Merge "[spatial svc]Don't skip motion search in first pass encoding"
diff --git a/examples/vp9_spatial_svc_encoder.c b/examples/vp9_spatial_svc_encoder.c
index 5bc6575..8c87b2a 100644
--- a/examples/vp9_spatial_svc_encoder.c
+++ b/examples/vp9_spatial_svc_encoder.c
@@ -296,6 +296,7 @@
int frame_duration = 1; /* 1 timebase tick per frame */
FILE *infile = NULL;
int end_of_stream = 0;
+ int frame_size;
memset(&svc_ctx, 0, sizeof(svc_ctx));
svc_ctx.log_print = 1;
@@ -351,11 +352,10 @@
die_codec(&codec, "Failed to encode frame");
}
if (!(app_input.passes == 2 && app_input.pass == 1)) {
- if (vpx_svc_get_frame_size(&svc_ctx) > 0) {
+ while ((frame_size = vpx_svc_get_frame_size(&svc_ctx)) > 0) {
vpx_video_writer_write_frame(writer,
vpx_svc_get_buffer(&svc_ctx),
- vpx_svc_get_frame_size(&svc_ctx),
- pts);
+ frame_size, pts);
}
}
if (vpx_svc_get_rc_stats_buffer_size(&svc_ctx) > 0) {
diff --git a/test/invalid_file_test.cc b/test/invalid_file_test.cc
index 6ec5564..4933658 100644
--- a/test/invalid_file_test.cc
+++ b/test/invalid_file_test.cc
@@ -95,7 +95,8 @@
const char *const kVP9InvalidFileTests[] = {
"invalid-vp90-01.webm",
- "invalid-vp90-02.webm"
+ "invalid-vp90-02.webm",
+ "invalid-vp90-2-00-quantizer-00.webm.ivf.s5861_r01-05_b6-.ivf",
};
#define NELEMENTS(x) static_cast<int>(sizeof(x) / sizeof(x[0]))
diff --git a/test/svc_test.cc b/test/svc_test.cc
index db26a8e..f831e75 100644
--- a/test/svc_test.cc
+++ b/test/svc_test.cc
@@ -265,9 +265,17 @@
video.duration(), VPX_DL_GOOD_QUALITY);
EXPECT_EQ(VPX_CODEC_OK, res);
+ if (vpx_svc_get_frame_size(&svc_) == 0) {
+ // Flush encoder
+ res = vpx_svc_encode(&svc_, &codec_, NULL, 0,
+ video.duration(), VPX_DL_GOOD_QUALITY);
+ EXPECT_EQ(VPX_CODEC_OK, res);
+ }
+
+ int frame_size = vpx_svc_get_frame_size(&svc_);
+ EXPECT_GT(frame_size, 0);
const vpx_codec_err_t res_dec = decoder_->DecodeFrame(
- static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)),
- vpx_svc_get_frame_size(&svc_));
+ static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)), frame_size);
// this test fails with a decoder error
ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
@@ -277,6 +285,9 @@
svc_.spatial_layers = 2;
vpx_svc_set_scale_factors(&svc_, "4/16,16/16");
vpx_svc_set_quantizers(&svc_, "40,30", 0);
+ int decoded_frames = 0;
+ vpx_codec_err_t res_dec;
+ int frame_size;
vpx_codec_err_t res =
vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
@@ -291,13 +302,14 @@
// This frame is a keyframe.
res = vpx_svc_encode(&svc_, &codec_, video.img(), video.pts(),
video.duration(), VPX_DL_GOOD_QUALITY);
- ASSERT_EQ(VPX_CODEC_OK, res);
- EXPECT_EQ(1, vpx_svc_is_keyframe(&svc_));
- vpx_codec_err_t res_dec = decoder_->DecodeFrame(
- static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)),
- vpx_svc_get_frame_size(&svc_));
- ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ if ((frame_size = vpx_svc_get_frame_size(&svc_)) > 0) {
+ EXPECT_EQ((decoded_frames == 0), vpx_svc_is_keyframe(&svc_));
+ res_dec = decoder_->DecodeFrame(
+ static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)), frame_size);
+ ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ ++decoded_frames;
+ }
// FRAME 1
video.Next();
@@ -305,12 +317,14 @@
res = vpx_svc_encode(&svc_, &codec_, video.img(), video.pts(),
video.duration(), VPX_DL_GOOD_QUALITY);
ASSERT_EQ(VPX_CODEC_OK, res);
- EXPECT_EQ(0, vpx_svc_is_keyframe(&svc_));
- res_dec = decoder_->DecodeFrame(
- static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)),
- vpx_svc_get_frame_size(&svc_));
- ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ if ((frame_size = vpx_svc_get_frame_size(&svc_)) > 0) {
+ EXPECT_EQ((decoded_frames == 0), vpx_svc_is_keyframe(&svc_));
+ res_dec = decoder_->DecodeFrame(
+ static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)), frame_size);
+ ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ ++decoded_frames;
+ }
// FRAME 2
video.Next();
@@ -318,12 +332,29 @@
res = vpx_svc_encode(&svc_, &codec_, video.img(), video.pts(),
video.duration(), VPX_DL_GOOD_QUALITY);
ASSERT_EQ(VPX_CODEC_OK, res);
- EXPECT_EQ(0, vpx_svc_is_keyframe(&svc_));
- res_dec = decoder_->DecodeFrame(
- static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)),
- vpx_svc_get_frame_size(&svc_));
- ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ if ((frame_size = vpx_svc_get_frame_size(&svc_)) > 0) {
+ EXPECT_EQ((decoded_frames == 0), vpx_svc_is_keyframe(&svc_));
+ res_dec = decoder_->DecodeFrame(
+ static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)), frame_size);
+ ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ ++decoded_frames;
+ }
+
+ // Flush encoder
+ res = vpx_svc_encode(&svc_, &codec_, NULL, 0,
+ video.duration(), VPX_DL_GOOD_QUALITY);
+ EXPECT_EQ(VPX_CODEC_OK, res);
+
+ while ((frame_size = vpx_svc_get_frame_size(&svc_)) > 0) {
+ EXPECT_EQ((decoded_frames == 0), vpx_svc_is_keyframe(&svc_));
+ res_dec = decoder_->DecodeFrame(
+ static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)), frame_size);
+ ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ ++decoded_frames;
+ }
+
+ EXPECT_EQ(decoded_frames, 3);
}
TEST_F(SvcTest, GetLayerResolution) {
@@ -413,6 +444,9 @@
vpx_codec_destroy(&codec_);
// Second pass encode
+ int decoded_frames = 0;
+ vpx_codec_err_t res_dec;
+ int frame_size;
codec_enc_.g_pass = VPX_RC_LAST_PASS;
codec_enc_.rc_twopass_stats_in.buf = &stats_buf[0];
codec_enc_.rc_twopass_stats_in.sz = stats_buf.size();
@@ -427,12 +461,14 @@
res = vpx_svc_encode(&svc_, &codec_, video.img(), video.pts(),
video.duration(), VPX_DL_GOOD_QUALITY);
ASSERT_EQ(VPX_CODEC_OK, res);
- EXPECT_EQ(1, vpx_svc_is_keyframe(&svc_));
- vpx_codec_err_t res_dec = decoder_->DecodeFrame(
- static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)),
- vpx_svc_get_frame_size(&svc_));
- ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ if ((frame_size = vpx_svc_get_frame_size(&svc_)) > 0) {
+ EXPECT_EQ((decoded_frames == 0), vpx_svc_is_keyframe(&svc_));
+ res_dec = decoder_->DecodeFrame(
+ static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)), frame_size);
+ ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ ++decoded_frames;
+ }
// FRAME 1
video.Next();
@@ -440,12 +476,14 @@
res = vpx_svc_encode(&svc_, &codec_, video.img(), video.pts(),
video.duration(), VPX_DL_GOOD_QUALITY);
ASSERT_EQ(VPX_CODEC_OK, res);
- EXPECT_EQ(0, vpx_svc_is_keyframe(&svc_));
- res_dec = decoder_->DecodeFrame(
- static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)),
- vpx_svc_get_frame_size(&svc_));
- ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ if ((frame_size = vpx_svc_get_frame_size(&svc_)) > 0) {
+ EXPECT_EQ((decoded_frames == 0), vpx_svc_is_keyframe(&svc_));
+ res_dec = decoder_->DecodeFrame(
+ static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)), frame_size);
+ ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ ++decoded_frames;
+ }
// FRAME 2
video.Next();
@@ -453,12 +491,29 @@
res = vpx_svc_encode(&svc_, &codec_, video.img(), video.pts(),
video.duration(), VPX_DL_GOOD_QUALITY);
ASSERT_EQ(VPX_CODEC_OK, res);
- EXPECT_EQ(0, vpx_svc_is_keyframe(&svc_));
- res_dec = decoder_->DecodeFrame(
- static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)),
- vpx_svc_get_frame_size(&svc_));
- ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ if ((frame_size = vpx_svc_get_frame_size(&svc_)) > 0) {
+ EXPECT_EQ((decoded_frames == 0), vpx_svc_is_keyframe(&svc_));
+ res_dec = decoder_->DecodeFrame(
+ static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)), frame_size);
+ ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ ++decoded_frames;
+ }
+
+ // Flush encoder
+ res = vpx_svc_encode(&svc_, &codec_, NULL, 0,
+ video.duration(), VPX_DL_GOOD_QUALITY);
+ EXPECT_EQ(VPX_CODEC_OK, res);
+
+ while ((frame_size = vpx_svc_get_frame_size(&svc_)) > 0) {
+ EXPECT_EQ((decoded_frames == 0), vpx_svc_is_keyframe(&svc_));
+ res_dec = decoder_->DecodeFrame(
+ static_cast<const uint8_t *>(vpx_svc_get_buffer(&svc_)), frame_size);
+ ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
+ ++decoded_frames;
+ }
+
+ EXPECT_EQ(decoded_frames, 3);
}
} // namespace
diff --git a/test/test-data.sha1 b/test/test-data.sha1
index 2755835..bc6f77e 100644
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -644,3 +644,5 @@
e3ab35d4316c5e81325c50f5236ceca4bc0d35df vp90-2-15-segkey.webm.md5
