Merge "Rework non-RD mode decision coding flow"
diff --git a/docs.mk b/docs.mk
index 9426f76..797b466 100644
--- a/docs.mk
+++ b/docs.mk
@@ -30,7 +30,9 @@
EXAMPLE_PATH += $(SRC_PATH_BARE) #for CHANGELOG, README, etc
+EXAMPLE_PATH += $(SRC_PATH_BARE)/examples
+doxyfile: $(if $(findstring examples, $(ALL_TARGETS)),examples.doxy)
doxyfile: libs.doxy_template libs.doxy
@echo " [CREATE] $@"
@cat $^ > $@
diff --git a/examples.mk b/examples.mk
index e4abcf7..40756e1 100644
--- a/examples.mk
+++ b/examples.mk
@@ -285,3 +285,36 @@
$(addprefix bin/$(p)/,$(ALL_EXAMPLES_BASENAME:.c=.exe)))
$(foreach proj,$(call enabled,PROJECTS),\
$(eval $(call vcproj_template,$(proj))))
+
+#
+# Documentation Rules
+#
+%.dox: %.c
+ @echo " [DOXY] $@"
+ @echo "/*!\page example_$(@F:.dox=) $(@F:.dox=)" > $@
+ @echo " \includelineno $(<F)" >> $@
+ @echo "*/" >> $@
+
+samples.dox: examples.mk
+ @echo " [DOXY] $@"
+ @echo "/*!\page samples Sample Code" > $@
+ @echo " This SDK includes a number of sample applications."\
+ "Each sample documents a feature of the SDK in both prose"\
+ "and the associated C code."\
+ "The following samples are included: ">>$@
+ @$(foreach ex,$(sort $(notdir $(EXAMPLES:.c=))),\
+ echo " - \subpage example_$(ex) $($(ex).DESCRIPTION)" >> $@;)
+ @echo >> $@
+ @echo " In addition, the SDK contains a number of utilities."\
+ "Since these utilities are built upon the concepts described"\
+ "in the sample code listed above, they are not documented in"\
+ "pieces like the samples are. Their source is included here"\
+ "for reference. The following utilities are included:" >> $@
+ @$(foreach ex,$(sort $(UTILS:.c=)),\
+ echo " - \subpage example_$(ex) $($(ex).DESCRIPTION)" >> $@;)
+ @echo "*/" >> $@
+
+CLEAN-OBJS += examples.doxy samples.dox $(ALL_EXAMPLES:.c=.dox)
+DOCS-yes += examples.doxy samples.dox
+examples.doxy: samples.dox $(ALL_EXAMPLES:.c=.dox)
+ @echo "INPUT += $^" > $@
diff --git a/examples/decode_to_md5.c b/examples/decode_to_md5.c
index aabac60..28d1ad5 100644
--- a/examples/decode_to_md5.c
+++ b/examples/decode_to_md5.c
@@ -115,7 +115,7 @@
size_t frame_size = 0;
const unsigned char *frame = vpx_video_reader_get_frame(reader,
&frame_size);
- if (vpx_codec_decode(&codec, frame, frame_size, NULL, 0))
+ if (vpx_codec_decode(&codec, frame, (unsigned int)frame_size, NULL, 0))
die_codec(&codec, "Failed to decode frame");
while ((img = vpx_codec_get_frame(&codec, &iter)) != NULL) {
diff --git a/examples/decode_with_drops.c b/examples/decode_with_drops.c
index c6f7d43..af1aa63 100644
--- a/examples/decode_with_drops.c
+++ b/examples/decode_with_drops.c
@@ -120,7 +120,7 @@
int skip;
const unsigned char *frame = vpx_video_reader_get_frame(reader,
&frame_size);
- if (vpx_codec_decode(&codec, frame, frame_size, NULL, 0))
+ if (vpx_codec_decode(&codec, frame, (unsigned int)frame_size, NULL, 0))
die_codec(&codec, "Failed to decode frame.");
++frame_cnt;
diff --git a/examples/error_resilient.c b/examples/error_resilient.c
index ef0a6c38..19235c8 100644
--- a/examples/error_resilient.c
+++ b/examples/error_resilient.c
@@ -118,7 +118,7 @@
return;
pts = pkt->data.frame.pts;
- mem_put_le32(header, pkt->data.frame.sz);
+ mem_put_le32(header, (unsigned int)pkt->data.frame.sz);
mem_put_le32(header+4, pts&0xFFFFFFFF);
mem_put_le32(header+8, pts >> 32);
diff --git a/examples/force_keyframe.c b/examples/force_keyframe.c
index f03b3d0..6531e47 100644
--- a/examples/force_keyframe.c
+++ b/examples/force_keyframe.c
@@ -119,7 +119,7 @@
return;
pts = pkt->data.frame.pts;
- mem_put_le32(header, pkt->data.frame.sz);
+ mem_put_le32(header, (unsigned int)pkt->data.frame.sz);
mem_put_le32(header+4, pts&0xFFFFFFFF);
mem_put_le32(header+8, pts >> 32);
diff --git a/examples/postproc.c b/examples/postproc.c
index 2912fe6..be08e92 100644
--- a/examples/postproc.c
+++ b/examples/postproc.c
@@ -118,7 +118,7 @@
};
// Decode the frame with 15ms deadline
- if (vpx_codec_decode(&codec, frame, frame_size, NULL, 15000))
+ if (vpx_codec_decode(&codec, frame, (unsigned int)frame_size, NULL, 15000))
die_codec(&codec, "Failed to decode frame");
while ((img = vpx_codec_get_frame(&codec, &iter)) != NULL) {
diff --git a/examples/simple_decoder.c b/examples/simple_decoder.c
index b0ca77d..8c15051 100644
--- a/examples/simple_decoder.c
+++ b/examples/simple_decoder.c
@@ -134,7 +134,7 @@
size_t frame_size = 0;
const unsigned char *frame = vpx_video_reader_get_frame(reader,
&frame_size);
- if (vpx_codec_decode(&codec, frame, frame_size, NULL, 0))
+ if (vpx_codec_decode(&codec, frame, (unsigned int)frame_size, NULL, 0))
die_codec(&codec, "Failed to decode frame.");
while ((img = vpx_codec_get_frame(&codec, &iter)) != NULL) {
diff --git a/examples/twopass_encoder.c b/examples/twopass_encoder.c
index f16db66..8bca18c 100644
--- a/examples/twopass_encoder.c
+++ b/examples/twopass_encoder.c
@@ -69,9 +69,9 @@
static void get_frame_stats(vpx_codec_ctx_t *ctx,
const vpx_image_t *img,
vpx_codec_pts_t pts,
- uint64_t duration,
+ unsigned int duration,
vpx_enc_frame_flags_t flags,
- uint64_t deadline,
+ unsigned int deadline,
vpx_fixed_buf_t *stats) {
vpx_codec_iter_t iter = NULL;
const vpx_codec_cx_pkt_t *pkt = NULL;
@@ -94,9 +94,9 @@
static void encode_frame(vpx_codec_ctx_t *ctx,
const vpx_image_t *img,
vpx_codec_pts_t pts,
- uint64_t duration,
+ unsigned int duration,
vpx_enc_frame_flags_t flags,
- uint64_t deadline,
+ unsigned int deadline,
VpxVideoWriter *writer) {
vpx_codec_iter_t iter = NULL;
const vpx_codec_cx_pkt_t *pkt = NULL;
diff --git a/examples/vp8_set_maps.c b/examples/vp8_set_maps.c
index ec9bc65..f3cc9a7 100644
--- a/examples/vp8_set_maps.c
+++ b/examples/vp8_set_maps.c
@@ -62,7 +62,7 @@
static void set_roi_map(const vpx_codec_enc_cfg_t *cfg,
vpx_codec_ctx_t *codec) {
- int i;
+ unsigned int i;
vpx_roi_map_t roi = {0};
roi.rows = cfg->g_h / 16;
@@ -95,7 +95,7 @@
static void set_active_map(const vpx_codec_enc_cfg_t *cfg,
vpx_codec_ctx_t *codec) {
- int i;
+ unsigned int i;
vpx_active_map_t map = {0};
map.rows = cfg->g_h / 16;
diff --git a/examples/vp8cx_set_ref.c b/examples/vp8cx_set_ref.c
index 5a67578..f87dd35 100644
--- a/examples/vp8cx_set_ref.c
+++ b/examples/vp8cx_set_ref.c
@@ -139,7 +139,7 @@
return;
pts = pkt->data.frame.pts;
- mem_put_le32(header, pkt->data.frame.sz);
+ mem_put_le32(header, (unsigned int)pkt->data.frame.sz);
mem_put_le32(header+4, pts&0xFFFFFFFF);
mem_put_le32(header+8, pts >> 32);
diff --git a/examples/vpx_temporal_scalable_patterns.c b/examples/vpx_temporal_scalable_patterns.c
index 28bb7ff..32e88e3 100644
--- a/examples/vpx_temporal_scalable_patterns.c
+++ b/examples/vpx_temporal_scalable_patterns.c
@@ -41,23 +41,23 @@
// Number of encoded non-key frames per layer.
int layer_enc_frames[VPX_TS_MAX_LAYERS];
// Framerate per layer layer (cumulative).
- float layer_framerate[VPX_TS_MAX_LAYERS];
+ double layer_framerate[VPX_TS_MAX_LAYERS];
// Target average frame size per layer (per-frame-bandwidth per layer).
- float layer_pfb[VPX_TS_MAX_LAYERS];
+ double layer_pfb[VPX_TS_MAX_LAYERS];
// Actual average frame size per layer.
