Cosmetic changes in the supertx expt

Converts most negative !CONFIG_SUPERTX checks to positive ones.

Change-Id: I80b7f8c5d3483a7861f0de7fc7ebc425b9c68766

build/make/gen_msvs_proj.sh

@@ -137,7 +137,9 @@ for opt in "$@"; do
         ;;
         --lib) proj_kind="lib"
         ;;
-        --src-path-bare=*) src_path_bare=$(fix_path "$optval")
+        --src-path-bare=*)
+            src_path_bare=$(fix_path "$optval")
+            src_path_bare=${src_path_bare%/}
         ;;
         --static-crt) use_static_runtime=true
         ;;
@@ -151,9 +153,9 @@ for opt in "$@"; do
             esac
         ;;
         -I*)
-            opt="${opt%/}"
             opt=${opt##-I}
             opt=$(fix_path "$opt")
+            opt="${opt%/}"
             incs="${incs}${incs:+;}"${opt}""
             yasmincs="${yasmincs} -I"${opt}""
         ;;
@@ -414,7 +416,7 @@ generate_vcproj() {
                     vpx)
                         tag Tool \
                             Name="VCPreBuildEventTool" \
-                            CommandLine="call obj_int_extract.bat $src_path_bare $plat_no_ws\\\$(ConfigurationName)" \
+                            CommandLine="call obj_int_extract.bat "$src_path_bare" $plat_no_ws\\\$(ConfigurationName)" \
 
                         tag Tool \
                             Name="VCCLCompilerTool" \

build/make/gen_msvs_vcxproj.sh

@@ -157,7 +157,9 @@ for opt in "$@"; do
         ;;
         --lib) proj_kind="lib"
         ;;
-        --src-path-bare=*) src_path_bare=$(fix_path "$optval")
+        --src-path-bare=*)
+            src_path_bare=$(fix_path "$optval")
+            src_path_bare=${src_path_bare%/}
         ;;
         --static-crt) use_static_runtime=true
         ;;
@@ -173,9 +175,9 @@ for opt in "$@"; do
             esac
         ;;
         -I*)
-            opt="${opt%/}"
             opt=${opt##-I}
             opt=$(fix_path "$opt")
+            opt="${opt%/}"
             incs="${incs}${incs:+;}"${opt}""
             yasmincs="${yasmincs} -I"${opt}""
         ;;

build/make/iosbuild.sh

@@ -50,9 +50,79 @@ build_target() {
   vlog "***Done building target: ${target}***"
 }
 
+# Returns the preprocessor symbol for the target specified by $1.
+target_to_preproc_symbol() {
+  target="$1"
+  case "${target}" in
+    armv6-*)
+      echo "__ARM_ARCH_6__"
+      ;;
+    armv7-*)
+      echo "__ARM_ARCH_7__"
+      ;;
+    armv7s-*)
+      echo "__ARM_ARCH_7S__"
+      ;;
+    x86-*)
+      echo "__i386__"
+      ;;
+    x86_64-*)
+      echo "__x86_64__"
+      ;;
+    *)
+      echo "#error ${target} unknown/unsupported"
+      return 1
+      ;;
+  esac
+}
+
+# Create a vpx_config.h shim that, based on preprocessor settings for the
+# current target CPU, includes the real vpx_config.h for the current target.
+# $1 is the list of targets.
+create_vpx_framework_config_shim() {
+  local targets="$1"
+  local config_file="${HEADER_DIR}/vpx_config.h"
+  local preproc_symbol=""
+  local target=""
+  local include_guard="VPX_FRAMEWORK_HEADERS_VPX_VPX_CONFIG_H_"
+
+  local file_header="/*
+ *  Copyright (c) $(date +%Y) The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+/* GENERATED FILE: DO NOT EDIT! */
+
+#ifndef ${include_guard}
+#define ${include_guard}
+
+#if defined"
+
+  printf "%s" "${file_header}" > "${config_file}"
+  for target in ${targets}; do
+    preproc_symbol=$(target_to_preproc_symbol "${target}")
+    printf " ${preproc_symbol}\n" >> "${config_file}"
+    printf "#include \"VPX/vpx/${target}/vpx_config.h\"\n" >> "${config_file}"
+    printf "#elif defined" >> "${config_file}"
+    mkdir "${HEADER_DIR}/${target}"
+    cp -p "${BUILD_ROOT}/${target}/vpx_config.h" "${HEADER_DIR}/${target}"
+  done
+
+  # Consume the last line of output from the loop: We don't want it.
+  sed -i '' -e '$d' "${config_file}"
+
+  printf "#endif\n\n" >> "${config_file}"
+  printf "#endif  // ${include_guard}" >> "${config_file}"
+}
+
 # Configures and builds each target specified by $1, and then builds
 # VPX.framework.
-build_targets() {
+build_framework() {
   local lib_list=""
   local targets="$1"
   local target=""
@@ -75,15 +145,20 @@ build_targets() {
 
   cd "${ORIG_PWD}"
 
-  # Includes are identical for all platforms, and according to dist target
-  # behavior vpx_config.h and vpx_version.h aren't actually necessary for user
-  # apps built with libvpx. So, just copy the includes from the last target
-  # built.
-  # TODO(tomfinegan): The above is a lame excuse. Build common config/version
-  # includes that use the preprocessor to include the correct file.
+  # The basic libvpx API includes are all the same; just grab the most recent
+  # set.
   cp -p "${target_dist_dir}"/include/vpx/* "${HEADER_DIR}"
+
+  # Build the fat library.
   ${LIPO} -create ${lib_list} -output ${FRAMEWORK_DIR}/VPX
 
+  # Create the vpx_config.h shim that allows usage of vpx_config.h from
+  # within VPX.framework.
+  create_vpx_framework_config_shim "${targets}"
+
+  # Copy in vpx_version.h.
+  cp -p "${BUILD_ROOT}/${target}/vpx_version.h" "${HEADER_DIR}"
+
   vlog "Created fat library ${FRAMEWORK_DIR}/VPX containing:"
   for lib in ${lib_list}; do
     vlog "  $(echo ${lib} | awk -F / '{print $2, $NF}')"
@@ -166,4 +241,4 @@ cat << EOF
 EOF
 fi
 
-build_targets "${TARGETS}"
+build_framework "${TARGETS}"

configure

@@ -273,6 +273,13 @@ EXPERIMENT_LIST="
     multiple_arf
     spatial_svc
     denoising
+    masked_interinter
+    interintra
+    masked_interintra
+    filterintra
+    ext_tx
+    supertx
+    copy_coding
 "
 CONFIG_LIST="
     external_build

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) {

test/decode_test_driver.cc

@@ -15,13 +15,27 @@
 
 namespace libvpx_test {
 
+const char kVP8Name[] = "WebM Project VP8";
+
+vpx_codec_err_t Decoder::PeekStream(const uint8_t *cxdata, size_t size,
+                                    vpx_codec_stream_info_t *stream_info) {
+  return vpx_codec_peek_stream_info(CodecInterface(),
+                                    cxdata, static_cast<unsigned int>(size),
+                                    stream_info);
+}
+
 vpx_codec_err_t Decoder::DecodeFrame(const uint8_t *cxdata, size_t size) {
+  return DecodeFrame(cxdata, size, NULL);
+}
+
+vpx_codec_err_t Decoder::DecodeFrame(const uint8_t *cxdata, size_t size,
+                                     void *user_priv) {
   vpx_codec_err_t res_dec;
   InitOnce();
   REGISTER_STATE_CHECK(
       res_dec = vpx_codec_decode(&decoder_,
                                  cxdata, static_cast<unsigned int>(size),
-                                 NULL, 0));
+                                 user_priv, 0));
   return res_dec;
 }
 
@@ -29,13 +43,37 @@ void DecoderTest::RunLoop(CompressedVideoSource *video) {
   vpx_codec_dec_cfg_t dec_cfg = {0};
   Decoder* const decoder = codec_->CreateDecoder(dec_cfg, 0);
   ASSERT_TRUE(decoder != NULL);
+  const char *codec_name = decoder->GetDecoderName();
+  const bool is_vp8 = strncmp(kVP8Name, codec_name, sizeof(kVP8Name) - 1) == 0;
 
   // Decode frames.
-  for (video->Begin(); video->cxdata(); video->Next()) {
+  for (video->Begin(); !::testing::Test::HasFailure() && video->cxdata();
+       video->Next()) {
     PreDecodeFrameHook(*video, decoder);
+
+    vpx_codec_stream_info_t stream_info;
+    stream_info.sz = sizeof(stream_info);
+    const vpx_codec_err_t res_peek = decoder->PeekStream(video->cxdata(),
+                                                         video->frame_size(),
+                                                         &stream_info);
+    if (is_vp8) {
+      /* Vp8's implementation of PeekStream returns an error if the frame you
+       * pass it is not a keyframe, so we only expect VPX_CODEC_OK on the first
+       * frame, which must be a keyframe. */
+      if (video->frame_number() == 0)
+        ASSERT_EQ(VPX_CODEC_OK, res_peek) << "Peek return failed: "
+            << vpx_codec_err_to_string(res_peek);
+    } else {
+      /* The Vp9 implementation of PeekStream returns an error only if the
+       * data passed to it isn't a valid Vp9 chunk. */
+      ASSERT_EQ(VPX_CODEC_OK, res_peek) << "Peek return failed: "
+          << vpx_codec_err_to_string(res_peek);
+    }
+
     vpx_codec_err_t res_dec = decoder->DecodeFrame(video->cxdata(),
                                                    video->frame_size());
-    ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder->DecodeError();
+    if (!HandleDecodeResult(res_dec, *video, decoder))
+      break;
 
     DxDataIterator dec_iter = decoder->GetDxData();
     const vpx_image_t *img = NULL;

test/decode_test_driver.h

@@ -49,8 +49,14 @@ class Decoder {
     vpx_codec_destroy(&decoder_);
   }
 
+  vpx_codec_err_t PeekStream(const uint8_t *cxdata, size_t size,
+                             vpx_codec_stream_info_t *stream_info);
+
   vpx_codec_err_t DecodeFrame(const uint8_t *cxdata, size_t size);
 
+  vpx_codec_err_t DecodeFrame(const uint8_t *cxdata, size_t size,
+                              void *user_priv);
+
   DxDataIterator GetDxData() {
     return DxDataIterator(&decoder_);
   }
@@ -85,6 +91,10 @@ class Decoder {
         &decoder_, cb_get, cb_release, user_priv);
   }
 
+  const char* GetDecoderName() {
+    return vpx_codec_iface_name(CodecInterface());
+  }
+
  protected:
   virtual vpx_codec_iface_t* CodecInterface() const = 0;
 
@@ -114,6 +124,14 @@ class DecoderTest {
   virtual void PreDecodeFrameHook(const CompressedVideoSource& video,
                                   Decoder *decoder) {}
 
+  // Hook to be called to handle decode result. Return true to continue.
+  virtual bool HandleDecodeResult(const vpx_codec_err_t res_dec,
+                                  const CompressedVideoSource& /* video */,
+                                  Decoder *decoder) {
+    EXPECT_EQ(VPX_CODEC_OK, res_dec) << decoder->DecodeError();
+    return VPX_CODEC_OK == res_dec;
+  }
+
   // Hook to be called on every decompressed frame.
   virtual void DecompressedFrameHook(const vpx_image_t& img,
                                      const unsigned int frame_number) {}

test/invalid_file_test.cc

@@ -0,0 +1,109 @@
+/*
+ *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <cstdio>
+#include <cstdlib>
+#include <string>
+#include <vector>
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "./vpx_config.h"
+#include "test/codec_factory.h"
+#include "test/decode_test_driver.h"
+#include "test/ivf_video_source.h"
+#include "test/util.h"
+#if CONFIG_WEBM_IO
+#include "test/webm_video_source.h"
+#endif
+#include "vpx_mem/vpx_mem.h"
+
+namespace {
+
+class InvalidFileTest
+    : public ::libvpx_test::DecoderTest,
+      public ::libvpx_test::CodecTestWithParam<const char*> {
+ protected:
+  InvalidFileTest() : DecoderTest(GET_PARAM(0)), res_file_(NULL) {}
+
+  virtual ~InvalidFileTest() {
+    if (res_file_ != NULL)
+      fclose(res_file_);
+  }
+
+  void OpenResFile(const std::string &res_file_name_) {
+    res_file_ = libvpx_test::OpenTestDataFile(res_file_name_);
+    ASSERT_TRUE(res_file_ != NULL) << "Result file open failed. Filename: "
+        << res_file_name_;
+  }
+
+  virtual bool HandleDecodeResult(
+      const vpx_codec_err_t res_dec,
+      const libvpx_test::CompressedVideoSource &video,
+      libvpx_test::Decoder *decoder) {
+    EXPECT_TRUE(res_file_ != NULL);
+    int expected_res_dec;
+
+    // Read integer result.
+    const int res = fscanf(res_file_, "%d", &expected_res_dec);
+    EXPECT_NE(res, EOF) << "Read result data failed";
+
+    // Check results match.
+    EXPECT_EQ(expected_res_dec, res_dec)
+        << "Results don't match: frame number = " << video.frame_number();
+
+    return !HasFailure();
+  }
+
+ private:
+  FILE *res_file_;
+};
+
+TEST_P(InvalidFileTest, ReturnCode) {
+  const std::string filename = GET_PARAM(1);
+  libvpx_test::CompressedVideoSource *video = NULL;
+
+  // Open compressed video file.
+  if (filename.substr(filename.length() - 3, 3) == "ivf") {
+    video = new libvpx_test::IVFVideoSource(filename);
+  } else if (filename.substr(filename.length() - 4, 4) == "webm") {
+#if CONFIG_WEBM_IO
+    video = new libvpx_test::WebMVideoSource(filename);
+#else
+    fprintf(stderr, "WebM IO is disabled, skipping test vector %s\n",
+            filename.c_str());
+    return;
+#endif
+  }
+  video->Init();
+
+  // Construct result file name. The file holds a list of expected integer
+  // results, one for each decoded frame.  Any result that doesn't match
+  // the files list will cause a test failure.
+  const std::string res_filename = filename + ".res";
+  OpenResFile(res_filename);
+
+  // Decode frame, and check the md5 matching.
+  ASSERT_NO_FATAL_FAILURE(RunLoop(video));
+  delete video;
+}
+
+const char *const kVP9InvalidFileTests[] = {
+  "invalid-vp90-01.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]))
+
+VP9_INSTANTIATE_TEST_CASE(InvalidFileTest,
+                          ::testing::ValuesIn(kVP9InvalidFileTests,
+                                              kVP9InvalidFileTests +
+                                              NELEMENTS(kVP9InvalidFileTests)));
+
+}  // namespace

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_));
+  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_));
 
+  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_)),
-      vpx_svc_get_frame_size(&svc_));
+        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_));
 
+  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_)),
-      vpx_svc_get_frame_size(&svc_));
+        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_));
+  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_));
 
+  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_)),
-      vpx_svc_get_frame_size(&svc_));
+        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_));
 
+  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_)),
-      vpx_svc_get_frame_size(&svc_));
+        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

test/test-data.sha1

@@ -1,5 +1,9 @@
 d5dfb0151c9051f8c85999255645d7a23916d3c0  hantro_collage_w352h288.yuv
 b87815bf86020c592ccc7a846ba2e28ec8043902  hantro_odd.yuv
+fe346136b9b8c1e6f6084cc106485706915795e4  invalid-vp90-01.webm
+25751f5d3b05ff03f0719ad42cd625348eb8961e  invalid-vp90-01.webm.res
+d78e2fceba5ac942246503ec8366f879c4775ca5  invalid-vp90-02.webm
+2dadee5306245fa5eeb0f99652d0e17afbcba96d  invalid-vp90-02.webm.res
 b1f1c3ec79114b9a0651af24ce634afb44a9a419  rush_hour_444.y4m
 5184c46ddca8b1fadd16742e8500115bc8f749da  vp80-00-comprehensive-001.ivf
 65bf1bbbced81b97bd030f376d1b7f61a224793f  vp80-00-comprehensive-002.ivf
@@ -576,6 +580,8 @@ d48c5db1b0f8e60521a7c749696b8067886033a3  vp90-2-09-aq2.webm
 54638c38009198c38c8f3b25c182b709b6c1fd2e  vp90-2-09-lf_deltas.webm.md5
 510d95f3beb3b51c572611fdaeeece12277dac30  vp90-2-10-show-existing-frame.webm
 14d631096f4bfa2d71f7f739aec1448fb3c33bad  vp90-2-10-show-existing-frame.webm.md5
+d2feea7728e8d2c615981d0f47427a4a5a45d881  vp90-2-10-show-existing-frame2.webm
+5f7c7811baa3e4f03be1dd78c33971b727846821  vp90-2-10-show-existing-frame2.webm.md5
 b4318e75f73a6a08992c7326de2fb589c2a794c7  vp90-2-11-size-351x287.webm
 b3c48382cf7d0454e83a02497c229d27720f9e20  vp90-2-11-size-351x287.webm.md5
 8e0096475ea2535bac71d3e2fc09e0c451c444df  vp90-2-11-size-351x288.webm
@@ -638,5 +644,5 @@ e615575ded499ea1d992f3b38e3baa434509cdcd  vp90-2-15-segkey.webm
 e3ab35d4316c5e81325c50f5236ceca4bc0d35df  vp90-2-15-segkey.webm.md5
 9b7ca2cac09d34c4a5d296c1900f93b1e2f69d0d  vp90-2-15-segkey_adpq.webm
 8f46ba5f785d0c2170591a153e0d0d146a7c8090  vp90-2-15-segkey_adpq.webm.md5
-d78e2fceba5ac942246503ec8366f879c4775ca5  vp90-2-15-fuzz-flicker.webm
-bbd7dd15f43a703ff0a332fee4959e7b23bf77dc  vp90-2-15-fuzz-flicker.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

test/test.mk

@@ -30,6 +30,7 @@ LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += cq_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += keyframe_test.cc
 
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += external_frame_buffer_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += user_priv_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += active_map_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += borders_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += cpu_speed_test.cc
@@ -54,6 +55,7 @@ LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += ../webmdec.h
 LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += webm_video_source.h
 endif
 
+LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += invalid_file_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += test_vector_test.cc
 
 # Currently we only support decoder perf tests for vp9. Also they read from WebM
@@ -690,6 +692,8 @@ LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-09-subpixel-00.ivf
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-09-subpixel-00.ivf.md5
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-10-show-existing-frame.webm
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-10-show-existing-frame.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-10-show-existing-frame2.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-10-show-existing-frame2.webm.md5
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-11-size-351x287.webm
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-11-size-351x287.webm.md5
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-11-size-351x288.webm
@@ -754,8 +758,14 @@ LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-15-segkey.webm
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-15-segkey.webm.md5
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-15-segkey_adpq.webm
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-15-segkey_adpq.webm.md5
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-15-fuzz-flicker.webm
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-15-fuzz-flicker.webm.md5
+
+# Invalid files for testing libvpx error checking.
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-01.webm
+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

test/test_vectors.cc

@@ -161,6 +161,7 @@ const char *const kVP9TestVectors[] = {
   "vp90-2-08-tile-4x1.webm", "vp90-2-09-subpixel-00.ivf",
   "vp90-2-02-size-lf-1920x1080.webm", "vp90-2-09-aq2.webm",
   "vp90-2-09-lf_deltas.webm", "vp90-2-10-show-existing-frame.webm",
+  "vp90-2-10-show-existing-frame2.webm",
   "vp90-2-11-size-351x287.webm", "vp90-2-11-size-351x288.webm",
   "vp90-2-11-size-352x287.webm", "vp90-2-12-droppable_1.ivf",
   "vp90-2-12-droppable_2.ivf", "vp90-2-12-droppable_3.ivf",
@@ -179,7 +180,6 @@ const char *const kVP9TestVectors[] = {
   "vp90-2-14-resize-fp-tiles-8-16.webm", "vp90-2-14-resize-fp-tiles-8-1.webm",
   "vp90-2-14-resize-fp-tiles-8-2.webm", "vp90-2-14-resize-fp-tiles-8-4.webm",
   "vp90-2-15-segkey.webm", "vp90-2-15-segkey_adpq.webm",
-  "vp90-2-15-fuzz-flicker.webm"
 };
 const int kNumVP9TestVectors = NELEMENTS(kVP9TestVectors);
 #endif  // CONFIG_VP9_DECODER

test/user_priv_test.cc

@@ -0,0 +1,100 @@
+/*
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <cstdio>
+#include <cstdlib>
+#include <string>
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "./vpx_config.h"
+#include "test/acm_random.h"
+#include "test/codec_factory.h"
+#include "test/decode_test_driver.h"
+#include "test/ivf_video_source.h"
+#include "test/md5_helper.h"
+#include "test/util.h"
+#if CONFIG_WEBM_IO
+#include "test/webm_video_source.h"
+#endif
+#include "vpx_mem/vpx_mem.h"
+#include "vpx/vp8.h"
+
+namespace {
+
+using std::string;
+using libvpx_test::ACMRandom;
+
+#if CONFIG_WEBM_IO
+
+void CheckUserPrivateData(void *user_priv, int *target) {
+  // actual pointer value should be the same as expected.
+  EXPECT_EQ(reinterpret_cast<void *>(target), user_priv) <<
+      "user_priv pointer value does not match.";
+}
+
+// Decodes |filename|. Passes in user_priv data when calling DecodeFrame and
+// compares the user_priv from return img with the original user_priv to see if
+// they match. Both the pointer values and the values inside the addresses
+// should match.
+string DecodeFile(const string &filename) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  libvpx_test::WebMVideoSource video(filename);
+  video.Init();
+
+  vpx_codec_dec_cfg_t cfg = {0};
+  libvpx_test::VP9Decoder decoder(cfg, 0);
+
+  libvpx_test::MD5 md5;
+  int frame_num = 0;
+  for (video.Begin(); !::testing::Test::HasFailure() && video.cxdata();
+       video.Next()) {
+    void *user_priv = reinterpret_cast<void *>(&frame_num);
+    const vpx_codec_err_t res =
+        decoder.DecodeFrame(video.cxdata(), video.frame_size(),
+                            (frame_num == 0) ? NULL : user_priv);
+    if (res != VPX_CODEC_OK) {
+      EXPECT_EQ(VPX_CODEC_OK, res) << decoder.DecodeError();
+      break;
+    }
+    libvpx_test::DxDataIterator dec_iter = decoder.GetDxData();
+    const vpx_image_t *img = NULL;
+
+    // Get decompressed data.
+    while ((img = dec_iter.Next())) {
+      if (frame_num == 0) {
+        CheckUserPrivateData(img->user_priv, NULL);
+      } else {
+        CheckUserPrivateData(img->user_priv, &frame_num);
+
+        // Also test ctrl_get_reference api.
+        struct vp9_ref_frame ref;
+        // Randomly fetch a reference frame.
+        ref.idx = rnd.Rand8() % 3;
+        decoder.Control(VP9_GET_REFERENCE, &ref);
+
+        CheckUserPrivateData(ref.img.user_priv, &frame_num);
+      }
+      md5.Add(img);
+    }
+
+    frame_num++;
+  }
+  return string(md5.Get());
+}
+
+TEST(UserPrivTest, VideoDecode) {
+  // no tiles or frame parallel; this exercises the decoding to test the
+  // user_priv.
+  EXPECT_STREQ("b35a1b707b28e82be025d960aba039bc",
+               DecodeFile("vp90-2-03-size-226x226.webm").c_str());
+}
+
+#endif  // CONFIG_WEBM_IO
+
+}  // namespace