9b7ca2cac09d34c4a5d296c1900f93b1e2f69d0d vp90-2-15-segkey_adpq.webm
8f46ba5f785d0c2170591a153e0d0d146a7c8090 vp90-2-15-segkey_adpq.webm.md5
+76024eb753cdac6a5e5703aaea189d35c3c30ac7 invalid-vp90-2-00-quantizer-00.webm.ivf.s5861_r01-05_b6-.ivf
+d3964f9dad9f60363c81b688324d95b4ec7c8038 invalid-vp90-2-00-quantizer-00.webm.ivf.s5861_r01-05_b6-.ivf.res
diff --git a/test/test.mk b/test/test.mk
index a839723..af344e5 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -766,6 +766,8 @@
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-01.webm.res
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-02.webm
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-02.webm.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-00-quantizer-00.webm.ivf.s5861_r01-05_b6-.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-00-quantizer-00.webm.ivf.s5861_r01-05_b6-.ivf.res
ifeq ($(CONFIG_DECODE_PERF_TESTS),yes)
# BBB VP9 streams
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index f36105f..9220a9e 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -685,6 +685,10 @@
while (max_ones-- && vp9_rb_read_bit(rb))
cm->log2_tile_cols++;
+ if (cm->log2_tile_cols > 6)
+ vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
+ "Invalid number of tile columns");
+
// rows
cm->log2_tile_rows = vp9_rb_read_bit(rb);
if (cm->log2_tile_rows)
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index ed52999..c69ed16 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -904,8 +904,7 @@
(cpi->refresh_golden_frame << cpi->alt_fb_idx);
} else {
int arf_idx = cpi->alt_fb_idx;
-
- if ((cpi->pass == 2) && cpi->multi_arf_enabled) {
+ if (cpi->pass == 2) {
const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
arf_idx = gf_group->arf_update_idx[gf_group->index];
}
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index f031c18..b9349a4 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -3318,7 +3318,10 @@
vp9_setup_pre_planes(xd, ref, cfg, mi_row, mi_col,
&xd->block_refs[ref]->sf);
}
- vp9_build_inter_predictors_sb(xd, mi_row, mi_col, MAX(bsize, BLOCK_8X8));
+ if (!cpi->sf.reuse_inter_pred_sby)
+ vp9_build_inter_predictors_sby(xd, mi_row, mi_col, MAX(bsize, BLOCK_8X8));
+
+ vp9_build_inter_predictors_sbuv(xd, mi_row, mi_col, MAX(bsize, BLOCK_8X8));
if (!x->skip) {
mbmi->skip = 1;
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index be29201..11fd584 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -106,7 +106,7 @@
}
}
-static void set_high_precision_mv(VP9_COMP *cpi, int allow_high_precision_mv) {
+void vp9_set_high_precision_mv(VP9_COMP *cpi, int allow_high_precision_mv) {
MACROBLOCK *const mb = &cpi->mb;
cpi->common.allow_high_precision_mv = allow_high_precision_mv;
if (cpi->common.allow_high_precision_mv) {
@@ -572,7 +572,7 @@
cm->reset_frame_context = 0;
vp9_reset_segment_features(&cm->seg);
- set_high_precision_mv(cpi, 0);
+ vp9_set_high_precision_mv(cpi, 0);
{
int i;
@@ -1514,7 +1514,7 @@
} else { /* For non key/golden frames */
if (cpi->refresh_alt_ref_frame) {
int arf_idx = cpi->alt_fb_idx;
- if ((cpi->pass == 2) && cpi->multi_arf_enabled) {
+ if (cpi->pass == 2) {
const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
arf_idx = gf_group->arf_update_idx[gf_group->index];
}
@@ -2117,7 +2117,7 @@
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);
+ vp9_set_high_precision_mv(cpi, q < HIGH_PRECISION_MV_QTHRESH);
}
if (cpi->sf.recode_loop == DISALLOW_RECODE) {
@@ -2298,12 +2298,22 @@
int res = 0;
const int subsampling_x = sd->uv_width < sd->y_width;
const int subsampling_y = sd->uv_height < sd->y_height;
+ const int is_spatial_svc = cpi->use_svc &&
+ (cpi->svc.number_temporal_layers == 1);
check_initial_width(cpi, subsampling_x, subsampling_y);
vpx_usec_timer_start(&timer);
- if (vp9_lookahead_push(cpi->lookahead,
- sd, time_stamp, end_time, frame_flags))
+
+#ifdef CONFIG_SPATIAL_SVC
+ if (is_spatial_svc)
+ res = vp9_svc_lookahead_push(cpi, cpi->lookahead, sd, time_stamp, end_time,
+ frame_flags);
+ else
+#endif
+ res = vp9_lookahead_push(cpi->lookahead,
+ sd, time_stamp, end_time, frame_flags);
+ if (res)
res = -1;
vpx_usec_timer_mark(&timer);
cpi->time_receive_data += vpx_usec_timer_elapsed(&timer);
@@ -2419,11 +2429,14 @@
YV12_BUFFER_CONFIG *force_src_buffer = NULL;
MV_REFERENCE_FRAME ref_frame;
int arf_src_index;
+ const int is_spatial_svc = cpi->use_svc &&
+ (cpi->svc.number_temporal_layers == 1);
if (!cpi)
return -1;
- if (cpi->svc.number_spatial_layers > 1 && cpi->pass == 2) {
+ if (is_spatial_svc && cpi->pass == 2) {
+ vp9_svc_lookahead_peek(cpi, cpi->lookahead, 0, 1);
vp9_restore_layer_context(cpi);
}
@@ -2432,7 +2445,7 @@
cpi->source = NULL;
cpi->last_source = NULL;
- set_high_precision_mv(cpi, ALTREF_HIGH_PRECISION_MV);
+ vp9_set_high_precision_mv(cpi, ALTREF_HIGH_PRECISION_MV);
// Normal defaults
cm->reset_frame_context = 0;
@@ -2446,7 +2459,14 @@
if (arf_src_index) {
assert(arf_src_index <= rc->frames_to_key);
- if ((cpi->source = vp9_lookahead_peek(cpi->lookahead, arf_src_index))) {
+#ifdef CONFIG_SPATIAL_SVC
+ if (is_spatial_svc)
+ cpi->source = vp9_svc_lookahead_peek(cpi, cpi->lookahead,
+ arf_src_index, 1);
+ else
+#endif
+ cpi->source = vp9_lookahead_peek(cpi->lookahead, arf_src_index);
+ if (cpi->source != NULL) {
cpi->alt_ref_source = cpi->source;
if (cpi->oxcf.arnr_max_frames > 0) {
@@ -2472,12 +2492,24 @@
if (!cpi->source) {
// Get last frame source.
if (cm->current_video_frame > 0) {
- if ((cpi->last_source = vp9_lookahead_peek(cpi->lookahead, -1)) == NULL)
+#ifdef CONFIG_SPATIAL_SVC
+ if (is_spatial_svc)
+ cpi->last_source = vp9_svc_lookahead_peek(cpi, cpi->lookahead, -1, 0);
+ else
+#endif
+ cpi->last_source = vp9_lookahead_peek(cpi->lookahead, -1);
+ if (cpi->last_source == NULL)
return -1;
}
// Read in the source frame.
- if ((cpi->source = vp9_lookahead_pop(cpi->lookahead, flush))) {
+#ifdef CONFIG_SPATIAL_SVC
+ if (is_spatial_svc)
+ cpi->source = vp9_svc_lookahead_pop(cpi, cpi->lookahead, flush);
+ else
+#endif
+ cpi->source = vp9_lookahead_pop(cpi->lookahead, flush);
+ if (cpi->source != NULL) {
cm->show_frame = 1;
cm->intra_only = 0;
@@ -2499,7 +2531,9 @@
*time_stamp = cpi->source->ts_start;
*time_end = cpi->source->ts_end;
- *frame_flags = cpi->source->flags;
+ *frame_flags =
+ (cpi->source->flags & VPX_EFLAG_FORCE_KF) ? FRAMEFLAGS_KEY : 0;
+
} else {
*size = 0;
if (flush && cpi->pass == 1 && !cpi->twopass.first_pass_done) {
@@ -2830,3 +2864,42 @@
int vp9_get_quantizer(VP9_COMP *cpi) {
return cpi->common.base_qindex;
}
+
+void vp9_apply_encoding_flags(VP9_COMP *cpi, vpx_enc_frame_flags_t flags) {
+ if (flags & (VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_GF |
+ VP8_EFLAG_NO_REF_ARF)) {
+ int ref = 7;
+
+ if (flags & VP8_EFLAG_NO_REF_LAST)
+ ref ^= VP9_LAST_FLAG;
+
+ if (flags & VP8_EFLAG_NO_REF_GF)
+ ref ^= VP9_GOLD_FLAG;
+
+ if (flags & VP8_EFLAG_NO_REF_ARF)
+ ref ^= VP9_ALT_FLAG;
+
+ vp9_use_as_reference(cpi, ref);
+ }
+
+ if (flags & (VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF |
+ VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_FORCE_GF |
+ VP8_EFLAG_FORCE_ARF)) {
+ int upd = 7;
+
+ if (flags & VP8_EFLAG_NO_UPD_LAST)
+ upd ^= VP9_LAST_FLAG;
+
+ if (flags & VP8_EFLAG_NO_UPD_GF)
+ upd ^= VP9_GOLD_FLAG;
+
+ if (flags & VP8_EFLAG_NO_UPD_ARF)
+ upd ^= VP9_ALT_FLAG;
+
+ vp9_update_reference(cpi, upd);
+ }
+
+ if (flags & VP8_EFLAG_NO_UPD_ENTROPY) {
+ vp9_update_entropy(cpi, 0);
+ }
+}
diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index fc2e007..47649a8 100644
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -517,10 +517,14 @@
int64_t vp9_rescale(int64_t val, int64_t num, int denom);
+void vp9_set_high_precision_mv(VP9_COMP *cpi, int allow_high_precision_mv);
+
YV12_BUFFER_CONFIG *vp9_scale_if_required(VP9_COMMON *cm,
YV12_BUFFER_CONFIG *unscaled,
YV12_BUFFER_CONFIG *scaled);
+void vp9_apply_encoding_flags(VP9_COMP *cpi, vpx_enc_frame_flags_t flags);
+
static INLINE void set_ref_ptrs(VP9_COMMON *cm, MACROBLOCKD *xd,
MV_REFERENCE_FRAME ref0,
MV_REFERENCE_FRAME ref1) {
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index e1a20dd..0261b98 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -1283,6 +1283,18 @@
return MAX((int)(((int64_t)boost * total_group_bits) / allocation_chunks), 0);
}
+// Current limit on maximum number of active arfs in a GF/ARF group.