- float layer_avg_frame_size[VPX_TS_MAX_LAYERS];
+ double layer_avg_frame_size[VPX_TS_MAX_LAYERS];
// Average rate mismatch per layer (|target - actual| / target).
- float layer_avg_rate_mismatch[VPX_TS_MAX_LAYERS];
+ double layer_avg_rate_mismatch[VPX_TS_MAX_LAYERS];
// Actual encoding bitrate per layer (cumulative).
- float layer_encoding_bitrate[VPX_TS_MAX_LAYERS];
+ double layer_encoding_bitrate[VPX_TS_MAX_LAYERS];
};
static void set_rate_control_metrics(struct RateControlMetrics *rc,
vpx_codec_enc_cfg_t *cfg) {
- int i = 0;
+ unsigned int i = 0;
// Set the layer (cumulative) framerate and the target layer (non-cumulative)
// per-frame-bandwidth, for the rate control encoding stats below.
- float framerate = cfg->g_timebase.den / cfg->g_timebase.num;
+ const double framerate = cfg->g_timebase.den / cfg->g_timebase.num;
rc->layer_framerate[0] = framerate / cfg->ts_rate_decimator[0];
rc->layer_pfb[0] = 1000.0 * cfg->ts_target_bitrate[0] /
rc->layer_framerate[0];
@@ -80,7 +80,7 @@
static void printout_rate_control_summary(struct RateControlMetrics *rc,
vpx_codec_enc_cfg_t *cfg,
int frame_cnt) {
- int i = 0;
+ unsigned int i = 0;
int check_num_frames = 0;
printf("Total number of processed frames: %d\n\n", frame_cnt -1);
printf("Rate control layer stats for %d layer(s):\n\n",
@@ -432,7 +432,7 @@
int frame_avail;
int got_data;
int flags = 0;
- int i;
+ unsigned int i;
int pts = 0; // PTS starts at 0.
int frame_duration = 1; // 1 timebase tick per frame.
int layering_mode = 0;
@@ -492,7 +492,7 @@
cfg.g_timebase.num = strtol(argv[6], NULL, 0);
cfg.g_timebase.den = strtol(argv[7], NULL, 0);
- for (i = 10; i < 10 + mode_to_num_layers[layering_mode]; ++i) {
+ for (i = 10; (int)i < 10 + mode_to_num_layers[layering_mode]; ++i) {
cfg.ts_target_bitrate[i - 10] = strtol(argv[i], NULL, 0);
}
@@ -516,9 +516,6 @@
// Disable automatic keyframe placement.
cfg.kf_min_dist = cfg.kf_max_dist = 3000;
- // Default setting for bitrate: used in special case of 1 layer (case 0).
- cfg.rc_target_bitrate = cfg.ts_target_bitrate[0];
-
set_temporal_layer_pattern(layering_mode,
&cfg,
layer_flags,
@@ -526,6 +523,10 @@
set_rate_control_metrics(&rc, &cfg);
+ // Target bandwidth for the whole stream.
+ // Set to ts_target_bitrate for highest layer (total bitrate).
+ cfg.rc_target_bitrate = cfg.ts_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]);
diff --git a/test/ivf_video_source.h b/test/ivf_video_source.h
index 3fbafbd..824a39d 100644
--- a/test/ivf_video_source.h
+++ b/test/ivf_video_source.h
@@ -94,14 +94,14 @@
virtual const uint8_t *cxdata() const {
return end_of_file_ ? NULL : compressed_frame_buf_;
}
- virtual const unsigned int frame_size() const { return frame_sz_; }
- virtual const unsigned int frame_number() const { return frame_; }
+ virtual size_t frame_size() const { return frame_sz_; }
+ virtual unsigned int frame_number() const { return frame_; }
protected:
std::string file_name_;
FILE *input_file_;
uint8_t *compressed_frame_buf_;
- unsigned int frame_sz_;
+ size_t frame_sz_;
unsigned int frame_;
bool end_of_file_;
};
diff --git a/test/video_source.h b/test/video_source.h
index 3d01d39..6d1855a 100644
--- a/test/video_source.h
+++ b/test/video_source.h
@@ -184,9 +184,9 @@
virtual const uint8_t *cxdata() const = 0;
- virtual const unsigned int frame_size() const = 0;
+ virtual size_t frame_size() const = 0;
- virtual const unsigned int frame_number() const = 0;
+ virtual unsigned int frame_number() const = 0;
};
} // namespace libvpx_test
diff --git a/test/webm_video_source.h b/test/webm_video_source.h
index 53b0ba2..f21cf98 100644
--- a/test/webm_video_source.h
+++ b/test/webm_video_source.h
@@ -169,8 +169,8 @@
virtual const uint8_t *cxdata() const {
return end_of_file_ ? NULL : buf_;
}
- virtual const unsigned int frame_size() const { return buf_sz_; }
- virtual const unsigned int frame_number() const { return frame_; }
+ virtual size_t frame_size() const { return buf_sz_; }
+ virtual unsigned int frame_number() const { return frame_; }
protected:
std::string file_name_;
diff --git a/vp9/common/vp9_entropy.c b/vp9/common/vp9_entropy.c
index 13e954e..bc12f9a 100644
--- a/vp9/common/vp9_entropy.c
+++ b/vp9/common/vp9_entropy.c
@@ -16,7 +16,7 @@
#include "vpx/vpx_integer.h"
-DECLARE_ALIGNED(16, const uint8_t, vp9_coefband_trans_8x8plus[1024]) = {
+const uint8_t vp9_coefband_trans_8x8plus[1024] = {
0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4,
4, 4, 4, 4, 4, 5,
// beyond MAXBAND_INDEX+1 all values are filled as 5
@@ -85,11 +85,11 @@
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
};
-DECLARE_ALIGNED(16, const uint8_t, vp9_coefband_trans_4x4[16]) = {
+const uint8_t vp9_coefband_trans_4x4[16] = {
0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 5, 5, 5,
};
-DECLARE_ALIGNED(16, const uint8_t, vp9_pt_energy_class[ENTROPY_TOKENS]) = {
+const uint8_t vp9_pt_energy_class[ENTROPY_TOKENS] = {
0, 1, 2, 3, 3, 4, 4, 5, 5, 5, 5, 5
};
diff --git a/vp9/common/vp9_entropy.h b/vp9/common/vp9_entropy.h
index d6b380f..aab8b53 100644
--- a/vp9/common/vp9_entropy.h
+++ b/vp9/common/vp9_entropy.h
@@ -42,7 +42,7 @@
#define ENTROPY_NODES 11
-extern DECLARE_ALIGNED(16, const uint8_t, vp9_pt_energy_class[ENTROPY_TOKENS]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_pt_energy_class[ENTROPY_TOKENS]);
#define EOB_MODEL_TOKEN 3
extern const vp9_tree_index vp9_coefmodel_tree[];
@@ -116,8 +116,8 @@
// This macro is currently unused but may be used by certain implementations
#define MAXBAND_INDEX 21
-extern DECLARE_ALIGNED(16, const uint8_t, vp9_coefband_trans_8x8plus[1024]);
-extern DECLARE_ALIGNED(16, const uint8_t, vp9_coefband_trans_4x4[16]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_coefband_trans_8x8plus[1024]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_coefband_trans_4x4[16]);
static INLINE const uint8_t *get_band_translate(TX_SIZE tx_size) {
return tx_size == TX_4X4 ? vp9_coefband_trans_4x4
diff --git a/vp9/common/vp9_filter.c b/vp9/common/vp9_filter.c
index 546f603..7474a88 100644
--- a/vp9/common/vp9_filter.c
+++ b/vp9/common/vp9_filter.c
@@ -10,12 +10,9 @@
#include <assert.h>
-#include "vpx_ports/mem.h"
-
#include "vp9/common/vp9_filter.h"
-DECLARE_ALIGNED(256, const InterpKernel,
- vp9_bilinear_filters[SUBPEL_SHIFTS]) = {
+const InterpKernel vp9_bilinear_filters[SUBPEL_SHIFTS] = {
{ 0, 0, 0, 128, 0, 0, 0, 0 },
{ 0, 0, 0, 120, 8, 0, 0, 0 },
{ 0, 0, 0, 112, 16, 0, 0, 0 },
@@ -35,8 +32,7 @@
};
// Lagrangian interpolation filter
-DECLARE_ALIGNED(256, const InterpKernel,
- vp9_sub_pel_filters_8[SUBPEL_SHIFTS]) = {
+const InterpKernel vp9_sub_pel_filters_8[SUBPEL_SHIFTS] = {
{ 0, 0, 0, 128, 0, 0, 0, 0},
{ 0, 1, -5, 126, 8, -3, 1, 0},
{ -1, 3, -10, 122, 18, -6, 2, 0},
@@ -56,8 +52,7 @@
};
// DCT based filter
-DECLARE_ALIGNED(256, const InterpKernel,
- vp9_sub_pel_filters_8s[SUBPEL_SHIFTS]) = {
+const InterpKernel vp9_sub_pel_filters_8s[SUBPEL_SHIFTS] = {
{0, 0, 0, 128, 0, 0, 0, 0},
{-1, 3, -7, 127, 8, -3, 1, 0},
{-2, 5, -13, 125, 17, -6, 3, -1},
@@ -77,8 +72,7 @@
};
// freqmultiplier = 0.5
-DECLARE_ALIGNED(256, const InterpKernel,
- vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS]) = {
+const InterpKernel vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS] = {
{ 0, 0, 0, 128, 0, 0, 0, 0},
{-3, -1, 32, 64, 38, 1, -3, 0},
{-2, -2, 29, 63, 41, 2, -3, 0},
diff --git a/vp9/common/vp9_filter.h b/vp9/common/vp9_filter.h
index 15610d7..29d3867 100644
--- a/vp9/common/vp9_filter.h
+++ b/vp9/common/vp9_filter.h
@@ -13,6 +13,8 @@
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+
#ifdef __cplusplus
extern "C" {
@@ -37,10 +39,14 @@
const InterpKernel *vp9_get_interp_kernel(INTERP_FILTER filter);
-extern const InterpKernel vp9_bilinear_filters[SUBPEL_SHIFTS];
-extern const InterpKernel vp9_sub_pel_filters_8[SUBPEL_SHIFTS];
-extern const InterpKernel vp9_sub_pel_filters_8s[SUBPEL_SHIFTS];
-extern const InterpKernel vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS];
+DECLARE_ALIGNED(256, extern const InterpKernel,
+ vp9_bilinear_filters[SUBPEL_SHIFTS]);
+DECLARE_ALIGNED(256, extern const InterpKernel,
+ vp9_sub_pel_filters_8[SUBPEL_SHIFTS]);
+DECLARE_ALIGNED(256, extern const InterpKernel,
+ vp9_sub_pel_filters_8s[SUBPEL_SHIFTS]);
+DECLARE_ALIGNED(256, extern const InterpKernel,
+ vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS]);
// The VP9_BILINEAR_FILTERS_2TAP macro returns a pointer to the bilinear
// filter kernel as a 2 tap filter.
