From: Minghai Shang <minghai@google.com>
Date: Tue, 24 Jun 2014 21:01:17 +0000 (-0700)
Subject: [spatial svc]Implement lag in frames for spatial svc
X-Git-Tag: v1.4.0~1340^2
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=277338f7488d40d27a9ea8d3a5b2bf1a0f32aa95;p=platform%2Fupstream%2Flibvpx.git

[spatial svc]Implement lag in frames for spatial svc

Change-Id: I930dced169c9d53f8044d2754a04332138347409
---

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 main(int argc, const char **argv) {
   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 @@ int main(int argc, const char **argv) {
       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/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 @@ TEST_F(SvcTest, FirstFrameHasLayers) {
                        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 @@ TEST_F(SvcTest, EncodeThreeFrames) {
   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 @@ TEST_F(SvcTest, EncodeThreeFrames) {
   // 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 @@ TEST_F(SvcTest, EncodeThreeFrames) {
   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 @@ TEST_F(SvcTest, EncodeThreeFrames) {
   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 @@ TEST_F(SvcTest, TwoPassEncode) {
   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 @@ TEST_F(SvcTest, TwoPassEncode) {
   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 @@ TEST_F(SvcTest, TwoPassEncode) {
   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 @@ TEST_F(SvcTest, TwoPassEncode) {
   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/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index be29201..eb1742b 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -106,7 +106,7 @@ static INLINE void Scale2Ratio(VPX_SCALING mode, int *hr, int *hs) {
   }
 }
 
-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 @@ void vp9_change_config(struct VP9_COMP *cpi, const VP9EncoderConfig *oxcf) {
   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;
@@ -2117,7 +2117,7 @@ static void encode_frame_to_data_rate(VP9_COMP *cpi,
   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 vp9_receive_raw_frame(VP9_COMP *cpi, unsigned int frame_flags,
   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 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
   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 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
   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 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
   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 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
   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 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
 
     *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_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b) {
 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 @@ void vp9_update_reference_frames(VP9_COMP *cpi);
 
 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_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 @@ struct lookahead_entry {
   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_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 @@ int vp9_is_upper_layer_key_frame(const VP9_COMP *const cpi) {
          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 @@ typedef struct {
   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 @@ void vp9_inc_frame_in_layer(SVC *svc);
 // 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 @@ struct vpx_codec_alg_priv {
   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 @@ static vpx_codec_err_t encoder_encode(vpx_codec_alg_priv_t  *ctx,
     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 @@ static vpx_codec_err_t encoder_encode(vpx_codec_alg_priv_t  *ctx,
 
   // 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 @@ static vpx_codec_err_t encoder_encode(vpx_codec_alg_priv_t  *ctx,
     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 @@ static vpx_codec_err_t encoder_encode(vpx_codec_alg_priv_t  *ctx,
     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 @@ static vpx_codec_err_t encoder_encode(vpx_codec_alg_priv_t  *ctx,
 
     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 @@ static vpx_codec_err_t encoder_encode(vpx_codec_alg_priv_t  *ctx,
         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 @@ static vpx_codec_err_t encoder_encode(vpx_codec_alg_priv_t  *ctx,
           / 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 @@ static vpx_codec_err_t ctrl_set_svc_parameters(vpx_codec_alg_priv_t *ctx,
   VP9_COMP *const cpi = ctx->cpi;
   vpx_svc_parameters_t *const params = va_arg(args, vpx_svc_parameters_t *);
 
-  if (params == NULL)
-    return VPX_CODEC_INVALID_PARAM;
-
-  cpi->svc.spatial_layer_id = params->spatial_layer;
-  cpi->svc.temporal_layer_id = params->temporal_layer;
-
-  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)
+  if (params == NULL || params->spatial_layer < 0 ||
+      params->spatial_layer >= cpi->svc.number_spatial_layers)
     return VPX_CODEC_INVALID_PARAM;
 
-  ctx->cfg.rc_max_quantizer = params->max_quantizer;
-  ctx->cfg.rc_min_quantizer = params->min_quantizer;
+  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;
+    }
+  }
 
-  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 @@ _CRTIMP char *__cdecl strtok_s(char *str, const char *delim, char **context);
 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 @@ typedef struct SvcInternal {
 
   // 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 @@ typedef struct SvcInternal {
   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 @@ vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
   // 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 @@ vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
   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 vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
   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 @@ vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
     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 @@ vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
             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;
-          }
-          ld_list_add(&cx_layer_list, layer_data);
-
-          // save layer size in superframe index
-          superframe.sizes[superframe.count++] = frame_pkt_size;
-          superframe.magnitude |= frame_pkt_size;
-          break;
+  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];
         }
-        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];
+        ++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;
           }
-          break;
+          si->rc_stats_buf = p;
+          si->rc_stats_buf_size = new_size;
         }
-        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;
       }
-    }
-    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;
+      default: {
+        break;
       }
-      // 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 @@ const char *vpx_svc_get_message(const SvcContext *svc_ctx) {
   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_get_encode_frame_count(const SvcContext *svc_ctx) {
 
 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 @@ void vpx_svc_release(SvcContext *svc_ctx) {
   // 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_dump_statistics(SvcContext *svc_ctx);
 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