITE_RECON_BUFFER 0
-#if WRITE_RECON_BUFFER
-void write_cx_frame_to_file(YV12_BUFFER_CONFIG *frame, int this_frame) {
- FILE *yframe;
- int i;
- char filename[255];
-
- snprintf(filename, sizeof(filename), "cx\\y%04d.raw", this_frame);
- yframe = fopen(filename, "wb");
-
- for (i = 0; i < frame->y_height; i++)
- fwrite(frame->y_buffer + i * frame->y_stride,
- frame->y_width, 1, yframe);
-
- fclose(yframe);
- snprintf(filename, sizeof(filename), "cx\\u%04d.raw", this_frame);
- yframe = fopen(filename, "wb");
-
- for (i = 0; i < frame->uv_height; i++)
- fwrite(frame->u_buffer + i * frame->uv_stride,
- frame->uv_width, 1, yframe);
-
- fclose(yframe);
- snprintf(filename, sizeof(filename), "cx\\v%04d.raw", this_frame);
- yframe = fopen(filename, "wb");
-
- for (i = 0; i < frame->uv_height; i++)
- fwrite(frame->v_buffer + i * frame->uv_stride,
- frame->uv_width, 1, yframe);
-
- fclose(yframe);
-}
-#endif
-
// Function to test for conditions that indicate we should loop
// back and recode a frame.
static int recode_loop_test(const VP9_COMP *cpi,
@@ -2013,18 +1979,16 @@
}
}
-static void configure_skippable_frame(VP9_COMP *cpi) {
+static int is_skippable_frame(const VP9_COMP *cpi) {
// If the current frame does not have non-zero motion vector detected in the
// first pass, and so do its previous and forward frames, then this frame
// can be skipped for partition check, and the partition size is assigned
// according to the variance
+ const SVC *const svc = &cpi->svc;
+ const TWO_PASS *const twopass = is_spatial_svc(cpi) ?
+ &svc->layer_context[svc->spatial_layer_id].twopass : &cpi->twopass;
- SVC *const svc = &cpi->svc;
- TWO_PASS *const twopass = is_spatial_svc(cpi) ?
- &svc->layer_context[svc->spatial_layer_id].twopass
- : &cpi->twopass;
-
- cpi->skippable_frame = (!frame_is_intra_only(&cpi->common) &&
+ return (!frame_is_intra_only(&cpi->common) &&
twopass->stats_in - 2 > twopass->stats_in_start &&
twopass->stats_in < twopass->stats_in_end &&
(twopass->stats_in - 1)->pcnt_inter - (twopass->stats_in - 1)->pcnt_motion
@@ -2198,7 +2162,7 @@
// second pass according to the first pass stats
if (oxcf->pass == 2 &&
(!cpi->use_svc || is_spatial_svc(cpi))) {
- configure_skippable_frame(cpi);
+ cpi->skippable_frame = is_skippable_frame(cpi);
}
// For 1 pass CBR, check if we are dropping this frame.
@@ -2280,27 +2244,9 @@
cm->frame_to_show = get_frame_new_buffer(cm);
-#if WRITE_RECON_BUFFER
- if (cm->show_frame)
- write_cx_frame_to_file(cm->frame_to_show,
- cm->current_video_frame);
- else
- write_cx_frame_to_file(cm->frame_to_show,
- cm->current_video_frame + 1000);
-#endif
-
// Pick the loop filter level for the frame.
loopfilter_frame(cpi, cm);
-#if WRITE_RECON_BUFFER
- if (cm->show_frame)
- write_cx_frame_to_file(cm->frame_to_show,
- cm->current_video_frame + 2000);
- else
- write_cx_frame_to_file(cm->frame_to_show,
- cm->current_video_frame + 3000);
-#endif
-
// build the bitstream
cpi->dummy_packing = 0;
vp9_pack_bitstream(cpi, dest, size);
diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index 2dba67c..9f82a5a 100644
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -203,7 +203,6 @@
int arnr_max_frames;
int arnr_strength;
- int arnr_type;
int tile_columns;
int tile_rows;
@@ -523,6 +522,10 @@
return frame_index & 0x1;
}
+static INLINE int *cond_sad_list(const struct VP9_COMP *cpi, int *sad_list) {
+ return cpi->sf.mv.subpel_search_method != SUBPEL_TREE ? sad_list : NULL;
+}
+
#ifdef __cplusplus
} // extern "C"
#endif
diff --git a/vp9/encoder/vp9_mbgraph.c b/vp9/encoder/vp9_mbgraph.c
index 6e04e2a..b8e7164 100644
--- a/vp9/encoder/vp9_mbgraph.c
+++ b/vp9/encoder/vp9_mbgraph.c
@@ -34,6 +34,7 @@
const int tmp_row_min = x->mv_row_min;
const int tmp_row_max = x->mv_row_max;
MV ref_full;
+ int sad_list[5];
// Further step/diamond searches as necessary
int step_param = mv_sf->reduce_first_step_size;
@@ -45,8 +46,9 @@
ref_full.row = ref_mv->row >> 3;
/*cpi->sf.search_method == HEX*/
- vp9_hex_search(x, &ref_full, step_param, x->errorperbit, 0, &v_fn_ptr, 0,
- ref_mv, dst_mv);
+ vp9_hex_search(x, &ref_full, step_param, x->errorperbit, 0,
+ cond_sad_list(cpi, sad_list),
+ &v_fn_ptr, 0, ref_mv, dst_mv);
// Try sub-pixel MC
// if (bestsme > error_thresh && bestsme < INT_MAX)
@@ -55,8 +57,10 @@
unsigned int sse;
cpi->find_fractional_mv_step(
x, dst_mv, ref_mv, cpi->common.allow_high_precision_mv, x->errorperbit,
- &v_fn_ptr, 0, mv_sf->subpel_iters_per_step, NULL, NULL, &distortion,
- &sse, NULL, 0, 0);
+ &v_fn_ptr, 0, mv_sf->subpel_iters_per_step,
+ cond_sad_list(cpi, sad_list),
+ NULL, NULL,
+ &distortion, &sse, NULL, 0, 0);
}
xd->mi[0]->mbmi.mode = NEWMV;
diff --git a/vp9/encoder/vp9_mcomp.c b/vp9/encoder/vp9_mcomp.c
index ae924d5..d6f6b25 100644
--- a/vp9/encoder/vp9_mcomp.c
+++ b/vp9/encoder/vp9_mcomp.c
@@ -256,6 +256,137 @@
} \
}
+#define SETUP_SUBPEL_SEARCH \
+ const uint8_t *const z = x->plane[0].src.buf; \
+ const int src_stride = x->plane[0].src.stride; \
+ const MACROBLOCKD *xd = &x->e_mbd; \
+ unsigned int besterr = INT_MAX; \
+ unsigned int sse; \
+ unsigned int whichdir; \
+ int thismse; \
+ const unsigned int halfiters = iters_per_step; \
+ const unsigned int quarteriters = iters_per_step; \
+ const unsigned int eighthiters = iters_per_step; \
+ const int y_stride = xd->plane[0].pre[0].stride; \
+ const int offset = bestmv->row * y_stride + bestmv->col; \
+ const uint8_t *const y = xd->plane[0].pre[0].buf; \
+ \
+ int rr = ref_mv->row; \
+ int rc = ref_mv->col; \
+ int br = bestmv->row * 8; \
+ int bc = bestmv->col * 8; \
+ int hstep = 4; \
+ const int minc = MAX(x->mv_col_min * 8, ref_mv->col - MV_MAX); \
+ const int maxc = MIN(x->mv_col_max * 8, ref_mv->col + MV_MAX); \
+ const int minr = MAX(x->mv_row_min * 8, ref_mv->row - MV_MAX); \
+ const int maxr = MIN(x->mv_row_max * 8, ref_mv->row + MV_MAX); \
+ int tr = br; \
+ int tc = bc; \
+ \
+ bestmv->row *= 8; \
+ bestmv->col *= 8; \