+#define MAX_ACTIVE_ARFS 2
+#define ARF_SLOT1 2
+#define ARF_SLOT2 3
+// This function indirects the choice of buffers for arfs.
+// At the moment the values are fixed but this may change as part of
+// the integration process with other codec features that swap buffers around.
+static void get_arf_buffer_indices(unsigned char *arf_buffer_indices) {
+ arf_buffer_indices[0] = ARF_SLOT1;
+ arf_buffer_indices[1] = ARF_SLOT2;
+}
+
static void allocate_gf_group_bits(VP9_COMP *cpi, int64_t gf_group_bits,
double group_error, int gf_arf_bits) {
RATE_CONTROL *const rc = &cpi->rc;
@@ -1299,10 +1311,13 @@
double err_fraction;
int mid_boost_bits = 0;
int middle_frame_idx;
+ unsigned char arf_buffer_indices[MAX_ACTIVE_ARFS];
key_frame = cpi->common.frame_type == KEY_FRAME ||
vp9_is_upper_layer_key_frame(cpi);
+ get_arf_buffer_indices(arf_buffer_indices);
+
// For key frames the frame target rate is already set and it
// is also the golden frame.
if (!key_frame) {
@@ -1310,14 +1325,14 @@
twopass->gf_group.update_type[0] = OVERLAY_UPDATE;
twopass->gf_group.rf_level[0] = INTER_NORMAL;
twopass->gf_group.bit_allocation[0] = 0;
- twopass->gf_group.arf_update_idx[0] = 2;
- twopass->gf_group.arf_ref_idx[0] = 2;
+ twopass->gf_group.arf_update_idx[0] = arf_buffer_indices[0];
+ twopass->gf_group.arf_ref_idx[0] = arf_buffer_indices[0];
} else {
twopass->gf_group.update_type[0] = GF_UPDATE;
twopass->gf_group.rf_level[0] = GF_ARF_STD;
twopass->gf_group.bit_allocation[0] = gf_arf_bits;
- twopass->gf_group.arf_update_idx[0] = 2;
- twopass->gf_group.arf_ref_idx[0] = 2;
+ twopass->gf_group.arf_update_idx[0] = arf_buffer_indices[0];
+ twopass->gf_group.arf_ref_idx[0] = arf_buffer_indices[0];
}
// Step over the golden frame / overlay frame
@@ -1332,18 +1347,20 @@
// Store the bits to spend on the ARF if there is one.
if (rc->source_alt_ref_pending) {
- // A portion of the gf / arf extra bits are set asside for lower level
- // boosted frames in the middle of the group.
- mid_boost_bits = gf_arf_bits >> 5;
- gf_arf_bits -= (gf_arf_bits >> 5);
+ if (cpi->multi_arf_enabled) {
+ // A portion of the gf / arf extra bits are set asside for lower level
+ // boosted frames in the middle of the group.
+ mid_boost_bits += gf_arf_bits >> 5;
+ gf_arf_bits -= (gf_arf_bits >> 5);
+ }
twopass->gf_group.update_type[frame_index] = ARF_UPDATE;
twopass->gf_group.rf_level[frame_index] = GF_ARF_STD;
twopass->gf_group.bit_allocation[frame_index] = gf_arf_bits;
twopass->gf_group.arf_src_offset[frame_index] =
(unsigned char)(rc->baseline_gf_interval - 1);
- twopass->gf_group.arf_update_idx[frame_index] = 2;
- twopass->gf_group.arf_ref_idx[frame_index] = 2;
+ twopass->gf_group.arf_update_idx[frame_index] = arf_buffer_indices[0];
+ twopass->gf_group.arf_ref_idx[frame_index] = arf_buffer_indices[0];
++frame_index;
if (cpi->multi_arf_enabled) {
@@ -1352,8 +1369,8 @@
twopass->gf_group.rf_level[frame_index] = GF_ARF_LOW;
twopass->gf_group.arf_src_offset[frame_index] =
(unsigned char)((rc->baseline_gf_interval >> 1) - 1);
- twopass->gf_group.arf_update_idx[frame_index] = 3;
- twopass->gf_group.arf_ref_idx[frame_index] = 2;
+ twopass->gf_group.arf_update_idx[frame_index] = arf_buffer_indices[1];
+ twopass->gf_group.arf_ref_idx[frame_index] = arf_buffer_indices[0];
++frame_index;
}
}
@@ -1363,6 +1380,7 @@
// Allocate bits to the other frames in the group.
for (i = 0; i < rc->baseline_gf_interval - 1; ++i) {
+ int arf_idx = 0;
if (EOF == input_stats(twopass, &frame_stats))
break;
@@ -1379,17 +1397,11 @@
mid_boost_bits += (target_frame_size >> 4);
target_frame_size -= (target_frame_size >> 4);
- if (frame_index <= middle_frame_idx) {
- twopass->gf_group.arf_update_idx[frame_index] = 3;
- twopass->gf_group.arf_ref_idx[frame_index] = 3;
- } else {
- twopass->gf_group.arf_update_idx[frame_index] = 2;
- twopass->gf_group.arf_ref_idx[frame_index] = 2;
- }
- } else {
- twopass->gf_group.arf_update_idx[frame_index] = 2;
- twopass->gf_group.arf_ref_idx[frame_index] = 2;
+ if (frame_index <= middle_frame_idx)
+ arf_idx = 1;
}
+ twopass->gf_group.arf_update_idx[frame_index] = arf_buffer_indices[arf_idx];
+ twopass->gf_group.arf_ref_idx[frame_index] = arf_buffer_indices[arf_idx];
target_frame_size = clamp(target_frame_size, 0,
MIN(max_bits, (int)total_group_bits));
@@ -1401,23 +1413,27 @@
++frame_index;
}
+ // Note:
+ // We need to configure the frame at the end of the sequence + 1 that will be
+ // the start frame for the next group. Otherwise prior to the call to
+ // vp9_rc_get_second_pass_params() the data will be undefined.
+ twopass->gf_group.arf_update_idx[frame_index] = arf_buffer_indices[0];
+ twopass->gf_group.arf_ref_idx[frame_index] = arf_buffer_indices[0];
+
if (rc->source_alt_ref_pending) {
+ twopass->gf_group.update_type[frame_index] = OVERLAY_UPDATE;
+ twopass->gf_group.rf_level[frame_index] = INTER_NORMAL;
+
+ // Final setup for second arf and its overlay.
if (cpi->multi_arf_enabled) {
twopass->gf_group.bit_allocation[2] =
twopass->gf_group.bit_allocation[middle_frame_idx] + mid_boost_bits;
twopass->gf_group.update_type[middle_frame_idx] = OVERLAY_UPDATE;
twopass->gf_group.bit_allocation[middle_frame_idx] = 0;
}
-
- // Configure the overlay frame at the end of the sequence that will also
- // be the start frame of the next group. The reason for doing this here
- // is that on entry to vp9_get_compressed_data() for the overlay
- // frame, but before the call to vp9_rc_get_second_pass_params() the
- // data will otherwise be undefined.
- twopass->gf_group.update_type[frame_index] = OVERLAY_UPDATE;
- twopass->gf_group.rf_level[frame_index] = INTER_NORMAL;
- twopass->gf_group.arf_update_idx[frame_index] = 2;
- twopass->gf_group.arf_ref_idx[frame_index] = 2;
+ } else {
+ twopass->gf_group.update_type[frame_index] = GF_UPDATE;
+ twopass->gf_group.rf_level[frame_index] = GF_ARF_STD;
}
}
diff --git a/vp9/encoder/vp9_firstpass.h b/vp9/encoder/vp9_firstpass.h
index 61268ff..1ee56a3 100644
--- a/vp9/encoder/vp9_firstpass.h
+++ b/vp9/encoder/vp9_firstpass.h
@@ -51,12 +51,12 @@
typedef struct {
unsigned char index;
- RATE_FACTOR_LEVEL rf_level[MAX_LAG_BUFFERS * 2];
- FRAME_UPDATE_TYPE update_type[MAX_LAG_BUFFERS * 2];
- unsigned char arf_src_offset[MAX_LAG_BUFFERS * 2];
- unsigned char arf_update_idx[MAX_LAG_BUFFERS * 2];
- unsigned char arf_ref_idx[MAX_LAG_BUFFERS * 2];
- int bit_allocation[MAX_LAG_BUFFERS * 2];
+ RATE_FACTOR_LEVEL rf_level[(MAX_LAG_BUFFERS * 2) + 1];
+ FRAME_UPDATE_TYPE update_type[(MAX_LAG_BUFFERS * 2) + 1];
+ unsigned char arf_src_offset[(MAX_LAG_BUFFERS * 2) + 1];
+ unsigned char arf_update_idx[(MAX_LAG_BUFFERS * 2) + 1];
+ unsigned char arf_ref_idx[(MAX_LAG_BUFFERS * 2) + 1];
+ int bit_allocation[(MAX_LAG_BUFFERS * 2) + 1];
} GF_GROUP;
typedef struct {
diff --git a/vp9/encoder/vp9_lookahead.c b/vp9/encoder/vp9_lookahead.c
index abe71e6..e743517 100644
--- a/vp9/encoder/vp9_lookahead.c
+++ b/vp9/encoder/vp9_lookahead.c
@@ -18,18 +18,6 @@
#include "vp9/encoder/vp9_extend.h"
#include "vp9/encoder/vp9_lookahead.h"
-// The max of past frames we want to keep in the queue.