diff --git a/vp9/common/vp9_pred_common.h b/vp9/common/vp9_pred_common.h
index 7f9e563..6c7a0d3 100644
--- a/vp9/common/vp9_pred_common.h
+++ b/vp9/common/vp9_pred_common.h
@@ -39,7 +39,7 @@
return above_sip + left_sip;
}
-static INLINE vp9_prob vp9_get_pred_prob_seg_id(struct segmentation *seg,
+static INLINE vp9_prob vp9_get_pred_prob_seg_id(const struct segmentation *seg,
const MACROBLOCKD *xd) {
return seg->pred_probs[vp9_get_pred_context_seg_id(xd)];
}
diff --git a/vp9/common/vp9_prob.c b/vp9/common/vp9_prob.c
index f9bc06e..a1befc6 100644
--- a/vp9/common/vp9_prob.c
+++ b/vp9/common/vp9_prob.c
@@ -10,7 +10,7 @@
#include "vp9/common/vp9_prob.h"
-DECLARE_ALIGNED(16, const uint8_t, vp9_norm[256]) = {
+const uint8_t vp9_norm[256] = {
0, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4,
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
diff --git a/vp9/decoder/vp9_decodemv.c b/vp9/decoder/vp9_decodemv.c
index 856c8b5..4de8db3 100644
--- a/vp9/decoder/vp9_decodemv.c
+++ b/vp9/decoder/vp9_decodemv.c
@@ -356,6 +356,11 @@
mbmi->uv_mode = read_intra_mode_uv(cm, r, mbmi->mode);
}
+static INLINE int is_mv_valid(const MV *mv) {
+ return mv->row > MV_LOW && mv->row < MV_UPP &&
+ mv->col > MV_LOW && mv->col < MV_UPP;
+}
+
static INLINE int assign_mv(VP9_COMMON *cm, MB_PREDICTION_MODE mode,
int_mv mv[2], int_mv ref_mv[2],
int_mv nearest_mv[2], int_mv near_mv[2],
@@ -367,14 +372,10 @@
case NEWMV: {
nmv_context_counts *const mv_counts = cm->frame_parallel_decoding_mode ?
NULL : &cm->counts.mv;
- read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv,
- &cm->fc.nmvc, mv_counts, allow_hp);
- if (is_compound)
- read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv,
- &cm->fc.nmvc, mv_counts, allow_hp);
for (i = 0; i < 1 + is_compound; ++i) {
- ret = ret && mv[i].as_mv.row < MV_UPP && mv[i].as_mv.row > MV_LOW;
- ret = ret && mv[i].as_mv.col < MV_UPP && mv[i].as_mv.col > MV_LOW;
+ read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, &cm->fc.nmvc, mv_counts,
+ allow_hp);
+ ret = ret && is_mv_valid(&mv[i].as_mv);
}
break;
}
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index 2a1836f..31ec069 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -34,7 +34,6 @@
#include "vp9/encoder/vp9_write_bit_buffer.h"
#ifdef ENTROPY_STATS
-vp9_coeff_stats tree_update_hist[TX_SIZES][PLANE_TYPES];
extern unsigned int active_section;
#endif
@@ -247,15 +246,15 @@
const nmv_context *nmvc = &cm->fc.nmvc;
MACROBLOCK *const x = &cpi->mb;
MACROBLOCKD *const xd = &x->e_mbd;
- struct segmentation *seg = &cm->seg;
+ const struct segmentation *const seg = &cm->seg;
MB_MODE_INFO *const mi = &m->mbmi;
- const MV_REFERENCE_FRAME rf = mi->ref_frame[0];
- const MV_REFERENCE_FRAME sec_rf = mi->ref_frame[1];
+ const MV_REFERENCE_FRAME ref0 = mi->ref_frame[0];
+ const MV_REFERENCE_FRAME ref1 = mi->ref_frame[1];
const MB_PREDICTION_MODE mode = mi->mode;
const int segment_id = mi->segment_id;
- int skip;
const BLOCK_SIZE bsize = mi->sb_type;
const int allow_hp = cm->allow_high_precision_mv;
+ int skip;
#ifdef ENTROPY_STATS
active_section = 9;
@@ -276,15 +275,15 @@
skip = write_skip(cpi, segment_id, m, bc);
if (!vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME))
- vp9_write(bc, rf != INTRA_FRAME, vp9_get_intra_inter_prob(cm, xd));
+ vp9_write(bc, ref0 != INTRA_FRAME, vp9_get_intra_inter_prob(cm, xd));
if (bsize >= BLOCK_8X8 && cm->tx_mode == TX_MODE_SELECT &&
- !(rf != INTRA_FRAME &&
+ !(ref0 != INTRA_FRAME &&
(skip || vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP)))) {
write_selected_tx_size(cpi, m, mi->tx_size, bsize, bc);
}
- if (rf == INTRA_FRAME) {
+ if (ref0 == INTRA_FRAME) {
#ifdef ENTROPY_STATS
active_section = 6;
#endif
@@ -306,7 +305,7 @@
} else {
vp9_prob *mv_ref_p;
encode_ref_frame(cpi, bc);
- mv_ref_p = cpi->common.fc.inter_mode_probs[mi->mode_context[rf]];
+ mv_ref_p = cm->fc.inter_mode_probs[mi->mode_context[ref0]];
#ifdef ENTROPY_STATS
active_section = 3;
@@ -316,7 +315,7 @@
if (!vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP)) {
if (bsize >= BLOCK_8X8) {
write_inter_mode(bc, mode, mv_ref_p);
- ++cm->counts.inter_mode[mi->mode_context[rf]][INTER_OFFSET(mode)];
+ ++cm->counts.inter_mode[mi->mode_context[ref0]][INTER_OFFSET(mode)];
}
}
@@ -336,21 +335,19 @@
for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
const int j = idy * 2 + idx;
- const MB_PREDICTION_MODE blockmode = m->bmi[j].as_mode;
- write_inter_mode(bc, blockmode, mv_ref_p);
- ++cm->counts.inter_mode[mi->mode_context[rf]]
- [INTER_OFFSET(blockmode)];
-
- if (blockmode == NEWMV) {
+ const MB_PREDICTION_MODE b_mode = m->bmi[j].as_mode;
+ write_inter_mode(bc, b_mode, mv_ref_p);
+ ++cm->counts.inter_mode[mi->mode_context[ref0]][INTER_OFFSET(b_mode)];
+ if (b_mode == NEWMV) {
#ifdef ENTROPY_STATS
active_section = 11;
#endif
vp9_encode_mv(cpi, bc, &m->bmi[j].as_mv[0].as_mv,
- &mi->ref_mvs[rf][0].as_mv, nmvc, allow_hp);
+ &mi->ref_mvs[ref0][0].as_mv, nmvc, allow_hp);
if (has_second_ref(mi))
vp9_encode_mv(cpi, bc, &m->bmi[j].as_mv[1].as_mv,
- &mi->ref_mvs[sec_rf][0].as_mv, nmvc, allow_hp);
+ &mi->ref_mvs[ref1][0].as_mv, nmvc, allow_hp);
}
}
}
@@ -359,11 +356,11 @@
active_section = 5;
#endif
vp9_encode_mv(cpi, bc, &mi->mv[0].as_mv,
- &mi->ref_mvs[rf][0].as_mv, nmvc, allow_hp);
+ &mi->ref_mvs[ref0][0].as_mv, nmvc, allow_hp);
if (has_second_ref(mi))
vp9_encode_mv(cpi, bc, &mi->mv[1].as_mv,
- &mi->ref_mvs[sec_rf][0].as_mv, nmvc, allow_hp);
+ &mi->ref_mvs[ref1][0].as_mv, nmvc, allow_hp);
}
}
}
@@ -628,10 +625,6 @@
if (s > 0 && newp != *oldp)
u = 1;
vp9_write(bc, u, upd);
-#ifdef ENTROPY_STATS
- if (!cpi->dummy_packing)
- ++tree_update_hist[tx_size][i][j][k][l][t][u];
-#endif
if (u) {
/* send/use new probability */
vp9_write_prob_diff_update(bc, newp, *oldp);
@@ -683,10 +676,6 @@
updates += u;
if (u == 0 && updates == 0) {
noupdates_before_first++;
-#ifdef ENTROPY_STATS
- if (!cpi->dummy_packing)
- ++tree_update_hist[tx_size][i][j][k][l][t][u];
-#endif
continue;
}
if (u == 1 && updates == 1) {
@@ -697,10 +686,6 @@
vp9_write(bc, 0, upd);
}
vp9_write(bc, u, upd);
-#ifdef ENTROPY_STATS
- if (!cpi->dummy_packing)
- ++tree_update_hist[tx_size][i][j][k][l][t][u];
-#endif
if (u) {
/* send/use new probability */
vp9_write_prob_diff_update(bc, newp, *oldp);
@@ -1272,7 +1257,7 @@
active_section = 7;
#endif
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
first_part_size = write_compressed_header(cpi, data);
data += first_part_size;
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index fa7cf1a..e2afb1a 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -398,7 +398,6 @@
MB_MODE_INFO *const mbmi = &xd->mi_8x8[0]->mbmi;
MODE_INFO *mi_addr = xd->mi_8x8[0];
- const int mb_mode_index = ctx->best_mode_index;
const int mis = cm->mode_info_stride;
const int mi_width = num_8x8_blocks_wide_lookup[bsize];
const int mi_height = num_8x8_blocks_high_lookup[bsize];
@@ -470,8 +469,8 @@
cpi->rd_tx_select_diff[i] += ctx->tx_rd_diff[i];
}
- if (frame_is_intra_only(cm)) {
#if CONFIG_INTERNAL_STATS
+ if (frame_is_intra_only(cm)) {
static const int kf_mode_index[] = {
THR_DC /*DC_PRED*/,
THR_V_PRED /*V_PRED*/,
@@ -484,12 +483,13 @@
THR_D63_PRED /*D63_PRED*/,
THR_TM /*TM_PRED*/,
};
- cpi->mode_chosen_counts[kf_mode_index[mbmi->mode]]++;
-#endif
+ ++cpi->mode_chosen_counts[kf_mode_index[mbmi->mode]];
} else {
// Note how often each mode chosen as best
- cpi->mode_chosen_counts[mb_mode_index]++;
-
+ ++cpi->mode_chosen_counts[ctx->best_mode_index];
+ }
+#endif
+ if (!frame_is_intra_only(cm)) {
if (is_inter_block(mbmi)) {
if (mbmi->sb_type < BLOCK_8X8 || mbmi->mode == NEWMV) {
int_mv best_mv[2];
@@ -627,7 +627,7 @@
int orig_rdmult = x->rdmult;
double rdmult_ratio;
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
rdmult_ratio = 1.0; // avoid uninitialized warnings
// Use the lower precision, but faster, 32x32 fdct for mode selection.