+ if (second_pred != NULL) { \
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, comp_pred, 64 * 64); \
+ vp9_comp_avg_pred(comp_pred, second_pred, w, h, y + offset, y_stride); \
+ besterr = vfp->vf(comp_pred, w, z, src_stride, sse1); \
+ } else { \
+ besterr = vfp->vf(y + offset, y_stride, z, src_stride, sse1); \
+ } \
+ *distortion = besterr; \
+ besterr += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost, error_per_bit);
+
+int vp9_find_best_sub_pixel_tree_pruned(const MACROBLOCK *x,
+ MV *bestmv, const MV *ref_mv,
+ int allow_hp,
+ int error_per_bit,
+ const vp9_variance_fn_ptr_t *vfp,
+ int forced_stop,
+ int iters_per_step,
+ int *sad_list,
+ int *mvjcost, int *mvcost[2],
+ int *distortion,
+ unsigned int *sse1,
+ const uint8_t *second_pred,
+ int w, int h) {
+ SETUP_SUBPEL_SEARCH;
+
+ if (sad_list &&
+ sad_list[0] != INT_MAX && sad_list[1] != INT_MAX &&
+ sad_list[2] != INT_MAX && sad_list[3] != INT_MAX &&
+ sad_list[4] != INT_MAX) {
+ unsigned int left, right, up, down, diag;
+ whichdir = (sad_list[1] < sad_list[3] ? 0 : 1) +
+ (sad_list[2] < sad_list[4] ? 0 : 2);
+ switch (whichdir) {
+ case 0:
+ CHECK_BETTER(left, tr, tc - hstep);
+ CHECK_BETTER(up, tr - hstep, tc);
+ CHECK_BETTER(diag, tr - hstep, tc - hstep);
+ break;
+ case 1:
+ CHECK_BETTER(right, tr, tc + hstep);
+ CHECK_BETTER(up, tr - hstep, tc);
+ CHECK_BETTER(diag, tr - hstep, tc + hstep);
+ break;
+ case 2:
+ CHECK_BETTER(left, tr, tc - hstep);
+ CHECK_BETTER(down, tr + hstep, tc);
+ CHECK_BETTER(diag, tr + hstep, tc - hstep);
+ break;
+ case 3:
+ CHECK_BETTER(right, tr, tc + hstep);
+ CHECK_BETTER(down, tr + hstep, tc);
+ CHECK_BETTER(diag, tr + hstep, tc + hstep);
+ break;
+ }
+ } else {
+ FIRST_LEVEL_CHECKS;
+ if (halfiters > 1) {
+ SECOND_LEVEL_CHECKS;
+ }
+ }
+
+ tr = br;
+ tc = bc;
+
+ // Each subsequent iteration checks at least one point in common with
+ // the last iteration could be 2 ( if diag selected) 1/4 pel
+
+ // Note forced_stop: 0 - full, 1 - qtr only, 2 - half only
+ if (forced_stop != 2) {
+ hstep >>= 1;
+ FIRST_LEVEL_CHECKS;
+ if (quarteriters > 1) {
+ SECOND_LEVEL_CHECKS;
+ }
+ tr = br;
+ tc = bc;
+ }
+
+ if (allow_hp && vp9_use_mv_hp(ref_mv) && forced_stop == 0) {
+ hstep >>= 1;
+ FIRST_LEVEL_CHECKS;
+ if (eighthiters > 1) {
+ SECOND_LEVEL_CHECKS;
+ }
+ tr = br;
+ tc = bc;
+ }
+ // These lines insure static analysis doesn't warn that
+ // tr and tc aren't used after the above point.
+ (void) tr;
+ (void) tc;
+
+ bestmv->row = br;
+ bestmv->col = bc;
+
+ if ((abs(bestmv->col - ref_mv->col) > (MAX_FULL_PEL_VAL << 3)) ||
+ (abs(bestmv->row - ref_mv->row) > (MAX_FULL_PEL_VAL << 3)))
+ return INT_MAX;
+
+ return besterr;
+}
+
int vp9_find_best_sub_pixel_tree(const MACROBLOCK *x,
MV *bestmv, const MV *ref_mv,
int allow_hp,
@@ -263,55 +394,14 @@
const vp9_variance_fn_ptr_t *vfp,
int forced_stop,
int iters_per_step,
+ int *sad_list,
int *mvjcost, int *mvcost[2],
int *distortion,
unsigned int *sse1,
const uint8_t *second_pred,
int w, int h) {
- const uint8_t *const z = x->plane[0].src.buf;
- const int src_stride = x->plane[0].src.stride;
- const MACROBLOCKD *xd = &x->e_mbd;
- unsigned int besterr = INT_MAX;
- unsigned int sse;
- unsigned int whichdir;
- int thismse;
- const unsigned int halfiters = iters_per_step;
- const unsigned int quarteriters = iters_per_step;
- const unsigned int eighthiters = iters_per_step;
-
- const int y_stride = xd->plane[0].pre[0].stride;
- const int offset = bestmv->row * y_stride + bestmv->col;
- const uint8_t *const y = xd->plane[0].pre[0].buf;
-
- int rr = ref_mv->row;
- int rc = ref_mv->col;
- int br = bestmv->row * 8;
- int bc = bestmv->col * 8;
- int hstep = 4;
- const int minc = MAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);
- const int maxc = MIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);
- const int minr = MAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);
- const int maxr = MIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);
-
- int tr = br;
- int tc = bc;
-
- // central mv
- bestmv->row *= 8;
- bestmv->col *= 8;
-
- // calculate central point error
- // TODO(yunqingwang): central pointer error was already calculated in full-
- // pixel search, and can be passed in this function.
- if (second_pred != NULL) {
- DECLARE_ALIGNED_ARRAY(16, uint8_t, comp_pred, 64 * 64);
- vp9_comp_avg_pred(comp_pred, second_pred, w, h, y + offset, y_stride);
- besterr = vfp->vf(comp_pred, w, z, src_stride, sse1);
- } else {
- besterr = vfp->vf(y + offset, y_stride, z, src_stride, sse1);
- }
- *distortion = besterr;
- besterr += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost, error_per_bit);
+ SETUP_SUBPEL_SEARCH;
+ (void) sad_list; // to silence compiler warning
// Each subsequent iteration checks at least one point in
// common with the last iteration could be 2 ( if diag selected)
@@ -398,14 +488,17 @@
// Each scale can have a different number of candidates and shape of
// candidates as indicated in the num_candidates and candidates arrays
// passed into this function
+//
static int vp9_pattern_search(const MACROBLOCK *x,
MV *ref_mv,
int search_param,
int sad_per_bit,
- int do_init_search, int do_refine,
+ int do_init_search,
+ int *sad_list,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost,
- const MV *center_mv, MV *best_mv,
+ const MV *center_mv,
+ MV *best_mv,
const int num_candidates[MAX_PATTERN_SCALES],
const MV candidates[MAX_PATTERN_SCALES]
[MAX_PATTERN_CANDIDATES]) {
@@ -413,7 +506,7 @@
static const int search_param_to_steps[MAX_MVSEARCH_STEPS] = {
10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
};
- int i, j, s, t;
+ int i, s, t;
const struct buf_2d *const what = &x->plane[0].src;
const struct buf_2d *const in_what = &xd->plane[0].pre[0];
int br, bc;
@@ -552,47 +645,38 @@
} while (s--);
}
- // Check 4 1-away neighbors if do_refine is true.
- // For most well-designed schemes do_refine will not be necessary.