third_party/libmkv/EbmlIDs.h

@@ -0,0 +1,231 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#ifndef MKV_DEFS_HPP
+#define MKV_DEFS_HPP 1
+
+/* Commenting out values not available in webm, but available in matroska */
+
+enum mkv {
+  EBML = 0x1A45DFA3,
+  EBMLVersion = 0x4286,
+  EBMLReadVersion = 0x42F7,
+  EBMLMaxIDLength = 0x42F2,
+  EBMLMaxSizeLength = 0x42F3,
+  DocType = 0x4282,
+  DocTypeVersion = 0x4287,
+  DocTypeReadVersion = 0x4285,
+/* CRC_32 = 0xBF, */
+  Void = 0xEC,
+  SignatureSlot = 0x1B538667,
+  SignatureAlgo = 0x7E8A,
+  SignatureHash = 0x7E9A,
+  SignaturePublicKey = 0x7EA5,
+  Signature = 0x7EB5,
+  SignatureElements = 0x7E5B,
+  SignatureElementList = 0x7E7B,
+  SignedElement = 0x6532,
+  /* segment */
+  Segment = 0x18538067,
+  /* Meta Seek Information */
+  SeekHead = 0x114D9B74,
+  Seek = 0x4DBB,
+  SeekID = 0x53AB,
+  SeekPosition = 0x53AC,
+  /* Segment Information */
+  Info = 0x1549A966,
+/* SegmentUID = 0x73A4, */
+/* SegmentFilename = 0x7384, */
+/* PrevUID = 0x3CB923, */
+/* PrevFilename = 0x3C83AB, */
+/* NextUID = 0x3EB923, */
+/* NextFilename = 0x3E83BB, */
+/* SegmentFamily = 0x4444, */
+/* ChapterTranslate = 0x6924, */
+/* ChapterTranslateEditionUID = 0x69FC, */
+/* ChapterTranslateCodec = 0x69BF, */
+/* ChapterTranslateID = 0x69A5, */
+  TimecodeScale = 0x2AD7B1,
+  Segment_Duration = 0x4489,
+  DateUTC = 0x4461,
+/* Title = 0x7BA9, */
+  MuxingApp = 0x4D80,
+  WritingApp = 0x5741,
+  /* Cluster */
+  Cluster = 0x1F43B675,
+  Timecode = 0xE7,
+/* SilentTracks = 0x5854, */
+/* SilentTrackNumber = 0x58D7, */
+/* Position = 0xA7, */
+  PrevSize = 0xAB,
+  BlockGroup = 0xA0,
+  Block = 0xA1,
+/* BlockVirtual = 0xA2, */
+  BlockAdditions = 0x75A1,
+  BlockMore = 0xA6,
+  BlockAddID = 0xEE,
+  BlockAdditional = 0xA5,
+  BlockDuration = 0x9B,
+/* ReferencePriority = 0xFA, */
+  ReferenceBlock = 0xFB,
+/* ReferenceVirtual = 0xFD, */
+/* CodecState = 0xA4, */
+/* Slices = 0x8E, */
+/* TimeSlice = 0xE8, */
+  LaceNumber = 0xCC,
+/* FrameNumber = 0xCD, */
+/* BlockAdditionID = 0xCB, */
+/* MkvDelay = 0xCE, */
+/* Cluster_Duration = 0xCF, */
+  SimpleBlock = 0xA3,
+/* EncryptedBlock = 0xAF, */
+  /* Track */
+  Tracks = 0x1654AE6B,
+  TrackEntry = 0xAE,
+  TrackNumber = 0xD7,
+  TrackUID = 0x73C5,
+  TrackType = 0x83,
+  FlagEnabled = 0xB9,
+  FlagDefault = 0x88,
+  FlagForced = 0x55AA,
+  FlagLacing = 0x9C,
+/* MinCache = 0x6DE7, */
+/* MaxCache = 0x6DF8, */
+  DefaultDuration = 0x23E383,
+/* TrackTimecodeScale = 0x23314F, */
+/* TrackOffset = 0x537F, */
+  MaxBlockAdditionID = 0x55EE,
+  Name = 0x536E,
+  Language = 0x22B59C,
+  CodecID = 0x86,
+  CodecPrivate = 0x63A2,
+  CodecName = 0x258688,
+/* AttachmentLink = 0x7446, */
+/* CodecSettings = 0x3A9697, */
+/* CodecInfoURL = 0x3B4040, */
+/* CodecDownloadURL = 0x26B240, */
+/* CodecDecodeAll = 0xAA, */
+/* TrackOverlay = 0x6FAB, */
+/* TrackTranslate = 0x6624, */
+/* TrackTranslateEditionUID = 0x66FC, */
+/* TrackTranslateCodec = 0x66BF, */
+/* TrackTranslateTrackID = 0x66A5, */
+  /* video */
+  Video = 0xE0,
+  FlagInterlaced = 0x9A,
+  StereoMode = 0x53B8,
+  AlphaMode = 0x53C0,
+  PixelWidth = 0xB0,
+  PixelHeight = 0xBA,
+  PixelCropBottom = 0x54AA,
+  PixelCropTop = 0x54BB,
+  PixelCropLeft = 0x54CC,
+  PixelCropRight = 0x54DD,
+  DisplayWidth = 0x54B0,
+  DisplayHeight = 0x54BA,
+  DisplayUnit = 0x54B2,
+  AspectRatioType = 0x54B3,
+/* ColourSpace = 0x2EB524, */
+/* GammaValue = 0x2FB523, */
+  FrameRate = 0x2383E3,
+  /* end video */
+  /* audio */
+  Audio = 0xE1,
+  SamplingFrequency = 0xB5,
+  OutputSamplingFrequency = 0x78B5,
+  Channels = 0x9F,
+/* ChannelPositions = 0x7D7B, */
+  BitDepth = 0x6264,
+  /* end audio */
+  /* content encoding */
+/* ContentEncodings = 0x6d80, */
+/* ContentEncoding = 0x6240, */
+/* ContentEncodingOrder = 0x5031, */
+/* ContentEncodingScope = 0x5032, */
+/* ContentEncodingType = 0x5033, */
+/* ContentCompression = 0x5034, */
+/* ContentCompAlgo = 0x4254, */
+/* ContentCompSettings = 0x4255, */
+/* ContentEncryption = 0x5035, */
+/* ContentEncAlgo = 0x47e1, */
+/* ContentEncKeyID = 0x47e2, */
+/* ContentSignature = 0x47e3, */
+/* ContentSigKeyID = 0x47e4, */
+/* ContentSigAlgo = 0x47e5, */
+/* ContentSigHashAlgo = 0x47e6, */
+  /* end content encoding */
+  /* Cueing Data */
+  Cues = 0x1C53BB6B,
+  CuePoint = 0xBB,
+  CueTime = 0xB3,
+  CueTrackPositions = 0xB7,
+  CueTrack = 0xF7,
+  CueClusterPosition = 0xF1,
+  CueBlockNumber = 0x5378
+/* CueCodecState = 0xEA, */
+/* CueReference = 0xDB, */
+/* CueRefTime = 0x96, */
+/* CueRefCluster = 0x97, */
+/* CueRefNumber = 0x535F, */
+/* CueRefCodecState = 0xEB, */
+  /* Attachment */
+/* Attachments = 0x1941A469, */
+/* AttachedFile = 0x61A7, */
+/* FileDescription = 0x467E, */
+/* FileName = 0x466E, */
+/* FileMimeType = 0x4660, */
+/* FileData = 0x465C, */
+/* FileUID = 0x46AE, */
+/* FileReferral = 0x4675, */
+  /* Chapters */
+/* Chapters = 0x1043A770, */
+/* EditionEntry = 0x45B9, */
+/* EditionUID = 0x45BC, */
+/* EditionFlagHidden = 0x45BD, */
+/* EditionFlagDefault = 0x45DB, */
+/* EditionFlagOrdered = 0x45DD, */
+/* ChapterAtom = 0xB6, */
+/* ChapterUID = 0x73C4, */
+/* ChapterTimeStart = 0x91, */
+/* ChapterTimeEnd = 0x92, */
+/* ChapterFlagHidden = 0x98, */
+/* ChapterFlagEnabled = 0x4598, */
+/* ChapterSegmentUID = 0x6E67, */
+/* ChapterSegmentEditionUID = 0x6EBC, */
+/* ChapterPhysicalEquiv = 0x63C3, */
+/* ChapterTrack = 0x8F, */
+/* ChapterTrackNumber = 0x89, */
+/* ChapterDisplay = 0x80, */
+/* ChapString = 0x85, */
+/* ChapLanguage = 0x437C, */
+/* ChapCountry = 0x437E, */
+/* ChapProcess = 0x6944, */
+/* ChapProcessCodecID = 0x6955, */
+/* ChapProcessPrivate = 0x450D, */
+/* ChapProcessCommand = 0x6911, */
+/* ChapProcessTime = 0x6922, */
+/* ChapProcessData = 0x6933, */
+  /* Tagging */
+/* Tags = 0x1254C367, */
+/* Tag = 0x7373, */
+/* Targets = 0x63C0, */
+/* TargetTypeValue = 0x68CA, */
+/* TargetType = 0x63CA, */
+/* Tagging_TrackUID = 0x63C5, */
+/* Tagging_EditionUID = 0x63C9, */
+/* Tagging_ChapterUID = 0x63C4, */
+/* AttachmentUID = 0x63C6, */
+/* SimpleTag = 0x67C8, */
+/* TagName = 0x45A3, */
+/* TagLanguage = 0x447A, */
+/* TagDefault = 0x4484, */
+/* TagString = 0x4487, */
+/* TagBinary = 0x4485, */
+};
+#endif

third_party/libmkv/EbmlWriter.c

@@ -0,0 +1,157 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include "EbmlWriter.h"
+#include <stdlib.h>
+#include <wchar.h>
+#include <string.h>
+#include <limits.h>
+#if defined(_MSC_VER)
+#define LITERALU64(n) n
+#else
+#define LITERALU64(n) n##LLU
+#endif
+
+void Ebml_WriteLen(EbmlGlobal *glob, int64_t val) {
+  /* TODO check and make sure we are not > than 0x0100000000000000LLU */
+  unsigned char size = 8; /* size in bytes to output */
+
+  /* mask to compare for byte size */
+  int64_t minVal = 0xff;
+
+  for (size = 1; size < 8; size ++) {
+    if (val < minVal)
+      break;
+
+    minVal = (minVal << 7);
+  }
+
+  val |= (((uint64_t)0x80) << ((size - 1) * 7));
+
+  Ebml_Serialize(glob, (void *) &val, sizeof(val), size);
+}
+
+void Ebml_WriteString(EbmlGlobal *glob, const char *str) {
+  const size_t size_ = strlen(str);
+  const uint64_t  size = size_;
+  Ebml_WriteLen(glob, size);
+  /* TODO: it's not clear from the spec whether the nul terminator
+   * should be serialized too.  For now we omit the null terminator.
+   */
+  Ebml_Write(glob, str, (unsigned long)size);
+}
+
+void Ebml_WriteUTF8(EbmlGlobal *glob, const wchar_t *wstr) {
+  const size_t strlen = wcslen(wstr);
+
+  /* TODO: it's not clear from the spec whether the nul terminator
+   * should be serialized too.  For now we include it.
+   */
+  const uint64_t  size = strlen;
+
+  Ebml_WriteLen(glob, size);
+  Ebml_Write(glob, wstr, (unsigned long)size);
+}
+
+void Ebml_WriteID(EbmlGlobal *glob, unsigned long class_id) {
+  int len;
+
+  if (class_id >= 0x01000000)
+    len = 4;
+  else if (class_id >= 0x00010000)
+    len = 3;
+  else if (class_id >= 0x00000100)
+    len = 2;
+  else
+    len = 1;
+
+  Ebml_Serialize(glob, (void *)&class_id, sizeof(class_id), len);
+}
+
+void Ebml_SerializeUnsigned64(EbmlGlobal *glob, unsigned long class_id, uint64_t ui) {
+  unsigned char sizeSerialized = 8 | 0x80;
+  Ebml_WriteID(glob, class_id);
+  Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
+  Ebml_Serialize(glob, &ui, sizeof(ui), 8);
+}
+
+void Ebml_SerializeUnsigned(EbmlGlobal *glob, unsigned long class_id, unsigned long ui) {
+  unsigned char size = 8; /* size in bytes to output */
+  unsigned char sizeSerialized = 0;
+  unsigned long minVal;
+
+  Ebml_WriteID(glob, class_id);
+  minVal = 0x7fLU; /* mask to compare for byte size */
+
+  for (size = 1; size < 4; size ++) {
+    if (ui < minVal) {
+      break;
+    }
+
+    minVal <<= 7;
+  }
+
+  sizeSerialized = 0x80 | size;
+  Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
+  Ebml_Serialize(glob, &ui, sizeof(ui), size);
+}
+/* TODO: perhaps this is a poor name for this id serializer helper function */
+void Ebml_SerializeBinary(EbmlGlobal *glob, unsigned long class_id, unsigned long bin) {
+  int size;
+  for (size = 4; size > 1; size--) {
+    if (bin & (unsigned int)0x000000ff << ((size - 1) * 8))
+      break;
+  }
+  Ebml_WriteID(glob, class_id);
+  Ebml_WriteLen(glob, size);
+  Ebml_WriteID(glob, bin);
+}
+
+void Ebml_SerializeFloat(EbmlGlobal *glob, unsigned long class_id, double d) {
+  unsigned char len = 0x88;
+
+  Ebml_WriteID(glob, class_id);
+  Ebml_Serialize(glob, &len, sizeof(len), 1);
+  Ebml_Serialize(glob,  &d, sizeof(d), 8);
+}
+
+void Ebml_WriteSigned16(EbmlGlobal *glob, short val) {
+  signed long out = ((val & 0x003FFFFF) | 0x00200000) << 8;
+  Ebml_Serialize(glob, &out, sizeof(out), 3);
+}
+
+void Ebml_SerializeString(EbmlGlobal *glob, unsigned long class_id, const char *s) {
+  Ebml_WriteID(glob, class_id);
+  Ebml_WriteString(glob, s);
+}
+
+void Ebml_SerializeUTF8(EbmlGlobal *glob, unsigned long class_id, wchar_t *s) {
+  Ebml_WriteID(glob,  class_id);
+  Ebml_WriteUTF8(glob,  s);
+}
+
+void Ebml_SerializeData(EbmlGlobal *glob, unsigned long class_id, unsigned char *data, unsigned long data_length) {
+  Ebml_WriteID(glob, class_id);
+  Ebml_WriteLen(glob, data_length);
+  Ebml_Write(glob,  data, data_length);
+}
+
+void Ebml_WriteVoid(EbmlGlobal *glob, unsigned long vSize) {
+  unsigned char tmp = 0;
+  unsigned long i = 0;
+
+  Ebml_WriteID(glob, 0xEC);
+  Ebml_WriteLen(glob, vSize);
+
+  for (i = 0; i < vSize; i++) {
+    Ebml_Write(glob, &tmp, 1);
+  }
+}
+
+/* TODO Serialize Date */

third_party/libmkv/EbmlWriter.h

@@ -0,0 +1,42 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#ifndef EBMLWRITER_HPP
+#define EBMLWRITER_HPP
+#include <stddef.h>
+#include "vpx/vpx_integer.h"
+
+/* note: you must define write and serialize functions as well as your own
+ * EBML_GLOBAL
+ *
+ * These functions MUST be implemented
+ */
+
+typedef struct EbmlGlobal EbmlGlobal;
+void  Ebml_Serialize(EbmlGlobal *glob, const void *, int, unsigned long);
+void  Ebml_Write(EbmlGlobal *glob, const void *, unsigned long);
+
+/*****/
+
+void Ebml_WriteLen(EbmlGlobal *glob, int64_t val);
+void Ebml_WriteString(EbmlGlobal *glob, const char *str);
+void Ebml_WriteUTF8(EbmlGlobal *glob, const wchar_t *wstr);
+void Ebml_WriteID(EbmlGlobal *glob, unsigned long class_id);
+void Ebml_SerializeUnsigned64(EbmlGlobal *glob, unsigned long class_id, uint64_t ui);
+void Ebml_SerializeUnsigned(EbmlGlobal *glob, unsigned long class_id, unsigned long ui);
+void Ebml_SerializeBinary(EbmlGlobal *glob, unsigned long class_id, unsigned long ui);
+void Ebml_SerializeFloat(EbmlGlobal *glob, unsigned long class_id, double d);
+/* TODO make this more generic to signed */
+void Ebml_WriteSigned16(EbmlGlobal *glob, short val);
+void Ebml_SerializeString(EbmlGlobal *glob, unsigned long class_id, const char *s);
+void Ebml_SerializeUTF8(EbmlGlobal *glob, unsigned long class_id, wchar_t *s);
+void Ebml_SerializeData(EbmlGlobal *glob, unsigned long class_id, unsigned char *data, unsigned long data_length);
+void Ebml_WriteVoid(EbmlGlobal *glob, unsigned long vSize);
+/* TODO need date function */
+#endif

vp8/common/rtcd_defs.pl

@@ -463,9 +463,7 @@ $vp8_short_walsh4x4_neon_asm=vp8_short_walsh4x4_neon;
 # Quantizer
 #
 add_proto qw/void vp8_regular_quantize_b/, "struct block *, struct blockd *";
-specialize qw/vp8_regular_quantize_b sse2/;
-# TODO(johann) Update sse4 implementation and re-enable
-#$vp8_regular_quantize_b_sse4_1=vp8_regular_quantize_b_sse4;
+specialize qw/vp8_regular_quantize_b sse2 sse4_1/;
 
 add_proto qw/void vp8_fast_quantize_b/, "struct block *, struct blockd *";
 specialize qw/vp8_fast_quantize_b sse2 ssse3 media neon_asm/;

vp8/encoder/x86/quantize_sse2.c

@@ -26,11 +26,10 @@
         int cmp = (x[z] < boost) | (y[z] == 0); \
         zbin_boost_ptr++; \
         if (cmp) \
-            goto select_eob_end_##i; \
+            break; \
         qcoeff_ptr[z] = y[z]; \
         eob = i; \
         zbin_boost_ptr = b->zrun_zbin_boost; \
-        select_eob_end_##i:; \
     } while (0)
 
 void vp8_regular_quantize_b_sse2(BLOCK *b, BLOCKD *d)

vp8/encoder/x86/quantize_sse4.asm

@@ -1,256 +0,0 @@
-;
-;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-;  Use of this source code is governed by a BSD-style license and patent
-;  grant that can be found in the LICENSE file in the root of the source
-;  tree. All contributing project authors may be found in the AUTHORS
-;  file in the root of the source tree.
-;
-
-
-%include "vpx_ports/x86_abi_support.asm"
-%include "vp8_asm_enc_offsets.asm"
-
-
-; void vp8_regular_quantize_b_sse4 | arg
-;  (BLOCK  *b,                     |  0
-;   BLOCKD *d)                     |  1
-
-global sym(vp8_regular_quantize_b_sse4) PRIVATE
-sym(vp8_regular_quantize_b_sse4):
-
-%if ABI_IS_32BIT
-    push        rbp
-    mov         rbp, rsp
-    GET_GOT     rbx
-    push        rdi
-    push        rsi
-
-    ALIGN_STACK 16, rax
-    %define qcoeff      0 ; 32
-    %define stack_size 32
-    sub         rsp, stack_size
-%else
-  %if LIBVPX_YASM_WIN64
-    SAVE_XMM 8, u
-    push        rdi
-    push        rsi
-  %endif
-%endif
-    ; end prolog
-
-%if ABI_IS_32BIT
-    mov         rdi, arg(0)                 ; BLOCK *b
-    mov         rsi, arg(1)                 ; BLOCKD *d
-%else
-  %if LIBVPX_YASM_WIN64
-    mov         rdi, rcx                    ; BLOCK *b
-    mov         rsi, rdx                    ; BLOCKD *d
-  %else
-    ;mov         rdi, rdi                    ; BLOCK *b
-    ;mov         rsi, rsi                    ; BLOCKD *d
-  %endif
-%endif
-
-    mov         rax, [rdi + vp8_block_coeff]
-    mov         rcx, [rdi + vp8_block_zbin]
-    mov         rdx, [rdi + vp8_block_round]
-    movd        xmm7, [rdi + vp8_block_zbin_extra]
-
-    ; z
-    movdqa      xmm0, [rax]
-    movdqa      xmm1, [rax + 16]
-
-    ; duplicate zbin_oq_value
-    pshuflw     xmm7, xmm7, 0
-    punpcklwd   xmm7, xmm7
-
-    movdqa      xmm2, xmm0
-    movdqa      xmm3, xmm1
-
-    ; sz
-    psraw       xmm0, 15
-    psraw       xmm1, 15
-
-    ; (z ^ sz)
-    pxor        xmm2, xmm0
-    pxor        xmm3, xmm1
-
-    ; x = abs(z)
-    psubw       xmm2, xmm0
-    psubw       xmm3, xmm1
-
-    ; zbin
-    movdqa      xmm4, [rcx]
-    movdqa      xmm5, [rcx + 16]
-
-    ; *zbin_ptr + zbin_oq_value
-    paddw       xmm4, xmm7
-    paddw       xmm5, xmm7
-
-    movdqa      xmm6, xmm2
-    movdqa      xmm7, xmm3
-
-    ; x - (*zbin_ptr + zbin_oq_value)
-    psubw       xmm6, xmm4
-    psubw       xmm7, xmm5
-
-    ; round
-    movdqa      xmm4, [rdx]
-    movdqa      xmm5, [rdx + 16]
-
-    mov         rax, [rdi + vp8_block_quant_shift]
-    mov         rcx, [rdi + vp8_block_quant]
-    mov         rdx, [rdi + vp8_block_zrun_zbin_boost]
-
-    ; x + round
-    paddw       xmm2, xmm4
-    paddw       xmm3, xmm5
-
-    ; quant
-    movdqa      xmm4, [rcx]
-    movdqa      xmm5, [rcx + 16]
-
-    ; y = x * quant_ptr >> 16
-    pmulhw      xmm4, xmm2
-    pmulhw      xmm5, xmm3
-
-    ; y += x
-    paddw       xmm2, xmm4
-    paddw       xmm3, xmm5
-
-    pxor        xmm4, xmm4
-%if ABI_IS_32BIT
-    movdqa      [rsp + qcoeff], xmm4
-    movdqa      [rsp + qcoeff + 16], xmm4
-%else
-    pxor        xmm8, xmm8
-%endif
-
-    ; quant_shift
-    movdqa      xmm5, [rax]
-
-    ; zrun_zbin_boost
-    mov         rax, rdx
-
-%macro ZIGZAG_LOOP 5
-    ; x
-    pextrw      ecx, %4, %2
-
-    ; if (x >= zbin)
-    sub         cx, WORD PTR[rdx]           ; x - zbin
-    lea         rdx, [rdx + 2]              ; zbin_boost_ptr++
-    jl          .rq_zigzag_loop_%1          ; x < zbin
-
-    pextrw      edi, %3, %2                 ; y
-
-    ; downshift by quant_shift[rc]
-    pextrb      ecx, xmm5, %1               ; quant_shift[rc]
-    sar         edi, cl                     ; also sets Z bit
-    je          .rq_zigzag_loop_%1          ; !y
-%if ABI_IS_32BIT
-    mov         WORD PTR[rsp + qcoeff + %1 *2], di
-%else
-    pinsrw      %5, edi, %2                 ; qcoeff[rc]
-%endif
-    mov         rdx, rax                    ; reset to b->zrun_zbin_boost
-.rq_zigzag_loop_%1:
-%endmacro
-; in vp8_default_zig_zag1d order: see vp8/common/entropy.c
-ZIGZAG_LOOP  0, 0, xmm2, xmm6, xmm4
-ZIGZAG_LOOP  1, 1, xmm2, xmm6, xmm4
-ZIGZAG_LOOP  4, 4, xmm2, xmm6, xmm4
-ZIGZAG_LOOP  8, 0, xmm3, xmm7, xmm8
-ZIGZAG_LOOP  5, 5, xmm2, xmm6, xmm4
-ZIGZAG_LOOP  2, 2, xmm2, xmm6, xmm4
-ZIGZAG_LOOP  3, 3, xmm2, xmm6, xmm4
-ZIGZAG_LOOP  6, 6, xmm2, xmm6, xmm4
-ZIGZAG_LOOP  9, 1, xmm3, xmm7, xmm8
-ZIGZAG_LOOP 12, 4, xmm3, xmm7, xmm8
-ZIGZAG_LOOP 13, 5, xmm3, xmm7, xmm8
-ZIGZAG_LOOP 10, 2, xmm3, xmm7, xmm8
-ZIGZAG_LOOP  7, 7, xmm2, xmm6, xmm4
-ZIGZAG_LOOP 11, 3, xmm3, xmm7, xmm8
-ZIGZAG_LOOP 14, 6, xmm3, xmm7, xmm8
-ZIGZAG_LOOP 15, 7, xmm3, xmm7, xmm8
-
-    mov         rcx, [rsi + vp8_blockd_dequant]
-    mov         rdi, [rsi + vp8_blockd_dqcoeff]
-
-%if ABI_IS_32BIT
-    movdqa      xmm4, [rsp + qcoeff]
-    movdqa      xmm5, [rsp + qcoeff + 16]
-%else
-    %define     xmm5 xmm8
-%endif
-
-    ; y ^ sz
-    pxor        xmm4, xmm0
-    pxor        xmm5, xmm1
-    ; x = (y ^ sz) - sz
-    psubw       xmm4, xmm0
-    psubw       xmm5, xmm1
-
-    ; dequant
-    movdqa      xmm0, [rcx]
-    movdqa      xmm1, [rcx + 16]
-
-    mov         rcx, [rsi + vp8_blockd_qcoeff]
-
-    pmullw      xmm0, xmm4
-    pmullw      xmm1, xmm5
-
-    ; store qcoeff
-    movdqa      [rcx], xmm4
-    movdqa      [rcx + 16], xmm5
-
-    ; store dqcoeff
-    movdqa      [rdi], xmm0
-    movdqa      [rdi + 16], xmm1
-
-    mov         rcx, [rsi + vp8_blockd_eob]
-
-    ; select the last value (in zig_zag order) for EOB
-    pxor        xmm6, xmm6
-    pcmpeqw     xmm4, xmm6
-    pcmpeqw     xmm5, xmm6
-
-    packsswb    xmm4, xmm5
-    pshufb      xmm4, [GLOBAL(zig_zag1d)]
-    pmovmskb    edx, xmm4
-    xor         rdi, rdi
-    mov         eax, -1
-    xor         dx, ax
-    bsr         eax, edx
-    sub         edi, edx
-    sar         edi, 31
-    add         eax, 1
-    and         eax, edi
-
-    mov         BYTE PTR [rcx], al          ; store eob
-
-    ; begin epilog
-%if ABI_IS_32BIT
-    add         rsp, stack_size
-    pop         rsp
-
-    pop         rsi
-    pop         rdi
-    RESTORE_GOT
-    pop         rbp
-%else
-  %undef xmm5
-  %if LIBVPX_YASM_WIN64
-    pop         rsi
-    pop         rdi
-    RESTORE_XMM
-  %endif
-%endif
-
-    ret
-
-SECTION_RODATA
-align 16
-; vp8/common/entropy.c: vp8_default_zig_zag1d
-zig_zag1d:
-    db 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15