@@ -683,8 +683,8 @@
activity_masking(cpi, x);
if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
- vp9_clear_system_state(); // __asm emms;
- x->rdmult = round(x->rdmult * rdmult_ratio);
+ vp9_clear_system_state();
+ x->rdmult = (int)round(x->rdmult * rdmult_ratio);
} else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
const int mi_offset = mi_row * cm->mi_cols + mi_col;
unsigned char complexity = cpi->complexity_map[mi_offset];
@@ -713,8 +713,8 @@
if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
x->rdmult = orig_rdmult;
if (*totalrate != INT_MAX) {
- vp9_clear_system_state(); // __asm emms;
- *totalrate = round(*totalrate * rdmult_ratio);
+ vp9_clear_system_state();
+ *totalrate = (int)round(*totalrate * rdmult_ratio);
}
}
else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
@@ -1049,11 +1049,10 @@
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mi_8x8[0]->mbmi;
- const int mb_mode_index = ctx->best_mode_index;
x->skip = ctx->skip;
- if (frame_is_intra_only(cm)) {
#if CONFIG_INTERNAL_STATS
+ if (frame_is_intra_only(cm)) {
static const int kf_mode_index[] = {
THR_DC /*DC_PRED*/,
THR_V_PRED /*V_PRED*/,
@@ -1067,10 +1066,12 @@
THR_TM /*TM_PRED*/,
};
++cpi->mode_chosen_counts[kf_mode_index[mbmi->mode]];
-#endif
} else {
// Note how often each mode chosen as best
- cpi->mode_chosen_counts[mb_mode_index]++;
+ ++cpi->mode_chosen_counts[ctx->best_mode_index];
+ }
+#endif
+ if (!frame_is_intra_only(cm)) {
if (is_inter_block(mbmi)) {
if (mbmi->sb_type < BLOCK_8X8 || mbmi->mode == NEWMV) {
int_mv best_mv[2];
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index 6f976f5..19421aa 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -345,6 +345,15 @@
pd->above_context, pd->left_context,
num_4x4_w, num_4x4_h);
}
+
+static INLINE void fdct32x32(int rd_transform,
+ const int16_t *src, int16_t *dst, int src_stride) {
+ if (rd_transform)
+ vp9_fdct32x32_rd(src, dst, src_stride);
+ else
+ vp9_fdct32x32(src, dst, src_stride);
+}
+
void vp9_xform_quant(MACROBLOCK *x, int plane, int block,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
MACROBLOCKD *const xd = &x->e_mbd;
@@ -364,10 +373,7 @@
switch (tx_size) {
case TX_32X32:
scan_order = &vp9_default_scan_orders[TX_32X32];
- if (x->use_lp32x32fdct)
- vp9_fdct32x32_rd(src_diff, coeff, diff_stride);
- else
- vp9_fdct32x32(src_diff, coeff, diff_stride);
+ fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
vp9_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, p->zbin_extra, eob, scan_order->scan,
@@ -533,10 +539,12 @@
uint8_t *src, *dst;
int16_t *src_diff;
uint16_t *eob = &p->eobs[block];
+ const int src_stride = p->src.stride;
+ const int dst_stride = pd->dst.stride;
int i, j;
txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
- dst = &pd->dst.buf[4 * (j * pd->dst.stride + i)];
- src = &p->src.buf[4 * (j * p->src.stride + i)];
+ dst = &pd->dst.buf[4 * (j * dst_stride + i)];
+ src = &p->src.buf[4 * (j * src_stride + i)];
src_diff = &p->src_diff[4 * (j * diff_stride + i)];
// if (x->optimize)
@@ -548,22 +556,19 @@
mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
vp9_predict_intra_block(xd, block >> 6, bwl, TX_32X32, mode,
x->skip_encode ? src : dst,
- x->skip_encode ? p->src.stride : pd->dst.stride,
- dst, pd->dst.stride, i, j, plane);
+ x->skip_encode ? src_stride : dst_stride,
+ dst, dst_stride, i, j, plane);
if (!x->skip_recode) {
vp9_subtract_block(32, 32, src_diff, diff_stride,
- src, p->src.stride, dst, pd->dst.stride);
- if (x->use_lp32x32fdct)
- vp9_fdct32x32_rd(src_diff, coeff, diff_stride);
- else
- vp9_fdct32x32(src_diff, coeff, diff_stride);
+ src, src_stride, dst, dst_stride);
+ fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
vp9_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, p->zbin_extra, eob, scan_order->scan,
scan_order->iscan);
}
if (!x->skip_encode && *eob)
- vp9_idct32x32_add(dqcoeff, dst, pd->dst.stride, *eob);
+ vp9_idct32x32_add(dqcoeff, dst, dst_stride, *eob);
break;
case TX_16X16:
tx_type = get_tx_type_16x16(pd->plane_type, xd);
@@ -571,11 +576,11 @@
mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
vp9_predict_intra_block(xd, block >> 4, bwl, TX_16X16, mode,
x->skip_encode ? src : dst,
- x->skip_encode ? p->src.stride : pd->dst.stride,
- dst, pd->dst.stride, i, j, plane);
+ x->skip_encode ? src_stride : dst_stride,
+ dst, dst_stride, i, j, plane);
if (!x->skip_recode) {
vp9_subtract_block(16, 16, src_diff, diff_stride,
- src, p->src.stride, dst, pd->dst.stride);
+ src, src_stride, dst, dst_stride);
vp9_fht16x16(src_diff, coeff, diff_stride, tx_type);
vp9_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
@@ -583,7 +588,7 @@
scan_order->iscan);
}
if (!x->skip_encode && *eob)
- vp9_iht16x16_add(tx_type, dqcoeff, dst, pd->dst.stride, *eob);
+ vp9_iht16x16_add(tx_type, dqcoeff, dst, dst_stride, *eob);
break;
case TX_8X8:
tx_type = get_tx_type_8x8(pd->plane_type, xd);
@@ -591,11 +596,11 @@
mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
vp9_predict_intra_block(xd, block >> 2, bwl, TX_8X8, mode,
x->skip_encode ? src : dst,
- x->skip_encode ? p->src.stride : pd->dst.stride,
- dst, pd->dst.stride, i, j, plane);
+ x->skip_encode ? src_stride : dst_stride,
+ dst, dst_stride, i, j, plane);
if (!x->skip_recode) {
vp9_subtract_block(8, 8, src_diff, diff_stride,
- src, p->src.stride, dst, pd->dst.stride);
+ src, src_stride, dst, dst_stride);
vp9_fht8x8(src_diff, coeff, diff_stride, tx_type);
vp9_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff,
@@ -603,7 +608,7 @@
scan_order->iscan);
}
if (!x->skip_encode && *eob)
- vp9_iht8x8_add(tx_type, dqcoeff, dst, pd->dst.stride, *eob);
+ vp9_iht8x8_add(tx_type, dqcoeff, dst, dst_stride, *eob);
break;
case TX_4X4:
tx_type = get_tx_type_4x4(pd->plane_type, xd, block);
@@ -615,12 +620,12 @@
vp9_predict_intra_block(xd, block, bwl, TX_4X4, mode,
x->skip_encode ? src : dst,
- x->skip_encode ? p->src.stride : pd->dst.stride,
- dst, pd->dst.stride, i, j, plane);
+ x->skip_encode ? src_stride : dst_stride,
+ dst, dst_stride, i, j, plane);
if (!x->skip_recode) {
vp9_subtract_block(4, 4, src_diff, diff_stride,
- src, p->src.stride, dst, pd->dst.stride);
+ src, src_stride, dst, dst_stride);
if (tx_type != DCT_DCT)
vp9_fht4x4(src_diff, coeff, diff_stride, tx_type);
else
@@ -636,9 +641,9 @@
// this is like vp9_short_idct4x4 but has a special case around eob<=1
// which is significant (not just an optimization) for the lossless
// case.