-#define MAX_PRE_FRAMES 1
-
-struct lookahead_ctx {
- unsigned int max_sz; /* Absolute size of the queue */
- unsigned int sz; /* Number of buffers currently in the queue */
- unsigned int read_idx; /* Read index */
- unsigned int write_idx; /* Write index */
- struct lookahead_entry *buf; /* Buffer list */
-};
-
-
/* Return the buffer at the given absolute index and increment the index */
static struct lookahead_entry *pop(struct lookahead_ctx *ctx,
unsigned int *idx) {
diff --git a/vp9/encoder/vp9_lookahead.h b/vp9/encoder/vp9_lookahead.h
index ff63c0d..f9cc3c8 100644
--- a/vp9/encoder/vp9_lookahead.h
+++ b/vp9/encoder/vp9_lookahead.h
@@ -14,6 +14,11 @@
#include "vpx_scale/yv12config.h"
#include "vpx/vpx_integer.h"
+#ifdef CONFIG_SPATIAL_SVC
+#include "vpx/vp8cx.h"
+#include "vpx/vpx_encoder.h"
+#endif
+
#ifdef __cplusplus
extern "C" {
#endif
@@ -25,10 +30,22 @@
int64_t ts_start;
int64_t ts_end;
unsigned int flags;
+
+#ifdef CONFIG_SPATIAL_SVC
+ vpx_svc_parameters_t svc_params[VPX_SS_MAX_LAYERS];
+#endif
};
+// The max of past frames we want to keep in the queue.
+#define MAX_PRE_FRAMES 1
-struct lookahead_ctx;
+struct lookahead_ctx {
+ unsigned int max_sz; /* Absolute size of the queue */
+ unsigned int sz; /* Number of buffers currently in the queue */
+ unsigned int read_idx; /* Read index */
+ unsigned int write_idx; /* Write index */
+ struct lookahead_entry *buf; /* Buffer list */
+};
/**\brief Initializes the lookahead stage
*
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 29896aa..149fafd 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -23,6 +23,7 @@
#include "vp9/common/vp9_reconintra.h"
#include "vp9/encoder/vp9_encoder.h"
+#include "vp9/encoder/vp9_pickmode.h"
#include "vp9/encoder/vp9_ratectrl.h"
#include "vp9/encoder/vp9_rdopt.h"
@@ -183,6 +184,22 @@
*out_dist_sum += dist << 4;
}
+static int get_pred_buffer(PRED_BUFFER *p, int len) {
+ int i;
+
+ for (i = 0; i < len; i++) {
+ if (!p[i].in_use) {
+ p[i].in_use = 1;
+ return i;
+ }
+ }
+ return -1;
+}
+
+static void free_pred_buffer(PRED_BUFFER *p) {
+ p->in_use = 0;
+}
+
// TODO(jingning) placeholder for inter-frame non-RD mode decision.
// this needs various further optimizations. to be continued..
int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
@@ -229,6 +246,31 @@
const int pred_filter_search = (((mi_row + mi_col) >> bsl) +
get_chessboard_index(cm)) % 2;
+ // For speed 6, the result of interp filter is reused later in actual encoding
+ // process.
+ int bh = num_4x4_blocks_high_lookup[bsize] << 2;
+ int bw = num_4x4_blocks_wide_lookup[bsize] << 2;
+ int pixels_in_block = bh * bw;
+ // tmp[3] points to dst buffer, and the other 3 point to allocated buffers.
+ PRED_BUFFER tmp[4];
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, pred_buf, 3 * 64 * 64);
+ struct buf_2d orig_dst = pd->dst;
+ PRED_BUFFER *best_pred = NULL;
+ PRED_BUFFER *this_mode_pred = NULL;
+ int i;
+
+ if (cpi->sf.reuse_inter_pred_sby) {
+ for (i = 0; i < 3; i++) {
+ tmp[i].data = &pred_buf[pixels_in_block * i];
+ tmp[i].stride = bw;
+ tmp[i].in_use = 0;
+ }
+
+ tmp[3].data = pd->dst.buf;
+ tmp[3].stride = pd->dst.stride;
+ tmp[3].in_use = 0;
+ }
+
x->skip_encode = cpi->sf.skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
x->skip = 0;
@@ -241,9 +283,9 @@
mbmi->ref_frame[0] = NONE;
mbmi->ref_frame[1] = NONE;
mbmi->tx_size = MIN(max_txsize_lookup[bsize],
- tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
- mbmi->interp_filter = cpi->common.interp_filter == SWITCHABLE ?
- EIGHTTAP : cpi->common.interp_filter;
+ tx_mode_to_biggest_tx_size[cm->tx_mode]);
+ mbmi->interp_filter = cm->interp_filter == SWITCHABLE ?
+ EIGHTTAP : cm->interp_filter;
mbmi->skip = 0;
mbmi->segment_id = segment_id;
@@ -324,6 +366,16 @@
// Search for the best prediction filter type, when the resulting
// motion vector is at sub-pixel accuracy level for luma component, i.e.,
// the last three bits are all zeros.
+ if (cpi->sf.reuse_inter_pred_sby) {
+ if (this_mode == NEARESTMV) {
+ this_mode_pred = &tmp[3];
+ } else {
+ this_mode_pred = &tmp[get_pred_buffer(tmp, 3)];
+ pd->dst.buf = this_mode_pred->data;
+ pd->dst.stride = bw;
+ }
+ }
+
if ((this_mode == NEWMV || filter_ref == SWITCHABLE) &&
pred_filter_search &&
((mbmi->mv[0].as_mv.row & 0x07) != 0 ||
@@ -334,6 +386,7 @@
unsigned int pf_sse[3];
int64_t best_cost = INT64_MAX;
INTERP_FILTER best_filter = SWITCHABLE, filter;
+ PRED_BUFFER *current_pred = this_mode_pred;
for (filter = EIGHTTAP; filter <= EIGHTTAP_SHARP; ++filter) {
int64_t cost;
@@ -345,12 +398,28 @@
vp9_get_switchable_rate(cpi) + pf_rate[filter],
pf_dist[filter]);
if (cost < best_cost) {
- best_filter = filter;
- best_cost = cost;
- skip_txfm = x->skip_txfm;
+ best_filter = filter;
+ best_cost = cost;
+ skip_txfm = x->skip_txfm;
+
+ if (cpi->sf.reuse_inter_pred_sby) {
+ if (this_mode_pred != current_pred) {
+ free_pred_buffer(this_mode_pred);
+ this_mode_pred = current_pred;
+ }
+
+ if (filter < EIGHTTAP_SHARP) {
+ current_pred = &tmp[get_pred_buffer(tmp, 3)];
+ pd->dst.buf = current_pred->data;
+ pd->dst.stride = bw;
+ }
+ }
}
}
+ if (cpi->sf.reuse_inter_pred_sby && this_mode_pred != current_pred)
+ free_pred_buffer(current_pred);
+
mbmi->interp_filter = best_filter;
rate = pf_rate[mbmi->interp_filter];
dist = pf_dist[mbmi->interp_filter];
@@ -449,6 +518,16 @@
best_pred_filter = mbmi->interp_filter;
best_ref_frame = ref_frame;
skip_txfm = x->skip_txfm;
+
+ if (cpi->sf.reuse_inter_pred_sby) {
+ if (best_pred != NULL)
+ free_pred_buffer(best_pred);
+
+ best_pred = this_mode_pred;
+ }
+ } else {
+ if (cpi->sf.reuse_inter_pred_sby)
+ free_pred_buffer(this_mode_pred);
}
if (x->skip)
@@ -456,6 +535,19 @@
}
}
+ // If best prediction is not in dst buf, then copy the prediction block from
+ // temp buf to dst buf.
+ if (cpi->sf.reuse_inter_pred_sby && best_pred->data != orig_dst.buf) {
+ uint8_t *copy_from, *copy_to;
+
+ pd->dst = orig_dst;
+ copy_to = pd->dst.buf;
+
+ copy_from = best_pred->data;
+
+ vp9_convolve_copy(copy_from, bw, copy_to, pd->dst.stride, NULL, 0, NULL, 0,
+ bw, bh);
+ }
mbmi->mode = best_mode;
mbmi->interp_filter = best_pred_filter;
@@ -469,12 +561,21 @@
if (!x->skip && best_rd > inter_mode_thresh &&
bsize <= cpi->sf.max_intra_bsize) {
for (this_mode = DC_PRED; this_mode <= DC_PRED; ++this_mode) {
+ if (cpi->sf.reuse_inter_pred_sby) {
+ pd->dst.buf = tmp[0].data;
+ pd->dst.stride = bw;
+ }
+
vp9_predict_intra_block(xd, 0, b_width_log2(bsize),
mbmi->tx_size, this_mode,
&p->src.buf[0], p->src.stride,
&pd->dst.buf[0], pd->dst.stride, 0, 0, 0);
model_rd_for_sb_y(cpi, bsize, x, xd, &rate, &dist, &var_y, &sse_y);
+
+ if (cpi->sf.reuse_inter_pred_sby)
+ pd->dst = orig_dst;
+
rate += cpi->mbmode_cost[this_mode];
rate += intra_cost_penalty;
this_rd = RDCOST(x->rdmult, x->rddiv, rate, dist);
@@ -492,6 +593,7 @@
}
}
}
+
#if CONFIG_DENOISING
vp9_denoiser_denoise(&cpi->denoiser, x, mi_row, mi_col, bsize);
#endif
diff --git a/vp9/encoder/vp9_pickmode.h b/vp9/encoder/vp9_pickmode.h
index a9c948d..3d89974 100644
--- a/vp9/encoder/vp9_pickmode.h
+++ b/vp9/encoder/vp9_pickmode.h
@@ -17,6 +17,12 @@
extern "C" {
#endif
+typedef struct {
+ uint8_t *data;
+ int stride;
+ int in_use;
+} PRED_BUFFER;
+
int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
const struct TileInfo *const tile,
int mi_row, int mi_col,
diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index 533f599..e110983 100644
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -892,21 +892,20 @@
*bottom_index = active_best_quality;
#if LIMIT_QRANGE_FOR_ALTREF_AND_KEY
+ vp9_clear_system_state();
{
- int qdelta = 0;
- vp9_clear_system_state();
-
- // Limit Q range for the adaptive loop.