- if (do_refine) {
- static const MV neighbors[4] = {{0, -1}, { -1, 0}, {1, 0}, {0, 1}};
-
- for (j = 0; j < 16; j++) {
- int best_site = -1;
- if (check_bounds(x, br, bc, 1)) {
- for (i = 0; i < 4; i++) {
- const MV this_mv = {br + neighbors[i].row,
- bc + neighbors[i].col};
- thissad = vfp->sdf(what->buf, what->stride,
- get_buf_from_mv(in_what, &this_mv),
- in_what->stride);
- CHECK_BETTER
- }
- } else {
- for (i = 0; i < 4; i++) {
- const MV this_mv = {br + neighbors[i].row,
- bc + neighbors[i].col};
- if (!is_mv_in(x, &this_mv))
- continue;
- thissad = vfp->sdf(what->buf, what->stride,
- get_buf_from_mv(in_what, &this_mv),
- in_what->stride);
- CHECK_BETTER
- }
+ // Returns the one-away integer pel sad values around the best as follows:
+ // sad_list[0]: sad at the best integer pel
+ // sad_list[1]: sad at delta {0, -1} (left) from the best integer pel
+ // sad_list[2]: sad at delta {-1, 0} (top) from the best integer pel
+ // sad_list[3]: sad at delta { 0, 1} (right) from the best integer pel
+ // sad_list[4]: sad at delta { 1, 0} (bottom) from the best integer pel
+ if (sad_list) {
+ static const MV neighbors[4] = {{0, -1}, {-1, 0}, {0, 1}, {1, 0}};
+ sad_list[0] = bestsad;
+ if (check_bounds(x, br, bc, 1)) {
+ for (i = 0; i < 4; i++) {
+ const MV this_mv = {br + neighbors[i].row,
+ bc + neighbors[i].col};
+ sad_list[i + 1] = vfp->sdf(what->buf, what->stride,
+ get_buf_from_mv(in_what, &this_mv),
+ in_what->stride);
}
-
- if (best_site == -1) {
- break;
- } else {
- br += neighbors[best_site].row;
- bc += neighbors[best_site].col;
+ } else {
+ for (i = 0; i < 4; i++) {
+ const MV this_mv = {br + neighbors[i].row,
+ bc + neighbors[i].col};
+ if (!is_mv_in(x, &this_mv))
+ sad_list[i + 1] = INT_MAX;
+ else
+ sad_list[i + 1] = vfp->sdf(what->buf, what->stride,
+ get_buf_from_mv(in_what, &this_mv),
+ in_what->stride);
}
}
}
-
best_mv->row = br;
best_mv->col = bc;
-
return bestsad;
}
@@ -634,6 +718,7 @@
int search_param,
int sad_per_bit,
int do_init_search,
+ int *sad_list,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost,
const MV *center_mv, MV *best_mv) {
@@ -658,7 +743,7 @@
{ -1024, 0}},
};
return vp9_pattern_search(x, ref_mv, search_param, sad_per_bit,
- do_init_search, 0, vfp, use_mvcost,
+ do_init_search, sad_list, vfp, use_mvcost,
center_mv, best_mv,
hex_num_candidates, hex_candidates);
}
@@ -668,6 +753,7 @@
int search_param,
int sad_per_bit,
int do_init_search,
+ int *sad_list,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost,
const MV *center_mv,
@@ -699,7 +785,7 @@
{-512, 512}, {-1024, 0}},
};
return vp9_pattern_search(x, ref_mv, search_param, sad_per_bit,
- do_init_search, 0, vfp, use_mvcost,
+ do_init_search, sad_list, vfp, use_mvcost,
center_mv, best_mv,
bigdia_num_candidates, bigdia_candidates);
}
@@ -709,6 +795,7 @@
int search_param,
int sad_per_bit,
int do_init_search,
+ int *sad_list,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost,
const MV *center_mv,
@@ -740,7 +827,7 @@
{0, 1024}, {-1024, 1024}, {-1024, 0}},
};
return vp9_pattern_search(x, ref_mv, search_param, sad_per_bit,
- do_init_search, 0, vfp, use_mvcost,
+ do_init_search, sad_list, vfp, use_mvcost,
center_mv, best_mv,
square_num_candidates, square_candidates);
}
@@ -750,12 +837,13 @@
int search_param,
int sad_per_bit,
int do_init_search, // must be zero for fast_hex
+ int *sad_list,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost,
const MV *center_mv,
MV *best_mv) {
return vp9_hex_search(x, ref_mv, MAX(MAX_MVSEARCH_STEPS - 2, search_param),
- sad_per_bit, do_init_search, vfp, use_mvcost,
+ sad_per_bit, do_init_search, sad_list, vfp, use_mvcost,
center_mv, best_mv);
}
@@ -764,13 +852,14 @@
int search_param,
int sad_per_bit,
int do_init_search,
+ int *sad_list,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost,
const MV *center_mv,
MV *best_mv) {
return vp9_bigdia_search(x, ref_mv, MAX(MAX_MVSEARCH_STEPS - 2, search_param),
- sad_per_bit, do_init_search, vfp, use_mvcost,
- center_mv, best_mv);
+ sad_per_bit, do_init_search, sad_list, vfp,
+ use_mvcost, center_mv, best_mv);
}
#undef CHECK_BETTER
@@ -1368,33 +1457,41 @@
int vp9_full_pixel_search(VP9_COMP *cpi, MACROBLOCK *x,
BLOCK_SIZE bsize, MV *mvp_full,
int step_param, int error_per_bit,
+ int *sad_list,
const MV *ref_mv, MV *tmp_mv,
int var_max, int rd) {
const SPEED_FEATURES *const sf = &cpi->sf;
const SEARCH_METHODS method = sf->mv.search_method;
vp9_variance_fn_ptr_t *fn_ptr = &cpi->fn_ptr[bsize];
int var = 0;
+ if (sad_list) {
+ sad_list[0] = INT_MAX;
+ sad_list[1] = INT_MAX;
+ sad_list[2] = INT_MAX;
+ sad_list[3] = INT_MAX;
+ sad_list[4] = INT_MAX;
+ }
switch (method) {
case FAST_DIAMOND:
var = vp9_fast_dia_search(x, mvp_full, step_param, error_per_bit, 0,
- fn_ptr, 1, ref_mv, tmp_mv);
+ sad_list, fn_ptr, 1, ref_mv, tmp_mv);
break;
case FAST_HEX:
var = vp9_fast_hex_search(x, mvp_full, step_param, error_per_bit, 0,
- fn_ptr, 1, ref_mv, tmp_mv);
+ sad_list, fn_ptr, 1, ref_mv, tmp_mv);
break;
case HEX:
var = vp9_hex_search(x, mvp_full, step_param, error_per_bit, 1,
- fn_ptr, 1, ref_mv, tmp_mv);
+ sad_list, fn_ptr, 1, ref_mv, tmp_mv);
break;
case SQUARE:
var = vp9_square_search(x, mvp_full, step_param, error_per_bit, 1,
- fn_ptr, 1, ref_mv, tmp_mv);
+ sad_list, fn_ptr, 1, ref_mv, tmp_mv);
break;
case BIGDIA:
var = vp9_bigdia_search(x, mvp_full, step_param, error_per_bit, 1,
- fn_ptr, 1, ref_mv, tmp_mv);
+ sad_list, fn_ptr, 1, ref_mv, tmp_mv);
break;
case NSTEP:
var = vp9_full_pixel_diamond(cpi, x, mvp_full, step_param, error_per_bit,
diff --git a/vp9/encoder/vp9_mcomp.h b/vp9/encoder/vp9_mcomp.h
index 298fbb6..9b4734a 100644
--- a/vp9/encoder/vp9_mcomp.h
+++ b/vp9/encoder/vp9_mcomp.h
@@ -79,6 +79,7 @@
int search_param,
int error_per_bit,
int do_init_search,
+ int *sad_list,
const vp9_variance_fn_ptr_t *vf,
int use_mvcost,
const MV *center_mv,
@@ -98,12 +99,14 @@
const vp9_variance_fn_ptr_t *vfp,
int forced_stop, // 0 - full, 1 - qtr only, 2 - half only
int iters_per_step,
+ int *sad_list,
int *mvjcost, int *mvcost[2],
int *distortion, unsigned int *sse1,
const uint8_t *second_pred,
int w, int h);
extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree;
+extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree_pruned;
typedef int (*vp9_full_search_fn_t)(const MACROBLOCK *x,
const MV *ref_mv, int sad_per_bit,
@@ -136,8 +139,10 @@
int vp9_full_pixel_search(struct VP9_COMP *cpi, MACROBLOCK *x,
BLOCK_SIZE bsize, MV *mvp_full,
int step_param, int error_per_bit,
+ int *sad_list,
const MV *ref_mv, MV *tmp_mv,
int var_max, int rd);
+
#ifdef __cplusplus
} // extern "C"
#endif
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 7a7bb28..28d8302 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -126,6 +126,7 @@
const int tmp_row_min = x->mv_row_min;
const int tmp_row_max = x->mv_row_max;
int rv = 0;
+ int sad_list[5];
const YV12_BUFFER_CONFIG *scaled_ref_frame = vp9_get_scaled_ref_frame(cpi,
ref);
if (cpi->common.show_frame &&
@@ -152,8 +153,9 @@
mvp_full.col >>= 3;
mvp_full.row >>= 3;
- vp9_full_pixel_search(cpi, x, bsize, &mvp_full, step_param, sadpb, &ref_mv,
- &tmp_mv->as_mv, INT_MAX, 0);
+ vp9_full_pixel_search(cpi, x, bsize, &mvp_full, step_param, sadpb,
+ cond_sad_list(cpi, sad_list),
+ &ref_mv, &tmp_mv->as_mv, INT_MAX, 0);
x->mv_col_min = tmp_col_min;
x->mv_col_max = tmp_col_max;
@@ -179,6 +181,7 @@
&cpi->fn_ptr[bsize],
cpi->sf.mv.subpel_force_stop,
cpi->sf.mv.subpel_iters_per_step,
+ cond_sad_list(cpi, sad_list),
x->nmvjointcost, x->mvcost,
&dis, &x->pred_sse[ref], NULL, 0, 0);
x->pred_mv[ref] = tmp_mv->as_mv;
diff --git a/vp9/encoder/vp9_rd.c b/vp9/encoder/vp9_rd.c
index 2841efa..b9e4408 100644
--- a/vp9/encoder/vp9_rd.c
+++ b/vp9/encoder/vp9_rd.c
@@ -364,18 +364,15 @@
int ref_frame, BLOCK_SIZE block_size) {
MACROBLOCKD *xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
- int_mv this_mv;
int i;
int zero_seen = 0;
int best_index = 0;
int best_sad = INT_MAX;
int this_sad = INT_MAX;
int max_mv = 0;
-
uint8_t *src_y_ptr = x->plane[0].src.buf;
uint8_t *ref_y_ptr;
- int row_offset, col_offset;
- int num_mv_refs = MAX_MV_REF_CANDIDATES +
+ const int num_mv_refs = MAX_MV_REF_CANDIDATES +
(cpi->sf.adaptive_motion_search &&
cpi->common.show_frame &&
block_size < cpi->sf.max_partition_size);
@@ -387,19 +384,16 @@
// Get the sad for each candidate reference mv.
for (i = 0; i < num_mv_refs; ++i) {
- this_mv.as_mv = pred_mv[i];
+ const MV *this_mv = &pred_mv[i];
- max_mv = MAX(max_mv,
- MAX(abs(this_mv.as_mv.row), abs(this_mv.as_mv.col)) >> 3);
- // Only need to check zero mv once.
- if (!this_mv.as_int && zero_seen)
+ max_mv = MAX(max_mv, MAX(abs(this_mv->row), abs(this_mv->col)) >> 3);
+ if (is_zero_mv(this_mv) && zero_seen)
continue;
- zero_seen = zero_seen || !this_mv.as_int;
+ zero_seen |= is_zero_mv(this_mv);
- row_offset = this_mv.as_mv.row >> 3;
- col_offset = this_mv.as_mv.col >> 3;
- ref_y_ptr = ref_y_buffer + (ref_y_stride * row_offset) + col_offset;
+ ref_y_ptr =
+ &ref_y_buffer[ref_y_stride * (this_mv->row >> 3) + (this_mv->col >> 3)];
// Find sad for current vector.
this_sad = cpi->fn_ptr[block_size].sdf(src_y_ptr, x->plane[0].src.stride,
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 5f1b0a5..f73006e 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -171,19 +171,43 @@
int64_t dist_sum = 0;
const int ref = xd->mi[0]->mbmi.ref_frame[0];
unsigned int sse;
+ unsigned int var = 0;
const int shift = 8;
+ int rate;
+ int64_t dist;
+
+ x->pred_sse[ref] = 0;
for (i = 0; i < MAX_MB_PLANE; ++i) {
struct macroblock_plane *const p = &x->plane[i];
struct macroblockd_plane *const pd = &xd->plane[i];
const BLOCK_SIZE bs = get_plane_block_size(bsize, pd);
+ const TX_SIZE max_tx_size = max_txsize_lookup[bs];
+ const BLOCK_SIZE unit_size = txsize_to_bsize[max_tx_size];
+ int bw = 1 << (b_width_log2_lookup[bs] - b_width_log2_lookup[unit_size]);
+ int bh = 1 << (b_height_log2_lookup[bs] - b_width_log2_lookup[unit_size]);
+ int idx, idy;
+ int lw = b_width_log2_lookup[unit_size] + 2;
+ int lh = b_height_log2_lookup[unit_size] + 2;
- const unsigned int var = cpi->fn_ptr[bs].vf(p->src.buf, p->src.stride,
- pd->dst.buf, pd->dst.stride,
- &sse);
+ x->bsse[i] = 0;
+
+ for (idy = 0; idy < bh; ++idy) {
+ for (idx = 0; idx < bw; ++idx) {
+ uint8_t *src = p->src.buf + (idy * p->src.stride << lh) + (idx << lw);
+ uint8_t *dst = pd->dst.buf + (idy * pd->dst.stride << lh) + (idx << lh);
+
+ var += cpi->fn_ptr[unit_size].vf(src , p->src.stride,
+ dst, pd->dst.stride, &sse);
+
+ x->bsse[i] += sse;
+ if (i == 0)
+ x->pred_sse[ref] += sse;
+ }
+ }
if (!x->select_tx_size) {
- if (sse < p->quant_thred[0] >> shift)
+ if (x->bsse[i] < p->quant_thred[0] >> shift)
x->skip_txfm[i] = 1;
else if (var < p->quant_thred[1] >> shift)
x->skip_txfm[i] = 2;
@@ -191,10 +215,6 @@
x->skip_txfm[i] = 0;
}
- x->bsse[i] = sse;
- if (i == 0)
- x->pred_sse[ref] = sse;
-
// Fast approximate the modelling function.