- xd->itxm_add(dqcoeff, dst, pd->dst.stride, *eob);
+ xd->itxm_add(dqcoeff, dst, dst_stride, *eob);
else
- vp9_iht4x4_16_add(dqcoeff, dst, pd->dst.stride, tx_type);
+ vp9_iht4x4_16_add(dqcoeff, dst, dst_stride, tx_type);
}
break;
default:
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index c2aac3e..ddb901d 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -495,7 +495,7 @@
struct twopass_rc *const twopass = &cpi->twopass;
const MV zero_mv = {0, 0};
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
vp9_setup_src_planes(x, cpi->Source, 0, 0);
setup_pre_planes(xd, 0, lst_yv12, 0, 0, NULL);
@@ -544,7 +544,7 @@
double error_weight = 1.0;
const BLOCK_SIZE bsize = get_bsize(cm, mb_row, mb_col);
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
xd->plane[0].dst.buf = new_yv12->y_buffer + recon_yoffset;
xd->plane[1].dst.buf = new_yv12->u_buffer + recon_uvoffset;
@@ -565,7 +565,7 @@
// Do intra 16x16 prediction.
this_error = vp9_encode_intra(x, use_dc_pred);
if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
this_error = (int)(this_error * error_weight);
}
@@ -601,7 +601,7 @@
first_pass_motion_search(cpi, x, &best_ref_mv.as_mv, &mv.as_mv,
&motion_error);
if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
motion_error = (int)(motion_error * error_weight);
}
@@ -612,7 +612,7 @@
first_pass_motion_search(cpi, x, &zero_mv, &tmp_mv.as_mv,
&tmp_err);
if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
tmp_err = (int)(tmp_err * error_weight);
}
@@ -633,7 +633,7 @@
first_pass_motion_search(cpi, x, &zero_mv, &tmp_mv.as_mv,
&gf_motion_error);
if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
gf_motion_error = (int)(gf_motion_error * error_weight);
}
@@ -742,10 +742,10 @@
x->plane[2].src.buf += uv_mb_height * x->plane[1].src.stride -
uv_mb_height * cm->mb_cols;
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
}
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
{
FIRSTPASS_STATS fps;
@@ -1409,7 +1409,7 @@
twopass->gf_group_bits = 0;
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
start_pos = twopass->stats_in;
@@ -1907,7 +1907,7 @@
vp9_zero(next_frame);
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
start_position = twopass->stats_in;
cpi->common.frame_type = KEY_FRAME;
diff --git a/vp9/encoder/vp9_mbgraph.c b/vp9/encoder/vp9_mbgraph.c
index 7eacda2..44c1f90 100644
--- a/vp9/encoder/vp9_mbgraph.c
+++ b/vp9/encoder/vp9_mbgraph.c
@@ -419,7 +419,7 @@
golden_ref, cpi->Source);
}
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
separate_arf_mbs(cpi);
}
diff --git a/vp9/encoder/vp9_mcomp.c b/vp9/encoder/vp9_mcomp.c
index 94e49bd..10dee52e 100644
--- a/vp9/encoder/vp9_mcomp.c
+++ b/vp9/encoder/vp9_mcomp.c
@@ -855,6 +855,184 @@
square_num_candidates, square_candidates);
};
+// Number of candidates in first hex search
+#define FIRST_HEX_CANDIDATES 6
+// Index of previous hex search's best match
+#define PRE_BEST_CANDIDATE 6
+// Number of candidates in following hex search
+#define NEXT_HEX_CANDIDATES 3
+// Number of candidates in refining search
+#define REFINE_CANDIDATES 4
+
+int vp9_fast_hex_search(const MACROBLOCK *x,
+ MV *ref_mv,
+ int search_param,
+ int sad_per_bit,
+ const vp9_variance_fn_ptr_t *vfp,
+ int use_mvcost,
+ const MV *center_mv,
+ MV *best_mv) {
+ const MACROBLOCKD* const xd = &x->e_mbd;
+ static const MV hex[FIRST_HEX_CANDIDATES] = {
+ { -1, -2}, {1, -2}, {2, 0}, {1, 2}, { -1, 2}, { -2, 0}
+ };
+ static const MV next_chkpts[PRE_BEST_CANDIDATE][NEXT_HEX_CANDIDATES] = {
+ {{ -2, 0}, { -1, -2}, {1, -2}},
+ {{ -1, -2}, {1, -2}, {2, 0}},
+ {{1, -2}, {2, 0}, {1, 2}},
+ {{2, 0}, {1, 2}, { -1, 2}},
+ {{1, 2}, { -1, 2}, { -2, 0}},
+ {{ -1, 2}, { -2, 0}, { -1, -2}}
+ };
+ static const MV neighbors[REFINE_CANDIDATES] = {
+ {0, -1}, { -1, 0}, {1, 0}, {0, 1}
+ };
+ int i, j;
+
+ const uint8_t *what = x->plane[0].src.buf;
+ const int what_stride = x->plane[0].src.stride;
+ const int in_what_stride = xd->plane[0].pre[0].stride;
+ int br, bc;
+ MV this_mv;
+ unsigned int bestsad = 0x7fffffff;
+ unsigned int thissad;
+ const uint8_t *base_offset;
+ const uint8_t *this_offset;
+ int k = -1;
+ int best_site = -1;
+ const int max_hex_search = 512;
+ const int max_dia_search = 32;
+
+ const int *mvjsadcost = x->nmvjointsadcost;
+ int *mvsadcost[2] = {x->nmvsadcost[0], x->nmvsadcost[1]};
+
+ const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
+
+ // Adjust ref_mv to make sure it is within MV range
+ clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+ br = ref_mv->row;
+ bc = ref_mv->col;
+
+ // Check the start point
+ base_offset = xd->plane[0].pre[0].buf;
+ this_offset = base_offset + (br * in_what_stride) + bc;
+ this_mv.row = br;
+ this_mv.col = bc;
+ bestsad = vfp->sdf(what, what_stride, this_offset, in_what_stride, 0x7fffffff)
+ + mvsad_err_cost(&this_mv, &fcenter_mv, mvjsadcost, mvsadcost,
+ sad_per_bit);
+
+ // Initial 6-point hex search
+ if (check_bounds(x, br, bc, 2)) {
+ for (i = 0; i < FIRST_HEX_CANDIDATES; i++) {
+ this_mv.row = br + hex[i].row;
+ this_mv.col = bc + hex[i].col;
+ this_offset = base_offset + (this_mv.row * in_what_stride) + this_mv.col;
+ thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
+ bestsad);
+ CHECK_BETTER
+ }
+ } else {
+ for (i = 0; i < FIRST_HEX_CANDIDATES; i++) {
+ this_mv.row = br + hex[i].row;
+ this_mv.col = bc + hex[i].col;
+ if (!is_mv_in(x, &this_mv))
+ continue;
+ this_offset = base_offset + (this_mv.row * in_what_stride) + this_mv.col;
+ thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
+ bestsad);
+ CHECK_BETTER
+ }
+ }
+
+ // Continue hex search if we find a better match in first round
+ if (best_site != -1) {
+ br += hex[best_site].row;
+ bc += hex[best_site].col;
+ k = best_site;
+
+ // Allow search covering maximum MV range
+ for (j = 1; j < max_hex_search; j++) {
+ best_site = -1;
+
+ if (check_bounds(x, br, bc, 2)) {
+ for (i = 0; i < 3; i++) {
+ this_mv.row = br + next_chkpts[k][i].row;
+ this_mv.col = bc + next_chkpts[k][i].col;
+ this_offset = base_offset + (this_mv.row * in_what_stride) +
+ this_mv.col;
+ thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
+ bestsad);
+ CHECK_BETTER
+ }
+ } else {
+ for (i = 0; i < 3; i++) {
+ this_mv.row = br + next_chkpts[k][i].row;
+ this_mv.col = bc + next_chkpts[k][i].col;
+ if (!is_mv_in(x, &this_mv))
+ continue;
+ this_offset = base_offset + (this_mv.row * in_what_stride) +
+ this_mv.col;
+ thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
+ bestsad);
+ CHECK_BETTER
+ }
+ }
+
+ if (best_site == -1) {
+ break;
+ } else {
+ br += next_chkpts[k][best_site].row;
+ bc += next_chkpts[k][best_site].col;
+ k += 5 + best_site;
+ if (k >= 12) k -= 12;
+ else if (k >= 6) k -= 6;
+ }
+ }
+ }
+
+ // Check 4 1-away neighbors
+ for (j = 0; j < max_dia_search; j++) {
+ best_site = -1;
+
+ if (check_bounds(x, br, bc, 1)) {
+ for (i = 0; i < REFINE_CANDIDATES; i++) {
+ this_mv.row = br + neighbors[i].row;
+ this_mv.col = bc + neighbors[i].col;
+ this_offset = base_offset + (this_mv.row * in_what_stride) +
+ this_mv.col;
+ thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
+ bestsad);
+ CHECK_BETTER
+ }
+ } else {
+ for (i = 0; i < REFINE_CANDIDATES; i++) {
+ this_mv.row = br + neighbors[i].row;
+ this_mv.col = bc + neighbors[i].col;
+ if (!is_mv_in(x, &this_mv))
+ continue;
+ this_offset = base_offset + (this_mv.row * in_what_stride) +
+ this_mv.col;
+ thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
+ bestsad);
+ CHECK_BETTER
+ }
+ }
+
+ if (best_site == -1) {
+ break;
+ } else {
+ br += neighbors[best_site].row;
+ bc += neighbors[best_site].col;
+ }
+ }
+
+ best_mv->row = br;
+ best_mv->col = bc;
+
+ return bestsad;
+}
+
#undef CHECK_BETTER
int vp9_full_range_search_c(const MACROBLOCK *x, MV *ref_mv, MV *best_mv,
diff --git a/vp9/encoder/vp9_mcomp.h b/vp9/encoder/vp9_mcomp.h