- if ((cm->frame_type == KEY_FRAME || vp9_is_upper_layer_key_frame(cpi)) &&
- !rc->this_key_frame_forced) {
- qdelta = vp9_compute_qdelta_by_rate(&cpi->rc, cm->frame_type,
- active_worst_quality, 2.0);
- } else if (!rc->is_src_frame_alt_ref &&
- (oxcf->rc_mode != VPX_CBR) &&
- (cpi->refresh_alt_ref_frame)) {
- qdelta = vp9_compute_qdelta_by_rate(&cpi->rc, cm->frame_type,
- active_worst_quality, 1.75);
- }
+ const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+ const double rate_factor_deltas[RATE_FACTOR_LEVELS] = {
+ 1.00, // INTER_NORMAL
+ 1.00, // INTER_HIGH
+ 1.50, // GF_ARF_LOW
+ 1.75, // GF_ARF_STD
+ 2.00, // KF_STD
+ };
+ const double rate_factor =
+ rate_factor_deltas[gf_group->rf_level[gf_group->index]];
+ int qdelta = vp9_compute_qdelta_by_rate(&cpi->rc, cm->frame_type,
+ active_worst_quality, rate_factor);
*top_index = active_worst_quality + qdelta;
*top_index = (*top_index > *bottom_index) ? *top_index : *bottom_index;
}
@@ -1357,6 +1356,8 @@
// Extended interval for genuinely static scenes
rc->static_scene_max_gf_interval = oxcf->key_freq >> 1;
+ if (rc->static_scene_max_gf_interval > (MAX_LAG_BUFFERS * 2))
+ rc->static_scene_max_gf_interval = MAX_LAG_BUFFERS * 2;
if (is_altref_enabled(oxcf)) {
if (rc->static_scene_max_gf_interval > oxcf->lag_in_frames - 1)
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 9bd9bc6..9402d4a 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -2066,9 +2066,9 @@
return bsi->segment_rd;
}
-static void mv_pred(VP9_COMP *cpi, MACROBLOCK *x,
- uint8_t *ref_y_buffer, int ref_y_stride,
- int ref_frame, BLOCK_SIZE block_size ) {
+void vp9_mv_pred(VP9_COMP *cpi, MACROBLOCK *x,
+ uint8_t *ref_y_buffer, int ref_y_stride,
+ int ref_frame, BLOCK_SIZE block_size) {
MACROBLOCKD *xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
int_mv this_mv;
@@ -2207,12 +2207,12 @@
sizeof(*best_filter_diff) * SWITCHABLE_FILTER_CONTEXTS);
}
-static void setup_pred_block(const MACROBLOCKD *xd,
- struct buf_2d dst[MAX_MB_PLANE],
- const YV12_BUFFER_CONFIG *src,
- int mi_row, int mi_col,
- const struct scale_factors *scale,
- const struct scale_factors *scale_uv) {
+void vp9_setup_pred_block(const MACROBLOCKD *xd,
+ struct buf_2d dst[MAX_MB_PLANE],
+ const YV12_BUFFER_CONFIG *src,
+ int mi_row, int mi_col,
+ const struct scale_factors *scale,
+ const struct scale_factors *scale_uv) {
int i;
dst[0].buf = src->y_buffer;
@@ -2250,7 +2250,7 @@
// TODO(jkoleszar): Is the UV buffer ever used here? If so, need to make this
// use the UV scaling factors.
- setup_pred_block(xd, yv12_mb[ref_frame], yv12, mi_row, mi_col, sf, sf);
+ vp9_setup_pred_block(xd, yv12_mb[ref_frame], yv12, mi_row, mi_col, sf, sf);
// Gets an initial list of candidate vectors from neighbours and orders them
vp9_find_mv_refs(cm, xd, tile, mi, ref_frame, candidates, mi_row, mi_col);
@@ -2264,8 +2264,8 @@
// in full and choose the best as the centre point for subsequent searches.
// The current implementation doesn't support scaling.
if (!vp9_is_scaled(sf) && block_size >= BLOCK_8X8)
- mv_pred(cpi, x, yv12_mb[ref_frame][0].buf, yv12->y_stride,
- ref_frame, block_size);
+ vp9_mv_pred(cpi, x, yv12_mb[ref_frame][0].buf, yv12->y_stride,
+ ref_frame, block_size);
}
const YV12_BUFFER_CONFIG *vp9_get_scaled_ref_frame(const VP9_COMP *cpi,
diff --git a/vp9/encoder/vp9_rdopt.h b/vp9/encoder/vp9_rdopt.h
index 4f57293..3dfe2d0 100644
--- a/vp9/encoder/vp9_rdopt.h
+++ b/vp9/encoder/vp9_rdopt.h
@@ -191,6 +191,16 @@
return best_rd < ((int64_t)thresh * thresh_fact >> 5) || thresh == INT_MAX;
}
+void vp9_mv_pred(struct VP9_COMP *cpi, MACROBLOCK *x,
+ uint8_t *ref_y_buffer, int ref_y_stride,
+ int ref_frame, BLOCK_SIZE block_size);
+
+void vp9_setup_pred_block(const MACROBLOCKD *xd,
+ struct buf_2d dst[MAX_MB_PLANE],
+ const YV12_BUFFER_CONFIG *src,
+ int mi_row, int mi_col,
+ const struct scale_factors *scale,
+ const struct scale_factors *scale_uv);
#ifdef __cplusplus
} // extern "C"
#endif
diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index 9c3fb5e..d7017f2 100644
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -277,6 +277,9 @@
// is checked for a partition block. Later, we can try to allow large
// partitions to do intra mode checking.
sf->max_intra_bsize = BLOCK_8X8;
+
+ // This feature is only enabled when partition search is disabled.
+ sf->reuse_inter_pred_sby = 1;
}
if (speed >= 7) {
@@ -342,6 +345,7 @@
for (i = 0; i < BLOCK_SIZES; ++i)
sf->inter_mode_mask[i] = INTER_ALL;
sf->max_intra_bsize = BLOCK_64X64;
+ sf->reuse_inter_pred_sby = 0;
// This setting only takes effect when partition_search_type is set
// to FIXED_PARTITION.
sf->always_this_block_size = BLOCK_16X16;
diff --git a/vp9/encoder/vp9_speed_features.h b/vp9/encoder/vp9_speed_features.h
index f5d0b85..75070a7 100644
--- a/vp9/encoder/vp9_speed_features.h
+++ b/vp9/encoder/vp9_speed_features.h
@@ -353,6 +353,11 @@
// The threshold used in SOURCE_VAR_BASED_PARTITION search type.
unsigned int source_var_thresh;
+
+ // When partition is pre-set, the inter prediction result from pick_inter_mode
+ // can be reused in final block encoding process. It is enabled only for real-
+ // time mode speed 6.
+ int reuse_inter_pred_sby;
} SPEED_FEATURES;
struct VP9_COMP;
diff --git a/vp9/encoder/vp9_svc_layercontext.c b/vp9/encoder/vp9_svc_layercontext.c
index 84f3449..07c17b2 100644
--- a/vp9/encoder/vp9_svc_layercontext.c
+++ b/vp9/encoder/vp9_svc_layercontext.c
@@ -12,6 +12,7 @@
#include "vp9/encoder/vp9_encoder.h"
#include "vp9/encoder/vp9_svc_layercontext.h"
+#include "vp9/encoder/vp9_extend.h"
void vp9_init_layer_context(VP9_COMP *const cpi) {
SVC *const svc = &cpi->svc;
@@ -209,3 +210,101 @@
cpi->svc.spatial_layer_id > 0 &&
cpi->svc.layer_context[cpi->svc.spatial_layer_id].is_key_frame;
}
+
+int vp9_svc_lookahead_push(const VP9_COMP *const cpi, struct lookahead_ctx *ctx,
+ YV12_BUFFER_CONFIG *src, int64_t ts_start,
+ int64_t ts_end, unsigned int flags) {
+ struct lookahead_entry *buf;
+ int i, index;
+
+ if (vp9_lookahead_push(ctx, src, ts_start, ts_end, flags))
+ return 1;
+
+ index = ctx->write_idx - 1;
+ if (index < 0)
+ index += ctx->max_sz;
+
+ buf = ctx->buf + index;
+
+ if (buf == NULL)
+ return 1;
+
+ // Store svc parameters for each layer
+ for (i = 0; i < cpi->svc.number_spatial_layers; ++i)
+ buf->svc_params[i] = cpi->svc.layer_context[i].svc_params_received;
+
+ return 0;
+}
+
+static int copy_svc_params(VP9_COMP *const cpi, struct lookahead_entry *buf) {
+ int layer_id;
+ vpx_svc_parameters_t *layer_param;
+ vpx_enc_frame_flags_t flags;
+
+ // Find the next layer to be encoded
+ for (layer_id = 0; layer_id < cpi->svc.number_spatial_layers; ++layer_id) {
+ if (buf->svc_params[layer_id].spatial_layer >=0)
+ break;
+ }
+
+ if (layer_id == cpi->svc.number_spatial_layers)
+ return 1;
+
+ layer_param = &buf->svc_params[layer_id];
+ buf->flags = flags = layer_param->flags;
+ cpi->svc.spatial_layer_id = layer_param->spatial_layer;
+ cpi->svc.temporal_layer_id = layer_param->temporal_layer;
+ cpi->lst_fb_idx = layer_param->lst_fb_idx;
+ cpi->gld_fb_idx = layer_param->gld_fb_idx;
+ cpi->alt_fb_idx = layer_param->alt_fb_idx;
+
+ if (vp9_set_size_literal(cpi, layer_param->width, layer_param->height) != 0)
+ return VPX_CODEC_INVALID_PARAM;
+
+ cpi->oxcf.worst_allowed_q =
+ vp9_quantizer_to_qindex(layer_param->max_quantizer);
+ cpi->oxcf.best_allowed_q =
+ vp9_quantizer_to_qindex(layer_param->min_quantizer);
+
+ vp9_change_config(cpi, &cpi->oxcf);
+
+ vp9_set_high_precision_mv(cpi, 1);
+
+ // Retrieve the encoding flags for each layer and apply it to encoder.
+ // It includes reference frame flags and update frame flags.