if (cpi->oxcf.speed > 4) {
int64_t rate;
@@ -210,9 +230,7 @@
rate_sum += rate;
dist_sum += dist;
} else {
- int rate;
- int64_t dist;
- vp9_model_rd_from_var_lapndz(sse, 1 << num_pels_log2_lookup[bs],
+ vp9_model_rd_from_var_lapndz(x->bsse[i], 1 << num_pels_log2_lookup[bs],
pd->dequant[1] >> 3, &rate, &dist);
rate_sum += rate;
dist_sum += dist;
@@ -1360,6 +1378,7 @@
int sadpb = x->sadperbit4;
MV mvp_full;
int max_mv;
+ int sad_list[5];
/* Is the best so far sufficiently good that we cant justify doing
* and new motion search. */
@@ -1403,9 +1422,11 @@
vp9_set_mv_search_range(x, &bsi->ref_mv[0]->as_mv);
- bestsme = vp9_full_pixel_search(cpi, x, bsize, &mvp_full, step_param,
- sadpb, &bsi->ref_mv[0]->as_mv, new_mv,
- INT_MAX, 1);
+ bestsme = vp9_full_pixel_search(
+ cpi, x, bsize, &mvp_full, step_param, sadpb,
+ cpi->sf.mv.subpel_search_method != SUBPEL_TREE ? sad_list : NULL,
+ &bsi->ref_mv[0]->as_mv, new_mv,
+ INT_MAX, 1);
// Should we do a full search (best quality only)
if (cpi->oxcf.mode == BEST) {
@@ -1417,6 +1438,7 @@
sadpb, 16, &cpi->fn_ptr[bsize],
&bsi->ref_mv[0]->as_mv,
&best_mv->as_mv);
+ sad_list[1] = sad_list[2] = sad_list[3] = sad_list[4] = INT_MAX;
if (thissme < bestsme) {
bestsme = thissme;
*new_mv = best_mv->as_mv;
@@ -1429,17 +1451,19 @@
if (bestsme < INT_MAX) {
int distortion;
- cpi->find_fractional_mv_step(x,
- new_mv,
- &bsi->ref_mv[0]->as_mv,
- cm->allow_high_precision_mv,
- x->errorperbit, &cpi->fn_ptr[bsize],
- cpi->sf.mv.subpel_force_stop,
- cpi->sf.mv.subpel_iters_per_step,
- x->nmvjointcost, x->mvcost,
- &distortion,
- &x->pred_sse[mbmi->ref_frame[0]],
- NULL, 0, 0);
+ cpi->find_fractional_mv_step(
+ x,
+ new_mv,
+ &bsi->ref_mv[0]->as_mv,
+ cm->allow_high_precision_mv,
+ x->errorperbit, &cpi->fn_ptr[bsize],
+ cpi->sf.mv.subpel_force_stop,
+ cpi->sf.mv.subpel_iters_per_step,
+ cond_sad_list(cpi, sad_list),
+ x->nmvjointcost, x->mvcost,
+ &distortion,
+ &x->pred_sse[mbmi->ref_frame[0]],
+ NULL, 0, 0);
// save motion search result for use in compound prediction
seg_mvs[i][mbmi->ref_frame[0]].as_mv = *new_mv;
@@ -1696,12 +1720,14 @@
int mode_index,
int64_t comp_pred_diff[REFERENCE_MODES],
const int64_t tx_size_diff[TX_MODES],
- int64_t best_filter_diff[SWITCHABLE_FILTER_CONTEXTS]) {
+ int64_t best_filter_diff[SWITCHABLE_FILTER_CONTEXTS],
+ int skippable) {
MACROBLOCKD *const xd = &x->e_mbd;
// Take a snapshot of the coding context so it can be
// restored if we decide to encode this way
ctx->skip = x->skip;
+ ctx->skippable = skippable;
ctx->best_mode_index = mode_index;
ctx->mic = *xd->mi[0];
ctx->single_pred_diff = (int)comp_pred_diff[SINGLE_REFERENCE];
@@ -1767,6 +1793,7 @@
int tmp_col_max = x->mv_col_max;
int tmp_row_min = x->mv_row_min;
int tmp_row_max = x->mv_row_max;
+ int sad_list[5];
const YV12_BUFFER_CONFIG *scaled_ref_frame = vp9_get_scaled_ref_frame(cpi,
ref);
@@ -1839,6 +1866,7 @@
mvp_full.row >>= 3;
bestsme = vp9_full_pixel_search(cpi, x, bsize, &mvp_full, step_param, sadpb,
+ cond_sad_list(cpi, sad_list),
&ref_mv, &tmp_mv->as_mv, INT_MAX, 1);
x->mv_col_min = tmp_col_min;
@@ -1854,6 +1882,7 @@
&cpi->fn_ptr[bsize],
cpi->sf.mv.subpel_force_stop,
cpi->sf.mv.subpel_iters_per_step,
+ cond_sad_list(cpi, sad_list),
x->nmvjointcost, x->mvcost,
&dis, &x->pred_sse[ref], NULL, 0, 0);
}
@@ -1978,6 +2007,7 @@
x->errorperbit,
&cpi->fn_ptr[bsize],
0, cpi->sf.mv.subpel_iters_per_step,
+ NULL,
x->nmvjointcost, x->mvcost,
&dis, &sse, second_pred,
pw, ph);
@@ -2130,7 +2160,7 @@
DECLARE_ALIGNED_ARRAY(16, uint8_t, tmp_buf, MAX_MB_PLANE * 64 * 64);
int pred_exists = 0;
int intpel_mv;
- int64_t rd, best_rd = INT64_MAX;
+ int64_t rd, tmp_rd, best_rd = INT64_MAX;
int best_needs_copy = 0;
uint8_t *orig_dst[MAX_MB_PLANE];
int orig_dst_stride[MAX_MB_PLANE];
@@ -2220,6 +2250,10 @@
* if the first is known */
*rate2 += cost_mv_ref(cpi, this_mode, mbmi->mode_context[refs[0]]);
+ if (RDCOST(x->rdmult, x->rddiv, *rate2, 0) > ref_best_rd &&
+ mbmi->mode != NEARESTMV)
+ return INT64_MAX;
+
pred_exists = 0;
// Are all MVs integer pel for Y and UV
intpel_mv = !mv_has_subpel(&mbmi->mv[0].as_mv);
@@ -2315,6 +2349,7 @@
(cm->interp_filter != SWITCHABLE &&
cm->interp_filter == mbmi->interp_filter)) {
pred_exists = 1;
+ tmp_rd = best_rd;
}
}
restore_dst_buf(xd, orig_dst, orig_dst_stride);
@@ -2333,17 +2368,19 @@
xd->plane[i].dst.stride = 64;
}
}
+ rd = tmp_rd + RDCOST(x->rdmult, x->rddiv, rs, 0);
} else {
+ int tmp_rate;
+ int64_t tmp_dist;
// Handles the special case when a filter that is not in the
- // switchable list (ex. bilinear, 6-tap) is indicated at the frame level
+ // switchable list (ex. bilinear) is indicated at the frame level, or
+ // skip condition holds.
vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
+ model_rd_for_sb(cpi, bsize, x, xd, &tmp_rate, &tmp_dist);
+ rd = RDCOST(x->rdmult, x->rddiv, rs + tmp_rate, tmp_dist);
}
if (cpi->sf.use_rd_breakout && ref_best_rd < INT64_MAX) {
- int tmp_rate;
- int64_t tmp_dist;
- model_rd_for_sb(cpi, bsize, x, xd, &tmp_rate, &tmp_dist);
- rd = RDCOST(x->rdmult, x->rddiv, rs + tmp_rate, tmp_dist);
// if current pred_error modeled rd is substantially more than the best
// so far, do not bother doing full rd
if (rd / 2 > ref_best_rd) {
@@ -2353,7 +2390,7 @@
}
if (cm->interp_filter == SWITCHABLE)
- *rate2 += vp9_get_switchable_rate(cpi);
+ *rate2 += rs;
if (!is_comp_pred) {
if (cpi->allow_encode_breakout)
@@ -2521,6 +2558,7 @@
int64_t best_filter_rd[SWITCHABLE_FILTER_CONTEXTS];
int64_t best_filter_diff[SWITCHABLE_FILTER_CONTEXTS];
MB_MODE_INFO best_mbmode;
+ int best_mode_skippable = 0;
int mode_index, best_mode_index = -1;
unsigned int ref_costs_single[MAX_REF_FRAMES], ref_costs_comp[MAX_REF_FRAMES];
vp9_prob comp_mode_p;
@@ -2928,6 +2966,8 @@
best_rd = this_rd;
best_mbmode = *mbmi;
best_skip2 = this_skip2;
+ best_mode_skippable = skippable;
+
if (!x->select_tx_size)