index 4414f3d..ff4b1df 100644
--- a/vp9/encoder/vp9_mcomp.h
+++ b/vp9/encoder/vp9_mcomp.h
@@ -75,6 +75,14 @@
int use_mvcost,
const MV *center_mv,
MV *best_mv);
+int vp9_fast_hex_search(const MACROBLOCK *x,
+ MV *ref_mv,
+ int search_param,
+ int sad_per_bit,
+ const vp9_variance_fn_ptr_t *vfp,
+ int use_mvcost,
+ const MV *center_mv,
+ MV *best_mv);
typedef int (fractional_mv_step_fp) (
const MACROBLOCK *x,
diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c
index 23274fc..73ac9bc 100644
--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -154,20 +154,22 @@
}
static void dealloc_compressor_data(VP9_COMP *cpi) {
+ VP9_COMMON *const cm = &cpi->common;
+
// Delete sementation map
vpx_free(cpi->segmentation_map);
- cpi->segmentation_map = 0;
- vpx_free(cpi->common.last_frame_seg_map);
- cpi->common.last_frame_seg_map = 0;
+ cpi->segmentation_map = NULL;
+ vpx_free(cm->last_frame_seg_map);
+ cm->last_frame_seg_map = NULL;
vpx_free(cpi->coding_context.last_frame_seg_map_copy);
- cpi->coding_context.last_frame_seg_map_copy = 0;
+ cpi->coding_context.last_frame_seg_map_copy = NULL;
vpx_free(cpi->complexity_map);
cpi->complexity_map = 0;
vpx_free(cpi->active_map);
cpi->active_map = 0;
- vp9_free_frame_buffers(&cpi->common);
+ vp9_free_frame_buffers(cm);
vp9_free_frame_buffer(&cpi->last_frame_uf);
vp9_free_frame_buffer(&cpi->scaled_source);
@@ -194,19 +196,20 @@
// to a target value
// target q value
int vp9_compute_qdelta(const VP9_COMP *cpi, double qstart, double qtarget) {
+ const RATE_CONTROL *const rc = &cpi->rc;
+ int start_index = rc->worst_quality;
+ int target_index = rc->worst_quality;
int i;
- int start_index = cpi->rc.worst_quality;
- int target_index = cpi->rc.worst_quality;
// Convert the average q value to an index.
- for (i = cpi->rc.best_quality; i < cpi->rc.worst_quality; i++) {
+ for (i = rc->best_quality; i < rc->worst_quality; ++i) {
start_index = i;
if (vp9_convert_qindex_to_q(i) >= qstart)
break;
}
// Convert the q target to an index
- for (i = cpi->rc.best_quality; i < cpi->rc.worst_quality; i++) {
+ for (i = rc->best_quality; i < rc->worst_quality; ++i) {
target_index = i;
if (vp9_convert_qindex_to_q(i) >= qtarget)
break;
@@ -221,25 +224,20 @@
static int compute_qdelta_by_rate(VP9_COMP *cpi, int base_q_index,
double rate_target_ratio) {
int i;
- int base_bits_per_mb;
- int target_bits_per_mb;
int target_index = cpi->rc.worst_quality;
- // Make SURE use of floating point in this function is safe.
- vp9_clear_system_state();
-
// Look up the current projected bits per block for the base index
- base_bits_per_mb = vp9_rc_bits_per_mb(cpi->common.frame_type,
- base_q_index, 1.0);
+ const int base_bits_per_mb = vp9_rc_bits_per_mb(cpi->common.frame_type,
+ base_q_index, 1.0);
// Find the target bits per mb based on the base value and given ratio.
- target_bits_per_mb = (int)(rate_target_ratio * base_bits_per_mb);
+ const int target_bits_per_mb = (int)(rate_target_ratio * base_bits_per_mb);
// Convert the q target to an index
- for (i = cpi->rc.best_quality; i < cpi->rc.worst_quality; i++) {
+ for (i = cpi->rc.best_quality; i < cpi->rc.worst_quality; ++i) {
target_index = i;
- if (vp9_rc_bits_per_mb(cpi->common.frame_type,
- i, 1.0) <= target_bits_per_mb )
+ if (vp9_rc_bits_per_mb(cpi->common.frame_type, i, 1.0) <=
+ target_bits_per_mb )
break;
}
@@ -249,11 +247,8 @@
// This function sets up a set of segments with delta Q values around
// the baseline frame quantizer.
static void setup_in_frame_q_adj(VP9_COMP *cpi) {
- VP9_COMMON *cm = &cpi->common;
- struct segmentation *seg = &cm->seg;
- // double q_ratio;
- int segment;
- int qindex_delta;
+ VP9_COMMON *const cm = &cpi->common;
+ struct segmentation *const seg = &cm->seg;
// Make SURE use of floating point in this function is safe.
vp9_clear_system_state();
@@ -261,6 +256,8 @@
if (cm->frame_type == KEY_FRAME ||
cpi->refresh_alt_ref_frame ||
(cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref)) {
+ int segment;
+
// Clear down the segment map
vpx_memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
@@ -278,16 +275,16 @@
// Use some of the segments for in frame Q adjustment
for (segment = 1; segment < 2; segment++) {
- qindex_delta = compute_qdelta_by_rate(cpi, cm->base_qindex,
- in_frame_q_adj_ratio[segment]);
+ const int qindex_delta = compute_qdelta_by_rate(cpi, cm->base_qindex,
+ in_frame_q_adj_ratio[segment]);
vp9_enable_segfeature(seg, segment, SEG_LVL_ALT_Q);
vp9_set_segdata(seg, segment, SEG_LVL_ALT_Q, qindex_delta);
}
}
}
static void configure_static_seg_features(VP9_COMP *cpi) {
- VP9_COMMON *cm = &cpi->common;
- struct segmentation *seg = &cm->seg;
+ VP9_COMMON *const cm = &cpi->common;
+ struct segmentation *const seg = &cm->seg;
int high_q = (int)(cpi->rc.avg_q > 48.0);
int qi_delta;
@@ -431,13 +428,13 @@
static void update_reference_segmentation_map(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
+ MODE_INFO **mi_8x8_ptr = cm->mi_grid_visible;
+ uint8_t *cache_ptr = cm->last_frame_seg_map;
int row, col;
- MODE_INFO **mi_8x8, **mi_8x8_ptr = cm->mi_grid_visible;
- uint8_t *cache_ptr = cm->last_frame_seg_map, *cache;
for (row = 0; row < cm->mi_rows; row++) {
- mi_8x8 = mi_8x8_ptr;
- cache = cache_ptr;
+ MODE_INFO **mi_8x8 = mi_8x8_ptr;
+ uint8_t *cache = cache_ptr;
for (col = 0; col < cm->mi_cols; col++, mi_8x8++, cache++)
cache[0] = mi_8x8[0]->mbmi.segment_id;
mi_8x8_ptr += cm->mode_info_stride;
@@ -860,6 +857,8 @@
}
sf->frame_parameter_update = 0;
sf->encode_breakout_thresh = 1000;
+
+ sf->search_method = FAST_HEX;
}
if (speed >= 6) {
sf->always_this_block_size = BLOCK_16X16;
@@ -878,8 +877,10 @@
if (speed < 0)
speed = -speed;
+#if CONFIG_INTERNAL_STATS
for (i = 0; i < MAX_MODES; ++i)
cpi->mode_chosen_counts[i] = 0;
+#endif
// best quality defaults
sf->frame_parameter_update = 1;
@@ -975,16 +976,17 @@
static void alloc_raw_frame_buffers(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
+ const VP9_CONFIG *oxcf = &cpi->oxcf;
- cpi->lookahead = vp9_lookahead_init(cpi->oxcf.width, cpi->oxcf.height,
+ cpi->lookahead = vp9_lookahead_init(oxcf->width, oxcf->height,
cm->subsampling_x, cm->subsampling_y,
- cpi->oxcf.lag_in_frames);
+ oxcf->lag_in_frames);
if (!cpi->lookahead)
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate lag buffers");
if (vp9_realloc_frame_buffer(&cpi->alt_ref_buffer,
- cpi->oxcf.width, cpi->oxcf.height,
+ oxcf->width, oxcf->height,
cm->subsampling_x, cm->subsampling_y,
VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
@@ -2221,7 +2223,7 @@
psnr->samples[1 + i] = samples;
psnr->psnr[1 + i] = vp9_mse2psnr(samples, 255.0, (double) sse);
- total_sse += (uint64_t)sse;
+ total_sse += sse;
total_samples += samples;
}
@@ -2740,7 +2742,7 @@
FILE *const f = fopen("tmp.stt", cm->current_video_frame ? "a" : "w");
int recon_err;
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
recon_err = vp9_calc_ss_err(cpi->Source, get_frame_new_buffer(cm));
@@ -2784,8 +2786,6 @@
for (i = 0; i < MAX_MODES; ++i)
fprintf(fmodes, "%5d ", cpi->mode_chosen_counts[i]);
- for (i = 0; i < MAX_REFS; ++i)
- fprintf(fmodes, "%5d ", cpi->sub8x8_mode_chosen_counts[i]);
fprintf(fmodes, "\n");
@@ -2799,7 +2799,7 @@
uint8_t *dest,
int q) {
VP9_COMMON *const cm = &cpi->common;
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
vp9_set_quantizer(cpi, q);
// Set up entropy context depending on frame type. The decoder mandates
@@ -2810,7 +2810,7 @@
if (cm->frame_type == KEY_FRAME) {
vp9_setup_key_frame(cpi);
} else {
- if (!cm->intra_only && !cm->error_resilient_mode) {
+ if (!cm->intra_only && !cm->error_resilient_mode && !cpi->use_svc) {
cpi->common.frame_context_idx = cpi->refresh_alt_ref_frame;
}
vp9_setup_inter_frame(cpi);
@@ -2828,7 +2828,7 @@
// Update the skip mb flag probabilities based on the distribution
// seen in the last encoder iteration.