+ vp9_apply_encoding_flags(cpi, flags);
+
+ return 0;
+}
+
+struct lookahead_entry *vp9_svc_lookahead_peek(VP9_COMP *const cpi,
+ struct lookahead_ctx *ctx,
+ int index, int copy_params) {
+ struct lookahead_entry *buf = vp9_lookahead_peek(ctx, index);
+
+ if (buf != NULL && copy_params != 0) {
+ if (copy_svc_params(cpi, buf) != 0)
+ return NULL;
+ }
+ return buf;
+}
+
+struct lookahead_entry *vp9_svc_lookahead_pop(VP9_COMP *const cpi,
+ struct lookahead_ctx *ctx,
+ int drain) {
+ struct lookahead_entry *buf = NULL;
+
+ if (ctx->sz && (drain || ctx->sz == ctx->max_sz - MAX_PRE_FRAMES)) {
+ buf = vp9_svc_lookahead_peek(cpi, ctx, 0, 1);
+ if (buf != NULL) {
+ // Only remove the buffer when pop the highest layer. Simply set the
+ // spatial_layer to -1 for lower layers.
+ buf->svc_params[cpi->svc.spatial_layer_id].spatial_layer = -1;
+ if (cpi->svc.spatial_layer_id == cpi->svc.number_spatial_layers - 1) {
+ vp9_lookahead_pop(ctx, drain);
+ }
+ }
+ }
+
+ return buf;
+}
diff --git a/vp9/encoder/vp9_svc_layercontext.h b/vp9/encoder/vp9_svc_layercontext.h
index 36e2027..3ebb831 100644
--- a/vp9/encoder/vp9_svc_layercontext.h
+++ b/vp9/encoder/vp9_svc_layercontext.h
@@ -28,6 +28,7 @@
struct vpx_fixed_buf rc_twopass_stats_in;
unsigned int current_video_frame_in_layer;
int is_key_frame;
+ vpx_svc_parameters_t svc_params_received;
} LAYER_CONTEXT;
typedef struct {
@@ -74,6 +75,23 @@
// Check if current layer is key frame in spatial upper layer
int vp9_is_upper_layer_key_frame(const struct VP9_COMP *const cpi);
+// Copy the source image, flags and svc parameters into a new framebuffer
+// with the expected stride/border
+int vp9_svc_lookahead_push(const struct VP9_COMP *const cpi,
+ struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG *src,
+ int64_t ts_start, int64_t ts_end,
+ unsigned int flags);
+
+// Get the next source buffer to encode
+struct lookahead_entry *vp9_svc_lookahead_pop(struct VP9_COMP *const cpi,
+ struct lookahead_ctx *ctx,
+ int drain);
+
+// Get a future source buffer to encode
+struct lookahead_entry *vp9_svc_lookahead_peek(struct VP9_COMP *const cpi,
+ struct lookahead_ctx *ctx,
+ int index, int copy_params);
+
#ifdef __cplusplus
} // extern "C"
#endif
diff --git a/vp9/vp9_cx_iface.c b/vp9/vp9_cx_iface.c
index edd59ab..b150161 100644
--- a/vp9/vp9_cx_iface.c
+++ b/vp9/vp9_cx_iface.c
@@ -88,8 +88,8 @@
size_t pending_frame_magnitude;
vpx_image_t preview_img;
vp8_postproc_cfg_t preview_ppcfg;
- vpx_codec_pkt_list_decl(64) pkt_list;
- unsigned int fixed_kf_cntr;
+ vpx_codec_pkt_list_decl(128) pkt_list;
+ unsigned int fixed_kf_cntr;
};
static VP9_REFFRAME ref_frame_to_vp9_reframe(vpx_ref_frame_type_t frame) {
@@ -795,42 +795,7 @@
return VPX_CODEC_INVALID_PARAM;
}
- if (flags & (VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_GF |
- VP8_EFLAG_NO_REF_ARF)) {
- int ref = 7;
-
- if (flags & VP8_EFLAG_NO_REF_LAST)
- ref ^= VP9_LAST_FLAG;
-
- if (flags & VP8_EFLAG_NO_REF_GF)
- ref ^= VP9_GOLD_FLAG;
-
- if (flags & VP8_EFLAG_NO_REF_ARF)
- ref ^= VP9_ALT_FLAG;
-
- vp9_use_as_reference(ctx->cpi, ref);
- }
-
- if (flags & (VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF |
- VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_FORCE_GF |
- VP8_EFLAG_FORCE_ARF)) {
- int upd = 7;
-
- if (flags & VP8_EFLAG_NO_UPD_LAST)
- upd ^= VP9_LAST_FLAG;
-
- if (flags & VP8_EFLAG_NO_UPD_GF)
- upd ^= VP9_GOLD_FLAG;
-
- if (flags & VP8_EFLAG_NO_UPD_ARF)
- upd ^= VP9_ALT_FLAG;
-
- vp9_update_reference(ctx->cpi, upd);
- }
-
- if (flags & VP8_EFLAG_NO_UPD_ENTROPY) {
- vp9_update_entropy(ctx->cpi, 0);
- }
+ vp9_apply_encoding_flags(ctx->cpi, flags);
// Handle fixed keyframe intervals
if (ctx->cfg.kf_mode == VPX_KF_AUTO &&
@@ -843,7 +808,7 @@
// Initialize the encoder instance on the first frame.
if (res == VPX_CODEC_OK && ctx->cpi != NULL) {
- unsigned int lib_flags;
+ unsigned int lib_flags = 0;
YV12_BUFFER_CONFIG sd;
int64_t dst_time_stamp, dst_end_time_stamp;
size_t size, cx_data_sz;
@@ -853,9 +818,6 @@
if (ctx->base.init_flags & VPX_CODEC_USE_PSNR)
((VP9_COMP *)ctx->cpi)->b_calculate_psnr = 1;
- // Convert API flags to internal codec lib flags
- lib_flags = (flags & VPX_EFLAG_FORCE_KF) ? FRAMEFLAGS_KEY : 0;
-
/* vp9 use 10,000,000 ticks/second as time stamp */
dst_time_stamp = (pts * 10000000 * ctx->cfg.g_timebase.num)
/ ctx->cfg.g_timebase.den;
@@ -865,7 +827,9 @@
if (img != NULL) {
res = image2yuvconfig(img, &sd);
- if (vp9_receive_raw_frame(ctx->cpi, lib_flags,
+ // Store the original flags in to the frame buffer. Will extract the
+ // key frame flag when we actually encode this frame.
+ if (vp9_receive_raw_frame(ctx->cpi, flags,
&sd, dst_time_stamp, dst_end_time_stamp)) {
VP9_COMP *cpi = (VP9_COMP *)ctx->cpi;
res = update_error_state(ctx, &cpi->common.error);
@@ -874,7 +838,6 @@
cx_data = ctx->cx_data;
cx_data_sz = ctx->cx_data_sz;
- lib_flags = 0;
/* Any pending invisible frames? */
if (ctx->pending_cx_data) {
@@ -902,7 +865,12 @@
VP9_COMP *const cpi = (VP9_COMP *)ctx->cpi;
// Pack invisible frames with the next visible frame
- if (cpi->common.show_frame == 0) {
+ if (cpi->common.show_frame == 0
+#ifdef CONFIG_SPATIAL_SVC
+ || (cpi->use_svc && cpi->svc.number_temporal_layers == 1 &&
+ cpi->svc.spatial_layer_id < cpi->svc.number_spatial_layers - 1)
+#endif
+ ) {
if (ctx->pending_cx_data == 0)
ctx->pending_cx_data = cx_data;
ctx->pending_cx_data_sz += size;
@@ -925,7 +893,12 @@
/ ctx->cfg.g_timebase.num / 10000000);
pkt.data.frame.flags = lib_flags << 16;
- if (lib_flags & FRAMEFLAGS_KEY)
+ if (lib_flags & FRAMEFLAGS_KEY
+#ifdef CONFIG_SPATIAL_SVC
+ || (cpi->use_svc && cpi->svc.number_temporal_layers == 1 &&
+ cpi->svc.layer_context[0].is_key_frame)
+#endif
+ )
pkt.data.frame.flags |= VPX_FRAME_IS_KEY;
if (cpi->common.show_frame == 0) {
@@ -1165,24 +1138,19 @@
VP9_COMP *const cpi = ctx->cpi;
vpx_svc_parameters_t *const params = va_arg(args, vpx_svc_parameters_t *);
- if (params == NULL)
+ if (params == NULL || params->spatial_layer < 0 ||
+ params->spatial_layer >= cpi->svc.number_spatial_layers)
return VPX_CODEC_INVALID_PARAM;
- cpi->svc.spatial_layer_id = params->spatial_layer;
- cpi->svc.temporal_layer_id = params->temporal_layer;
+ if (params->spatial_layer == 0) {
+ int i;
+ for (i = 0; i < cpi->svc.number_spatial_layers; ++i) {
+ cpi->svc.layer_context[i].svc_params_received.spatial_layer = -1;
+ }
+ }
- cpi->lst_fb_idx = params->lst_fb_idx;
- cpi->gld_fb_idx = params->gld_fb_idx;
- cpi->alt_fb_idx = params->alt_fb_idx;
-
- if (vp9_set_size_literal(ctx->cpi, params->width, params->height) != 0)
- return VPX_CODEC_INVALID_PARAM;
-
- ctx->cfg.rc_max_quantizer = params->max_quantizer;
- ctx->cfg.rc_min_quantizer = params->min_quantizer;
-
- set_encoder_config(&ctx->oxcf, &ctx->cfg, &ctx->extra_cfg);
- vp9_change_config(ctx->cpi, &ctx->oxcf);
+ cpi->svc.layer_context[params->spatial_layer].svc_params_received =
+ *params;
return VPX_CODEC_OK;
}
diff --git a/vpx/src/svc_encodeframe.c b/vpx/src/svc_encodeframe.c
index 17e165b..4efba9c 100644
--- a/vpx/src/svc_encodeframe.c
+++ b/vpx/src/svc_encodeframe.c
@@ -24,6 +24,7 @@
#include "vpx/svc_context.h"
#include "vpx/vp8cx.h"
#include "vpx/vpx_encoder.h"
+#include "vpx_mem/vpx_mem.h"
#ifdef __MINGW32__
#define strtok_r strtok_s
@@ -47,6 +48,14 @@
static const char *DEFAULT_QUANTIZER_VALUES = "60,53,39,33,27";
static const char *DEFAULT_SCALE_FACTORS = "4/16,5/16,7/16,11/16,16/16";
+// One encoded frame
+typedef struct FrameData {
+ void *buf; // compressed data buffer
+ size_t size; // length of compressed data
+ vpx_codec_frame_flags_t flags; /**< flags for this frame */
+ struct FrameData *next;
+} FrameData;