vp8/encoder/x86/quantize_sse4.c

@@ -0,0 +1,128 @@
+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <smmintrin.h> /* SSE4.1 */
+
+#include "./vp8_rtcd.h"
+#include "vp8/encoder/block.h"
+#include "vp8/common/entropy.h" /* vp8_default_inv_zig_zag */
+
+#define SELECT_EOB(i, z, x, y, q) \
+    do { \
+        short boost = *zbin_boost_ptr; \
+        short x_z = _mm_extract_epi16(x, z); \
+        short y_z = _mm_extract_epi16(y, z); \
+        int cmp = (x_z < boost) | (y_z == 0); \
+        zbin_boost_ptr++; \
+        if (cmp) \
+            break; \
+        q = _mm_insert_epi16(q, y_z, z); \
+        eob = i; \
+        zbin_boost_ptr = b->zrun_zbin_boost; \
+    } while (0)
+
+void vp8_regular_quantize_b_sse4_1(BLOCK *b, BLOCKD *d) {
+    char eob = 0;
+    short *zbin_boost_ptr  = b->zrun_zbin_boost;
+
+    __m128i sz0, x0, sz1, x1, y0, y1, x_minus_zbin0, x_minus_zbin1,
+            dqcoeff0, dqcoeff1;
+    __m128i quant_shift0 = _mm_load_si128((__m128i *)(b->quant_shift));
+    __m128i quant_shift1 = _mm_load_si128((__m128i *)(b->quant_shift + 8));
+    __m128i z0 = _mm_load_si128((__m128i *)(b->coeff));
+    __m128i z1 = _mm_load_si128((__m128i *)(b->coeff+8));
+    __m128i zbin_extra = _mm_cvtsi32_si128(b->zbin_extra);
+    __m128i zbin0 = _mm_load_si128((__m128i *)(b->zbin));
+    __m128i zbin1 = _mm_load_si128((__m128i *)(b->zbin + 8));
+    __m128i round0 = _mm_load_si128((__m128i *)(b->round));
+    __m128i round1 = _mm_load_si128((__m128i *)(b->round + 8));
+    __m128i quant0 = _mm_load_si128((__m128i *)(b->quant));
+    __m128i quant1 = _mm_load_si128((__m128i *)(b->quant + 8));
+    __m128i dequant0 = _mm_load_si128((__m128i *)(d->dequant));
+    __m128i dequant1 = _mm_load_si128((__m128i *)(d->dequant + 8));
+    __m128i qcoeff0 = _mm_setzero_si128();
+    __m128i qcoeff1 = _mm_setzero_si128();
+
+    /* Duplicate to all lanes. */
+    zbin_extra = _mm_shufflelo_epi16(zbin_extra, 0);
+    zbin_extra = _mm_unpacklo_epi16(zbin_extra, zbin_extra);
+
+    /* Sign of z: z >> 15 */
+    sz0 = _mm_srai_epi16(z0, 15);
+    sz1 = _mm_srai_epi16(z1, 15);
+
+    /* x = abs(z): (z ^ sz) - sz */
+    x0 = _mm_xor_si128(z0, sz0);
+    x1 = _mm_xor_si128(z1, sz1);
+    x0 = _mm_sub_epi16(x0, sz0);
+    x1 = _mm_sub_epi16(x1, sz1);
+
+    /* zbin[] + zbin_extra */
+    zbin0 = _mm_add_epi16(zbin0, zbin_extra);
+    zbin1 = _mm_add_epi16(zbin1, zbin_extra);
+
+    /* In C x is compared to zbin where zbin = zbin[] + boost + extra. Rebalance
+     * the equation because boost is the only value which can change:
+     * x - (zbin[] + extra) >= boost */
+    x_minus_zbin0 = _mm_sub_epi16(x0, zbin0);
+    x_minus_zbin1 = _mm_sub_epi16(x1, zbin1);
+
+    /* All the remaining calculations are valid whether they are done now with
+     * simd or later inside the loop one at a time. */
+    x0 = _mm_add_epi16(x0, round0);
+    x1 = _mm_add_epi16(x1, round1);
+
+    y0 = _mm_mulhi_epi16(x0, quant0);
+    y1 = _mm_mulhi_epi16(x1, quant1);
+
+    y0 = _mm_add_epi16(y0, x0);
+    y1 = _mm_add_epi16(y1, x1);
+
+    /* Instead of shifting each value independently we convert the scaling
+     * factor with 1 << (16 - shift) so we can use multiply/return high half. */
+    y0 = _mm_mulhi_epi16(y0, quant_shift0);
+    y1 = _mm_mulhi_epi16(y1, quant_shift1);
+
+    /* Return the sign: (y ^ sz) - sz */
+    y0 = _mm_xor_si128(y0, sz0);
+    y1 = _mm_xor_si128(y1, sz1);
+    y0 = _mm_sub_epi16(y0, sz0);
+    y1 = _mm_sub_epi16(y1, sz1);
+
+    /* The loop gets unrolled anyway. Avoid the vp8_default_zig_zag1d lookup. */
+    SELECT_EOB(1, 0, x_minus_zbin0, y0, qcoeff0);
+    SELECT_EOB(2, 1, x_minus_zbin0, y0, qcoeff0);
+    SELECT_EOB(3, 4, x_minus_zbin0, y0, qcoeff0);
+    SELECT_EOB(4, 0, x_minus_zbin1, y1, qcoeff1);
+    SELECT_EOB(5, 5, x_minus_zbin0, y0, qcoeff0);
+    SELECT_EOB(6, 2, x_minus_zbin0, y0, qcoeff0);
+    SELECT_EOB(7, 3, x_minus_zbin0, y0, qcoeff0);
+    SELECT_EOB(8, 6, x_minus_zbin0, y0, qcoeff0);
+    SELECT_EOB(9, 1, x_minus_zbin1, y1, qcoeff1);
+    SELECT_EOB(10, 4, x_minus_zbin1, y1, qcoeff1);
+    SELECT_EOB(11, 5, x_minus_zbin1, y1, qcoeff1);
+    SELECT_EOB(12, 2, x_minus_zbin1, y1, qcoeff1);
+    SELECT_EOB(13, 7, x_minus_zbin0, y0, qcoeff0);
+    SELECT_EOB(14, 3, x_minus_zbin1, y1, qcoeff1);
+    SELECT_EOB(15, 6, x_minus_zbin1, y1, qcoeff1);
+    SELECT_EOB(16, 7, x_minus_zbin1, y1, qcoeff1);
+
+    _mm_store_si128((__m128i *)(d->qcoeff), qcoeff0);
+    _mm_store_si128((__m128i *)(d->qcoeff + 8), qcoeff1);
+
+    dqcoeff0 = _mm_mullo_epi16(qcoeff0, dequant0);
+    dqcoeff1 = _mm_mullo_epi16(qcoeff1, dequant1);
+
+    _mm_store_si128((__m128i *)(d->dqcoeff), dqcoeff0);
+    _mm_store_si128((__m128i *)(d->dqcoeff + 8), dqcoeff1);
+
+    *d->eob = eob;
+}

vp8/vp8cx.mk

@@ -89,6 +89,7 @@ VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/dct_sse2.asm
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/fwalsh_sse2.asm
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/quantize_sse2.c
 VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/quantize_ssse3.c
+VP8_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/quantize_sse4.c
 
 ifeq ($(CONFIG_TEMPORAL_DENOISING),yes)
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/denoising_sse2.c
@@ -97,7 +98,6 @@ endif
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/subtract_sse2.asm
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/temporal_filter_apply_sse2.asm
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp8_enc_stubs_sse2.c
-VP8_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/quantize_sse4.asm
 VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/quantize_mmx.asm
 VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/encodeopt.asm
 VP8_CX_SRCS-$(ARCH_X86_64) += encoder/x86/ssim_opt_x86_64.asm

vp9/common/arm/neon/vp9_convolve_neon.c

@@ -25,12 +25,14 @@ void vp9_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride,
   // Account for the vertical phase needing 3 lines prior and 4 lines post
   int intermediate_height = h + 7;
 
-  if (x_step_q4 != 16 || y_step_q4 != 16)
-    return vp9_convolve8_c(src, src_stride,
+  if (x_step_q4 != 16 || y_step_q4 != 16) {
+    vp9_convolve8_c(src, src_stride,
                     dst, dst_stride,
                     filter_x, x_step_q4,
                     filter_y, y_step_q4,
                     w, h);
+    return;
+  }
 
   /* Filter starting 3 lines back. The neon implementation will ignore the
    * given height and filter a multiple of 4 lines. Since this goes in to
@@ -57,12 +59,14 @@ void vp9_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride,
   DECLARE_ALIGNED_ARRAY(8, uint8_t, temp, 64 * 72);
   int intermediate_height = h + 7;
 
-  if (x_step_q4 != 16 || y_step_q4 != 16)
-    return vp9_convolve8_avg_c(src, src_stride,
+  if (x_step_q4 != 16 || y_step_q4 != 16) {
+    vp9_convolve8_avg_c(src, src_stride,
                         dst, dst_stride,
                         filter_x, x_step_q4,
                         filter_y, y_step_q4,
                         w, h);
+    return;
+  }
 
   /* This implementation has the same issues as above. In addition, we only want
    * to average the values after both passes.

vp9/common/arm/neon/vp9_loopfilter_16_neon.c

@@ -9,6 +9,7 @@
  */
 
 #include "./vp9_rtcd.h"
+#include "vpx/vpx_integer.h"
 
 void vp9_lpf_horizontal_8_dual_neon(uint8_t *s, int p /* pitch */,
                                     const uint8_t *blimit0,

vp9/common/vp9_alloccommon.c

@@ -109,7 +109,9 @@ void vp9_free_frame_buffers(VP9_COMMON *cm) {
   }
 
   vp9_free_frame_buffer(&cm->post_proc_buffer);
+}
 
+void vp9_free_context_buffers(VP9_COMMON *cm) {
   free_mi(cm);
 
   vpx_free(cm->last_frame_seg_map);
@@ -165,37 +167,55 @@ int vp9_resize_frame_buffers(VP9_COMMON *cm, int width, int height) {
 
  fail:
   vp9_free_frame_buffers(cm);
+  vp9_free_context_buffers(cm);
   return 1;
 }
 
+static void init_frame_bufs(VP9_COMMON *cm) {
+  int i;
+
+  cm->new_fb_idx = FRAME_BUFFERS - 1;
+  cm->frame_bufs[cm->new_fb_idx].ref_count = 1;
+
+  for (i = 0; i < REF_FRAMES; ++i) {
+    cm->ref_frame_map[i] = i;
+    cm->frame_bufs[i].ref_count = 1;
+  }
+}
+
 int vp9_alloc_frame_buffers(VP9_COMMON *cm, int width, int height) {
-  const int aligned_width = ALIGN_POWER_OF_TWO(width, MI_SIZE_LOG2);
-  const int aligned_height = ALIGN_POWER_OF_TWO(height, MI_SIZE_LOG2);
+  int i;
   const int ss_x = cm->subsampling_x;
   const int ss_y = cm->subsampling_y;
-  int i;
 
   vp9_free_frame_buffers(cm);
 
-  for (i = 0; i < FRAME_BUFFERS; i++) {
+  for (i = 0; i < FRAME_BUFFERS; ++i) {
     cm->frame_bufs[i].ref_count = 0;
     if (vp9_alloc_frame_buffer(&cm->frame_bufs[i].buf, width, height,
                                ss_x, ss_y, VP9_ENC_BORDER_IN_PIXELS) < 0)
       goto fail;
   }
 
-  cm->new_fb_idx = FRAME_BUFFERS - 1;
-  cm->frame_bufs[cm->new_fb_idx].ref_count = 1;
-
-  for (i = 0; i < REF_FRAMES; i++) {
-    cm->ref_frame_map[i] = i;
-    cm->frame_bufs[i].ref_count = 1;
-  }
+  init_frame_bufs(cm);
 
   if (vp9_alloc_frame_buffer(&cm->post_proc_buffer, width, height, ss_x, ss_y,
                              VP9_ENC_BORDER_IN_PIXELS) < 0)
     goto fail;
 
+  return 0;
+
+ fail:
+  vp9_free_frame_buffers(cm);
+  return 1;
+}
+
+int vp9_alloc_context_buffers(VP9_COMMON *cm, int width, int height) {
+  const int aligned_width = ALIGN_POWER_OF_TWO(width, MI_SIZE_LOG2);
+  const int aligned_height = ALIGN_POWER_OF_TWO(height, MI_SIZE_LOG2);
+
+  vp9_free_context_buffers(cm);
+
   set_mb_mi(cm, aligned_width, aligned_height);
 
   if (alloc_mi(cm, cm->mi_stride * (cm->mi_rows + MI_BLOCK_SIZE)))
@@ -224,12 +244,13 @@ int vp9_alloc_frame_buffers(VP9_COMMON *cm, int width, int height) {
   return 0;
 
  fail:
-  vp9_free_frame_buffers(cm);
+  vp9_free_context_buffers(cm);
   return 1;
 }
 
 void vp9_remove_common(VP9_COMMON *cm) {
   vp9_free_frame_buffers(cm);
+  vp9_free_context_buffers(cm);
   vp9_free_internal_frame_buffers(&cm->int_frame_buffers);
 }
 

vp9/common/vp9_alloccommon.h

@@ -23,8 +23,12 @@ void vp9_remove_common(struct VP9Common *cm);
 int vp9_resize_frame_buffers(struct VP9Common *cm, int width, int height);
 
 int vp9_alloc_frame_buffers(struct VP9Common *cm, int width, int height);
+int vp9_alloc_state_buffers(struct VP9Common *cm, int width, int height);
+int vp9_alloc_context_buffers(struct VP9Common *cm, int width, int height);
 
 void vp9_free_frame_buffers(struct VP9Common *cm);
+void vp9_free_state_buffers(struct VP9Common *cm);
+void vp9_free_context_buffers(struct VP9Common *cm);
 
 void vp9_update_frame_size(struct VP9Common *cm);
 

vp9/common/vp9_blockd.h

@@ -32,6 +32,9 @@ extern "C" {
 #define BLOCK_SIZE_GROUPS 4
 #define SKIP_CONTEXTS 3
 #define INTER_MODE_CONTEXTS 7
+#if CONFIG_COPY_CODING
+#define COPY_MODE_CONTEXTS 5
+#endif
 
 /* Segment Feature Masks */
 #define MAX_MV_REF_CANDIDATES 2
@@ -79,6 +82,16 @@ typedef enum {
   MB_MODE_COUNT
 } PREDICTION_MODE;
 
+#if CONFIG_COPY_CODING
+typedef enum {
+  NOREF,
+  REF0,
+  REF1,
+  REF2,
+  COPY_MODE_COUNT
+} COPY_MODE;
+#endif
+
 static INLINE int is_inter_mode(PREDICTION_MODE mode) {
   return mode >= NEARESTMV && mode <= NEWMV;
 }
@@ -118,11 +131,86 @@ static INLINE int mi_width_log2(BLOCK_SIZE sb_type) {
   return mi_width_log2_lookup[sb_type];
 }
 
+#if CONFIG_SUPERTX
+static INLINE TX_SIZE bsize_to_tx_size(BLOCK_SIZE bsize) {
+  const TX_SIZE tx_size_lookup[BLOCK_SIZES] = {
+    TX_4X4, TX_4X4, TX_4X4,
+    TX_8X8, TX_8X8, TX_8X8,
+    TX_16X16, TX_16X16, TX_16X16,
+    TX_32X32, TX_32X32, TX_32X32, TX_32X32};
+  return tx_size_lookup[bsize];
+}
+#endif
+
+#if CONFIG_MASKED_INTERINTER
+#define MASK_BITS_SML   3
+#define MASK_BITS_MED   4
+#define MASK_BITS_BIG   5
+#define MASK_NONE      -1
+
+static inline int get_mask_bits(BLOCK_SIZE sb_type) {
+  if (sb_type < BLOCK_8X8)
+    return 0;
+  if (sb_type <= BLOCK_8X8)
+    return MASK_BITS_SML;
+  else if (sb_type <= BLOCK_32X32)
+    return MASK_BITS_MED;
+  else
+    return MASK_BITS_BIG;
+}
+#endif
+
+#if CONFIG_INTERINTRA
+static INLINE TX_SIZE intra_size_log2_for_interintra(int bs) {
+  switch (bs) {
+    case 4:
+      return TX_4X4;
+      break;
+    case 8:
+      return TX_8X8;
+      break;
+    case 16:
+      return TX_16X16;
+      break;
+    case 32:
+      return TX_32X32;
+      break;
+    default:
+      return TX_32X32;
+      break;
+  }
+}
+
+static INLINE int is_interintra_allowed(BLOCK_SIZE sb_type) {
+  return ((sb_type >= BLOCK_8X8) && (sb_type < BLOCK_64X64));
+}
+
+#if CONFIG_MASKED_INTERINTRA
+#define MASK_BITS_SML_INTERINTRA   3
+#define MASK_BITS_MED_INTERINTRA   4
+#define MASK_BITS_BIG_INTERINTRA   5
+#define MASK_NONE_INTERINTRA      -1
+static INLINE int get_mask_bits_interintra(BLOCK_SIZE sb_type) {
+  if (sb_type == BLOCK_4X4)
+     return 0;
+  if (sb_type <= BLOCK_8X8)
+    return MASK_BITS_SML_INTERINTRA;
+  else if (sb_type <= BLOCK_32X32)
+    return MASK_BITS_MED_INTERINTRA;
+  else
+    return MASK_BITS_BIG_INTERINTRA;
+}
+#endif
+#endif
+
 // This structure now relates to 8x8 block regions.
 typedef struct {
   // Common for both INTER and INTRA blocks
   BLOCK_SIZE sb_type;
   PREDICTION_MODE mode;
+#if CONFIG_FILTERINTRA
+  int filterbit, uv_filterbit;
+#endif
   TX_SIZE tx_size;
   uint8_t skip;
   uint8_t segment_id;
@@ -137,10 +225,34 @@ typedef struct {
   int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
   uint8_t mode_context[MAX_REF_FRAMES];
   INTERP_FILTER interp_filter;
+
+#if CONFIG_EXT_TX
+  EXT_TX_TYPE ext_txfrm;
+#endif
+
+#if CONFIG_MASKED_INTERINTER
+  int use_masked_interinter;
+  int mask_index;
+#endif
+#if CONFIG_INTERINTRA
+  PREDICTION_MODE interintra_mode, interintra_uv_mode;
+#if CONFIG_MASKED_INTERINTRA
+  int interintra_mask_index;
+  int interintra_uv_mask_index;
+  int use_masked_interintra;
+#endif
+#endif
+#if CONFIG_COPY_CODING
+  COPY_MODE copy_mode;
+  int inter_ref_count;
+#endif
 } MB_MODE_INFO;
 
 typedef struct {
   MB_MODE_INFO mbmi;
+#if CONFIG_FILTERINTRA
+  int b_filter_info[4];
+#endif
   b_mode_info bmi[4];
 } MODE_INFO;
 
@@ -149,6 +261,16 @@ static INLINE PREDICTION_MODE get_y_mode(const MODE_INFO *mi, int block) {
                                       : mi->mbmi.mode;
 }
 
+#if CONFIG_FILTERINTRA
+static INLINE int is_filter_allowed(PREDICTION_MODE mode) {
+  return 1;
+}
+
+static INLINE int is_filter_enabled(TX_SIZE txsize) {
+  return (txsize <= TX_32X32);
+}
+#endif
+
 static INLINE int is_inter_block(const MB_MODE_INFO *mbmi) {
   return mbmi->ref_frame[0] > INTRA_FRAME;
 }
@@ -240,6 +362,13 @@ typedef struct macroblockd {
   PARTITION_CONTEXT left_seg_context[8];
 } MACROBLOCKD;
 
+#if CONFIG_SUPERTX
+static INLINE int supertx_enabled(const MB_MODE_INFO *mbmi) {
+  return mbmi->tx_size >
+         MIN(b_width_log2(mbmi->sb_type), b_height_log2(mbmi->sb_type));
+}
+#endif
+
 static INLINE BLOCK_SIZE get_subsize(BLOCK_SIZE bsize,
                                      PARTITION_TYPE partition) {
   const BLOCK_SIZE subsize = subsize_lookup[partition][bsize];
@@ -253,8 +382,20 @@ static INLINE TX_TYPE get_tx_type(PLANE_TYPE plane_type,
                                   const MACROBLOCKD *xd) {
   const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
 
+#if !CONFIG_EXT_TX
   if (plane_type != PLANE_TYPE_Y || is_inter_block(mbmi))
     return DCT_DCT;
+#else
+  if (plane_type != PLANE_TYPE_Y)
+      return DCT_DCT;
+
+  if (is_inter_block(mbmi)) {
+    if (mbmi->ext_txfrm == NORM || mbmi->tx_size >= TX_32X32)
+      return DCT_DCT;
+    else
+      return ADST_ADST;
+  }
+#endif
   return intra_mode_to_tx_type_lookup[mbmi->mode];
 }
 
@@ -262,8 +403,20 @@ static INLINE TX_TYPE get_tx_type_4x4(PLANE_TYPE plane_type,
                                       const MACROBLOCKD *xd, int ib) {
   const MODE_INFO *const mi = xd->mi[0];
 
+#if !CONFIG_EXT_TX
   if (plane_type != PLANE_TYPE_Y || xd->lossless || is_inter_block(&mi->mbmi))
     return DCT_DCT;
+#else
+  if (plane_type != PLANE_TYPE_Y || xd->lossless)
+      return DCT_DCT;
+
+  if (is_inter_block(&mi->mbmi)) {
+    if (mi->mbmi.ext_txfrm == NORM)
+      return DCT_DCT;
+    else
+      return ADST_ADST;
+  }
+#endif
 
   return intra_mode_to_tx_type_lookup[get_y_mode(mi, ib)];
 }
@@ -281,7 +434,15 @@ static INLINE TX_SIZE get_uv_tx_size_impl(TX_SIZE y_tx_size, BLOCK_SIZE bsize) {
 }
 
 static INLINE TX_SIZE get_uv_tx_size(const MB_MODE_INFO *mbmi) {
+#if CONFIG_SUPERTX
+  if (!supertx_enabled(mbmi)) {
+#endif
   return get_uv_tx_size_impl(mbmi->tx_size, mbmi->sb_type);
+#if CONFIG_SUPERTX
+  } else {
+    return uvsupertx_size_lookup[mbmi->tx_size];
+  }
+#endif
 }
 
 static INLINE BLOCK_SIZE get_plane_block_size(BLOCK_SIZE bsize,

vp9/common/vp9_common_data.c

@@ -133,6 +133,15 @@ const BLOCK_SIZE ss_size_lookup[BLOCK_SIZES][2][2] = {
   {{BLOCK_64X64, BLOCK_64X32},   {BLOCK_32X64,   BLOCK_32X32}},
 };
 
+#if CONFIG_SUPERTX
+const TX_SIZE uvsupertx_size_lookup[TX_SIZES] = {
+    TX_4X4,
+    TX_4X4,
+    TX_8X8,
+    TX_16X16
+};
+#endif
+
 // Generates 4 bit field in which each bit set to 1 represents
 // a blocksize partition  1111 means we split 64x64, 32x32, 16x16
 // and 8x8.  1000 means we just split the 64x64 to 32x32

vp9/common/vp9_common_data.h

@@ -31,6 +31,9 @@ extern const BLOCK_SIZE subsize_lookup[PARTITION_TYPES][BLOCK_SIZES];
 extern const TX_SIZE max_txsize_lookup[BLOCK_SIZES];
 extern const TX_SIZE tx_mode_to_biggest_tx_size[TX_MODES];
 extern const BLOCK_SIZE ss_size_lookup[BLOCK_SIZES][2][2];
+#if CONFIG_SUPERTX
+extern const TX_SIZE uvsupertx_size_lookup[TX_SIZES];
+#endif
 