swap_block_ptr(x, ctx, 1, 0, 0, max_plane);
vpx_memcpy(ctx->zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
@@ -3084,8 +3124,8 @@
}
set_ref_ptrs(cm, xd, mbmi->ref_frame[0], mbmi->ref_frame[1]);
- store_coding_context(x, ctx, best_mode_index,
- best_pred_diff, best_tx_diff, best_filter_diff);
+ store_coding_context(x, ctx, best_mode_index, best_pred_diff,
+ best_tx_diff, best_filter_diff, best_mode_skippable);
return best_rd;
}
@@ -3190,7 +3230,7 @@
if (!x->select_tx_size)
swap_block_ptr(x, ctx, 1, 0, 0, MAX_MB_PLANE);
store_coding_context(x, ctx, THR_ZEROMV,
- best_pred_diff, best_tx_diff, best_filter_diff);
+ best_pred_diff, best_tx_diff, best_filter_diff, 0);
return this_rd;
}
@@ -3795,7 +3835,7 @@
set_ref_ptrs(cm, xd, mbmi->ref_frame[0], mbmi->ref_frame[1]);
store_coding_context(x, ctx, best_ref_index,
- best_pred_diff, best_tx_diff, best_filter_diff);
+ best_pred_diff, best_tx_diff, best_filter_diff, 0);
return best_rd;
}
diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index 879c83c..0bead48 100644
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -92,6 +92,12 @@
sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V;
sf->tx_size_search_breakout = 1;
+
+ if (MIN(cm->width, cm->height) >= 720)
+ sf->partition_search_breakout_dist_thr = (1 << 23);
+ else
+ sf->partition_search_breakout_dist_thr = (1 << 21);
+ sf->partition_search_breakout_rate_thr = 500;
}
if (speed >= 2) {
@@ -120,6 +126,12 @@
sf->auto_min_max_partition_size = CONSTRAIN_NEIGHBORING_MIN_MAX;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION;
sf->adjust_partitioning_from_last_frame = 1;
+
+ if (MIN(cm->width, cm->height) >= 720)
+ sf->partition_search_breakout_dist_thr = (1 << 24);
+ else
+ sf->partition_search_breakout_dist_thr = (1 << 22);
+ sf->partition_search_breakout_rate_thr = 700;
}
if (speed >= 3) {
@@ -144,6 +156,12 @@
sf->intra_y_mode_mask[TX_32X32] = INTRA_DC;
sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC;
sf->adaptive_interp_filter_search = 1;
+
+ if (MIN(cm->width, cm->height) >= 720)
+ sf->partition_search_breakout_dist_thr = (1 << 25);
+ else
+ sf->partition_search_breakout_dist_thr = (1 << 23);
+ sf->partition_search_breakout_rate_thr = 1000;
}
if (speed >= 4) {
@@ -158,6 +176,12 @@
sf->use_lp32x32fdct = 1;
sf->use_fast_coef_updates = ONE_LOOP_REDUCED;
sf->use_fast_coef_costing = 1;
+
+ if (MIN(cm->width, cm->height) >= 720)
+ sf->partition_search_breakout_dist_thr = (1 << 26);
+ else
+ sf->partition_search_breakout_dist_thr = (1 << 24);
+ sf->partition_search_breakout_rate_thr = 1500;
}
if (speed >= 5) {
@@ -411,6 +435,8 @@
sf->recode_tolerance = 25;
sf->default_interp_filter = SWITCHABLE;
sf->tx_size_search_breakout = 0;
+ sf->partition_search_breakout_dist_thr = 0;
+ sf->partition_search_breakout_rate_thr = 0;
if (oxcf->mode == REALTIME)
set_rt_speed_feature(cpi, sf, oxcf->speed, oxcf->content);
@@ -436,6 +462,8 @@
if (sf->mv.subpel_search_method == SUBPEL_TREE) {
cpi->find_fractional_mv_step = vp9_find_best_sub_pixel_tree;
+ } else if (sf->mv.subpel_search_method == SUBPEL_TREE_PRUNED) {
+ cpi->find_fractional_mv_step = vp9_find_best_sub_pixel_tree_pruned;
}
cpi->mb.optimize = sf->optimize_coefficients == 1 && oxcf->pass != 1;
diff --git a/vp9/encoder/vp9_speed_features.h b/vp9/encoder/vp9_speed_features.h
index e2e5c1e..e31185e 100644
--- a/vp9/encoder/vp9_speed_features.h
+++ b/vp9/encoder/vp9_speed_features.h
@@ -40,6 +40,7 @@
typedef enum {
SUBPEL_TREE = 0,
+ SUBPEL_TREE_PRUNED = 1,
// Other methods to come
} SUBPEL_SEARCH_METHODS;
@@ -392,6 +393,10 @@
// mask for skip evaluation of certain interp_filter type.
INTERP_FILTER_MASK interp_filter_search_mask;
+
+ // Partition search early breakout thresholds.
+ int64_t partition_search_breakout_dist_thr;
+ int partition_search_breakout_rate_thr;
} SPEED_FEATURES;
struct VP9_COMP;
diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index 076d776..045e359 100644
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -145,6 +145,7 @@
int bestsme = INT_MAX;
int distortion;
unsigned int sse;
+ int sad_list[5];
MV best_ref_mv1 = {0, 0};
MV best_ref_mv1_full; /* full-pixel value of best_ref_mv1 */
@@ -168,6 +169,7 @@
// Ignore mv costing by sending NULL pointer instead of cost arrays
vp9_hex_search(x, &best_ref_mv1_full, step_param, sadpb, 1,
+ cond_sad_list(cpi, sad_list),
&cpi->fn_ptr[BLOCK_16X16], 0, &best_ref_mv1, ref_mv);
// Ignore mv costing by sending NULL pointer instead of cost array
@@ -177,6 +179,7 @@
x->errorperbit,
&cpi->fn_ptr[BLOCK_16X16],
0, mv_sf->subpel_iters_per_step,
+ cond_sad_list(cpi, sad_list),
NULL, NULL,
&distortion, &sse, NULL, 0, 0);
diff --git a/vp9/encoder/x86/vp9_variance_sse2.c b/vp9/encoder/x86/vp9_variance_sse2.c
index e935a23..88e3117 100644
--- a/vp9/encoder/x86/vp9_variance_sse2.c
+++ b/vp9/encoder/x86/vp9_variance_sse2.c
@@ -8,6 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <emmintrin.h> // SSE2
+
#include "./vpx_config.h"
#include "vp9/encoder/vp9_variance.h"
@@ -17,10 +19,37 @@
const unsigned char *ref, int ref_stride,
unsigned int *sse, int *sum);
-unsigned int vp9_get4x4var_mmx(const unsigned char *src, int src_stride,
- const unsigned char *ref, int ref_stride,
- unsigned int *sse, int *sum);
+#define READ64(p, stride, i) \
+ _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const uint32_t *)(p + i * stride)), \
+ _mm_cvtsi32_si128(*(const uint32_t *)(p + (i + 1) * stride)))
+unsigned int vp9_get4x4var_sse2(const uint8_t *src, int src_stride,
+ const uint8_t *ref, int ref_stride,
+ unsigned int *sse, int *sum) {
+ const __m128i zero = _mm_setzero_si128();
+ const __m128i src0 = _mm_unpacklo_epi8(READ64(src, src_stride, 0), zero);
+ const __m128i src1 = _mm_unpacklo_epi8(READ64(src, src_stride, 2), zero);
+ const __m128i ref0 = _mm_unpacklo_epi8(READ64(ref, ref_stride, 0), zero);