+
typedef struct SvcInternal {
char options[OPTION_BUFFER_SIZE]; // set by vpx_svc_set_options
char quantizers[OPTION_BUFFER_SIZE]; // set by vpx_svc_set_quantizers
@@ -72,15 +81,15 @@
// state variables
int encode_frame_count;
+ int frame_received;
int frame_within_gop;
vpx_enc_frame_flags_t enc_frame_flags;
int layers;
int layer;
int is_keyframe;
- size_t frame_size;
- size_t buffer_size;
- void *buffer;
+ FrameData *frame_list;
+ FrameData *frame_temp;
char *rc_stats_buf;
size_t rc_stats_buf_size;
@@ -90,128 +99,54 @@
vpx_codec_ctx_t *codec_ctx;
} SvcInternal;
-// Superframe is used to generate an index of individual frames (i.e., layers)
-struct Superframe {
- int count;
- uint32_t sizes[SUPERFRAME_SLOTS];
- uint32_t magnitude;
- uint8_t buffer[SUPERFRAME_BUFFER_SIZE];
- size_t index_size;
-};
-
-// One encoded frame layer
-struct LayerData {
- void *buf; // compressed data buffer
- size_t size; // length of compressed data
- struct LayerData *next;
-};
-
-// create LayerData from encoder output
-static struct LayerData *ld_create(void *buf, size_t size) {
- struct LayerData *const layer_data =
- (struct LayerData *)malloc(sizeof(*layer_data));
- if (layer_data == NULL) {
+// create FrameData from encoder output
+static struct FrameData *fd_create(void *buf, size_t size,
+ vpx_codec_frame_flags_t flags) {
+ struct FrameData *const frame_data =
+ (struct FrameData *)vpx_malloc(sizeof(*frame_data));
+ if (frame_data == NULL) {
return NULL;
}
- layer_data->buf = malloc(size);
- if (layer_data->buf == NULL) {
- free(layer_data);
+ frame_data->buf = vpx_malloc(size);
+ if (frame_data->buf == NULL) {
+ vpx_free(frame_data);
return NULL;
}
- memcpy(layer_data->buf, buf, size);
- layer_data->size = size;
- return layer_data;
+ vpx_memcpy(frame_data->buf, buf, size);
+ frame_data->size = size;
+ frame_data->flags = flags;
+ return frame_data;
}
-// free LayerData
-static void ld_free(struct LayerData *layer_data) {
- if (layer_data) {
- if (layer_data->buf) {
- free(layer_data->buf);
- layer_data->buf = NULL;
- }
- free(layer_data);
+// free FrameData
+static void fd_free(struct FrameData *p) {
+ if (p) {
+ if (p->buf)
+ vpx_free(p->buf);
+ vpx_free(p);
}
}
-// add layer data to list
-static void ld_list_add(struct LayerData **list, struct LayerData *layer_data) {
- struct LayerData **p = list;
+// add FrameData to list
+static void fd_list_add(struct FrameData **list, struct FrameData *layer_data) {
+ struct FrameData **p = list;
while (*p != NULL) p = &(*p)->next;
*p = layer_data;
layer_data->next = NULL;
}
-// get accumulated size of layer data
-static size_t ld_list_get_buffer_size(struct LayerData *list) {
- struct LayerData *p;
- size_t size = 0;
-
- for (p = list; p != NULL; p = p->next) {
- size += p->size;
- }
- return size;
-}
-
-// copy layer data to buffer
-static void ld_list_copy_to_buffer(struct LayerData *list, uint8_t *buffer) {
- struct LayerData *p;
-
- for (p = list; p != NULL; p = p->next) {
- buffer[0] = 1;
- memcpy(buffer, p->buf, p->size);
- buffer += p->size;
- }
-}
-
-// free layer data list
-static void ld_list_free(struct LayerData *list) {
- struct LayerData *p = list;
+// free FrameData list
+static void fd_free_list(struct FrameData *list) {
+ struct FrameData *p = list;
while (p) {
list = list->next;
- ld_free(p);
+ fd_free(p);
p = list;
}
}
-static void sf_create_index(struct Superframe *sf) {
- uint8_t marker = 0xc0;
- int i;
- uint32_t mag, mask;
- uint8_t *bufp;
-
- if (sf->count == 0 || sf->count >= 8) return;
-
- // Add the number of frames to the marker byte
- marker |= sf->count - 1;
-
- // Choose the magnitude
- for (mag = 0, mask = 0xff; mag < 4; ++mag) {
- if (sf->magnitude < mask) break;
- mask <<= 8;
- mask |= 0xff;
- }
- marker |= mag << 3;
-
- // Write the index
- sf->index_size = 2 + (mag + 1) * sf->count;
- bufp = sf->buffer;
-
- *bufp++ = marker;
- for (i = 0; i < sf->count; ++i) {
- int this_sz = sf->sizes[i];
- uint32_t j;
-
- for (j = 0; j <= mag; ++j) {
- *bufp++ = this_sz & 0xff;
- this_sz >>= 8;
- }
- }
- *bufp++ = marker;
-}
-
static SvcInternal *get_svc_internal(SvcContext *svc_ctx) {
if (svc_ctx == NULL) return NULL;
if (svc_ctx->internal == NULL) {
@@ -574,8 +509,6 @@
// modify encoder configuration
enc_cfg->ss_number_layers = si->layers;
enc_cfg->ts_number_layers = 1; // Temporal layers not used in this encoder.
- // Lag in frames not currently supported
- enc_cfg->g_lag_in_frames = 0;
// TODO(ivanmaltz): determine if these values need to be set explicitly for
// svc, or if the normal default/override mechanism can be used
@@ -608,6 +541,34 @@
return VPX_CODEC_OK;
}
+static void accumulate_frame_size_for_each_layer(SvcInternal *const si,
+ const uint8_t *const buf,
+ const size_t size) {
+ uint8_t marker = buf[size - 1];
+ if ((marker & 0xe0) == 0xc0) {
+ const uint32_t frames = (marker & 0x7) + 1;
+ const uint32_t mag = ((marker >> 3) & 0x3) + 1;
+ const size_t index_sz = 2 + mag * frames;
+
+ uint8_t marker2 = buf[size - index_sz];
+
+ if (size >= index_sz && marker2 == marker) {
+ // found a valid superframe index
+ uint32_t i, j;
+ const uint8_t *x = &buf[size - index_sz + 1];
+
+ // frames has a maximum of 8 and mag has a maximum of 4.
+ for (i = 0; i < frames; i++) {
+ uint32_t this_sz = 0;
+
+ for (j = 0; j < mag; j++)
+ this_sz |= (*x++) << (j * 8);
+ si->bytes_sum[i] += this_sz;
+ }
+ }
+ }
+}
+
// SVC Algorithm flags - these get mapped to VP8_EFLAG_* defined in vp8cx.h
// encoder should reference the last frame
@@ -846,15 +807,12 @@
vpx_codec_err_t res;
vpx_codec_iter_t iter;
const vpx_codec_cx_pkt_t *cx_pkt;
- struct LayerData *cx_layer_list = NULL;
- struct LayerData *layer_data;
- struct Superframe superframe;
+ int layer_for_psnr = 0;
SvcInternal *const si = get_svc_internal(svc_ctx);
if (svc_ctx == NULL || codec_ctx == NULL || si == NULL) {
return VPX_CODEC_INVALID_PARAM;
}
- memset(&superframe, 0, sizeof(superframe));
svc_log_reset(svc_ctx);
si->rc_stats_buf_used = 0;
@@ -863,7 +821,6 @@
si->frame_within_gop = 0;
}
si->is_keyframe = (si->frame_within_gop == 0);
- si->frame_size = 0;
if (rawimg != NULL) {
svc_log(svc_ctx, SVC_LOG_DEBUG,
@@ -872,124 +829,90 @@
si->frame_within_gop);
}
- // encode each layer
- for (si->layer = 0; si->layer < si->layers; ++si->layer) {
- if (svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP &&
- si->is_keyframe && (si->layer == 1 || si->layer == 3)) {
- svc_log(svc_ctx, SVC_LOG_DEBUG, "Skip encoding layer %d\n", si->layer);
- continue;
- }
-
- if (rawimg != NULL) {
+ if (rawimg != NULL) {
+ // encode each layer
+ for (si->layer = 0; si->layer < si->layers; ++si->layer) {
+ if (svc_ctx->encoding_mode == ALT_INTER_LAYER_PREDICTION_IP &&
+ si->is_keyframe && (si->layer == 1 || si->layer == 3)) {
+ svc_log(svc_ctx, SVC_LOG_DEBUG, "Skip encoding layer %d\n", si->layer);
+ continue;
+ }
calculate_enc_frame_flags(svc_ctx);
set_svc_parameters(svc_ctx, codec_ctx);
}
+ }
- res = vpx_codec_encode(codec_ctx, rawimg, pts, (uint32_t)duration,
- si->enc_frame_flags, deadline);
- if (res != VPX_CODEC_OK) {
- return res;
- }
- // save compressed data
- iter = NULL;
- while ((cx_pkt = vpx_codec_get_cx_data(codec_ctx, &iter))) {
- switch (cx_pkt->kind) {
- case VPX_CODEC_CX_FRAME_PKT: {
- const uint32_t frame_pkt_size = (uint32_t)(cx_pkt->data.frame.sz);
- si->bytes_sum[si->layer] += frame_pkt_size;
- svc_log(svc_ctx, SVC_LOG_DEBUG,
- "SVC frame: %d, layer: %d, size: %u\n",
- si->encode_frame_count, si->layer, frame_pkt_size);
- layer_data =
- ld_create(cx_pkt->data.frame.buf, (size_t)frame_pkt_size);
- if (layer_data == NULL) {