 #ifdef __cplusplus
 }  // extern "C"

vp9/common/vp9_convolve.c

@@ -117,17 +117,25 @@ static void convolve(const uint8_t *src, ptrdiff_t src_stride,
                      const InterpKernel *const y_filters,
                      int y0_q4, int y_step_q4,
                      int w, int h) {
-  // Fixed size intermediate buffer places limits on parameters.
-  // Maximum intermediate_height is 324, for y_step_q4 == 80,
-  // h == 64, taps == 8.
-  // y_step_q4 of 80 allows for 1/10 scale for 5 layer svc
-  uint8_t temp[64 * 324];
+  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
+  // 2d filtering proceeds in 2 steps:
+  //   (1) Interpolate horizontally into an intermediate buffer, temp.
+  //   (2) Interpolate temp vertically to derive the sub-pixel result.
+  // Deriving the maximum number of rows in the temp buffer (135):
+  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
+  // --Largest block size is 64x64 pixels.
+  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
+  //   original frame (in 1/16th pixel units).
+  // --Must round-up because block may be located at sub-pixel position.
+  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
+  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
+  uint8_t temp[135 * 64];
   int intermediate_height = (((h - 1) * y_step_q4 + 15) >> 4) + SUBPEL_TAPS;
 
   assert(w <= 64);
   assert(h <= 64);
-  assert(y_step_q4 <= 80);
-  assert(x_step_q4 <= 80);
+  assert(y_step_q4 <= 32);
+  assert(x_step_q4 <= 32);
 
   if (intermediate_height < h)
     intermediate_height = h;

vp9/common/vp9_entropymode.c

@@ -13,6 +13,84 @@
 #include "vp9/common/vp9_onyxc_int.h"
 #include "vp9/common/vp9_seg_common.h"
 
+#if CONFIG_MASKED_INTERINTER
+static const vp9_prob default_masked_interinter_prob[BLOCK_SIZES] = {
+    192, 192, 192, 192, 192, 192, 192, 192, 192, 192, 192, 192, 192
+};
+#endif
+
+#if CONFIG_INTERINTRA
+static const vp9_prob default_interintra_prob[BLOCK_SIZES] = {
+  192, 192, 192, 192, 192, 192, 192, 192, 192, 192, 192, 192, 192
+};
+#if CONFIG_MASKED_INTERINTRA
+static const vp9_prob default_masked_interintra_prob[BLOCK_SIZES] = {
+  192, 192, 192, 192, 192, 192, 192, 192, 192, 192, 192, 192, 192
+};
+#endif
+#endif
+
+#if CONFIG_FILTERINTRA
+static const vp9_prob default_filterintra_prob[TX_SIZES][INTRA_MODES] = {
+  // DC     V      H    D45   D135   D117   D153   D207    D63     TM
+  {153,   171,   147,   150,   129,   101,   100,   153,   132,   111},
+  {171,   173,   185,   131,    70,    53,    70,   148,   127,   114},
+  {175,   203,   213,    86,    45,    71,    41,   150,   125,   154},
+  {235,   230,   154,   202,   154,   205,    37,   128,     0,   202}
+};
+#endif
+
+#if CONFIG_EXT_TX
+static const vp9_prob default_ext_tx_prob = 178;
+#endif
+
+#if CONFIG_SUPERTX
+static const vp9_prob default_supertx_prob[TX_SIZES] = {
+  255, 160, 160, 160
+};
+
+static const vp9_prob default_supertxsplit_prob[TX_SIZES] = {
+  255, 200, 200, 200
+};
+#endif
+
+#if CONFIG_COPY_CODING
+static const vp9_prob default_copy_noref_prob[COPY_MODE_CONTEXTS]
+                                             [BLOCK_SIZES] = {
+  {255, 255, 255,  82, 148, 182,  65, 193, 158,  70, 138, 101,  23},
+  {255, 255, 255, 118, 153, 161, 123, 169, 157,  82, 101, 123,  88},
+  {255, 255, 255, 130, 178, 226, 194, 196, 174, 173, 135, 144, 141},
+  {255, 255, 255, 178, 218, 225, 197, 230, 222, 215, 220, 220, 220},
+  {255, 255, 255, 243, 248, 241, 233, 249, 249, 249, 249, 249, 249}
+};
+
+static const vp9_prob default_copy_mode_probs_l2[COPY_MODE_CONTEXTS][1] = {
+  {207},
+  {135},
+  {141},
+  {189},
+  {209}
+};
+
+const vp9_tree_index vp9_copy_mode_tree_l2[TREE_SIZE(2)] = {
+  -(REF0 - REF0), -(REF1 - REF0)
+};
+
+static const vp9_prob default_copy_mode_probs[COPY_MODE_CONTEXTS]
+                                             [COPY_MODE_COUNT - 2] = {
+  {130, 159},
+  {126, 176},
+  {120, 150},
+  {158, 183},
+  {149, 125}
+};
+
+const vp9_tree_index vp9_copy_mode_tree[TREE_SIZE(COPY_MODE_COUNT - 1)] = {
+  -(REF0 - REF0),  2,
+  -(REF1 - REF0),  -(REF2 - REF0)
+};
+#endif
+
 const vp9_prob vp9_kf_y_mode_prob[INTRA_MODES][INTRA_MODES][INTRA_MODES - 1] = {
   {  // above = dc
     { 137,  30,  42, 148, 151, 207,  70,  52,  91 },  // left = dc
@@ -245,7 +323,11 @@ const vp9_tree_index vp9_partition_tree[TREE_SIZE(PARTITION_TYPES)] = {
 };
 
 static const vp9_prob default_intra_inter_p[INTRA_INTER_CONTEXTS] = {
+#if !CONFIG_COPY_CODING
   9, 102, 187, 225
+#else
+  35, 112, 187, 225
+#endif
 };
 
 static const vp9_prob default_comp_inter_p[COMP_INTER_CONTEXTS] = {
@@ -326,6 +408,30 @@ void vp9_init_mode_probs(FRAME_CONTEXT *fc) {
   fc->tx_probs = default_tx_probs;
   vp9_copy(fc->skip_probs, default_skip_probs);
   vp9_copy(fc->inter_mode_probs, default_inter_mode_probs);
+#if CONFIG_MASKED_INTERINTER
+  vp9_copy(fc->masked_interinter_prob, default_masked_interinter_prob);
+#endif
+#if CONFIG_INTERINTRA
+  vp9_copy(fc->interintra_prob, default_interintra_prob);
+#if CONFIG_MASKED_INTERINTRA
+  vp9_copy(fc->masked_interintra_prob, default_masked_interintra_prob);
+#endif
+#endif
+#if CONFIG_FILTERINTRA
+  vp9_copy(fc->filterintra_prob, default_filterintra_prob);
+#endif
+#if CONFIG_EXT_TX
+  fc->ext_tx_prob = default_ext_tx_prob;
+#endif
+#if CONFIG_SUPERTX
+  vp9_copy(fc->supertx_prob, default_supertx_prob);
+  vp9_copy(fc->supertxsplit_prob, default_supertxsplit_prob);
+#endif
+#if CONFIG_COPY_CODING
+  vp9_copy(fc->copy_noref_prob, default_copy_noref_prob);
+  vp9_copy(fc->copy_mode_probs_l2, default_copy_mode_probs_l2);
+  vp9_copy(fc->copy_mode_probs, default_copy_mode_probs);
+#endif
 }
 
 const vp9_tree_index vp9_switchable_interp_tree
@@ -416,6 +522,73 @@ void vp9_adapt_mode_probs(VP9_COMMON *cm) {
 
   for (i = 0; i < SKIP_CONTEXTS; ++i)
     fc->skip_probs[i] = adapt_prob(pre_fc->skip_probs[i], counts->skip[i]);
+
+#if CONFIG_MASKED_INTERINTER
+  if (cm->use_masked_interinter) {
+    for (i = 0; i < BLOCK_SIZES; ++i) {
+      if (get_mask_bits(i))
+        fc->masked_interinter_prob[i] = adapt_prob
+                                      (pre_fc->masked_interinter_prob[i],
+                                       counts->masked_interinter[i]);
+    }
+  }
+#endif
+
+#if CONFIG_INTERINTRA
+  if (cm->use_interintra) {
+    for (i = 0; i < BLOCK_SIZES; ++i) {
+      if (is_interintra_allowed(i))
+        fc->interintra_prob[i] = adapt_prob(pre_fc->interintra_prob[i],
+                                            counts->interintra[i]);
+    }
+#if CONFIG_MASKED_INTERINTRA
+    if (cm->use_masked_interintra) {
+      for (i = 0; i < BLOCK_SIZES; ++i) {
+        if (is_interintra_allowed(i) && get_mask_bits_interintra(i))
+          fc->masked_interintra_prob[i] = adapt_prob(
+                                          pre_fc->masked_interintra_prob[i],
+                                          counts->masked_interintra[i]);
+      }
+    }
+#endif
+  }
+#endif
+
+#if CONFIG_FILTERINTRA
+  for (i = 0; i < TX_SIZES; ++i)
+    for (j = 0; j < INTRA_MODES; ++j)
+      fc->filterintra_prob[i][j] = adapt_prob(pre_fc->filterintra_prob[i][j],
+                                   counts->filterintra[i][j]);
+#endif
+
+#if CONFIG_EXT_TX
+  fc->ext_tx_prob = adapt_prob(pre_fc->ext_tx_prob, counts->ext_tx);
+#endif
+
+#if CONFIG_SUPERTX
+  for (i = 1; i < TX_SIZES; ++i) {
+    fc->supertx_prob[i] = adapt_prob(pre_fc->supertx_prob[i],
+                                     counts->supertx[i]);
+  }
+
+  for (i = 1; i < TX_SIZES; ++i) {
+    fc->supertxsplit_prob[i] = adapt_prob(pre_fc->supertxsplit_prob[i],
+                                          counts->supertxsplit[i]);
+  }
+#endif
+
+#if CONFIG_COPY_CODING
+  for (i = 0; i < COPY_MODE_CONTEXTS; i++) {
+    for (j = BLOCK_8X8; j < BLOCK_SIZES; j++) {
+      fc->copy_noref_prob[i][j] =
+          adapt_prob(pre_fc->copy_noref_prob[i][j], counts->copy_noref[i][j]);
+    }
+    adapt_probs(vp9_copy_mode_tree_l2, pre_fc->copy_mode_probs_l2[i],
+                counts->copy_mode_l2[i], fc->copy_mode_probs_l2[i]);
+    adapt_probs(vp9_copy_mode_tree, pre_fc->copy_mode_probs[i],
+                counts->copy_mode[i], fc->copy_mode_probs[i]);
+  }
+#endif
 }
 
 static void set_default_lf_deltas(struct loopfilter *lf) {

vp9/common/vp9_entropymode.h

@@ -52,6 +52,30 @@ typedef struct frame_contexts {
   struct tx_probs tx_probs;
   vp9_prob skip_probs[SKIP_CONTEXTS];
   nmv_context nmvc;
+#if CONFIG_MASKED_INTERINTER
+  vp9_prob masked_interinter_prob[BLOCK_SIZES];
+#endif
+#if CONFIG_INTERINTRA
+  vp9_prob interintra_prob[BLOCK_SIZES];
+#if CONFIG_MASKED_INTERINTRA
+  vp9_prob masked_interintra_prob[BLOCK_SIZES];
+#endif
+#endif
+#if CONFIG_FILTERINTRA
+  vp9_prob filterintra_prob[TX_SIZES][INTRA_MODES];
+#endif
+#if CONFIG_EXT_TX
+  vp9_prob ext_tx_prob;
+#endif
+#if CONFIG_SUPERTX
+  vp9_prob supertx_prob[TX_SIZES];
+  vp9_prob supertxsplit_prob[TX_SIZES];
+#endif
+#if CONFIG_COPY_CODING
+  vp9_prob copy_noref_prob[COPY_MODE_CONTEXTS][BLOCK_SIZES];
+  vp9_prob copy_mode_probs_l2[COPY_MODE_CONTEXTS][1];
+  vp9_prob copy_mode_probs[COPY_MODE_CONTEXTS][COPY_MODE_COUNT - 2];
+#endif
 } FRAME_CONTEXT;
 
 typedef struct {
@@ -71,6 +95,31 @@ typedef struct {
   struct tx_counts tx;
   unsigned int skip[SKIP_CONTEXTS][2];
   nmv_context_counts mv;
+#if CONFIG_MASKED_INTERINTER
+  unsigned int masked_interinter[BLOCK_SIZES][2];
+#endif
+#if CONFIG_INTERINTRA
+  unsigned int interintra[BLOCK_SIZES][2];
+#if CONFIG_MASKED_INTERINTRA
+  unsigned int masked_interintra[BLOCK_SIZES][2];
+#endif
+#endif
+#if CONFIG_FILTERINTRA
+  unsigned int filterintra[TX_SIZES][INTRA_MODES][2];
+#endif
+#if CONFIG_EXT_TX
+  unsigned int ext_tx[2];
+#endif
+#if CONFIG_SUPERTX
+  unsigned int supertx[TX_SIZES][2];
+  unsigned int supertxsplit[TX_SIZES][2];
+  unsigned int supertx_size[BLOCK_SIZES];
+#endif
+#if CONFIG_COPY_CODING
+  unsigned int copy_noref[COPY_MODE_CONTEXTS][BLOCK_SIZES][2];
+  unsigned int copy_mode_l2[COPY_MODE_CONTEXTS][2];
+  unsigned int copy_mode[COPY_MODE_CONTEXTS][COPY_MODE_COUNT - 1];
+#endif
 } FRAME_COUNTS;
 
 extern const vp9_prob vp9_kf_uv_mode_prob[INTRA_MODES][INTRA_MODES - 1];
@@ -83,6 +132,10 @@ extern const vp9_tree_index vp9_inter_mode_tree[TREE_SIZE(INTER_MODES)];
 extern const vp9_tree_index vp9_partition_tree[TREE_SIZE(PARTITION_TYPES)];
 extern const vp9_tree_index vp9_switchable_interp_tree
                                 [TREE_SIZE(SWITCHABLE_FILTERS)];
+#if CONFIG_COPY_CODING
+extern const vp9_tree_index vp9_copy_mode_tree_l2[TREE_SIZE(2)];
+extern const vp9_tree_index vp9_copy_mode_tree[TREE_SIZE(COPY_MODE_COUNT - 1)];
+#endif
 
 void vp9_setup_past_independence(struct VP9Common *cm);
 

vp9/common/vp9_enums.h

@@ -100,6 +100,14 @@ typedef enum {
   TX_TYPES = 4
 } TX_TYPE;
 
+#if CONFIG_EXT_TX
+typedef enum {
+  NORM   = 0,
+  ALT = 1,
+  EXT_TX_TYPES = 2
+} EXT_TX_TYPE;
+#endif
+
 typedef enum {
   UNKNOWN    = 0,
   BT_601     = 1,  // YUV

vp9/common/vp9_loopfilter.c

@@ -206,6 +206,13 @@ static const int mode_lf_lut[MB_MODE_COUNT] = {
   1, 1, 0, 1                     // INTER_MODES (ZEROMV == 0)
 };
 
+#if CONFIG_SUPERTX
+static int supertx_enabled_lpf(const MB_MODE_INFO *mbmi) {
+  return mbmi->tx_size >
+         MIN(b_width_log2(mbmi->sb_type), b_height_log2(mbmi->sb_type));
+}
+#endif
+
 static void update_sharpness(loop_filter_info_n *lfi, int sharpness_lvl) {
   int lvl;
 
@@ -572,6 +579,85 @@ static void build_masks(const loop_filter_info_n *const lfi_n,
     *int_4x4_uv |= (size_mask_uv[block_size] & 0xffff) << shift_uv;
 }
 
+#if CONFIG_SUPERTX
+static void build_masks_supertx(const loop_filter_info_n *const lfi_n,
+                                const MODE_INFO *mi, const int shift_y,
+                                const int shift_uv,
+                                LOOP_FILTER_MASK *lfm) {
+  const MB_MODE_INFO *mbmi = &mi->mbmi;
+  const TX_SIZE tx_size_y = mbmi->tx_size;
+  const TX_SIZE tx_size_uv = get_uv_tx_size(mbmi);
+  const BLOCK_SIZE block_size = 3 * (int)tx_size_y;
+  const int filter_level = get_filter_level(lfi_n, mbmi);
+  uint64_t *const left_y = &lfm->left_y[tx_size_y];
+  uint64_t *const above_y = &lfm->above_y[tx_size_y];
+  uint64_t *const int_4x4_y = &lfm->int_4x4_y;
+  uint16_t *const left_uv = &lfm->left_uv[tx_size_uv];
+  uint16_t *const above_uv = &lfm->above_uv[tx_size_uv];
+  uint16_t *const int_4x4_uv = &lfm->int_4x4_uv;
+  int i;
+
+  // If filter level is 0 we don't loop filter.
+  if (!filter_level) {
+    return;
+  } else {
+    const int w = num_8x8_blocks_wide_lookup[block_size];
+    const int h = num_8x8_blocks_high_lookup[block_size];
+    int index = shift_y;
+    for (i = 0; i < h; i++) {
+      vpx_memset(&lfm->lfl_y[index], filter_level, w);
+      index += 8;
+    }
+  }
+
+  // These set 1 in the current block size for the block size edges.
+  // For instance if the block size is 32x16,   we'll set :
+  //    above =   1111
+  //              0000
+  //    and
+  //    left  =   1000
+  //          =   1000
+  // NOTE : In this example the low bit is left most ( 1000 ) is stored as
+  //        1,  not 8...
+  //
+  // U and v set things on a 16 bit scale.
+  //
+  *above_y |= above_prediction_mask[block_size] << shift_y;
+  *above_uv |= above_prediction_mask_uv[block_size] << shift_uv;
+  *left_y |= left_prediction_mask[block_size] << shift_y;
+  *left_uv |= left_prediction_mask_uv[block_size] << shift_uv;
+
+  // If the block has no coefficients and is not intra we skip applying
+  // the loop filter on block edges.
+  if (mbmi->skip && is_inter_block(mbmi))
+    return;
+
+  // Here we are adding a mask for the transform size.  The transform
+  // size mask is set to be correct for a 64x64 prediction block size. We
+  // mask to match the size of the block we are working on and then shift it
+  // into place..
+  *above_y |= (size_mask[block_size] &
+               above_64x64_txform_mask[tx_size_y]) << shift_y;
+  *above_uv |= (size_mask_uv[block_size] &
+                above_64x64_txform_mask_uv[tx_size_uv]) << shift_uv;
+
+  *left_y |= (size_mask[block_size] &
+              left_64x64_txform_mask[tx_size_y]) << shift_y;
+  *left_uv |= (size_mask_uv[block_size] &
+               left_64x64_txform_mask_uv[tx_size_uv]) << shift_uv;
+
+  // Here we are trying to determine what to do with the internal 4x4 block
+  // boundaries.  These differ from the 4x4 boundaries on the outside edge of
+  // an 8x8 in that the internal ones can be skipped and don't depend on
+  // the prediction block size.
+  if (tx_size_y == TX_4X4)
+    *int_4x4_y |= (size_mask[block_size] & 0xffffffffffffffff) << shift_y;
+
+  if (tx_size_uv == TX_4X4)
+    *int_4x4_uv |= (size_mask_uv[block_size] & 0xffff) << shift_uv;
+}
+#endif
+
 // This function does the same thing as the one above with the exception that
 // it only affects the y masks.   It exists because for blocks < 16x16 in size,
 // we only update u and v masks on the first block.
@@ -615,6 +701,48 @@ static void build_y_mask(const loop_filter_info_n *const lfi_n,
     *int_4x4_y |= (size_mask[block_size] & 0xffffffffffffffff) << shift_y;
 }
 
+#if CONFIG_SUPERTX
+static void build_y_mask_supertx(const loop_filter_info_n *const lfi_n,
+                                 const MODE_INFO *mi, const int shift_y,
+                                 LOOP_FILTER_MASK *lfm) {
+  const MB_MODE_INFO *mbmi = &mi->mbmi;
+  const TX_SIZE tx_size_y = mbmi->tx_size;
+  const BLOCK_SIZE block_size = 3 * (int)tx_size_y;
+  const int filter_level = get_filter_level(lfi_n, mbmi);
+  uint64_t *const left_y = &lfm->left_y[tx_size_y];
+  uint64_t *const above_y = &lfm->above_y[tx_size_y];
+  uint64_t *const int_4x4_y = &lfm->int_4x4_y;
+  int i;
+
+  if (!filter_level) {
+    return;
+  } else {
+    const int w = num_8x8_blocks_wide_lookup[block_size];
+    const int h = num_8x8_blocks_high_lookup[block_size];
+    int index = shift_y;
+    for (i = 0; i < h; i++) {
+      vpx_memset(&lfm->lfl_y[index], filter_level, w);
+      index += 8;
+    }
+  }
+
+  *above_y |= above_prediction_mask[block_size] << shift_y;
+  *left_y |= left_prediction_mask[block_size] << shift_y;
+
+  if (mbmi->skip && is_inter_block(mbmi))
+    return;
+
+  *above_y |= (size_mask[block_size] &
+               above_64x64_txform_mask[tx_size_y]) << shift_y;
+
+  *left_y |= (size_mask[block_size] &
+              left_64x64_txform_mask[tx_size_y]) << shift_y;
+
+  if (tx_size_y == TX_4X4)
+    *int_4x4_y |= (size_mask[block_size] & 0xffffffffffffffff) << shift_y;
+}
+#endif
+
 // This function sets up the bit masks for the entire 64x64 region represented
 // by mi_row, mi_col.
 // TODO(JBB): This function only works for yv12.
@@ -650,6 +778,9 @@ void vp9_setup_mask(VP9_COMMON *const cm, const int mi_row, const int mi_col,
                         cm->mi_rows - mi_row : MI_BLOCK_SIZE);
   const int max_cols = (mi_col + MI_BLOCK_SIZE > cm->mi_cols ?
                         cm->mi_cols - mi_col : MI_BLOCK_SIZE);
+#if CONFIG_SUPERTX
+  int supertx;
+#endif
 
   vp9_zero(*lfm);
 
@@ -687,20 +818,43 @@ void vp9_setup_mask(VP9_COMMON *const cm, const int mi_row, const int mi_col,
             build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
             break;
           case BLOCK_32X16:
+#if CONFIG_SUPERTX
+            supertx = supertx_enabled_lpf(&mip[0]->mbmi);
+            if (!supertx) {
+#endif
             build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
             if (mi_32_row_offset + 2 >= max_rows)
               continue;
             mip2 = mip + mode_info_stride * 2;
             build_masks(lfi_n, mip2[0], shift_y + 16, shift_uv + 4, lfm);
+#if CONFIG_SUPERTX
+            } else {
+              build_masks_supertx(lfi_n, mip[0], shift_y, shift_uv, lfm);
+            }
+#endif
             break;
           case BLOCK_16X32:
+#if CONFIG_SUPERTX
+            supertx = supertx_enabled_lpf(&mip[0]->mbmi);
+            if (!supertx) {
+#endif
             build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
             if (mi_32_col_offset + 2 >= max_cols)
               continue;
             mip2 = mip + 2;
             build_masks(lfi_n, mip2[0], shift_y + 2, shift_uv + 1, lfm);
+#if CONFIG_SUPERTX
+            } else {
+              build_masks_supertx(lfi_n, mip[0], shift_y, shift_uv, lfm);
+            }
+#endif
             break;
           default:
+#if CONFIG_SUPERTX
+            if (mip[0]->mbmi.tx_size == TX_32X32) {
+              build_masks_supertx(lfi_n, mip[0], shift_y, shift_uv, lfm);
+            } else {
+#endif
             for (idx_16 = 0; idx_16 < 4; mip += offset_16[idx_16], ++idx_16) {
               const int shift_y = shift_32_y[idx_32] + shift_16_y[idx_16];
               const int shift_uv = shift_32_uv[idx_32] + shift_16_uv[idx_16];
@@ -717,24 +871,56 @@ void vp9_setup_mask(VP9_COMMON *const cm, const int mi_row, const int mi_col,
                   build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
                   break;
                 case BLOCK_16X8:
+#if CONFIG_SUPERTX
+                  supertx = supertx_enabled_lpf(&mip[0]->mbmi);
+                  if (!supertx) {
+#endif
                   build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
                   if (mi_16_row_offset + 1 >= max_rows)
                     continue;
                   mip2 = mip + mode_info_stride;
                   build_y_mask(lfi_n, mip2[0], shift_y+8, lfm);
+#if CONFIG_SUPERTX
+                  } else {
+                    build_masks_supertx(lfi_n, mip[0], shift_y, shift_uv, lfm);
+                  }
+#endif
                   break;
                 case BLOCK_8X16:
+#if CONFIG_SUPERTX
+                  supertx = supertx_enabled_lpf(&mip[0]->mbmi);
+                  if (!supertx) {
+#endif
                   build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
                   if (mi_16_col_offset +1 >= max_cols)
                     continue;
                   mip2 = mip + 1;
                   build_y_mask(lfi_n, mip2[0], shift_y+1, lfm);
+#if CONFIG_SUPERTX
+                  } else {
+                    build_masks_supertx(lfi_n, mip[0], shift_y, shift_uv, lfm);
+                  }
+#endif
                   break;
                 default: {
+#if CONFIG_SUPERTX
+                  if (mip[0]->mbmi.tx_size == TX_16X16) {
+                    build_masks_supertx(lfi_n, mip[0], shift_y, shift_uv, lfm);
+                  } else {
+#endif
                   const int shift_y = shift_32_y[idx_32] +
                                       shift_16_y[idx_16] +
                                       shift_8_y[0];
+#if CONFIG_SUPERTX
+                  supertx = supertx_enabled_lpf(&mip[0]->mbmi);
+                  if (!supertx) {
+#endif
                   build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
+#if CONFIG_SUPERTX
+                  } else {
+                    build_masks_supertx(lfi_n, mip[0], shift_y, shift_uv, lfm);
+                  }
+#endif
                   mip += offset[0];
                   for (idx_8 = 1; idx_8 < 4; mip += offset[idx_8], ++idx_8) {
                     const int shift_y = shift_32_y[idx_32] +
@@ -748,12 +934,26 @@ void vp9_setup_mask(VP9_COMMON *const cm, const int mi_row, const int mi_col,
                     if (mi_8_col_offset >= max_cols ||
                         mi_8_row_offset >= max_rows)
                       continue;
+#if CONFIG_SUPERTX
+                    supertx = supertx_enabled_lpf(&mip[0]->mbmi);
+                    if (!supertx)
+#endif
                     build_y_mask(lfi_n, mip[0], shift_y, lfm);
+#if CONFIG_SUPERTX
+                    else
+                      build_y_mask_supertx(lfi_n, mip[0], shift_y, lfm);
+#endif
                   }
+#if CONFIG_SUPERTX
+                  }
+#endif
                   break;
                 }
               }
             }
+#if CONFIG_SUPERTX
+            }
+#endif
             break;
         }
       }

vp9/common/vp9_mvref_common.c

@@ -11,181 +11,6 @@
 
 #include "vp9/common/vp9_mvref_common.h"
 