+ const __m128i ref1 = _mm_unpacklo_epi8(READ64(ref, ref_stride, 2), zero);
+ const __m128i diff0 = _mm_sub_epi16(src0, ref0);
+ const __m128i diff1 = _mm_sub_epi16(src1, ref1);
+
+ // sum
+ __m128i vsum = _mm_add_epi16(diff0, diff1);
+ vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 8));
+ vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 4));
+ vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 2));
+ *sum = (int16_t)_mm_extract_epi16(vsum, 0);
+
+ // sse
+ vsum = _mm_add_epi32(_mm_madd_epi16(diff0, diff0),
+ _mm_madd_epi16(diff1, diff1));
+ vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 8));
+ vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 4));
+ *sse = _mm_cvtsi128_si32(vsum);
+
+ return 0;
+}
unsigned int vp9_get8x8var_sse2(const unsigned char *src, int src_stride,
const unsigned char *ref, int ref_stride,
@@ -55,8 +84,7 @@
const unsigned char *ref, int ref_stride,
unsigned int *sse) {
int sum;
- variance_sse2(src, src_stride, ref, ref_stride, 4, 4,
- sse, &sum, vp9_get4x4var_mmx, 4);
+ vp9_get4x4var_sse2(src, src_stride, ref, ref_stride, sse, &sum);
return *sse - (((unsigned int)sum * sum) >> 4);
}
@@ -65,7 +93,7 @@
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 8, 4,
- sse, &sum, vp9_get4x4var_mmx, 4);
+ sse, &sum, vp9_get4x4var_sse2, 4);
return *sse - (((unsigned int)sum * sum) >> 5);
}
@@ -74,7 +102,7 @@
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 4, 8,
- sse, &sum, vp9_get4x4var_mmx, 4);
+ sse, &sum, vp9_get4x4var_sse2, 4);
return *sse - (((unsigned int)sum * sum) >> 5);
}
diff --git a/vp9/vp9_cx_iface.c b/vp9/vp9_cx_iface.c
index 5ff02d8..3dbf001 100644
--- a/vp9/vp9_cx_iface.c
+++ b/vp9/vp9_cx_iface.c
@@ -30,7 +30,6 @@
unsigned int tile_rows;
unsigned int arnr_max_frames;
unsigned int arnr_strength;
- unsigned int arnr_type;
vp8e_tuning tuning;
unsigned int cq_level; // constrained quality level
unsigned int rc_max_intra_bitrate_pct;
@@ -52,7 +51,6 @@
0, // tile_rows
7, // arnr_max_frames
5, // arnr_strength
- 3, // arnr_type
VP8_TUNE_PSNR, // tuning
10, // cq_level
0, // rc_max_intra_bitrate_pct
@@ -209,7 +207,6 @@
RANGE_CHECK_HI(extra_cfg, sharpness, 7);
RANGE_CHECK(extra_cfg, arnr_max_frames, 0, 15);
RANGE_CHECK_HI(extra_cfg, arnr_strength, 6);
- RANGE_CHECK(extra_cfg, arnr_type, 1, 3);
RANGE_CHECK(extra_cfg, cq_level, 0, 63);
RANGE_CHECK(extra_cfg, content,
VP9E_CONTENT_DEFAULT, VP9E_CONTENT_INVALID - 1);
@@ -390,7 +387,6 @@
oxcf->arnr_max_frames = extra_cfg->arnr_max_frames;
oxcf->arnr_strength = extra_cfg->arnr_strength;
- oxcf->arnr_type = extra_cfg->arnr_type;
oxcf->tuning = extra_cfg->tuning;
oxcf->content = extra_cfg->content;
@@ -586,9 +582,9 @@
static vpx_codec_err_t ctrl_set_arnr_type(vpx_codec_alg_priv_t *ctx,
va_list args) {
- struct vp9_extracfg extra_cfg = ctx->extra_cfg;
- extra_cfg.arnr_type = CAST(VP8E_SET_ARNR_TYPE, args);
- return update_extra_cfg(ctx, &extra_cfg);
+ (void)ctx;
+ (void)args;
+ return VPX_CODEC_OK;
}
static vpx_codec_err_t ctrl_set_tuning(vpx_codec_alg_priv_t *ctx,
@@ -648,41 +644,34 @@
(void)data;
if (ctx->priv == NULL) {
- struct vpx_codec_alg_priv *priv = calloc(1, sizeof(*priv));
+ vpx_codec_alg_priv_t *const alg_priv = calloc(1, sizeof(*alg_priv));
- if (priv == NULL)
+ if (alg_priv == NULL)
return VPX_CODEC_MEM_ERROR;
- ctx->priv = &priv->base;
- ctx->priv->sz = sizeof(*ctx->priv);
- ctx->priv->alg_priv = priv;
+ ctx->priv = (vpx_codec_priv_t *)alg_priv;
+ ctx->priv->sz = sizeof(vpx_codec_alg_priv_t);
ctx->priv->init_flags = ctx->init_flags;
ctx->priv->enc.total_encoders = 1;
if (ctx->config.enc) {
// Update the reference to the config structure to an internal copy.
- ctx->priv->alg_priv->cfg = *ctx->config.enc;
- ctx->config.enc = &ctx->priv->alg_priv->cfg;
+ alg_priv->cfg = *ctx->config.enc;
+ ctx->config.enc = &alg_priv->cfg;
}
- priv->extra_cfg = default_extra_cfg;
-
+ alg_priv->extra_cfg = default_extra_cfg;
vp9_initialize_enc();
- res = validate_config(priv, &priv->cfg, &priv->extra_cfg);
+ res = validate_config(alg_priv, &alg_priv->cfg, &alg_priv->extra_cfg);
if (res == VPX_CODEC_OK) {
- VP9_COMP *cpi;
- set_encoder_config(&ctx->priv->alg_priv->oxcf,
- &ctx->priv->alg_priv->cfg,
- &ctx->priv->alg_priv->extra_cfg);
- cpi = vp9_create_compressor(&ctx->priv->alg_priv->oxcf);
- if (cpi == NULL) {
+ set_encoder_config(&alg_priv->oxcf, &alg_priv->cfg, &alg_priv->extra_cfg);
+ alg_priv->cpi = vp9_create_compressor(&alg_priv->oxcf);
+ if (alg_priv->cpi == NULL)
res = VPX_CODEC_MEM_ERROR;
- } else {
- cpi->output_pkt_list = &priv->pkt_list.head;
- ctx->priv->alg_priv->cpi = cpi;
- }
+ else
+ alg_priv->cpi->output_pkt_list = &alg_priv->pkt_list.head;
}
}
diff --git a/vp9/vp9_dx_iface.c b/vp9/vp9_dx_iface.c
index 05d61c8..7909f53 100644
--- a/vp9/vp9_dx_iface.c
+++ b/vp9/vp9_dx_iface.c
@@ -58,28 +58,25 @@
(void)data;
if (!ctx->priv) {
- vpx_codec_alg_priv_t *alg_priv = vpx_memalign(32, sizeof(*alg_priv));
+ vpx_codec_alg_priv_t *const alg_priv = vpx_memalign(32, sizeof(*alg_priv));
if (alg_priv == NULL)
return VPX_CODEC_MEM_ERROR;
vp9_zero(*alg_priv);
ctx->priv = (vpx_codec_priv_t *)alg_priv;
- ctx->priv->sz = sizeof(*ctx->priv);
- ctx->priv->alg_priv = alg_priv;
- ctx->priv->alg_priv->si.sz = sizeof(ctx->priv->alg_priv->si);
+ ctx->priv->sz = sizeof(*alg_priv);
ctx->priv->init_flags = ctx->init_flags;
- ctx->priv->alg_priv->flushed = 0;
- ctx->priv->alg_priv->frame_parallel_decode =
- (ctx->init_flags & VPX_CODEC_USE_FRAME_THREADING);
- // Disable frame parallel decoding for now.
- ctx->priv->alg_priv->frame_parallel_decode = 0;
+ alg_priv->si.sz = sizeof(alg_priv->si);
+ alg_priv->flushed = 0;
+ alg_priv->frame_parallel_decode =
+ (ctx->init_flags & VPX_CODEC_USE_FRAME_THREADING);
+ alg_priv->frame_parallel_decode = 0; // Disable for now
if (ctx->config.dec) {
- // Update the reference to the config structure to an internal copy.