// update_base_skip_probs(cpi);
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
}
static void encode_with_recode_loop(VP9_COMP *cpi,
@@ -2838,6 +2838,7 @@
int bottom_index,
int top_index) {
VP9_COMMON *const cm = &cpi->common;
+ RATE_CONTROL *const rc = &cpi->rc;
int loop_count = 0;
int loop = 0;
int overshoot_seen = 0;
@@ -2847,12 +2848,12 @@
int frame_under_shoot_limit;
// Decide frame size bounds
- vp9_rc_compute_frame_size_bounds(cpi, cpi->rc.this_frame_target,
+ vp9_rc_compute_frame_size_bounds(cpi, rc->this_frame_target,
&frame_under_shoot_limit,
&frame_over_shoot_limit);
do {
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
vp9_set_quantizer(cpi, q);
@@ -2865,7 +2866,7 @@
if (cm->frame_type == KEY_FRAME) {
vp9_setup_key_frame(cpi);
} else {
- if (!cm->intra_only && !cm->error_resilient_mode) {
+ if (!cm->intra_only && !cm->error_resilient_mode && !cpi->use_svc) {
cpi->common.frame_context_idx = cpi->refresh_alt_ref_frame;
}
vp9_setup_inter_frame(cpi);
@@ -2887,7 +2888,7 @@
// seen in the last encoder iteration.
// update_base_skip_probs(cpi);
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
// Dummy pack of the bitstream using up to date stats to get an
// accurate estimate of output frame size to determine if we need
@@ -2898,7 +2899,7 @@
if (!cpi->sf.use_pick_mode)
vp9_pack_bitstream(cpi, dest, size);
- cpi->rc.projected_frame_size = (int)(*size) << 3;
+ rc->projected_frame_size = (int)(*size) << 3;
vp9_restore_coding_context(cpi);
if (frame_over_shoot_limit == 0)
@@ -2909,8 +2910,8 @@
loop = 0;
} else {
if ((cm->frame_type == KEY_FRAME) &&
- cpi->rc.this_key_frame_forced &&
- (cpi->rc.projected_frame_size < cpi->rc.max_frame_bandwidth)) {
+ rc->this_key_frame_forced &&
+ (rc->projected_frame_size < rc->max_frame_bandwidth)) {
int last_q = q;
int kf_err = vp9_calc_ss_err(cpi->Source, get_frame_new_buffer(cm));
@@ -2923,9 +2924,9 @@
// The key frame is not good enough or we can afford
// to make it better without undue risk of popping.
if ((kf_err > high_err_target &&
- cpi->rc.projected_frame_size <= frame_over_shoot_limit) ||
+ rc->projected_frame_size <= frame_over_shoot_limit) ||
(kf_err > low_err_target &&
- cpi->rc.projected_frame_size <= frame_under_shoot_limit)) {
+ rc->projected_frame_size <= frame_under_shoot_limit)) {
// Lower q_high
q_high = q > q_low ? q - 1 : q_low;
@@ -2933,7 +2934,7 @@
q = (q * high_err_target) / kf_err;
q = MIN(q, (q_high + q_low) >> 1);
} else if (kf_err < low_err_target &&
- cpi->rc.projected_frame_size >= frame_under_shoot_limit) {
+ rc->projected_frame_size >= frame_under_shoot_limit) {
// The key frame is much better than the previous frame
// Raise q_low
q_low = q < q_high ? q + 1 : q_high;
@@ -2959,10 +2960,10 @@
// Update correction factor & compute new Q to try...
// Frame is too large
- if (cpi->rc.projected_frame_size > cpi->rc.this_frame_target) {
+ if (rc->projected_frame_size > rc->this_frame_target) {
// Special case if the projected size is > the max allowed.
- if (cpi->rc.projected_frame_size >= cpi->rc.max_frame_bandwidth)
- q_high = cpi->rc.worst_quality;
+ if (rc->projected_frame_size >= rc->max_frame_bandwidth)
+ q_high = rc->worst_quality;
// Raise Qlow as to at least the current value
q_low = q < q_high ? q + 1 : q_high;
@@ -2976,12 +2977,12 @@
// Update rate_correction_factor unless
vp9_rc_update_rate_correction_factors(cpi, 0);
- q = vp9_rc_regulate_q(cpi, cpi->rc.this_frame_target,
+ q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
bottom_index, MAX(q_high, top_index));
while (q < q_low && retries < 10) {
vp9_rc_update_rate_correction_factors(cpi, 0);
- q = vp9_rc_regulate_q(cpi, cpi->rc.this_frame_target,
+ q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
bottom_index, MAX(q_high, top_index));
retries++;
}
@@ -2997,7 +2998,7 @@
q = (q_high + q_low) / 2;
} else {
vp9_rc_update_rate_correction_factors(cpi, 0);
- q = vp9_rc_regulate_q(cpi, cpi->rc.this_frame_target,
+ q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
bottom_index, top_index);
// Special case reset for qlow for constrained quality.
// This should only trigger where there is very substantial
@@ -3010,7 +3011,7 @@
while (q > q_high && retries < 10) {
vp9_rc_update_rate_correction_factors(cpi, 0);
- q = vp9_rc_regulate_q(cpi, cpi->rc.this_frame_target,
+ q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
bottom_index, top_index);
retries++;
}
@@ -3029,8 +3030,8 @@
}
// Special case for overlay frame.
- if (cpi->rc.is_src_frame_alt_ref &&
- (cpi->rc.projected_frame_size < cpi->rc.max_frame_bandwidth))
+ if (rc->is_src_frame_alt_ref &&
+ rc->projected_frame_size < rc->max_frame_bandwidth)
loop = 0;
if (loop) {
@@ -3098,8 +3099,8 @@
int top_index;
int bottom_index;
- SPEED_FEATURES *const sf = &cpi->sf;
- unsigned int max_mv_def = MIN(cm->width, cm->height);
+ const SPEED_FEATURES *const sf = &cpi->sf;
+ const unsigned int max_mv_def = MIN(cm->width, cm->height);
struct segmentation *const seg = &cm->seg;
set_ext_overrides(cpi);
@@ -3238,7 +3239,7 @@
if (!frame_is_intra_only(cm)) {
cm->interp_filter = DEFAULT_INTERP_FILTER;
/* TODO: Decide this more intelligently */
- set_high_precision_mv(cpi, (q < HIGH_PRECISION_MV_QTHRESH));
+ set_high_precision_mv(cpi, q < HIGH_PRECISION_MV_QTHRESH);
}
if (cpi->sf.recode_loop == DISALLOW_RECODE) {
@@ -3431,6 +3432,7 @@
static void check_initial_width(VP9_COMP *cpi, int subsampling_x,
int subsampling_y) {
VP9_COMMON *const cm = &cpi->common;
+
if (!cpi->initial_width) {
cm->subsampling_x = subsampling_x;
cm->subsampling_y = subsampling_y;
@@ -3444,12 +3446,12 @@
int vp9_receive_raw_frame(VP9_PTR ptr, unsigned int frame_flags,
YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
int64_t end_time) {
- VP9_COMP *cpi = (VP9_COMP *) ptr;
- VP9_COMMON *cm = &cpi->common;
- struct vpx_usec_timer timer;
- int res = 0;
- const int subsampling_x = sd->uv_width < sd->y_width;
- const int subsampling_y = sd->uv_height < sd->y_height;
+ VP9_COMP *cpi = (VP9_COMP *)ptr;
+ VP9_COMMON *cm = &cpi->common;
+ struct vpx_usec_timer timer;
+ int res = 0;
+ const int subsampling_x = sd->uv_width < sd->y_width;
+ const int subsampling_y = sd->uv_height < sd->y_height;
check_initial_width(cpi, subsampling_x, subsampling_y);
vpx_usec_timer_start(&timer);
@@ -3681,7 +3683,7 @@
*size = 0;
// Clear down mmx registers
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
/* find a free buffer for the new frame, releasing the reference previously
* held.