-#define MVREF_NEIGHBOURS 8
-
-typedef struct position {
-  int row;
-  int col;
-} POSITION;
-
-typedef enum {
-  BOTH_ZERO = 0,
-  ZERO_PLUS_PREDICTED = 1,
-  BOTH_PREDICTED = 2,
-  NEW_PLUS_NON_INTRA = 3,
-  BOTH_NEW = 4,
-  INTRA_PLUS_NON_INTRA = 5,
-  BOTH_INTRA = 6,
-  INVALID_CASE = 9
-} motion_vector_context;
-
-// This is used to figure out a context for the ref blocks. The code flattens
-// an array that would have 3 possible counts (0, 1 & 2) for 3 choices by
-// adding 9 for each intra block, 3 for each zero mv and 1 for each new
-// motion vector. This single number is then converted into a context
-// with a single lookup ( counter_to_context ).
-static const int mode_2_counter[MB_MODE_COUNT] = {
-  9,  // DC_PRED
-  9,  // V_PRED
-  9,  // H_PRED
-  9,  // D45_PRED
-  9,  // D135_PRED
-  9,  // D117_PRED
-  9,  // D153_PRED
-  9,  // D207_PRED
-  9,  // D63_PRED
-  9,  // TM_PRED
-  0,  // NEARESTMV
-  0,  // NEARMV
-  3,  // ZEROMV
-  1,  // NEWMV
-};
-
-// There are 3^3 different combinations of 3 counts that can be either 0,1 or
-// 2. However the actual count can never be greater than 2 so the highest
-// counter we need is 18. 9 is an invalid counter that's never used.
-static const int counter_to_context[19] = {
-  BOTH_PREDICTED,  // 0
-  NEW_PLUS_NON_INTRA,  // 1
-  BOTH_NEW,  // 2
-  ZERO_PLUS_PREDICTED,  // 3
-  NEW_PLUS_NON_INTRA,  // 4
-  INVALID_CASE,  // 5
-  BOTH_ZERO,  // 6
-  INVALID_CASE,  // 7
-  INVALID_CASE,  // 8
-  INTRA_PLUS_NON_INTRA,  // 9
-  INTRA_PLUS_NON_INTRA,  // 10
-  INVALID_CASE,  // 11
-  INTRA_PLUS_NON_INTRA,  // 12
-  INVALID_CASE,  // 13
-  INVALID_CASE,  // 14
-  INVALID_CASE,  // 15
-  INVALID_CASE,  // 16
-  INVALID_CASE,  // 17
-  BOTH_INTRA  // 18
-};
-
-static const POSITION mv_ref_blocks[BLOCK_SIZES][MVREF_NEIGHBOURS] = {
-  // 4X4
-  {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
-  // 4X8
-  {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
-  // 8X4
-  {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
-  // 8X8
-  {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
-  // 8X16
-  {{0, -1}, {-1, 0}, {1, -1}, {-1, -1}, {0, -2}, {-2, 0}, {-2, -1}, {-1, -2}},
-  // 16X8
-  {{-1, 0}, {0, -1}, {-1, 1}, {-1, -1}, {-2, 0}, {0, -2}, {-1, -2}, {-2, -1}},
-  // 16X16
-  {{-1, 0}, {0, -1}, {-1, 1}, {1, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-3, -3}},
-  // 16X32
-  {{0, -1}, {-1, 0}, {2, -1}, {-1, -1}, {-1, 1}, {0, -3}, {-3, 0}, {-3, -3}},
-  // 32X16
-  {{-1, 0}, {0, -1}, {-1, 2}, {-1, -1}, {1, -1}, {-3, 0}, {0, -3}, {-3, -3}},
-  // 32X32
-  {{-1, 1}, {1, -1}, {-1, 2}, {2, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-3, -3}},
-  // 32X64
-  {{0, -1}, {-1, 0}, {4, -1}, {-1, 2}, {-1, -1}, {0, -3}, {-3, 0}, {2, -1}},
-  // 64X32
-  {{-1, 0}, {0, -1}, {-1, 4}, {2, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-1, 2}},
-  // 64X64
-  {{-1, 3}, {3, -1}, {-1, 4}, {4, -1}, {-1, -1}, {-1, 0}, {0, -1}, {-1, 6}}
-};
-
-static const int idx_n_column_to_subblock[4][2] = {
-  {1, 2},
-  {1, 3},
-  {3, 2},
-  {3, 3}
-};
-
-// clamp_mv_ref
-#define MV_BORDER (16 << 3)  // Allow 16 pels in 1/8th pel units
-
-static void clamp_mv_ref(MV *mv, const MACROBLOCKD *xd) {
-  clamp_mv(mv, xd->mb_to_left_edge - MV_BORDER,
-               xd->mb_to_right_edge + MV_BORDER,
-               xd->mb_to_top_edge - MV_BORDER,
-               xd->mb_to_bottom_edge + MV_BORDER);
-}
-
-// This function returns either the appropriate sub block or block's mv
-// on whether the block_size < 8x8 and we have check_sub_blocks set.
-static INLINE int_mv get_sub_block_mv(const MODE_INFO *candidate, int which_mv,
-                                      int search_col, int block_idx) {
-  return block_idx >= 0 && candidate->mbmi.sb_type < BLOCK_8X8
-          ? candidate->bmi[idx_n_column_to_subblock[block_idx][search_col == 0]]
-              .as_mv[which_mv]
-          : candidate->mbmi.mv[which_mv];
-}
-
-
-// Performs mv sign inversion if indicated by the reference frame combination.
-static INLINE int_mv scale_mv(const MB_MODE_INFO *mbmi, int ref,
-                              const MV_REFERENCE_FRAME this_ref_frame,
-                              const int *ref_sign_bias) {
-  int_mv mv = mbmi->mv[ref];
-  if (ref_sign_bias[mbmi->ref_frame[ref]] != ref_sign_bias[this_ref_frame]) {
-    mv.as_mv.row *= -1;
-    mv.as_mv.col *= -1;
-  }
-  return mv;
-}
-
-// This macro is used to add a motion vector mv_ref list if it isn't
-// already in the list.  If it's the second motion vector it will also
-// skip all additional processing and jump to done!
-#define ADD_MV_REF_LIST(mv) \
-  do { \
-    if (refmv_count) { \
-      if ((mv).as_int != mv_ref_list[0].as_int) { \
-        mv_ref_list[refmv_count] = (mv); \
-        goto Done; \
-      } \
-    } else { \
-      mv_ref_list[refmv_count++] = (mv); \
-    } \
-  } while (0)
-
-// If either reference frame is different, not INTRA, and they
-// are different from each other scale and add the mv to our list.
-#define IF_DIFF_REF_FRAME_ADD_MV(mbmi) \
-  do { \
-    if (is_inter_block(mbmi)) { \
-      if ((mbmi)->ref_frame[0] != ref_frame) \
-        ADD_MV_REF_LIST(scale_mv((mbmi), 0, ref_frame, ref_sign_bias)); \
-      if (has_second_ref(mbmi) && \
-          (mbmi)->ref_frame[1] != ref_frame && \
-          (mbmi)->mv[1].as_int != (mbmi)->mv[0].as_int) \
-        ADD_MV_REF_LIST(scale_mv((mbmi), 1, ref_frame, ref_sign_bias)); \
-    } \
-  } while (0)
-
-
-// Checks that the given mi_row, mi_col and search point
-// are inside the borders of the tile.
-static INLINE int is_inside(const TileInfo *const tile,
-                            int mi_col, int mi_row, int mi_rows,
-                            const POSITION *mi_pos) {
-  return !(mi_row + mi_pos->row < 0 ||
-           mi_col + mi_pos->col < tile->mi_col_start ||
-           mi_row + mi_pos->row >= mi_rows ||
-           mi_col + mi_pos->col >= tile->mi_col_end);
-}
-
 // This function searches the neighbourhood of a given MB/SB
 // to try and find candidate reference vectors.
 static void find_mv_refs_idx(const VP9_COMMON *cm, const MACROBLOCKD *xd,
@@ -363,3 +188,176 @@ void vp9_append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd,
       assert("Invalid block index.");
   }
 }
+
+#if CONFIG_COPY_CODING
+static int compare_interinfo(MB_MODE_INFO *mbmi, MB_MODE_INFO *ref_mbmi) {
+  if (mbmi == ref_mbmi) {
+    return 1;
+  } else {
+    int is_same;
+#if CONFIG_INTERINTRA
+    MV_REFERENCE_FRAME mbmi_ref1_backup = mbmi->ref_frame[1];
+    MV_REFERENCE_FRAME refmbmi_ref1_backup = ref_mbmi->ref_frame[1];
+
+    if (mbmi->ref_frame[1] == INTRA_FRAME)
+      mbmi->ref_frame[1] = NONE;
+    if (ref_mbmi->ref_frame[1] == INTRA_FRAME)
+      ref_mbmi->ref_frame[1] = NONE;
+#endif
+    if (mbmi->ref_frame[0] == ref_mbmi->ref_frame[0] &&
+        mbmi->ref_frame[1] == ref_mbmi->ref_frame[1]) {
+      if (mbmi->ref_frame[1] > INTRA_FRAME)
+        is_same = mbmi->mv[0].as_int == ref_mbmi->mv[0].as_int &&
+                  mbmi->mv[1].as_int == ref_mbmi->mv[1].as_int &&
+                  mbmi->interp_filter == ref_mbmi->interp_filter;
+      else
+        is_same = mbmi->mv[0].as_int == ref_mbmi->mv[0].as_int &&
+                  mbmi->interp_filter == ref_mbmi->interp_filter;
+    } else {
+      is_same = 0;
+    }
+#if CONFIG_INTERINTRA
+    mbmi->ref_frame[1] = mbmi_ref1_backup;
+    ref_mbmi->ref_frame[1] = refmbmi_ref1_backup;
+#endif
+
+    return is_same;
+  }
+}
+
+static int check_inside(VP9_COMMON *cm, int mi_row, int mi_col) {
+  return mi_row >= 0 && mi_col >= 0 &&
+         mi_row < cm->mi_rows && mi_col < cm->mi_cols;
+}
+
+static int is_right_available(BLOCK_SIZE bsize, int mi_row, int mi_col) {
+  int depth, max_depth = 4 - MIN(b_width_log2(bsize), b_height_log2(bsize));
+  int block[4] = {0};
+
+  if (bsize == BLOCK_64X64)
+    return 1;
+  mi_row = mi_row % 8;
+  mi_col = mi_col % 8;
+  for (depth = 1; depth <= max_depth; depth++) {
+    block[depth] = (mi_row >> (3 - depth)) * 2 + (mi_col >> (3 - depth));
+    mi_row = mi_row % (8 >> depth);
+    mi_col = mi_col % (8 >> depth);
+  }
+
+  if (b_width_log2(bsize) < b_height_log2(bsize)) {
+    if (block[max_depth] == 0)
+      return 1;
+  } else if (b_width_log2(bsize) > b_height_log2(bsize)) {
+    if (block[max_depth] > 0)
+      return 0;
+  } else {
+    if (block[max_depth] == 0 || block[max_depth] == 2)
+      return 1;
+    else if (block[max_depth] == 3)
+      return 0;
+  }
+
+  for (depth = max_depth - 1; depth > 0; depth--) {
+    if (block[depth] == 0 || block[depth] == 2)
+      return 1;
+    else if (block[depth] == 3)
+      return 0;
+  }
+  return 1;
+}
+
+static int is_second_rec(int mi_row, int mi_col, BLOCK_SIZE bsize) {
+  int bw = 4 << b_width_log2(bsize);
+  int bh = 4 << b_height_log2(bsize);
+
+  if (bw < bh)
+    return (mi_col << 3) % (bw << 1) == 0 ? 0 : 1;
+  else if (bh < bw)
+    return (mi_row << 3) % (bh << 1) == 0 ? 0 : 2;
+  else
+    return 0;
+}
+
+int vp9_construct_ref_inter_list(VP9_COMMON *cm,  MACROBLOCKD *xd,
+                                 BLOCK_SIZE bsize, int mi_row, int mi_col,
+                                 MB_MODE_INFO *ref_list[18]) {
+  int bw = 4 << b_width_log2(bsize);
+  int bh = 4 << b_height_log2(bsize);
+  int row_offset, col_offset;
+  int mi_offset;
+  MB_MODE_INFO *ref_mbmi;
+  int ref_index, ref_num = 0;
+  int row_offset_cand[18], col_offset_cand[18];
+  int offset_num = 0, i, switchflag;
+  int is_sec_rec = is_second_rec(mi_row, mi_col, bsize);
+
+  if (is_sec_rec != 2) {
+    row_offset_cand[offset_num] = -1; col_offset_cand[offset_num] = 0;
+    offset_num++;
+  }
+  if (is_sec_rec != 1) {
+    row_offset_cand[offset_num] = bh / 16; col_offset_cand[offset_num] = -1;
+    offset_num++;
+  }
+
+  row_offset = bh / 8 - 1;
+  col_offset = 1;
+  if (is_sec_rec < 2)
+    switchflag = 1;
+  else
+    switchflag = 0;
+  while ((is_sec_rec == 0 && ((row_offset >=0) || col_offset < (bw / 8 + 1))) ||
+         (is_sec_rec == 1 && col_offset < (bw / 8 + 1)) ||
+         (is_sec_rec == 2 && row_offset >=0)) {
+    switch (switchflag) {
+      case 0:
+        if (row_offset >= 0) {
+          if (row_offset != bh / 16) {
+            row_offset_cand[offset_num] = row_offset;
+            col_offset_cand[offset_num] = -1;
+            offset_num++;
+          }
+          row_offset--;
+        }
+        break;
+      case 1:
+        if (col_offset < (bw / 8 + 1)) {
+          row_offset_cand[offset_num] = -1;
+          col_offset_cand[offset_num] = col_offset;
+          offset_num++;
+          col_offset++;
+        }
+        break;
+      default:
+        assert(0);
+    }
+    if (is_sec_rec == 0)
+      switchflag = 1 - switchflag;
+  }
+  row_offset_cand[offset_num] = -1;
+  col_offset_cand[offset_num] = -1;
+  offset_num++;
+
+  for (i = 0; i < offset_num; i++) {
+    row_offset = row_offset_cand[i];
+    col_offset = col_offset_cand[i];
+    if ((col_offset < (bw / 8) ||
+        (col_offset == (bw / 8) && is_right_available(bsize, mi_row, mi_col)))
+        && check_inside(cm, mi_row + row_offset, mi_col + col_offset)) {
+      mi_offset = row_offset * cm->mi_stride + col_offset;
+      ref_mbmi = &xd->mi[mi_offset]->mbmi;
+      if (is_inter_block(ref_mbmi)) {
+        for (ref_index = 0; ref_index < ref_num; ref_index++) {
+          if (compare_interinfo(ref_mbmi, ref_list[ref_index]))
+            break;
+        }
+        if (ref_index == ref_num) {
+          ref_list[ref_num] = ref_mbmi;
+          ref_num++;
+        }
+      }
+    }
+  }
+  return ref_num;
+}
+#endif

vp9/common/vp9_mvref_common.h

@@ -21,6 +21,181 @@ extern "C" {
 #define RIGHT_BOTTOM_MARGIN ((VP9_ENC_BORDER_IN_PIXELS -\
                                 VP9_INTERP_EXTEND) << 3)
 
+#define MVREF_NEIGHBOURS 8
+
+typedef struct position {
+  int row;
+  int col;
+} POSITION;
+
+typedef enum {
+  BOTH_ZERO = 0,
+  ZERO_PLUS_PREDICTED = 1,
+  BOTH_PREDICTED = 2,
+  NEW_PLUS_NON_INTRA = 3,
+  BOTH_NEW = 4,
+  INTRA_PLUS_NON_INTRA = 5,
+  BOTH_INTRA = 6,
+  INVALID_CASE = 9
+} motion_vector_context;
+
+// This is used to figure out a context for the ref blocks. The code flattens
+// an array that would have 3 possible counts (0, 1 & 2) for 3 choices by
+// adding 9 for each intra block, 3 for each zero mv and 1 for each new
+// motion vector. This single number is then converted into a context
+// with a single lookup ( counter_to_context ).
+static const int mode_2_counter[MB_MODE_COUNT] = {
+  9,  // DC_PRED
+  9,  // V_PRED
+  9,  // H_PRED
+  9,  // D45_PRED
+  9,  // D135_PRED
+  9,  // D117_PRED
+  9,  // D153_PRED
+  9,  // D207_PRED
+  9,  // D63_PRED
+  9,  // TM_PRED
+  0,  // NEARESTMV
+  0,  // NEARMV
+  3,  // ZEROMV
+  1,  // NEWMV
+};
+
+// There are 3^3 different combinations of 3 counts that can be either 0,1 or
+// 2. However the actual count can never be greater than 2 so the highest
+// counter we need is 18. 9 is an invalid counter that's never used.
+static const int counter_to_context[19] = {
+  BOTH_PREDICTED,  // 0
+  NEW_PLUS_NON_INTRA,  // 1
+  BOTH_NEW,  // 2
+  ZERO_PLUS_PREDICTED,  // 3
+  NEW_PLUS_NON_INTRA,  // 4
+  INVALID_CASE,  // 5
+  BOTH_ZERO,  // 6
+  INVALID_CASE,  // 7
+  INVALID_CASE,  // 8
+  INTRA_PLUS_NON_INTRA,  // 9
+  INTRA_PLUS_NON_INTRA,  // 10
+  INVALID_CASE,  // 11
+  INTRA_PLUS_NON_INTRA,  // 12
+  INVALID_CASE,  // 13
+  INVALID_CASE,  // 14
+  INVALID_CASE,  // 15
+  INVALID_CASE,  // 16
+  INVALID_CASE,  // 17
+  BOTH_INTRA  // 18
+};
+
+static const POSITION mv_ref_blocks[BLOCK_SIZES][MVREF_NEIGHBOURS] = {
+  // 4X4
+  {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
+  // 4X8
+  {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
+  // 8X4
+  {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
+  // 8X8
+  {{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
+  // 8X16
+  {{0, -1}, {-1, 0}, {1, -1}, {-1, -1}, {0, -2}, {-2, 0}, {-2, -1}, {-1, -2}},
+  // 16X8
+  {{-1, 0}, {0, -1}, {-1, 1}, {-1, -1}, {-2, 0}, {0, -2}, {-1, -2}, {-2, -1}},
+  // 16X16
+  {{-1, 0}, {0, -1}, {-1, 1}, {1, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-3, -3}},
+  // 16X32
+  {{0, -1}, {-1, 0}, {2, -1}, {-1, -1}, {-1, 1}, {0, -3}, {-3, 0}, {-3, -3}},
+  // 32X16
+  {{-1, 0}, {0, -1}, {-1, 2}, {-1, -1}, {1, -1}, {-3, 0}, {0, -3}, {-3, -3}},
+  // 32X32
+  {{-1, 1}, {1, -1}, {-1, 2}, {2, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-3, -3}},
+  // 32X64
+  {{0, -1}, {-1, 0}, {4, -1}, {-1, 2}, {-1, -1}, {0, -3}, {-3, 0}, {2, -1}},
+  // 64X32
+  {{-1, 0}, {0, -1}, {-1, 4}, {2, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-1, 2}},
+  // 64X64
+  {{-1, 3}, {3, -1}, {-1, 4}, {4, -1}, {-1, -1}, {-1, 0}, {0, -1}, {-1, 6}}
+};
+
+static const int idx_n_column_to_subblock[4][2] = {
+  {1, 2},
+  {1, 3},
+  {3, 2},
+  {3, 3}
+};
+
+// clamp_mv_ref
+#define MV_BORDER (16 << 3)  // Allow 16 pels in 1/8th pel units
+
+static void clamp_mv_ref(MV *mv, const MACROBLOCKD *xd) {
+  clamp_mv(mv, xd->mb_to_left_edge - MV_BORDER,
+               xd->mb_to_right_edge + MV_BORDER,
+               xd->mb_to_top_edge - MV_BORDER,
+               xd->mb_to_bottom_edge + MV_BORDER);
+}
+
+// This function returns either the appropriate sub block or block's mv
+// on whether the block_size < 8x8 and we have check_sub_blocks set.
+static INLINE int_mv get_sub_block_mv(const MODE_INFO *candidate, int which_mv,
+                                      int search_col, int block_idx) {
+  return block_idx >= 0 && candidate->mbmi.sb_type < BLOCK_8X8
+          ? candidate->bmi[idx_n_column_to_subblock[block_idx][search_col == 0]]
+              .as_mv[which_mv]
+          : candidate->mbmi.mv[which_mv];
+}
+
+
+// Performs mv sign inversion if indicated by the reference frame combination.
+static INLINE int_mv scale_mv(const MB_MODE_INFO *mbmi, int ref,
+                              const MV_REFERENCE_FRAME this_ref_frame,
+                              const int *ref_sign_bias) {
+  int_mv mv = mbmi->mv[ref];
+  if (ref_sign_bias[mbmi->ref_frame[ref]] != ref_sign_bias[this_ref_frame]) {
+    mv.as_mv.row *= -1;
+    mv.as_mv.col *= -1;
+  }
+  return mv;
+}
+
+// This macro is used to add a motion vector mv_ref list if it isn't
+// already in the list.  If it's the second motion vector it will also
+// skip all additional processing and jump to done!
+#define ADD_MV_REF_LIST(mv) \
+  do { \
+    if (refmv_count) { \
+      if ((mv).as_int != mv_ref_list[0].as_int) { \
+        mv_ref_list[refmv_count] = (mv); \
+        goto Done; \
+      } \
+    } else { \
+      mv_ref_list[refmv_count++] = (mv); \
+    } \
+  } while (0)
+
+// If either reference frame is different, not INTRA, and they
+// are different from each other scale and add the mv to our list.
+#define IF_DIFF_REF_FRAME_ADD_MV(mbmi) \
+  do { \
+    if (is_inter_block(mbmi)) { \
+      if ((mbmi)->ref_frame[0] != ref_frame) \
+        ADD_MV_REF_LIST(scale_mv((mbmi), 0, ref_frame, ref_sign_bias)); \
+      if (has_second_ref(mbmi) && \
+          (mbmi)->ref_frame[1] != ref_frame && \
+          (mbmi)->mv[1].as_int != (mbmi)->mv[0].as_int) \
+        ADD_MV_REF_LIST(scale_mv((mbmi), 1, ref_frame, ref_sign_bias)); \
+    } \
+  } while (0)
+
+
+// Checks that the given mi_row, mi_col and search point
+// are inside the borders of the tile.
+static INLINE int is_inside(const TileInfo *const tile,
+                            int mi_col, int mi_row, int mi_rows,
+                            const POSITION *mi_pos) {
+  return !(mi_row + mi_pos->row < 0 ||
+           mi_col + mi_pos->col < tile->mi_col_start ||
+           mi_row + mi_pos->row >= mi_rows ||
+           mi_col + mi_pos->col >= tile->mi_col_end);
+}
+
 // TODO(jingning): this mv clamping function should be block size dependent.
 static INLINE void clamp_mv2(MV *mv, const MACROBLOCKD *xd) {
   clamp_mv(mv, xd->mb_to_left_edge - LEFT_TOP_MARGIN,
@@ -45,6 +220,12 @@ void vp9_append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd,
                                    int block, int ref, int mi_row, int mi_col,
                                    int_mv *nearest, int_mv *near);
 