- ctx->priv->alg_priv->cfg = *ctx->config.dec;
- ctx->config.dec = &ctx->priv->alg_priv->cfg;
+ alg_priv->cfg = *ctx->config.dec;
+ ctx->config.dec = &alg_priv->cfg;
}
}
diff --git a/vpx/internal/vpx_codec_internal.h b/vpx/internal/vpx_codec_internal.h
index d7c5800..02f2079 100644
--- a/vpx/internal/vpx_codec_internal.h
+++ b/vpx/internal/vpx_codec_internal.h
@@ -336,7 +336,6 @@
*/
struct vpx_codec_priv {
unsigned int sz;
- struct vpx_codec_alg_priv *alg_priv;
const char *err_detail;
vpx_codec_flags_t init_flags;
struct {
diff --git a/vpx/src/vpx_codec.c b/vpx/src/vpx_codec.c
index d175eae..5a495ce 100644
--- a/vpx/src/vpx_codec.c
+++ b/vpx/src/vpx_codec.c
@@ -88,8 +88,7 @@
else if (!ctx->iface || !ctx->priv)
res = VPX_CODEC_ERROR;
else {
- if (ctx->priv->alg_priv)
- ctx->iface->destroy(ctx->priv->alg_priv);
+ ctx->iface->destroy((vpx_codec_alg_priv_t *)ctx->priv);
ctx->iface = NULL;
ctx->name = NULL;
@@ -125,7 +124,7 @@
va_list ap;
va_start(ap, ctrl_id);
- res = entry->fn(ctx->priv->alg_priv, ap);
+ res = entry->fn((vpx_codec_alg_priv_t *)ctx->priv, ap);
va_end(ap);
break;
}
diff --git a/vpx/src/vpx_decoder.c b/vpx/src/vpx_decoder.c
index fdcc9c1..802d8ed 100644
--- a/vpx/src/vpx_decoder.c
+++ b/vpx/src/vpx_decoder.c
@@ -18,6 +18,10 @@
#define SAVE_STATUS(ctx,var) (ctx?(ctx->err = var):var)
+static vpx_codec_alg_priv_t *get_alg_priv(vpx_codec_ctx_t *ctx) {
+ return (vpx_codec_alg_priv_t *)ctx->priv;
+}
+
vpx_codec_err_t vpx_codec_dec_init_ver(vpx_codec_ctx_t *ctx,
vpx_codec_iface_t *iface,
const vpx_codec_dec_cfg_t *cfg,
@@ -94,7 +98,7 @@
si->w = 0;
si->h = 0;
- res = ctx->iface->dec.get_si(ctx->priv->alg_priv, si);
+ res = ctx->iface->dec.get_si(get_alg_priv(ctx), si);
}
return SAVE_STATUS(ctx, res);
@@ -115,8 +119,8 @@
else if (!ctx->iface || !ctx->priv)
res = VPX_CODEC_ERROR;
else {
- res = ctx->iface->dec.decode(ctx->priv->alg_priv, data, data_sz,
- user_priv, deadline);
+ res = ctx->iface->dec.decode(get_alg_priv(ctx), data, data_sz, user_priv,
+ deadline);
}
return SAVE_STATUS(ctx, res);
@@ -129,7 +133,7 @@
if (!ctx || !iter || !ctx->iface || !ctx->priv)
img = NULL;
else
- img = ctx->iface->dec.get_frame(ctx->priv->alg_priv, iter);
+ img = ctx->iface->dec.get_frame(get_alg_priv(ctx), iter);
return img;
}
@@ -185,7 +189,7 @@
!(ctx->iface->caps & VPX_CODEC_CAP_EXTERNAL_FRAME_BUFFER)) {
res = VPX_CODEC_ERROR;
} else {
- res = ctx->iface->dec.set_fb_fn(ctx->priv->alg_priv, cb_get, cb_release,
+ res = ctx->iface->dec.set_fb_fn(get_alg_priv(ctx), cb_get, cb_release,
cb_priv);
}
diff --git a/vpx/src/vpx_encoder.c b/vpx/src/vpx_encoder.c
index 736a8da..1903b55 100644
--- a/vpx/src/vpx_encoder.c
+++ b/vpx/src/vpx_encoder.c
@@ -20,6 +20,10 @@
#define SAVE_STATUS(ctx,var) (ctx?(ctx->err = var):var)
+static vpx_codec_alg_priv_t *get_alg_priv(vpx_codec_ctx_t *ctx) {
+ return (vpx_codec_alg_priv_t *)ctx->priv;
+}
+
vpx_codec_err_t vpx_codec_enc_init_ver(vpx_codec_ctx_t *ctx,
vpx_codec_iface_t *iface,
const vpx_codec_enc_cfg_t *cfg,
@@ -216,7 +220,7 @@
FLOATING_POINT_INIT();
if (num_enc == 1)
- res = ctx->iface->enc.encode(ctx->priv->alg_priv, img, pts,
+ res = ctx->iface->enc.encode(get_alg_priv(ctx), img, pts,
duration, flags, deadline);
else {
/* Multi-resolution encoding:
@@ -230,7 +234,7 @@
if (img) img += num_enc - 1;
for (i = num_enc - 1; i >= 0; i--) {
- if ((res = ctx->iface->enc.encode(ctx->priv->alg_priv, img, pts,
+ if ((res = ctx->iface->enc.encode(get_alg_priv(ctx), img, pts,
duration, flags, deadline)))
break;
@@ -259,7 +263,7 @@
else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
ctx->err = VPX_CODEC_INCAPABLE;
else
- pkt = ctx->iface->enc.get_cx_data(ctx->priv->alg_priv, iter);
+ pkt = ctx->iface->enc.get_cx_data(get_alg_priv(ctx), iter);
}
if (pkt && pkt->kind == VPX_CODEC_CX_FRAME_PKT) {
@@ -327,7 +331,7 @@
else if (!ctx->iface->enc.get_preview)
ctx->err = VPX_CODEC_INCAPABLE;
else
- img = ctx->iface->enc.get_preview(ctx->priv->alg_priv);
+ img = ctx->iface->enc.get_preview(get_alg_priv(ctx));
}
return img;
@@ -345,7 +349,7 @@
else if (!ctx->iface->enc.get_glob_hdrs)
ctx->err = VPX_CODEC_INCAPABLE;
else
- buf = ctx->iface->enc.get_glob_hdrs(ctx->priv->alg_priv);
+ buf = ctx->iface->enc.get_glob_hdrs(get_alg_priv(ctx));
}
return buf;
@@ -361,7 +365,7 @@
else if (!(ctx->iface->caps & VPX_CODEC_CAP_ENCODER))
res = VPX_CODEC_INCAPABLE;
else
- res = ctx->iface->enc.cfg_set(ctx->priv->alg_priv, cfg);
+ res = ctx->iface->enc.cfg_set(get_alg_priv(ctx), cfg);
return SAVE_STATUS(ctx, res);
}
diff --git a/vpxenc.c b/vpxenc.c
index d8fc553..ab91d50 100644
--- a/vpxenc.c
+++ b/vpxenc.c
@@ -321,7 +321,7 @@
static const arg_def_t noise_sens = ARG_DEF(NULL, "noise-sensitivity", 1,
"Noise sensitivity (frames to blur)");
static const arg_def_t sharpness = ARG_DEF(NULL, "sharpness", 1,
- "Filter sharpness (0-7)");
+ "Loop filter sharpness (0..7)");
static const arg_def_t static_thresh = ARG_DEF(NULL, "static-thresh", 1,
"Motion detection threshold");
static const arg_def_t cpu_used = ARG_DEF(NULL, "cpu-used", 1,
@@ -329,11 +329,11 @@
static const arg_def_t auto_altref = ARG_DEF(NULL, "auto-alt-ref", 1,
"Enable automatic alt reference frames");
static const arg_def_t arnr_maxframes = ARG_DEF(NULL, "arnr-maxframes", 1,
- "AltRef Max Frames");
+ "AltRef max frames (0..15)");
static const arg_def_t arnr_strength = ARG_DEF(NULL, "arnr-strength", 1,
- "AltRef Strength");
+ "AltRef filter strength (0..6)");
static const arg_def_t arnr_type = ARG_DEF(NULL, "arnr-type", 1,
- "AltRef Type");
+ "AltRef type");
static const struct arg_enum_list tuning_enum[] = {
{"psnr", VP8_TUNE_PSNR},
{"ssim", VP8_TUNE_SSIM},
@@ -378,7 +378,7 @@
"Adaptive quantization mode (0: off (default), 1: variance 2: complexity, "
"3: cyclic refresh)");
static const arg_def_t frame_periodic_boost = ARG_DEF(
- NULL, "frame_boost", 1,
+ NULL, "frame-boost", 1,
"Enable frame periodic boost (0: off (default), 1: on)");
static const struct arg_enum_list tune_content_enum[] = {