- svc_log(svc_ctx, SVC_LOG_ERROR, "Error allocating LayerData\n");
- return VPX_CODEC_OK;
+ res = vpx_codec_encode(codec_ctx, rawimg, pts, (uint32_t)duration, 0,
+ deadline);
+ if (res != VPX_CODEC_OK) {
+ return res;
+ }
+ // save compressed data
+ iter = NULL;
+ while ((cx_pkt = vpx_codec_get_cx_data(codec_ctx, &iter))) {
+ switch (cx_pkt->kind) {
+ case VPX_CODEC_CX_FRAME_PKT: {
+ fd_list_add(&si->frame_list, fd_create(cx_pkt->data.frame.buf,
+ cx_pkt->data.frame.sz,
+ cx_pkt->data.frame.flags));
+ accumulate_frame_size_for_each_layer(si, cx_pkt->data.frame.buf,
+ cx_pkt->data.frame.sz);
+
+ svc_log(svc_ctx, SVC_LOG_DEBUG, "SVC frame: %d, kf: %d, size: %d, "
+ "pts: %d\n", si->frame_received,
+ (cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY) ? 1 : 0,
+ (int)cx_pkt->data.frame.sz, (int)cx_pkt->data.frame.pts);
+
+ ++si->frame_received;
+ layer_for_psnr = 0;
+ break;
+ }
+ case VPX_CODEC_PSNR_PKT: {
+ int i;
+ svc_log(svc_ctx, SVC_LOG_DEBUG,
+ "SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): "
+ "%2.3f %2.3f %2.3f %2.3f \n",
+ si->frame_received, layer_for_psnr,
+ cx_pkt->data.psnr.psnr[0], cx_pkt->data.psnr.psnr[1],
+ cx_pkt->data.psnr.psnr[2], cx_pkt->data.psnr.psnr[3]);
+ svc_log(svc_ctx, SVC_LOG_DEBUG,
+ "SVC frame: %d, layer: %d, SSE(Total/Y/U/V): "
+ "%2.3f %2.3f %2.3f %2.3f \n",
+ si->frame_received, layer_for_psnr,
+ cx_pkt->data.psnr.sse[0], cx_pkt->data.psnr.sse[1],
+ cx_pkt->data.psnr.sse[2], cx_pkt->data.psnr.sse[3]);
+ for (i = 0; i < COMPONENTS; i++) {
+ si->psnr_sum[layer_for_psnr][i] += cx_pkt->data.psnr.psnr[i];
+ si->sse_sum[layer_for_psnr][i] += cx_pkt->data.psnr.sse[i];
+ }
+ ++layer_for_psnr;
+ break;
+ }
+ case VPX_CODEC_STATS_PKT: {
+ size_t new_size = si->rc_stats_buf_used +
+ cx_pkt->data.twopass_stats.sz;
+
+ if (new_size > si->rc_stats_buf_size) {
+ char *p = (char*)realloc(si->rc_stats_buf, new_size);
+ if (p == NULL) {
+ svc_log(svc_ctx, SVC_LOG_ERROR, "Error allocating stats buf\n");
+ return VPX_CODEC_MEM_ERROR;
}
- ld_list_add(&cx_layer_list, layer_data);
+ si->rc_stats_buf = p;
+ si->rc_stats_buf_size = new_size;
+ }
- // save layer size in superframe index
- superframe.sizes[superframe.count++] = frame_pkt_size;
- superframe.magnitude |= frame_pkt_size;
- break;
- }
- case VPX_CODEC_PSNR_PKT: {
- int i;
- svc_log(svc_ctx, SVC_LOG_DEBUG,
- "SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): "
- "%2.3f %2.3f %2.3f %2.3f \n",
- si->encode_frame_count, si->layer,
- cx_pkt->data.psnr.psnr[0], cx_pkt->data.psnr.psnr[1],
- cx_pkt->data.psnr.psnr[2], cx_pkt->data.psnr.psnr[3]);
- svc_log(svc_ctx, SVC_LOG_DEBUG,
- "SVC frame: %d, layer: %d, SSE(Total/Y/U/V): "
- "%2.3f %2.3f %2.3f %2.3f \n",
- si->encode_frame_count, si->layer,
- cx_pkt->data.psnr.sse[0], cx_pkt->data.psnr.sse[1],
- cx_pkt->data.psnr.sse[2], cx_pkt->data.psnr.sse[3]);
- for (i = 0; i < COMPONENTS; i++) {
- si->psnr_sum[si->layer][i] += cx_pkt->data.psnr.psnr[i];
- si->sse_sum[si->layer][i] += cx_pkt->data.psnr.sse[i];
- }
- break;
- }
- case VPX_CODEC_STATS_PKT: {
- size_t new_size = si->rc_stats_buf_used +
- cx_pkt->data.twopass_stats.sz;
-
- if (new_size > si->rc_stats_buf_size) {
- char *p = (char*)realloc(si->rc_stats_buf, new_size);
- if (p == NULL) {
- svc_log(svc_ctx, SVC_LOG_ERROR, "Error allocating stats buf\n");
- break;
- }
- si->rc_stats_buf = p;
- si->rc_stats_buf_size = new_size;
- }
-
- memcpy(si->rc_stats_buf + si->rc_stats_buf_used,
- cx_pkt->data.twopass_stats.buf, cx_pkt->data.twopass_stats.sz);
- si->rc_stats_buf_used += cx_pkt->data.twopass_stats.sz;
- break;
- }
- default: {
- break;
- }
+ memcpy(si->rc_stats_buf + si->rc_stats_buf_used,
+ cx_pkt->data.twopass_stats.buf, cx_pkt->data.twopass_stats.sz);
+ si->rc_stats_buf_used += cx_pkt->data.twopass_stats.sz;
+ break;
+ }
+ default: {
+ break;
}
}
- if (rawimg == NULL) {
- break;
- }
}
- if (codec_ctx->config.enc->g_pass != VPX_RC_FIRST_PASS) {
- // add superframe index to layer data list
- sf_create_index(&superframe);
- layer_data = ld_create(superframe.buffer, superframe.index_size);
- ld_list_add(&cx_layer_list, layer_data);
- // get accumulated size of layer data
- si->frame_size = ld_list_get_buffer_size(cx_layer_list);
- if (si->frame_size > 0) {
- // all layers encoded, create single buffer with concatenated layers
- if (si->frame_size > si->buffer_size) {
- free(si->buffer);
- si->buffer = malloc(si->frame_size);
- if (si->buffer == NULL) {
- ld_list_free(cx_layer_list);
- return VPX_CODEC_MEM_ERROR;
- }
- si->buffer_size = si->frame_size;
- }
- // copy layer data into packet
- ld_list_copy_to_buffer(cx_layer_list, (uint8_t *)si->buffer);
-
- ld_list_free(cx_layer_list);
-
- svc_log(svc_ctx, SVC_LOG_DEBUG, "SVC frame: %d, kf: %d, size: %d, "
- "pts: %d\n", si->encode_frame_count, si->is_keyframe,
- (int)si->frame_size, (int)pts);
- }
- }
if (rawimg != NULL) {
++si->frame_within_gop;
++si->encode_frame_count;
@@ -1004,16 +927,27 @@
return si->message_buffer;
}
-void *vpx_svc_get_buffer(const SvcContext *svc_ctx) {
- const SvcInternal *const si = get_const_svc_internal(svc_ctx);
- if (svc_ctx == NULL || si == NULL) return NULL;
- return si->buffer;
+// We will maintain a list of output frame buffers since with lag_in_frame
+// we need to output all frame buffers at the end. vpx_svc_get_buffer() will
+// remove a frame buffer from the list the put it to a temporal pointer, which
+// will be removed at the next vpx_svc_get_buffer() or when closing encoder.
+void *vpx_svc_get_buffer(SvcContext *svc_ctx) {
+ SvcInternal *const si = get_svc_internal(svc_ctx);
+ if (svc_ctx == NULL || si == NULL || si->frame_list == NULL) return NULL;
+
+ if (si->frame_temp)
+ fd_free(si->frame_temp);
+
+ si->frame_temp = si->frame_list;
+ si->frame_list = si->frame_list->next;
+
+ return si->frame_temp->buf;
}
size_t vpx_svc_get_frame_size(const SvcContext *svc_ctx) {
const SvcInternal *const si = get_const_svc_internal(svc_ctx);
- if (svc_ctx == NULL || si == NULL) return 0;
- return si->frame_size;
+ if (svc_ctx == NULL || si == NULL || si->frame_list == NULL) return 0;
+ return si->frame_list->size;
}
int vpx_svc_get_encode_frame_count(const SvcContext *svc_ctx) {
@@ -1024,8 +958,8 @@
int vpx_svc_is_keyframe(const SvcContext *svc_ctx) {
const SvcInternal *const si = get_const_svc_internal(svc_ctx);
- if (svc_ctx == NULL || si == NULL) return 0;
- return si->is_keyframe;
+ if (svc_ctx == NULL || si == NULL || si->frame_list == NULL) return 0;
+ return (si->frame_list->flags & VPX_FRAME_IS_KEY) != 0;
}
void vpx_svc_set_keyframe(SvcContext *svc_ctx) {
@@ -1112,7 +1046,8 @@
// SvcInternal if it was not already allocated
si = (SvcInternal *)svc_ctx->internal;
if (si != NULL) {
- free(si->buffer);
+ fd_free(si->frame_temp);
+ fd_free_list(si->frame_list);
if (si->rc_stats_buf) {
free(si->rc_stats_buf);
}
diff --git a/vpx/svc_context.h b/vpx/svc_context.h
index 5d0fbbd..058ee20 100644
--- a/vpx/svc_context.h
+++ b/vpx/svc_context.h
@@ -104,14 +104,16 @@
const char *vpx_svc_get_message(const SvcContext *svc_ctx);
/**
- * return size of encoded data to be returned by vpx_svc_get_buffer
+ * return size of encoded data to be returned by vpx_svc_get_buffer.
+ * it needs to be called before vpx_svc_get_buffer.
*/
size_t vpx_svc_get_frame_size(const SvcContext *svc_ctx);
/**
- * return buffer with encoded data
+ * return buffer with encoded data. encoder will maintain a list of frame
+ * buffers. each call of vpx_svc_get_buffer() will return one frame.
*/
-void *vpx_svc_get_buffer(const SvcContext *svc_ctx);
+void *vpx_svc_get_buffer(SvcContext *svc_ctx);
/**
* return size of two pass rate control stats data to be returned by