+#if CONFIG_COPY_CODING
+int vp9_construct_ref_inter_list(VP9_COMMON *cm,  MACROBLOCKD *xd,
+                                 BLOCK_SIZE bsize, int mi_row, int mi_col,
+                                 MB_MODE_INFO *ref_list[18]);
+#endif
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif

vp9/common/vp9_onyxc_int.h

@@ -212,6 +212,16 @@ typedef struct VP9Common {
 
   PARTITION_CONTEXT *above_seg_context;
   ENTROPY_CONTEXT *above_context;
+
+#if CONFIG_MASKED_INTERINTER
+  int use_masked_interinter;
+#endif
+#if CONFIG_INTERINTRA
+  int use_interintra;
+#if CONFIG_MASKED_INTERINTRA
+  int use_masked_interintra;
+#endif
+#endif
 } VP9_COMMON;
 
 static INLINE YV12_BUFFER_CONFIG *get_frame_new_buffer(VP9_COMMON *cm) {

vp9/common/vp9_pred_common.c

@@ -383,3 +383,47 @@ int vp9_get_segment_id(VP9_COMMON *cm, const uint8_t *segment_ids,
   assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
   return segment_id;
 }
+
+#if CONFIG_COPY_CODING
+int vp9_get_copy_mode_context(const MACROBLOCKD *xd) {
+  const MB_MODE_INFO *const above_mbmi = get_mbmi(get_above_mi(xd));
+  const MB_MODE_INFO *const left_mbmi = get_mbmi(get_left_mi(xd));
+  const int has_above = above_mbmi != NULL;
+  const int has_left = left_mbmi != NULL;
+
+  if (has_above && has_left) {
+    const int above_intra = !is_inter_block(above_mbmi);
+    const int left_intra = !is_inter_block(left_mbmi);
+
+    if (above_intra && left_intra) {
+      return 4;
+    } else if (above_intra || left_intra) {
+      return 3;
+    } else {
+      const int above_predict = above_mbmi->copy_mode != NOREF;
+      const int left_predict = left_mbmi->copy_mode != NOREF;
+      if (above_predict && left_predict)
+        return 0;
+      else if (above_predict || left_predict)
+        return 1;
+      else
+        return 2;
+    }
+  } else if (has_above || has_left) {
+    const MB_MODE_INFO *const ref_mbmi = has_above ? above_mbmi : left_mbmi;
+    const int ref_intra = !is_inter_block(ref_mbmi);
+
+    if (ref_intra) {
+      return 3;
+    } else {
+     const int ref_predict = ref_mbmi != NOREF;
+      if (ref_predict)
+        return 0;
+      else
+        return 1;
+    }
+  } else {
+    return 0;
+  }
+}
+#endif

vp9/common/vp9_pred_common.h

@@ -134,6 +134,10 @@ static INLINE unsigned int *get_tx_counts(TX_SIZE max_tx_size, int ctx,
   }
 }
 
+#if CONFIG_COPY_CODING
+int vp9_get_copy_mode_context(const MACROBLOCKD *xd);
+#endif
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif

vp9/common/vp9_quant_common.c

@@ -12,7 +12,6 @@
 #include "vp9/common/vp9_quant_common.h"
 #include "vp9/common/vp9_seg_common.h"
 
-#if 1
 static const int16_t dc_qlookup[QINDEX_RANGE] = {
   4,       8,    8,    9,   10,   11,   12,   12,
   13,     14,   15,   16,   17,   18,   19,   19,
@@ -83,44 +82,6 @@ static const int16_t ac_qlookup[QINDEX_RANGE] = {
   1597, 1628, 1660, 1692, 1725, 1759, 1793, 1828,
 };
 
-void vp9_init_quant_tables(void) { }
-#else
-static int16_t dc_qlookup[QINDEX_RANGE];
-static int16_t ac_qlookup[QINDEX_RANGE];
-
-#define ACDC_MIN 8
-
-// TODO(dkovalev) move to common and reuse
-static double poly3(double a, double b, double c, double d, double x) {
-  return a*x*x*x + b*x*x + c*x + d;
-}
-
-void vp9_init_quant_tables() {
-  int i, val = 4;
-
-  // A "real" q of 1.0 forces lossless mode.
-  // In practice non lossless Q's between 1.0 and 2.0 (represented here by
-  // integer values from 5-7 give poor rd results (lower psnr and often
-  // larger size than the lossless encode. To block out those "not very useful"
-  // values we increment the ac and dc q lookup values by 4 after position 0.
-  ac_qlookup[0] = val;
-  dc_qlookup[0] = val;
-  val += 4;
-
-  for (i = 1; i < QINDEX_RANGE; i++) {
-    const int ac_val = val;
-
-    val = (int)(val * 1.01975);
-    if (val == ac_val)
-      ++val;
-
-    ac_qlookup[i] = (int16_t)ac_val;
-    dc_qlookup[i] = (int16_t)MAX(ACDC_MIN, poly3(0.000000305, -0.00065, 0.9,
-                                                 0.5, ac_val));
-  }
-}
-#endif
-
 int16_t vp9_dc_quant(int qindex, int delta) {
   return dc_qlookup[clamp(qindex + delta, 0, MAXQ)];
 }

vp9/common/vp9_quant_common.h

@@ -22,8 +22,6 @@ extern "C" {
 #define QINDEX_RANGE (MAXQ - MINQ + 1)
 #define QINDEX_BITS 8
 
-void vp9_init_quant_tables();
-
 int16_t vp9_dc_quant(int qindex, int delta);
 int16_t vp9_ac_quant(int qindex, int delta);
 

vp9/common/vp9_reconinter.c

@@ -139,9 +139,349 @@ MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd, const MV *src_mv,
   return clamped_mv;
 }
 
+#if CONFIG_MASKED_INTERINTER
+#define MASK_WEIGHT_BITS 6
+
+static int get_masked_weight(int m) {
+  #define SMOOTHER_LEN  32
+  static const uint8_t smoothfn[2 * SMOOTHER_LEN + 1] = {
+      0,  0,  0,  0,  0,  0,  0,  0,
+      0,  0,  0,  0,  0,  1,  1,  1,
+      1,  1,  2,  2,  3,  4,  5,  6,
+      8,  9, 12, 14, 17, 21, 24, 28,
+      32,
+      36, 40, 43, 47, 50, 52, 55, 56,
+      58, 59, 60, 61, 62, 62, 63, 63,
+      63, 63, 63, 64, 64, 64, 64, 64,
+      64, 64, 64, 64, 64, 64, 64, 64,
+  };
+  if (m < -SMOOTHER_LEN)
+    return 0;
+  else if (m > SMOOTHER_LEN)
+    return (1 << MASK_WEIGHT_BITS);
+  else
+    return smoothfn[m + SMOOTHER_LEN];
+}
+
+static int get_hard_mask(int m) {
+  return 1 << MASK_WEIGHT_BITS * (m > 0);
+}
+
+// Equation of line: f(x, y) = a[0]*(x - a[2]*w/4) + a[1]*(y - a[3]*h/4) = 0
+// The soft mask is obtained by computing f(x, y) and then calling
+// get_masked_weight(f(x, y)).
+static const int mask_params_sml[1 << MASK_BITS_SML][4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+};
+
+static const int mask_params_med_hgtw[1 << MASK_BITS_MED][4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  {-1,  2, 2, 1},
+  { 1, -2, 2, 1},
+  {-1,  2, 2, 3},
+  { 1, -2, 2, 3},
+  { 1,  2, 2, 1},
+  {-1, -2, 2, 1},
+  { 1,  2, 2, 3},
+  {-1, -2, 2, 3},
+};
+
+static const int mask_params_med_hltw[1 << MASK_BITS_MED][4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  {-2,  1, 1, 2},
+  { 2, -1, 1, 2},
+  {-2,  1, 3, 2},
+  { 2, -1, 3, 2},
+  { 2,  1, 1, 2},
+  {-2, -1, 1, 2},
+  { 2,  1, 3, 2},
+  {-2, -1, 3, 2},
+};
+
+static const int mask_params_med_heqw[1 << MASK_BITS_MED][4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  { 0,  2, 0, 1},
+  { 0, -2, 0, 1},
+  { 0,  2, 0, 3},
+  { 0, -2, 0, 3},
+  { 2,  0, 1, 0},
+  {-2,  0, 1, 0},
+  { 2,  0, 3, 0},
+  {-2,  0, 3, 0},
+};
+
+static const int mask_params_big_hgtw[1 << MASK_BITS_BIG][4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  {-1,  2, 2, 1},
+  { 1, -2, 2, 1},
+  {-1,  2, 2, 3},
+  { 1, -2, 2, 3},
+  { 1,  2, 2, 1},
+  {-1, -2, 2, 1},
+  { 1,  2, 2, 3},
+  {-1, -2, 2, 3},
+
+  {-2,  1, 1, 2},
+  { 2, -1, 1, 2},
+  {-2,  1, 3, 2},
+  { 2, -1, 3, 2},
+  { 2,  1, 1, 2},
+  {-2, -1, 1, 2},
+  { 2,  1, 3, 2},
+  {-2, -1, 3, 2},
+
+  { 0,  2, 0, 1},
+  { 0, -2, 0, 1},
+  { 0,  2, 0, 2},
+  { 0, -2, 0, 2},
+  { 0,  2, 0, 3},
+  { 0, -2, 0, 3},
+  { 2,  0, 2, 0},
+  {-2,  0, 2, 0},
+};
+
+static const int mask_params_big_hltw[1 << MASK_BITS_BIG][4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  {-1,  2, 2, 1},
+  { 1, -2, 2, 1},
+  {-1,  2, 2, 3},
+  { 1, -2, 2, 3},
+  { 1,  2, 2, 1},
+  {-1, -2, 2, 1},
+  { 1,  2, 2, 3},
+  {-1, -2, 2, 3},
+
+  {-2,  1, 1, 2},
+  { 2, -1, 1, 2},
+  {-2,  1, 3, 2},
+  { 2, -1, 3, 2},
+  { 2,  1, 1, 2},
+  {-2, -1, 1, 2},
+  { 2,  1, 3, 2},
+  {-2, -1, 3, 2},
+
+  { 0,  2, 0, 2},
+  { 0, -2, 0, 2},
+  { 2,  0, 1, 0},
+  {-2,  0, 1, 0},
+  { 2,  0, 2, 0},
+  {-2,  0, 2, 0},
+  { 2,  0, 3, 0},
+  {-2,  0, 3, 0},
+};
+
+static const int mask_params_big_heqw[1 << MASK_BITS_BIG][4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  {-1,  2, 2, 1},
+  { 1, -2, 2, 1},
+  {-1,  2, 2, 3},
+  { 1, -2, 2, 3},
+  { 1,  2, 2, 1},
+  {-1, -2, 2, 1},
+  { 1,  2, 2, 3},
+  {-1, -2, 2, 3},
+
+  {-2,  1, 1, 2},
+  { 2, -1, 1, 2},
+  {-2,  1, 3, 2},
+  { 2, -1, 3, 2},
+  { 2,  1, 1, 2},
+  {-2, -1, 1, 2},
+  { 2,  1, 3, 2},
+  {-2, -1, 3, 2},
+
+  { 0,  2, 0, 1},
+  { 0, -2, 0, 1},
+  { 0,  2, 0, 3},
+  { 0, -2, 0, 3},
+  { 2,  0, 1, 0},
+  {-2,  0, 1, 0},
+  { 2,  0, 3, 0},
+  {-2,  0, 3, 0},
+};
+
+static const int *get_mask_params(int mask_index,
+                                  BLOCK_SIZE sb_type,
+                                  int h, int w) {
+  const int *a;
+  const int mask_bits = get_mask_bits(sb_type);
+
+  if (mask_index == MASK_NONE)
+    return NULL;
+
+  if (mask_bits == MASK_BITS_SML) {
+    a = mask_params_sml[mask_index];
+  } else if (mask_bits == MASK_BITS_MED) {
+    if (h > w)
+      a = mask_params_med_hgtw[mask_index];
+    else if (h < w)
+      a = mask_params_med_hltw[mask_index];
+    else
+      a = mask_params_med_heqw[mask_index];
+  } else if (mask_bits == MASK_BITS_BIG) {
+    if (h > w)
+      a = mask_params_big_hgtw[mask_index];
+    else if (h < w)
+      a = mask_params_big_hltw[mask_index];
+    else
+      a = mask_params_big_heqw[mask_index];
+  } else {
+    assert(0);
+  }
+  return a;
+}
+
+void vp9_generate_masked_weight(int mask_index,
+                                BLOCK_SIZE sb_type,
+                                int h, int w,
+                                uint8_t *mask, int stride) {
+  int i, j;
+  const int *a = get_mask_params(mask_index, sb_type, h, w);
+  if (!a) return;
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) {
+      int x = (j - (a[2] * w) / 4);
+      int y = (i - (a[3] * h) / 4);
+      int m = a[0] * x + a[1] * y;
+      mask[i * stride + j] = get_masked_weight(m);
+    }
+}
+
+void vp9_generate_hard_mask(int mask_index, BLOCK_SIZE sb_type,
+                            int h, int w, uint8_t *mask, int stride) {
+  int i, j;
+  const int *a = get_mask_params(mask_index, sb_type, h, w);
+  if (!a) return;
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) {
+      int x = (j - (a[2] * w) / 4);
+      int y = (i - (a[3] * h) / 4);
+      int m = a[0] * x + a[1] * y;
+      mask[i * stride + j] = get_hard_mask(m);
+    }
+}
+
+static void build_masked_compound(uint8_t *dst, int dst_stride,
+                                  uint8_t *dst2, int dst2_stride,
+                                  int mask_index, BLOCK_SIZE sb_type,
+                                  int h, int w) {
+  int i, j;
+  uint8_t mask[4096];
+  vp9_generate_masked_weight(mask_index, sb_type, h, w, mask, 64);
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) {
+      int m = mask[i * 64 + j];
+      dst[i * dst_stride + j] = (dst[i * dst_stride + j] * m +
+                                 dst2[i * dst2_stride + j] *
+                                 ((1 << MASK_WEIGHT_BITS) - m) +
+                                 (1 << (MASK_WEIGHT_BITS - 1))) >>
+                                 MASK_WEIGHT_BITS;
+    }
+}
+
+#if CONFIG_SUPERTX
+void generate_masked_weight_extend(int mask_index, int plane,
+                                   BLOCK_SIZE sb_type, int h, int w,
+                                   int mask_offset_x, int mask_offset_y,
+                                   uint8_t *mask, int stride) {
+  int i, j;
+  int subh = (plane ? 2 : 4) << b_height_log2(sb_type);
+  int subw = (plane ? 2 : 4) << b_width_log2(sb_type);
+  const int *a = get_mask_params(mask_index, sb_type, subh, subw);
+  if (!a) return;
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) {
+      int x = (j - (a[2] * subw) / 4 - mask_offset_x);
+      int y = (i - (a[3] * subh) / 4 - mask_offset_y);
+      int m = a[0] * x + a[1] * y;
+      mask[i * stride + j] = get_masked_weight(m);
+    }
+}
+
+static void build_masked_compound_extend(uint8_t *dst, int dst_stride,
+                                         uint8_t *dst2, int dst2_stride,
+                                         int plane,
+                                         int mask_index, BLOCK_SIZE sb_type,
+                                         int mask_offset_x, int mask_offset_y,
+                                         int h, int w) {
+  int i, j;
+  uint8_t mask[4096];
+  generate_masked_weight_extend(mask_index, plane, sb_type, h, w,
+                                mask_offset_x, mask_offset_y, mask, 64);
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) {
+      int m = mask[i * 64 + j];
+      dst[i * dst_stride + j] = (dst[i * dst_stride + j] * m +
+                                 dst2[i * dst2_stride + j] *
+                                 ((1 << MASK_WEIGHT_BITS) - m) +
+                                 (1 << (MASK_WEIGHT_BITS - 1))) >>
+                                 MASK_WEIGHT_BITS;
+    }
+}
+#endif
+#endif
+
 static void build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
                                    int bw, int bh,
                                    int x, int y, int w, int h,
+#if CONFIG_SUPERTX && CONFIG_MASKED_INTERINTER
+                                   int mask_offset_x, int mask_offset_y,
+#endif
                                    int mi_x, int mi_y) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
   const MODE_INFO *mi = xd->mi[0];
@@ -193,8 +533,27 @@ static void build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
     pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride
            + (scaled_mv.col >> SUBPEL_BITS);
 
+#if CONFIG_MASKED_INTERINTER
+    if (ref && get_mask_bits(mi->mbmi.sb_type)
+        && mi->mbmi.use_masked_interinter) {
+      uint8_t tmp_dst[4096];
+      inter_predictor(pre, pre_buf->stride, tmp_dst, 64,
+                     subpel_x, subpel_y, sf, w, h, 0, kernel, xs, ys);
+#if CONFIG_SUPERTX
+      build_masked_compound_extend(dst, dst_buf->stride, tmp_dst, 64, plane,
+                                   mi->mbmi.mask_index, mi->mbmi.sb_type,
+                                   mask_offset_x, mask_offset_y, h, w);
+#else
+      build_masked_compound(dst, dst_buf->stride, tmp_dst, 64,
+                            mi->mbmi.mask_index, mi->mbmi.sb_type, h, w);
+#endif
+    } else {
+#endif
     inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
                     subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
+#if CONFIG_MASKED_INTERINTER
+    }
+#endif
   }
 }
 
@@ -218,10 +577,18 @@ static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
       for (y = 0; y < num_4x4_h; ++y)
         for (x = 0; x < num_4x4_w; ++x)
            build_inter_predictors(xd, plane, i++, bw, bh,
-                                  4 * x, 4 * y, 4, 4, mi_x, mi_y);
+                                  4 * x, 4 * y, 4, 4,
+#if CONFIG_SUPERTX && CONFIG_MASKED_INTERINTER
+                                  0, 0,
+#endif
+                                  mi_x, mi_y);
     } else {
       build_inter_predictors(xd, plane, 0, bw, bh,
-                             0, 0, bw, bh, mi_x, mi_y);
+                             0, 0, bw, bh,
+#if CONFIG_SUPERTX && CONFIG_MASKED_INTERINTER
+                             0, 0,
+#endif
+                             mi_x, mi_y);
     }
   }
 }
@@ -229,23 +596,323 @@ static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
 void vp9_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col,
                                     BLOCK_SIZE bsize) {
   build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0, 0);
+#if CONFIG_INTERINTRA
+  if (xd->mi[0]->mbmi.ref_frame[1] == INTRA_FRAME &&
+      is_interintra_allowed(xd->mi[0]->mbmi.sb_type))
+    vp9_build_interintra_predictors_sby(xd, xd->plane[0].dst.buf,
+                                        xd->plane[0].dst.stride, bsize);
+#endif
 }
 void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col,
                                      BLOCK_SIZE bsize) {
   build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 1,
                                     MAX_MB_PLANE - 1);
+#if CONFIG_INTERINTRA
+  if (xd->mi[0]->mbmi.ref_frame[1] == INTRA_FRAME &&
+      is_interintra_allowed(xd->mi[0]->mbmi.sb_type))
+    vp9_build_interintra_predictors_sbuv(xd, xd->plane[1].dst.buf,
+                                         xd->plane[2].dst.buf,
+                                         xd->plane[1].dst.stride,
+                                         xd->plane[2].dst.stride, bsize);
+#endif
 }
+
 void vp9_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
                                    BLOCK_SIZE bsize) {
   build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0,
                                     MAX_MB_PLANE - 1);
+#if CONFIG_INTERINTRA
+  if (xd->mi[0]->mbmi.ref_frame[1] == INTRA_FRAME &&
+      is_interintra_allowed(xd->mi[0]->mbmi.sb_type))
+    vp9_build_interintra_predictors(xd, xd->plane[0].dst.buf,
+                                    xd->plane[1].dst.buf, xd->plane[2].dst.buf,
+                                    xd->plane[0].dst.stride,
+                                    xd->plane[1].dst.stride,
+                                    xd->plane[2].dst.stride, bsize);
+#endif
 }
 