@@ -3843,22 +3845,23 @@
int vp9_get_preview_raw_frame(VP9_PTR comp, YV12_BUFFER_CONFIG *dest,
vp9_ppflags_t *flags) {
- VP9_COMP *cpi = (VP9_COMP *) comp;
+ VP9_COMP *cpi = (VP9_COMP *)comp;
+ VP9_COMMON *cm = &cpi->common;
- if (!cpi->common.show_frame) {
+ if (!cm->show_frame) {
return -1;
} else {
int ret;
#if CONFIG_VP9_POSTPROC
- ret = vp9_post_proc_frame(&cpi->common, dest, flags);
+ ret = vp9_post_proc_frame(cm, dest, flags);
#else
- if (cpi->common.frame_to_show) {
- *dest = *cpi->common.frame_to_show;
- dest->y_width = cpi->common.width;
- dest->y_height = cpi->common.height;
- dest->uv_width = cpi->common.width >> cpi->common.subsampling_x;
- dest->uv_height = cpi->common.height >> cpi->common.subsampling_y;
+ if (cm->frame_to_show) {
+ *dest = *cm->frame_to_show;
+ dest->y_width = cm->width;
+ dest->y_height = cm->height;
+ dest->uv_width = cm->width >> cm->subsampling_x;
+ dest->uv_height = cm->height >> cm->subsampling_y;
ret = 0;
} else {
ret = -1;
diff --git a/vp9/encoder/vp9_onyx_int.h b/vp9/encoder/vp9_onyx_int.h
index 7c0135e..7bcceed 100644
--- a/vp9/encoder/vp9_onyx_int.h
+++ b/vp9/encoder/vp9_onyx_int.h
@@ -136,7 +136,8 @@
NSTEP = 1,
HEX = 2,
BIGDIA = 3,
- SQUARE = 4
+ SQUARE = 4,
+ FAST_HEX = 5
} SEARCH_METHODS;
typedef enum {
@@ -486,9 +487,6 @@
// Ambient reconstruction err target for force key frames
int ambient_err;
- unsigned int mode_chosen_counts[MAX_MODES];
- unsigned int sub8x8_mode_chosen_counts[MAX_REFS];
-
int rd_threshes[MAX_SEGMENTS][BLOCK_SIZES][MAX_MODES];
int rd_thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
int rd_thresh_sub8x8[MAX_SEGMENTS][BLOCK_SIZES][MAX_REFS];
@@ -586,6 +584,8 @@
int fixed_divide[512];
#if CONFIG_INTERNAL_STATS
+ unsigned int mode_chosen_counts[MAX_MODES];
+
int count;
double total_y;
double total_u;
diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index a556063..f78ebfe 100644
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -354,7 +354,7 @@
int projected_size_based_on_q = 0;
// Clear down mmx registers to allow floating point in what follows
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
// Work out how big we would have expected the frame to be at this Q given
// the current correction factor.
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index a3e879b..70f9fab 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -274,7 +274,7 @@
MACROBLOCK *x = &cpi->mb;
int qindex, i;
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
// Further tests required to see if optimum is different
// for key frames, golden frames and arf frames.
@@ -431,7 +431,9 @@
// Note our transform coeffs are 8 times an orthogonal transform.
// Hence quantizer step is also 8 times. To get effective quantizer
// we need to divide by 8 before sending to modeling function.
- int i, rate_sum = 0, dist_sum = 0;
+ int i;
+ int64_t rate_sum = 0;
+ int64_t dist_sum = 0;
int ref = xd->mi_8x8[0]->mbmi.ref_frame[0];
unsigned int sse;
@@ -448,13 +450,13 @@
// Fast approximate the modelling function.
if (cpi->speed > 4) {
- int rate;
+ int64_t rate;
int64_t dist;
int64_t square_error = sse;
int quantizer = (pd->dequant[1] >> 3);
- if ( quantizer < 120)
- rate = (square_error * (280-quantizer) )>> 8;
+ if (quantizer < 120)
+ rate = (square_error * (280 - quantizer)) >> 8;
else
rate = 0;
dist = (square_error * quantizer) >> 8;
@@ -466,12 +468,12 @@
model_rd_from_var_lapndz(sse, 1 << num_pels_log2_lookup[bs],
pd->dequant[1] >> 3, &rate, &dist);
rate_sum += rate;
- dist_sum += (int)dist;
+ dist_sum += dist;
}
}
- *out_rate_sum = rate_sum;
- *out_dist_sum = (int64_t)dist_sum << 4;
+ *out_rate_sum = (int)rate_sum;
+ *out_dist_sum = dist_sum << 4;
}
static void model_rd_for_sb_y_tx(VP9_COMP *cpi, BLOCK_SIZE bsize,
@@ -2446,7 +2448,11 @@
// Further step/diamond searches as necessary
further_steps = (cpi->sf.max_step_search_steps - 1) - step_param;
- if (cpi->sf.search_method == HEX) {
+ if (cpi->sf.search_method == FAST_HEX) {
+ bestsme = vp9_fast_hex_search(x, &mvp_full, step_param, sadpb,
+ &cpi->fn_ptr[bsize], 1,
+ &ref_mv.as_mv, &tmp_mv->as_mv);
+ } else if (cpi->sf.search_method == HEX) {
bestsme = vp9_hex_search(x, &mvp_full, step_param, sadpb, 1,
&cpi->fn_ptr[bsize], 1,
&ref_mv.as_mv, &tmp_mv->as_mv);
@@ -2708,6 +2714,8 @@
int_mv tmp_mv;
single_motion_search(cpi, x, tile, bsize, mi_row, mi_col,
&tmp_mv, &rate_mv);
+ if (tmp_mv.as_int == INVALID_MV)
+ return INT64_MAX;
*rate2 += rate_mv;
frame_mv[refs[0]].as_int =
xd->mi_8x8[0]->bmi[0].as_mv[0].as_int = tmp_mv.as_int;
diff --git a/vp9/encoder/vp9_vaq.c b/vp9/encoder/vp9_vaq.c
index 896cd2c..600029b 100644
--- a/vp9/encoder/vp9_vaq.c
+++ b/vp9/encoder/vp9_vaq.c
@@ -44,7 +44,7 @@
double vp9_vaq_rdmult_ratio(int energy) {
ENERGY_IN_BOUNDS(energy);
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
return RDMULT_RATIO(energy);
}
@@ -52,7 +52,7 @@
double vp9_vaq_inv_q_ratio(int energy) {
ENERGY_IN_BOUNDS(energy);
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
return Q_RATIO(-energy);
}
@@ -63,7 +63,7 @@
assert(ENERGY_SPAN <= MAX_SEGMENTS);
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
base_ratio = 1.5;
@@ -88,7 +88,7 @@
seg->abs_delta = SEGMENT_DELTADATA;
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
for (i = ENERGY_MIN; i <= ENERGY_MAX; i++) {
int qindex_delta, segment_rdmult;
@@ -141,8 +141,8 @@
double energy;
unsigned int var = block_variance(cpi, x, bs);
- vp9_clear_system_state(); // __asm emms;
+ vp9_clear_system_state();
energy = 0.9 * (log(var + 1.0) - 10.0);
- return clamp(round(energy), ENERGY_MIN, ENERGY_MAX);
+ return clamp((int)round(energy), ENERGY_MIN, ENERGY_MAX);
}
diff --git a/vpx/src/svc_encodeframe.c b/vpx/src/svc_encodeframe.c
index 4f5ba6f..adce476 100644
--- a/vpx/src/svc_encodeframe.c
+++ b/vpx/src/svc_encodeframe.c
@@ -23,11 +23,13 @@
#include "vpx/vp8cx.h"
#include "vpx/vpx_encoder.h"
-#if defined(__MINGW32__) && !defined(MINGW_HAS_SECURE_API)
+#ifdef __MINGW32__
#define strtok_r strtok_s
+#ifndef MINGW_HAS_SECURE_API
// proto from /usr/x86_64-w64-mingw32/include/sec_api/string_s.h
_CRTIMP char *__cdecl strtok_s(char *str, const char *delim, char **context);
-#endif
+#endif /* MINGW_HAS_SECURE_API */
+#endif /* __MINGW32__ */
#ifdef _MSC_VER
#define strdup _strdup
diff --git a/vpx/vp8cx.h b/vpx/vp8cx.h
index d0ac1af..f7dde62 100644
--- a/vpx/vp8cx.h
+++ b/vpx/vp8cx.h
@@ -297,9 +297,16 @@
int alt_fb_idx; /**< alt reference frame frame buffer index */
} vpx_svc_parameters_t;
+/*!\brief vp9 svc layer parameters
+ *
+ * This defines the spatial and temporal layer id numbers for svc encoding.
+ * This is used with the #VP9E_SET_SVC_LAYER_ID control to set the spatial and
+ * temporal layer id for the current frame.
+ *
+ */
typedef struct vpx_svc_layer_id {
- int spatial_layer_id;
- int temporal_layer_id;
+ int spatial_layer_id; /**< Spatial layer id number. */
+ int temporal_layer_id; /**< Temporal layer id number. */
} vpx_svc_layer_id_t;
/*!\brief VP8 encoder control function parameter type
diff --git a/vpx/vpx_frame_buffer.h b/vpx/vpx_frame_buffer.h
index 6803759..e69df4b 100644
--- a/vpx/vpx_frame_buffer.h
+++ b/vpx/vpx_frame_buffer.h
@@ -11,6 +11,10 @@
#ifndef VPX_VPX_FRAME_BUFFER_H_
#define VPX_VPX_FRAME_BUFFER_H_
+/*!\file
+ * \brief Describes the decoder external frame buffer interface.
+ */
+
#ifdef __cplusplus
extern "C" {
#endif