+#if CONFIG_SUPERTX
+static int get_masked_weight_supertx(int m) {
+  #define SMOOTHER_LEN  32
+  static const uint8_t smoothfn[2 * SMOOTHER_LEN + 1] = {
+      0,  0,  0,  0,  0,  0,  0,  0,
+      0,  0,  0,  0,  0,  1,  1,  1,
+      1,  1,  2,  2,  3,  4,  5,  6,
+      8,  9, 12, 14, 17, 21, 24, 28,
+      32,
+      36, 40, 43, 47, 50, 52, 55, 56,
+      58, 59, 60, 61, 62, 62, 63, 63,
+      63, 63, 63, 64, 64, 64, 64, 64,
+      64, 64, 64, 64, 64, 64, 64, 64,
+  };
+  if (m < -SMOOTHER_LEN)
+    return 0;
+  else if (m > SMOOTHER_LEN)
+    return 64;
+  else
+    return smoothfn[m + SMOOTHER_LEN];
+}
+
+static const uint8_t mask_8[8] = {
+  64, 64, 62, 52, 12,  2,  0,  0
+};
+
+static const uint8_t mask_16[16] = {
+  63, 62, 60, 58, 55, 50, 43, 36, 28, 21, 14, 9, 6, 4, 2, 1
+};
+
+static const uint8_t mask_32[32] = {
+  64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 63, 61, 57, 52, 45, 36,
+  28, 19, 12,  7,  3,  1,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0
+};
+
+static void generate_1dmask(int length, uint8_t *mask) {
+  int i;
+  switch (length) {
+    case 8:
+      vpx_memcpy(mask, mask_8, length);
+      break;
+    case 16:
+      vpx_memcpy(mask, mask_16, length);
+      break;
+    case 32:
+      vpx_memcpy(mask, mask_32, length);
+      break;
+    default:
+      assert(0);
+  }
+  if (length > 16) {
+    for (i = 0; i < length; ++i)
+      mask[i] = get_masked_weight_supertx(-1 * (2 * i - length + 1));
+  }
+}
+
+void vp9_build_masked_inter_predictor_complex(uint8_t *dst, int dst_stride,
+                                              uint8_t *dst2, int dst2_stride,
+                                              int plane,
+                                              int mi_row, int mi_col,
+                                              int mi_row_ori, int mi_col_ori,
+                                              BLOCK_SIZE bsize,
+                                              BLOCK_SIZE top_bsize,
+                                              PARTITION_TYPE partition) {
+  int i, j;
+  uint8_t mask[32];
+  int top_w = 4 << b_width_log2(top_bsize),
+      top_h = 4 << b_height_log2(top_bsize);
+  int w = 4 << b_width_log2(bsize), h = 4 << b_height_log2(bsize);
+  int w_offset = (mi_col - mi_col_ori) << 3,
+      h_offset = (mi_row - mi_row_ori) << 3;
+  int m;
+
+  if (plane > 0) {
+    top_w = top_w >> 1; top_h = top_h >> 1;
+    w = w >> 1; h = h >> 1;
+    w_offset = w_offset >> 1; h_offset = h_offset >> 1;
+  }
+  switch (partition) {
+    case PARTITION_HORZ:
+      generate_1dmask(h, mask + h_offset);
+      vpx_memset(mask, 64, h_offset);
+      vpx_memset(mask + h_offset + h, 0, top_h - h_offset - h);
+      break;
+    case PARTITION_VERT:
+      generate_1dmask(w, mask + w_offset);
+      vpx_memset(mask, 64, w_offset);
+      vpx_memset(mask + w_offset + w, 0, top_w - w_offset - w);
+      break;
+    default:
+      assert(0);
+  }
+  for (i = 0; i < top_h; ++i)
+    for (j = 0; j < top_w; ++j) {
+      m = partition == PARTITION_HORZ ? mask[i] : mask[j];
+      if (m == 64)
+        continue;
+      if (m == 0)
+        dst[i * dst_stride + j] = dst2[i * dst2_stride + j];
+      else
+        dst[i * dst_stride + j] = (dst[i * dst_stride + j] * m +
+                                  dst2[i * dst2_stride + j] *
+                                  (64 - m) + 32) >> 6;
+    }
+}
+
+#if CONFIG_MASKED_INTERINTER
+void vp9_build_inter_predictors_sb_extend(MACROBLOCKD *xd,
+                                          int mi_row, int mi_col,
+                                          int mi_row_ori, int mi_col_ori,
+                                          BLOCK_SIZE bsize) {
+  int plane;
+  const int mi_x = mi_col_ori * MI_SIZE;
+  const int mi_y = mi_row_ori * MI_SIZE;
+  const int mask_offset_x = (mi_col - mi_col_ori) * MI_SIZE;
+  const int mask_offset_y = (mi_row - mi_row_ori) * MI_SIZE;
+  for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
+    const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize,
+                                                        &xd->plane[plane]);
+    const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
+    const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
+    const int bw = 4 * num_4x4_w;
+    const int bh = 4 * num_4x4_h;
+
+    if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8) {
+      int i = 0, x, y;
+      assert(bsize == BLOCK_8X8);
+      for (y = 0; y < num_4x4_h; ++y)
+        for (x = 0; x < num_4x4_w; ++x)
+           build_inter_predictors(xd, plane, i++, bw, bh, 4 * x, 4 * y, 4, 4,
+                                  mask_offset_x, mask_offset_y, mi_x, mi_y);
+    } else {
+      build_inter_predictors(xd, plane, 0, bw, bh, 0, 0, bw, bh,
+                             mask_offset_x, mask_offset_y, mi_x, mi_y);
+    }
+  }
+}
+#endif
+
+void vp9_build_inter_predictors_sby_sub8x8_extend(MACROBLOCKD *xd,
+                                                  int mi_row, int mi_col,
+                                                  int mi_row_ori,
+                                                  int mi_col_ori,
+                                                  BLOCK_SIZE top_bsize,
+                                                  PARTITION_TYPE partition) {
+  const int mi_x = mi_col_ori * MI_SIZE;
+  const int mi_y = mi_row_ori * MI_SIZE;
+#if CONFIG_MASKED_INTERINTER
+  const int mask_offset_x = (mi_col - mi_col_ori) * MI_SIZE;
+  const int mask_offset_y = (mi_row - mi_row_ori) * MI_SIZE;
+#endif
+  uint8_t *orig_dst;
+  int orig_dst_stride;
+  int bw = 4 << b_width_log2(top_bsize);
+  int bh = 4 << b_height_log2(top_bsize);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, tmp_buf, 32 * 32);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, tmp_buf1, 32 * 32);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, tmp_buf2, 32 * 32);
+
+  orig_dst = xd->plane[0].dst.buf;
+  orig_dst_stride = xd->plane[0].dst.stride;
+  build_inter_predictors(xd, 0, 0, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                         mask_offset_x, mask_offset_y,
+#endif
+                         mi_x, mi_y);
+
+  xd->plane[0].dst.buf = tmp_buf;
+  xd->plane[0].dst.stride = 32;
+  switch (partition) {
+    case PARTITION_HORZ:
+      build_inter_predictors(xd, 0, 2, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                             mask_offset_x, mask_offset_y,
+#endif
+                             mi_x, mi_y);
+      break;
+    case PARTITION_VERT:
+      build_inter_predictors(xd, 0, 1, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                             mask_offset_x, mask_offset_y,
+#endif
+                             mi_x, mi_y);
+      break;
+    case PARTITION_SPLIT:
+      build_inter_predictors(xd, 0, 1, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                             mask_offset_x, mask_offset_y,
+#endif
+                             mi_x, mi_y);
+      xd->plane[0].dst.buf = tmp_buf1;
+      xd->plane[0].dst.stride = 32;
+      build_inter_predictors(xd, 0, 2, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                             mask_offset_x, mask_offset_y,
+#endif
+                             mi_x, mi_y);
+      xd->plane[0].dst.buf = tmp_buf2;
+      xd->plane[0].dst.stride = 32;
+      build_inter_predictors(xd, 0, 3, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                             mask_offset_x, mask_offset_y,
+#endif
+                             mi_x, mi_y);
+      break;
+    default:
+      assert(0);
+  }
+
+  if (partition != PARTITION_SPLIT) {
+    vp9_build_masked_inter_predictor_complex(orig_dst, orig_dst_stride,
+                                             tmp_buf, 32,
+                                             0, mi_row, mi_col,
+                                             mi_row_ori, mi_col_ori,
+                                             BLOCK_8X8, top_bsize,
+                                             partition);
+    xd->plane[0].dst.buf = orig_dst;
+    xd->plane[0].dst.stride = orig_dst_stride;
+  } else {
+    vp9_build_masked_inter_predictor_complex(orig_dst, orig_dst_stride,
+                                             tmp_buf, 32,
+                                             0, mi_row, mi_col,
+                                             mi_row_ori, mi_col_ori,
+                                             BLOCK_8X8, top_bsize,
+                                             PARTITION_VERT);
+    vp9_build_masked_inter_predictor_complex(tmp_buf1, 32,
+                                             tmp_buf2, 32,
+                                             0, mi_row, mi_col,
+                                             mi_row_ori, mi_col_ori,
+                                             BLOCK_8X8, top_bsize,
+                                             PARTITION_VERT);
+    vp9_build_masked_inter_predictor_complex(orig_dst, orig_dst_stride,
+                                             tmp_buf1, 32,
+                                             0, mi_row, mi_col,
+                                             mi_row_ori, mi_col_ori,
+                                             BLOCK_8X8, top_bsize,
+                                             PARTITION_HORZ);
+    xd->plane[0].dst.buf = orig_dst;
+    xd->plane[0].dst.stride = orig_dst_stride;
+  }
+}
+
+void vp9_build_inter_predictors_sbuv_sub8x8_extend(MACROBLOCKD *xd,
+#if CONFIG_MASKED_INTERINTER
+                                                   int mi_row, int mi_col,
+#endif
+                                                   int mi_row_ori,
+                                                   int mi_col_ori,
+                                                   BLOCK_SIZE top_bsize) {
+  int plane;
+  const int mi_x = mi_col_ori * MI_SIZE;
+  const int mi_y = mi_row_ori * MI_SIZE;
+#if CONFIG_MASKED_INTERINTER
+  const int mask_offset_x = (mi_col - mi_col_ori) * MI_SIZE;
+  const int mask_offset_y = (mi_row - mi_row_ori) * MI_SIZE;
+#endif
+  for (plane = 1; plane < MAX_MB_PLANE; ++plane) {
+    const BLOCK_SIZE plane_bsize = get_plane_block_size(top_bsize,
+                                                        &xd->plane[plane]);
+    const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
+    const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
+    const int bw = 4 * num_4x4_w;
+    const int bh = 4 * num_4x4_h;
+
+    build_inter_predictors(xd, plane, 0, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                           mask_offset_x, mask_offset_y,
+#endif
+                           mi_x, mi_y);
+  }
+}
+#endif
+
 // TODO(jingning): This function serves as a placeholder for decoder prediction
 // using on demand border extension. It should be moved to /decoder/ directory.
 static void dec_build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
                                        int bw, int bh,
                                        int x, int y, int w, int h,
+#if CONFIG_SUPERTX && CONFIG_MASKED_INTERINTER
+                                       int mask_offset_x, int mask_offset_y,
+#endif
                                        int mi_x, int mi_y) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
   const MODE_INFO *mi = xd->mi[0];
@@ -377,8 +1044,27 @@ static void dec_build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
       }
     }
 
+#if CONFIG_MASKED_INTERINTER
+    if (ref && get_mask_bits(mi->mbmi.sb_type)
+        && mi->mbmi.use_masked_interinter) {
+      uint8_t tmp_dst[4096];
+      inter_predictor(buf_ptr, buf_stride, tmp_dst, 64,
+                     subpel_x, subpel_y, sf, w, h, 0, kernel, xs, ys);
+#if CONFIG_SUPERTX
+      build_masked_compound_extend(dst, dst_buf->stride, tmp_dst, 64, plane,
+                                   mi->mbmi.mask_index, mi->mbmi.sb_type,
+                                   mask_offset_x, mask_offset_y, h, w);
+#else
+      build_masked_compound(dst, dst_buf->stride, tmp_dst, 64,
+                            mi->mbmi.mask_index, mi->mbmi.sb_type, h, w);
+#endif
+    } else {
+#endif
     inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x,
                     subpel_y, sf, w, h, ref, kernel, xs, ys);
+#if CONFIG_MASKED_INTERINTER
+    }
+#endif
   }
 }
 
@@ -401,13 +1087,198 @@ void vp9_dec_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
       for (y = 0; y < num_4x4_h; ++y)
         for (x = 0; x < num_4x4_w; ++x)
           dec_build_inter_predictors(xd, plane, i++, bw, bh,
-                                     4 * x, 4 * y, 4, 4, mi_x, mi_y);
+                                     4 * x, 4 * y, 4, 4,
+#if CONFIG_SUPERTX && CONFIG_MASKED_INTERINTER
+                                     0, 0,
+#endif
+                                     mi_x, mi_y);
     } else {
       dec_build_inter_predictors(xd, plane, 0, bw, bh,
-                                 0, 0, bw, bh, mi_x, mi_y);
+                                 0, 0, bw, bh,
+#if CONFIG_SUPERTX && CONFIG_MASKED_INTERINTER
+                                 0, 0,
+#endif
+                                 mi_x, mi_y);
+    }
+  }
+#if CONFIG_INTERINTRA
+  if (xd->mi[0]->mbmi.ref_frame[1] == INTRA_FRAME &&
+      is_interintra_allowed(xd->mi[0]->mbmi.sb_type))
+    vp9_build_interintra_predictors(xd, xd->plane[0].dst.buf,
+                                    xd->plane[1].dst.buf, xd->plane[2].dst.buf,
+                                    xd->plane[0].dst.stride,
+                                    xd->plane[1].dst.stride,
+                                    xd->plane[2].dst.stride, bsize);
+#endif
+}
+
+#if CONFIG_SUPERTX
+#if CONFIG_MASKED_INTERINTER
+void vp9_dec_build_inter_predictors_sb_extend(MACROBLOCKD *xd,
+                                              int mi_row, int mi_col,
+                                              int mi_row_ori, int mi_col_ori,
+                                              BLOCK_SIZE bsize) {
+  int plane;
+  const int mi_x = mi_col_ori * MI_SIZE;
+  const int mi_y = mi_row_ori * MI_SIZE;
+  const int mask_offset_x = (mi_col - mi_col_ori) * MI_SIZE;
+  const int mask_offset_y = (mi_row - mi_row_ori) * MI_SIZE;
+  for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
+    const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize,
+                                                        &xd->plane[plane]);
+    const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
+    const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
+    const int bw = 4 * num_4x4_w;
+    const int bh = 4 * num_4x4_h;
+
+    if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8) {
+      int i = 0, x, y;
+      assert(bsize == BLOCK_8X8);
+      for (y = 0; y < num_4x4_h; ++y)
+        for (x = 0; x < num_4x4_w; ++x)
+          dec_build_inter_predictors(xd, plane, i++, bw, bh, 4 * x, 4 * y, 4, 4,
+                                     mask_offset_x, mask_offset_y, mi_x, mi_y);
+    } else {
+      dec_build_inter_predictors(xd, plane, 0, bw, bh, 0, 0, bw, bh,
+                                 mask_offset_x, mask_offset_y, mi_x, mi_y);
     }
   }
 }
+#endif
+
+void vp9_dec_build_inter_predictors_sby_sub8x8_extend(MACROBLOCKD *xd,
+                                                  int mi_row, int mi_col,
+                                                  int mi_row_ori,
+                                                  int mi_col_ori,
+                                                  BLOCK_SIZE top_bsize,
+                                                  PARTITION_TYPE partition) {
+  const int mi_x = mi_col_ori * MI_SIZE;
+  const int mi_y = mi_row_ori * MI_SIZE;
+#if CONFIG_MASKED_INTERINTER
+  const int mask_offset_x = (mi_col - mi_col_ori) * MI_SIZE;
+  const int mask_offset_y = (mi_row - mi_row_ori) * MI_SIZE;
+#endif
+  uint8_t *orig_dst;
+  int orig_dst_stride;
+  int bw = 4 << b_width_log2(top_bsize);
+  int bh = 4 << b_height_log2(top_bsize);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, tmp_buf, 32 * 32);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, tmp_buf1, 32 * 32);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, tmp_buf2, 32 * 32);
+
+  orig_dst = xd->plane[0].dst.buf;
+  orig_dst_stride = xd->plane[0].dst.stride;
+  dec_build_inter_predictors(xd, 0, 0, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                             mask_offset_x, mask_offset_y,
+#endif
+                             mi_x, mi_y);
+
+  xd->plane[0].dst.buf = tmp_buf;
+  xd->plane[0].dst.stride = 32;
+  switch (partition) {
+    case PARTITION_HORZ:
+      dec_build_inter_predictors(xd, 0, 2, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                                 mask_offset_x, mask_offset_y,
+#endif
+                                 mi_x, mi_y);
+      break;
+    case PARTITION_VERT:
+      dec_build_inter_predictors(xd, 0, 1, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                                 mask_offset_x, mask_offset_y,
+#endif
+                                 mi_x, mi_y);
+      break;
+    case PARTITION_SPLIT:
+      dec_build_inter_predictors(xd, 0, 1, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                                 mask_offset_x, mask_offset_y,
+#endif
+                                 mi_x, mi_y);
+      xd->plane[0].dst.buf = tmp_buf1;
+      xd->plane[0].dst.stride = 32;
+      dec_build_inter_predictors(xd, 0, 2, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                                 mask_offset_x, mask_offset_y,
+#endif
+                                 mi_x, mi_y);
+      xd->plane[0].dst.buf = tmp_buf2;
+      xd->plane[0].dst.stride = 32;
+      dec_build_inter_predictors(xd, 0, 3, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                                 mask_offset_x, mask_offset_y,
+#endif
+                                 mi_x, mi_y);
+      break;
+    default:
+      assert(0);
+  }
+
+  if (partition != PARTITION_SPLIT) {
+    vp9_build_masked_inter_predictor_complex(orig_dst, orig_dst_stride,
+                                             tmp_buf, 32,
+                                             0, mi_row, mi_col,
+                                             mi_row_ori, mi_col_ori,
+                                             BLOCK_8X8, top_bsize,
+                                             partition);
+    xd->plane[0].dst.buf = orig_dst;
+    xd->plane[0].dst.stride = orig_dst_stride;
+  } else {
+    vp9_build_masked_inter_predictor_complex(orig_dst, orig_dst_stride,
+                                             tmp_buf, 32,
+                                             0, mi_row, mi_col,
+                                             mi_row_ori, mi_col_ori,
+                                             BLOCK_8X8, top_bsize,
+                                             PARTITION_VERT);
+    vp9_build_masked_inter_predictor_complex(tmp_buf1, 32,
+                                             tmp_buf2, 32,
+                                             0, mi_row, mi_col,
+                                             mi_row_ori, mi_col_ori,
+                                             BLOCK_8X8, top_bsize,
+                                             PARTITION_VERT);
+    vp9_build_masked_inter_predictor_complex(orig_dst, orig_dst_stride,
+                                             tmp_buf1, 32,
+                                             0, mi_row, mi_col,
+                                             mi_row_ori, mi_col_ori,
+                                             BLOCK_8X8, top_bsize,
+                                             PARTITION_HORZ);
+    xd->plane[0].dst.buf = orig_dst;
+    xd->plane[0].dst.stride = orig_dst_stride;
+  }
+}
+
+void vp9_dec_build_inter_predictors_sbuv_sub8x8_extend(MACROBLOCKD *xd,
+#if CONFIG_MASKED_INTERINTER
+                                                       int mi_row, int mi_col,
+#endif
+                                                       int mi_row_ori,
+                                                       int mi_col_ori,
+                                                       BLOCK_SIZE top_bsize) {
+  int plane;
+  const int mi_x = mi_col_ori * MI_SIZE;
+  const int mi_y = mi_row_ori * MI_SIZE;
+#if CONFIG_MASKED_INTERINTER
+  const int mask_offset_x = (mi_col - mi_col_ori) * MI_SIZE;
+  const int mask_offset_y = (mi_row - mi_row_ori) * MI_SIZE;
+#endif
+  for (plane = 1; plane < MAX_MB_PLANE; ++plane) {
+    const BLOCK_SIZE plane_bsize = get_plane_block_size(top_bsize,
+                                                        &xd->plane[plane]);
+    const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
+    const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
+    const int bw = 4 * num_4x4_w;
+    const int bh = 4 * num_4x4_h;
+
+    dec_build_inter_predictors(xd, plane, 0, bw, bh, 0, 0, bw, bh,
+#if CONFIG_MASKED_INTERINTER
+                               mask_offset_x, mask_offset_y,
+#endif
+                               mi_x, mi_y);
+  }
+}
+#endif
 
 void vp9_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE],
                           const YV12_BUFFER_CONFIG *src,

vp9/common/vp9_reconinter.h

@@ -65,6 +65,60 @@ void vp9_setup_pre_planes(MACROBLOCKD *xd, int idx,
                           const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
                           const struct scale_factors *sf);
 
+#if CONFIG_MASKED_INTERINTER
+void vp9_generate_masked_weight(int mask_index, BLOCK_SIZE sb_type,
+                              int h, int w, uint8_t *mask, int stride);
+void vp9_generate_hard_mask(int mask_index, BLOCK_SIZE sb_type,
+                          int h, int w, uint8_t *mask, int stride);
+#endif
+
+#if CONFIG_SUPERTX
+void vp9_build_inter_predictors_sby_sub8x8_extend(MACROBLOCKD *xd,
+                                                  int mi_row, int mi_col,
+                                                  int mi_row_ori,
+                                                  int mi_col_ori,
+                                                  BLOCK_SIZE top_bsize,
+                                                  PARTITION_TYPE partition);
+void vp9_build_inter_predictors_sbuv_sub8x8_extend(MACROBLOCKD *xd,
+#if CONFIG_MASKED_INTERINTER
+                                                   int mi_row, int mi_col,
+#endif
+                                                   int mi_row_ori,
+                                                   int mi_col_ori,
+                                                   BLOCK_SIZE top_bsize);
+void vp9_build_masked_inter_predictor_complex(uint8_t *dst, int dst_stride,
+                                              uint8_t *dst2, int dst2_stride,
+                                              int plane,
+                                              int mi_row, int mi_col,
+                                              int mi_row_ori, int mi_col_ori,
+                                              BLOCK_SIZE bsize,
+                                              BLOCK_SIZE top_bsize,
+                                              PARTITION_TYPE partition);
+void vp9_dec_build_inter_predictors_sby_sub8x8_extend(MACROBLOCKD *xd,
+                                                      int mi_row, int mi_col,
+                                                      int mi_row_ori,
+                                                      int mi_col_ori,
+                                                      BLOCK_SIZE top_bsize,
+                                                      PARTITION_TYPE p);
+void vp9_dec_build_inter_predictors_sbuv_sub8x8_extend(MACROBLOCKD *xd,
+#if CONFIG_MASKED_INTERINTER
+                                                       int mi_row, int mi_col,
+#endif
+                                                       int mi_row_ori,
+                                                       int mi_col_ori,
+                                                       BLOCK_SIZE top_bsize);
+#if CONFIG_MASKED_INTERINTER
+void vp9_build_inter_predictors_sb_extend(MACROBLOCKD *xd,
+                                          int mi_row, int mi_col,
+                                          int mi_row_ori, int mi_col_ori,
+                                          BLOCK_SIZE bsize);
+void vp9_dec_build_inter_predictors_sb_extend(MACROBLOCKD *xd,
+                                              int mi_row, int mi_col,
+                                              int mi_row_ori, int mi_col_ori,
+                                              BLOCK_SIZE bsize);
+#endif
+#endif
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif

vp9/common/vp9_reconintra.c

@@ -444,8 +444,227 @@ static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
   }
 }
 
+#if CONFIG_FILTERINTRA
+static void filter_intra_predictors_4tap(uint8_t *ypred_ptr, int y_stride,
+                                         int bs,
+                                         const uint8_t *yabove_row,
+                                         const uint8_t *yleft_col,
+                                         int mode) {
+  static const int prec_bits = 10;
+  static const int round_val = 511;
+
+  int k, r, c;
+  int pred[33][33];
+  int mean, ipred;
+
+  int taps4_4[10][4] = {
+      {735, 881, -537, -54},
+      {1005, 519, -488, -11},
+      {383, 990, -343, -6},
+      {442, 805, -542, 319},
+      {658, 616, -133, -116},
+      {875, 442, -141, -151},
+      {386, 741, -23, -80},
+      {390, 1027, -446, 51},
+      {679, 606, -523, 262},
+      {903, 922, -778, -23}
+  };
+  int taps4_8[10][4] = {
+      {648, 803, -444, 16},
+      {972, 620, -576, 7},
+      {561, 967, -499, -5},
+      {585, 762, -468, 144},
+      {596, 619, -182, -9},
+      {895, 459, -176, -153},
+      {557, 722, -126, -129},
+      {601, 839, -523, 105},
+      {562, 709, -499, 251},
+      {803, 872, -695, 43}
+  };
+  int taps4_16[10][4] = {
+      {423, 728, -347, 111},
+      {963, 685, -665, 23},
+      {281, 1024, -480, 216},
+      {640, 596, -437, 78},
+      {429, 669, -259, 99},
+      {740, 646, -415, 23},
+      {568, 771, -346, 40},
+      {404, 833, -486, 209},
+      {398, 712, -423, 307},
+      {939, 935, -887, 17}
+  };
+  int taps4_32[10][4] = {
+      {477, 737, -393, 150},
+      {881, 630, -546, 67},
+      {506, 984, -443, -20},
+      {114, 459, -270, 528},
+      {433, 528, 14, 3},
+      {837, 470, -301, -30},
+      {181, 777, 89, -107},
+      {-29, 716, -232, 259},
+      {589, 646, -495, 255},
+      {740, 884, -728, 77}
+  };
+
+  const int c1 = (bs >= 32) ? taps4_32[mode][0] : ((bs >= 16) ?
+      taps4_16[mode][0] : ((bs >= 8) ? taps4_8[mode][0] : taps4_4[mode][0]));
+  const int c2 = (bs >= 32) ? taps4_32[mode][1] : ((bs >= 16) ?
+      taps4_16[mode][1] : ((bs >= 8) ? taps4_8[mode][1] : taps4_4[mode][1]));
+  const int c3 = (bs >= 32) ? taps4_32[mode][2] : ((bs >= 16) ?
+      taps4_16[mode][2] : ((bs >= 8) ? taps4_8[mode][2] : taps4_4[mode][2]));
+  const int c4 = (bs >= 32) ? taps4_32[mode][3] : ((bs >= 16) ?
+      taps4_16[mode][3] : ((bs >= 8) ? taps4_8[mode][3] : taps4_4[mode][3]));
+
+  k = 0;
+  mean = 0;
+  while (k < bs) {
+    mean = mean + (int)yleft_col[k];
+    mean = mean + (int)yabove_row[k];
+    k++;
+  }
+  mean = (mean + bs) / (2 * bs);
+
+  for (r = 0; r < bs; r++)
+    pred[r + 1][0] = (int)yleft_col[r] - mean;
+
+  for (c = 0; c < 2 * bs + 1; c++)
+    pred[0][c] = (int)yabove_row[c - 1] - mean;
+
+  for (r = 1; r < bs + 1; r++)
+    for (c = 1; c < 2 * bs + 1 - r; c++) {
+      ipred = c1 * pred[r - 1][c] + c2 * pred[r][c - 1]
+                    + c3 * pred[r - 1][c - 1] + c4 * pred[r - 1][c + 1];
+      pred[r][c] = ipred < 0 ? -((-ipred + round_val) >> prec_bits) :
+                               ((ipred + round_val) >> prec_bits);
+    }
+
+  for (r = 0; r < bs; r++) {
+    for (c = 0; c < bs; c++) {
+      ipred = pred[r + 1][c + 1] + mean;
+      ypred_ptr[c] = clip_pixel(ipred);
+    }
+    ypred_ptr += y_stride;
+  }
+}
+
+static void build_filter_intra_predictors(const MACROBLOCKD *xd,
+                                          const uint8_t *ref, int ref_stride,
+                                          uint8_t *dst, int dst_stride,
+                                          PREDICTION_MODE mode, TX_SIZE tx_size,
+                                          int up_available, int left_available,
+                                          int right_available, int x, int y,
+                                          int plane) {
+  int i;
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, left_col, 64);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, above_data, 128 + 16);
+  uint8_t *above_row = above_data + 16;
+  const uint8_t *const_above_row = above_row;
+  const int bs = 4 << tx_size;
+  int frame_width, frame_height;
+  int x0, y0;
+  const struct macroblockd_plane *const pd = &xd->plane[plane];
+
+  // Get current frame pointer, width and height.
+  if (plane == 0) {
+    frame_width = xd->cur_buf->y_width;
+    frame_height = xd->cur_buf->y_height;
+  } else {
+    frame_width = xd->cur_buf->uv_width;
+    frame_height = xd->cur_buf->uv_height;
+  }
+
+  // Get block position in current frame.
+  x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
+  y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
+
+  vpx_memset(left_col, 129, 64);
+
+  // left
+  if (left_available) {
+    if (xd->mb_to_bottom_edge < 0) {
+      /* slower path if the block needs border extension */
+      if (y0 + bs <= frame_height) {
+        for (i = 0; i < bs; ++i)
+          left_col[i] = ref[i * ref_stride - 1];
+      } else {
+        const int extend_bottom = frame_height - y0;
+        for (i = 0; i < extend_bottom; ++i)
+          left_col[i] = ref[i * ref_stride - 1];
+        for (; i < bs; ++i)
+          left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
+      }
+    } else {
+      /* faster path if the block does not need extension */
+      for (i = 0; i < bs; ++i)
+        left_col[i] = ref[i * ref_stride - 1];
+    }
+  }
+
+  // TODO(hkuang) do not extend 2*bs pixels for all modes.
+  // above
+  if (up_available) {
+    const uint8_t *above_ref = ref - ref_stride;
+    if (xd->mb_to_right_edge < 0) {
+      /* slower path if the block needs border extension */
+      if (x0 + 2 * bs <= frame_width) {
+        if (right_available && bs == 4) {
+          vpx_memcpy(above_row, above_ref, 2 * bs);
+        } else {
+          vpx_memcpy(above_row, above_ref, bs);
+          vpx_memset(above_row + bs, above_row[bs - 1], bs);
+        }
+      } else if (x0 + bs <= frame_width) {
+        const int r = frame_width - x0;
+        if (right_available && bs == 4) {
+          vpx_memcpy(above_row, above_ref, r);
+          vpx_memset(above_row + r, above_row[r - 1],
+                     x0 + 2 * bs - frame_width);
+        } else {
+          vpx_memcpy(above_row, above_ref, bs);
+          vpx_memset(above_row + bs, above_row[bs - 1], bs);
+        }
+      } else if (x0 <= frame_width) {
+        const int r = frame_width - x0;
+        if (right_available && bs == 4) {
+          vpx_memcpy(above_row, above_ref, r);
+          vpx_memset(above_row + r, above_row[r - 1],
+                     x0 + 2 * bs - frame_width);
+        } else {
+          vpx_memcpy(above_row, above_ref, r);
+          vpx_memset(above_row + r, above_row[r - 1],
+                     x0 + 2 * bs - frame_width);
+        }
+      }
+      above_row[-1] = left_available ? above_ref[-1] : 129;
+    } else {
+      /* faster path if the block does not need extension */
+      if (bs == 4 && right_available && left_available) {
+        const_above_row = above_ref;
+      } else {
+        vpx_memcpy(above_row, above_ref, bs);
+        if (bs == 4 && right_available)
+          vpx_memcpy(above_row + bs, above_ref + bs, bs);
+        else
+          vpx_memset(above_row + bs, above_row[bs - 1], bs);
+        above_row[-1] = left_available ? above_ref[-1] : 129;
+      }
+    }
+  } else {
+    vpx_memset(above_row, 127, bs * 2);
+    above_row[-1] = 127;
+  }
+
+  // predict
+  filter_intra_predictors_4tap(dst, dst_stride, bs, const_above_row, left_col,
+                               mode);
+}
+#endif
+
 void vp9_predict_intra_block(const MACROBLOCKD *xd, int block_idx, int bwl_in,
                              TX_SIZE tx_size, PREDICTION_MODE mode,
+#if CONFIG_FILTERINTRA
+                             int filterbit,
+#endif
                              const uint8_t *ref, int ref_stride,
                              uint8_t *dst, int dst_stride,
                              int aoff, int loff, int plane) {
@@ -456,8 +675,708 @@ void vp9_predict_intra_block(const MACROBLOCKD *xd, int block_idx, int bwl_in,
   const int have_right = ((block_idx & wmask) != wmask);
   const int x = aoff * 4;
   const int y = loff * 4;
+#if CONFIG_FILTERINTRA
+  const int filterflag = is_filter_allowed(mode) && is_filter_enabled(tx_size)
+                         && filterbit;
+#endif
 
   assert(bwl >= 0);
+#if CONFIG_FILTERINTRA
+  if (!filterflag) {
+#endif
   build_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode, tx_size,
                          have_top, have_left, have_right, x, y, plane);
+#if CONFIG_FILTERINTRA
+  } else {
+    build_filter_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode,
+                        tx_size, have_top, have_left, have_right, x, y, plane);
+  }
+#endif
 }
+
+#if CONFIG_INTERINTRA
+#if CONFIG_MASKED_INTERINTRA
+#define MASK_WEIGHT_BITS_INTERINTRA 6
+
+static int get_masked_weight_interintra(int m) {
+  #define SMOOTHER_LEN_INTERINTRA  32
+  static const uint8_t smoothfn[2 * SMOOTHER_LEN_INTERINTRA + 1] = {
+      0,  0,  0,  0,  0,  0,  0,  0,
+      0,  0,  0,  0,  0,  1,  1,  1,
+      1,  1,  2,  2,  3,  4,  5,  6,
+      8,  9, 12, 14, 17, 21, 24, 28,
+      32,
+      36, 40, 43, 47, 50, 52, 55, 56,
+      58, 59, 60, 61, 62, 62, 63, 63,
+      63, 63, 63, 64, 64, 64, 64, 64,
+      64, 64, 64, 64, 64, 64, 64, 64,
+  };
+  if (m < -SMOOTHER_LEN_INTERINTRA)
+    return 0;
+  else if (m > SMOOTHER_LEN_INTERINTRA)
+    return (1 << MASK_WEIGHT_BITS_INTERINTRA);
+  else
+    return smoothfn[m + SMOOTHER_LEN_INTERINTRA];
+}
+
+static int get_hard_mask_interintra(int m) {
+  return m > 0;
+}
+
+// Equation of line: f(x, y) = a[0]*(x - a[2]*w/4) + a[1]*(y - a[3]*h/4) = 0
+// The soft mask is obtained by computing f(x, y) and then calling
+// get_masked_weight(f(x, y)).
+static const int mask_params_sml_interintra[1 << MASK_BITS_SML_INTERINTRA]
+                                            [4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+};
+
+static const int mask_params_med_hgtw_interintra[1 << MASK_BITS_MED_INTERINTRA]
+                                                 [4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  {-1,  2, 2, 1},
+  { 1, -2, 2, 1},
+  {-1,  2, 2, 3},
+  { 1, -2, 2, 3},
+  { 1,  2, 2, 1},
+  {-1, -2, 2, 1},
+  { 1,  2, 2, 3},
+  {-1, -2, 2, 3},
+};
+
+static const int mask_params_med_hltw_interintra[1 << MASK_BITS_MED_INTERINTRA]
+                                                 [4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  {-2,  1, 1, 2},
+  { 2, -1, 1, 2},
+  {-2,  1, 3, 2},
+  { 2, -1, 3, 2},
+  { 2,  1, 1, 2},
+  {-2, -1, 1, 2},
+  { 2,  1, 3, 2},
+  {-2, -1, 3, 2},
+};
+
+static const int mask_params_med_heqw_interintra[1 << MASK_BITS_MED_INTERINTRA]
+                                                 [4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  { 0,  2, 0, 1},
+  { 0, -2, 0, 1},
+  { 0,  2, 0, 3},
+  { 0, -2, 0, 3},
+  { 2,  0, 1, 0},
+  {-2,  0, 1, 0},
+  { 2,  0, 3, 0},
+  {-2,  0, 3, 0},
+};
+
+static const int mask_params_big_hgtw_interintra[1 << MASK_BITS_BIG_INTERINTRA]
+                                                 [4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  {-1,  2, 2, 1},
+  { 1, -2, 2, 1},
+  {-1,  2, 2, 3},
+  { 1, -2, 2, 3},
+  { 1,  2, 2, 1},
+  {-1, -2, 2, 1},
+  { 1,  2, 2, 3},
+  {-1, -2, 2, 3},
+
+  {-2,  1, 1, 2},
+  { 2, -1, 1, 2},
+  {-2,  1, 3, 2},
+  { 2, -1, 3, 2},
+  { 2,  1, 1, 2},
+  {-2, -1, 1, 2},
+  { 2,  1, 3, 2},
+  {-2, -1, 3, 2},
+
+  { 0,  2, 0, 1},
+  { 0, -2, 0, 1},
+  { 0,  2, 0, 2},
+  { 0, -2, 0, 2},
+  { 0,  2, 0, 3},
+  { 0, -2, 0, 3},
+  { 2,  0, 2, 0},
+  {-2,  0, 2, 0},
+};
+
+static const int mask_params_big_hltw_interintra[1 << MASK_BITS_BIG_INTERINTRA]
+                                                 [4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  {-1,  2, 2, 1},
+  { 1, -2, 2, 1},
+  {-1,  2, 2, 3},
+  { 1, -2, 2, 3},
+  { 1,  2, 2, 1},
+  {-1, -2, 2, 1},
+  { 1,  2, 2, 3},
+  {-1, -2, 2, 3},
+
+  {-2,  1, 1, 2},
+  { 2, -1, 1, 2},
+  {-2,  1, 3, 2},
+  { 2, -1, 3, 2},
+  { 2,  1, 1, 2},
+  {-2, -1, 1, 2},
+  { 2,  1, 3, 2},
+  {-2, -1, 3, 2},
+
+  { 0,  2, 0, 2},
+  { 0, -2, 0, 2},
+  { 2,  0, 1, 0},
+  {-2,  0, 1, 0},
+  { 2,  0, 2, 0},
+  {-2,  0, 2, 0},
+  { 2,  0, 3, 0},
+  {-2,  0, 3, 0},
+};
+
+static const int mask_params_big_heqw_interintra[1 << MASK_BITS_BIG_INTERINTRA]
+                                                 [4] = {
+  {-1,  2, 2, 2},
+  { 1, -2, 2, 2},
+  {-2,  1, 2, 2},
+  { 2, -1, 2, 2},
+  { 2,  1, 2, 2},
+  {-2, -1, 2, 2},
+  { 1,  2, 2, 2},
+  {-1, -2, 2, 2},
+
+  {-1,  2, 2, 1},
+  { 1, -2, 2, 1},
+  {-1,  2, 2, 3},
+  { 1, -2, 2, 3},
+  { 1,  2, 2, 1},
+  {-1, -2, 2, 1},
+  { 1,  2, 2, 3},
+  {-1, -2, 2, 3},
+
+  {-2,  1, 1, 2},
+  { 2, -1, 1, 2},
+  {-2,  1, 3, 2},
+  { 2, -1, 3, 2},
+  { 2,  1, 1, 2},
+  {-2, -1, 1, 2},
+  { 2,  1, 3, 2},
+  {-2, -1, 3, 2},
+
+  { 0,  2, 0, 1},
+  { 0, -2, 0, 1},
+  { 0,  2, 0, 3},
+  { 0, -2, 0, 3},
+  { 2,  0, 1, 0},
+  {-2,  0, 1, 0},
+  { 2,  0, 3, 0},
+  {-2,  0, 3, 0},
+};
+
+static const int *get_mask_params_interintra(int mask_index,
+                                             BLOCK_SIZE sb_type,
+                                             int h, int w) {
+  const int *a;
+  const int mask_bits = get_mask_bits_interintra(sb_type);
+
+  if (mask_index == MASK_NONE_INTERINTRA)
+    return NULL;
+
+  if (mask_bits == MASK_BITS_SML_INTERINTRA) {
+    a = mask_params_sml_interintra[mask_index];
+  } else if (mask_bits == MASK_BITS_MED_INTERINTRA) {
+    if (h > w)
+      a = mask_params_med_hgtw_interintra[mask_index];
+    else if (h < w)
+      a = mask_params_med_hltw_interintra[mask_index];
+    else
+      a = mask_params_med_heqw_interintra[mask_index];
+  } else if (mask_bits == MASK_BITS_BIG_INTERINTRA) {
+    if (h > w)
+      a = mask_params_big_hgtw_interintra[mask_index];
+    else if (h < w)
+      a = mask_params_big_hltw_interintra[mask_index];
+    else
+      a = mask_params_big_heqw_interintra[mask_index];
+  } else {
+    assert(0);
+  }
+  return a;
+}
+
+void vp9_generate_masked_weight_interintra(int mask_index,
+                                           BLOCK_SIZE sb_type,
+                                           int h, int w,
+                                           uint8_t *mask, int stride) {
+  int i, j;
+  const int *a = get_mask_params_interintra(mask_index, sb_type, h, w);
+  if (!a) return;
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) {
+      int x = (j - (a[2] * w) / 4);
+      int y = (i - (a[3] * h) / 4);
+      int m = a[0] * x + a[1] * y;
+      mask[i * stride + j] = get_masked_weight_interintra(m);
+    }
+}
+
+void vp9_generate_hard_mask_interintra(int mask_index, BLOCK_SIZE sb_type,
+                            int h, int w, uint8_t *mask, int stride) {
+  int i, j;
+  const int *a = get_mask_params_interintra(mask_index, sb_type, h, w);
+  if (!a) return;
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) {
+      int x = (j - (a[2] * w) / 4);
+      int y = (i - (a[3] * h) / 4);
+      int m = a[0] * x + a[1] * y;
+      mask[i * stride + j] = get_hard_mask_interintra(m);
+    }
+}
+#endif
+
+static void combine_interintra(PREDICTION_MODE mode,
+#if CONFIG_MASKED_INTERINTRA
+                               int use_masked_interintra,
+                               int mask_index,
+                               BLOCK_SIZE bsize,
+#endif
+                               uint8_t *comppred,
+                               int compstride,
+                               uint8_t *interpred,
+                               int interstride,
+                               uint8_t *intrapred,
+                               int intrastride,
+                               int bw, int bh) {
+  static const int scale_bits = 8;
+  static const int scale_max = 256;
+  static const int scale_round = 127;
+  static const int weights1d[64] = {
+      128, 125, 122, 119, 116, 114, 111, 109,
+      107, 105, 103, 101,  99,  97,  96,  94,
+       93,  91,  90,  89,  88,  86,  85,  84,
+       83,  82,  81,  81,  80,  79,  78,  78,
+       77,  76,  76,  75,  75,  74,  74,  73,
+       73,  72,  72,  71,  71,  71,  70,  70,
+       70,  70,  69,  69,  69,  69,  68,  68,
+       68,  68,  68,  67,  67,  67,  67,  67,
+  };
+
+  int size = MAX(bw, bh);
+  int size_scale = (size >= 64 ? 1 :
+                    size == 32 ? 2 :
+                    size == 16 ? 4 :
+                    size == 8  ? 8 : 16);
+  int i, j;
+
+#if CONFIG_MASKED_INTERINTRA
+  if (use_masked_interintra && get_mask_bits_interintra(bsize)) {
+    uint8_t mask[4096];
+    vp9_generate_masked_weight_interintra(mask_index, bsize, bh, bw, mask, bw);
+    for (i = 0; i < bh; ++i) {
+      for (j = 0; j < bw; ++j) {
+        int m = mask[i * bw + j];
+        comppred[i * compstride + j] =
+            (intrapred[i * intrastride + j] * m +
+            interpred[i * interstride + j] *
+            ((1 << MASK_WEIGHT_BITS_INTERINTRA) - m) +
+            (1 << (MASK_WEIGHT_BITS_INTERINTRA - 1))) >>
+            MASK_WEIGHT_BITS_INTERINTRA;
+      }
+    }
+    return;
+  }
+#endif
+
+  switch (mode) {
+    case V_PRED:
+      for (i = 0; i < bh; ++i) {
+        for (j = 0; j < bw; ++j) {
+          int scale = weights1d[i * size_scale];
+            comppred[i * compstride + j] =
+              ((scale_max - scale) * interpred[i * interstride + j] +
+               scale * intrapred[i * intrastride + j] + scale_round)
+               >> scale_bits;
+        }
+      }
+     break;
+
+    case H_PRED:
+      for (i = 0; i < bh; ++i) {
+        for (j = 0; j < bw; ++j) {
+          int scale = weights1d[j * size_scale];
+            comppred[i * compstride + j] =
+              ((scale_max - scale) * interpred[i * interstride + j] +
+               scale * intrapred[i * intrastride + j] + scale_round)
+               >> scale_bits;
+        }
+      }
+     break;
+
+    case D63_PRED:
+    case D117_PRED:
+      for (i = 0; i < bh; ++i) {
+        for (j = 0; j < bw; ++j) {
+          int scale = (weights1d[i * size_scale] * 3 +
+                       weights1d[j * size_scale]) >> 2;
+            comppred[i * compstride + j] =
+              ((scale_max - scale) * interpred[i * interstride + j] +
+                  scale * intrapred[i * intrastride + j] + scale_round)
+                  >> scale_bits;
+        }
+      }
+     break;
+
+    case D207_PRED:
+    case D153_PRED:
+      for (i = 0; i < bh; ++i) {
+        for (j = 0; j < bw; ++j) {
+          int scale = (weights1d[j * size_scale] * 3 +
+                       weights1d[i * size_scale]) >> 2;
+            comppred[i * compstride + j] =
+              ((scale_max - scale) * interpred[i * interstride + j] +
+                  scale * intrapred[i * intrastride + j] + scale_round)
+                  >> scale_bits;
+        }
+      }
+     break;
+
+    case D135_PRED:
+      for (i = 0; i < bh; ++i) {
+        for (j = 0; j < bw; ++j) {
+          int scale = weights1d[(i < j ? i : j) * size_scale];
+            comppred[i * compstride + j] =
+              ((scale_max - scale) * interpred[i * interstride + j] +
+                  scale * intrapred[i * intrastride + j] + scale_round)
+                  >> scale_bits;
+        }
+      }
+     break;
+
+    case D45_PRED:
+      for (i = 0; i < bh; ++i) {
+        for (j = 0; j < bw; ++j) {
+          int scale = (weights1d[i * size_scale] +
+                       weights1d[j * size_scale]) >> 1;
+            comppred[i * compstride + j] =
+              ((scale_max - scale) * interpred[i * interstride + j] +
+                  scale * intrapred[i * intrastride + j] + scale_round)
+                  >> scale_bits;
+        }
+      }
+     break;
+
+    case TM_PRED:
+    case DC_PRED:
+    default:
+      for (i = 0; i < bh; ++i) {
+        for (j = 0; j < bw; ++j) {
+            comppred[i * compstride + j] = (interpred[i * interstride + j] +
+                intrapred[i * intrastride + j]) >> 1;
+        }
+      }
+      break;
+  }
+}
+
+
+static void build_intra_predictors_for_2nd_block_interintra
+                                  (const MACROBLOCKD *xd, const uint8_t *ref,
+                                   int ref_stride, uint8_t *dst, int dst_stride,
+                                   PREDICTION_MODE mode, TX_SIZE tx_size,
+                                   int up_available, int left_available,
+                                   int right_available, int bwltbh,
+                                   int x, int y, int plane) {
+  int i;
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, left_col, 64);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, above_data, 128 + 16);
+  uint8_t *above_row = above_data + 16;
+  const uint8_t *const_above_row = above_row;
+  const int bs = 4 << tx_size;
+  int frame_width, frame_height;
+  int x0, y0;
+  const struct macroblockd_plane *const pd = &xd->plane[plane];
+  const uint8_t *ref_fi;
+  int ref_stride_fi;
+
+  // 127 127 127 .. 127 127 127 127 127 127
+  // 129  A   B  ..  Y   Z
+  // 129  C   D  ..  W   X
+  // 129  E   F  ..  U   V
+  // 129  G   H  ..  S   T   T   T   T   T
+  // ..
+
+  once(init_intra_pred_fn_ptrs);
+
+  // Get current frame pointer, width and height.
+  if (plane == 0) {
+    frame_width = xd->cur_buf->y_width;
+    frame_height = xd->cur_buf->y_height;
+  } else {
+    frame_width = xd->cur_buf->uv_width;
+    frame_height = xd->cur_buf->uv_height;
+  }
+
+  // Get block position in current frame.
+  x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
+  y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
+
+  vpx_memset(left_col, 129, 64);
+
+  // left
+  if (left_available) {
+    if (bwltbh) {
+      ref_fi = ref;
+      ref_stride_fi = ref_stride;
+    } else {
+      ref_fi = dst;
+      ref_stride_fi = dst_stride;
+    }
+    if (xd->mb_to_bottom_edge < 0) {
+      /* slower path if the block needs border extension */
+      if (y0 + bs <= frame_height) {
+        for (i = 0; i < bs; ++i)
+          left_col[i] = ref_fi[i * ref_stride_fi - 1];
+      } else {
+        const int extend_bottom = frame_height - y0;
+        assert(extend_bottom >= 0);
+        for (i = 0; i < extend_bottom; ++i)
+          left_col[i] = ref_fi[i * ref_stride_fi - 1];
+        for (; i < bs; ++i)
+          left_col[i] = ref_fi[(extend_bottom - 1) * ref_stride_fi - 1];
+      }
+    } else {
+      /* faster path if the block does not need extension */
+      for (i = 0; i < bs; ++i)
+        left_col[i] = ref_fi[i * ref_stride_fi - 1];
+    }
+  }
+
+  // TODO(hkuang) do not extend 2*bs pixels for all modes.
+  // above
+  if (up_available) {
+    const uint8_t *above_ref;
+    if (bwltbh) {
+      ref_fi = dst;
+      ref_stride_fi = dst_stride;
+      above_row[-1] = left_available ? ref[-ref_stride-1] : 129;
+    } else {
+      ref_fi = ref;
+      ref_stride_fi = ref_stride;
+      above_row[-1] = ref[-ref_stride-1];
+    }
+    above_ref = ref_fi - ref_stride_fi;
+    if (xd->mb_to_right_edge < 0) {
+      /* slower path if the block needs border extension */
+      if (x0 + 2 * bs <= frame_width) {
+        if (right_available && bs == 4) {
+          vpx_memcpy(above_row, above_ref, 2 * bs);
+        } else {
+          vpx_memcpy(above_row, above_ref, bs);
+          vpx_memset(above_row + bs, above_row[bs - 1], bs);
+        }
+      } else if (x0 + bs <= frame_width) {
+        const int r = frame_width - x0;
+        if (right_available && bs == 4) {
+          vpx_memcpy(above_row, above_ref, r);
+          vpx_memset(above_row + r, above_row[r - 1],
+                     x0 + 2 * bs - frame_width);
+        } else {
+          vpx_memcpy(above_row, above_ref, bs);
+          vpx_memset(above_row + bs, above_row[bs - 1], bs);
+        }
+      } else if (x0 <= frame_width) {
+        const int r = frame_width - x0;
+        assert(r >= 0);
+        if (right_available && bs == 4) {
+          vpx_memcpy(above_row, above_ref, r);
+          vpx_memset(above_row + r, above_row[r - 1],
+                     x0 + 2 * bs - frame_width);
+        } else {
+          vpx_memcpy(above_row, above_ref, r);
+          vpx_memset(above_row + r, above_row[r - 1],
+                     x0 + 2 * bs - frame_width);
+        }
+      }
+    } else {
+      /* faster path if the block does not need extension */
+      if (bs == 4 && right_available && left_available) {
+        const_above_row = above_ref;
+      } else {
+        vpx_memcpy(above_row, above_ref, bs);
+        if (bs == 4 && right_available)
+          vpx_memcpy(above_row + bs, above_ref + bs, bs);
+        else
+          vpx_memset(above_row + bs, above_row[bs - 1], bs);
+      }
+    }
+  } else {
+    vpx_memset(above_row, 127, bs * 2);
+    above_row[-1] = 127;
+  }
+
+  // predict
+  if (mode == DC_PRED) {
+    dc_pred[left_available][up_available][tx_size](dst, dst_stride,
+                                                   const_above_row, left_col);
+  } else {
+    pred[mode][tx_size](dst, dst_stride, const_above_row, left_col);
+  }
+}
+
+// Break down rectangular intra prediction for joint spatio-temporal prediction
+// into two square intra predictions.
+static void build_intra_predictors_for_interintra(MACROBLOCKD *xd,
+                                           uint8_t *src, int src_stride,
+                                           uint8_t *pred_ptr, int stride,
+                                           PREDICTION_MODE mode,
+                                           int bw, int bh,
+                                           int up_available, int left_available,
+                                           int right_available, int plane) {
+  if (bw == bh) {
+    build_intra_predictors(xd, src, src_stride, pred_ptr, stride,
+                           mode, intra_size_log2_for_interintra(bw),
+                           up_available, left_available, right_available,
+                           0, 0, plane);
+  } else if (bw < bh) {
+    uint8_t *src_bottom = src + bw * src_stride;
+    uint8_t *pred_ptr_bottom = pred_ptr + bw * stride;
+    build_intra_predictors(xd, src, src_stride, pred_ptr, stride,
+                           mode, intra_size_log2_for_interintra(bw),
+                           up_available, left_available, right_available,
+                           0, 0, plane);
+    build_intra_predictors_for_2nd_block_interintra(xd, src_bottom, src_stride,
+                           pred_ptr_bottom, stride,
+                           mode, intra_size_log2_for_interintra(bw),
+                           up_available, left_available, 0, 1,
+                           0, bw, plane);
+  } else {
+    uint8_t *src_right = src + bh;
+    uint8_t *pred_ptr_right = pred_ptr + bh;
+    build_intra_predictors(xd, src, src_stride, pred_ptr, stride,
+                           mode, intra_size_log2_for_interintra(bh),
+                           up_available, left_available, 1,
+                           0, 0, plane);
+    build_intra_predictors_for_2nd_block_interintra(xd, src_right, src_stride,
+                           pred_ptr_right, stride,
+                           mode, intra_size_log2_for_interintra(bh),
+                           up_available, left_available, right_available, 0,
+                           bh, 0, plane);
+  }
+}
+
+void vp9_build_interintra_predictors_sby(MACROBLOCKD *xd,
+                                         uint8_t *ypred,
+                                         int ystride,
+                                         BLOCK_SIZE bsize) {
+  int bw = 4 << b_width_log2(bsize);
+  int bh = 4 << b_height_log2(bsize);
+  uint8_t intrapredictor[4096];
+  build_intra_predictors_for_interintra(
+      xd, xd->plane[0].dst.buf, xd->plane[0].dst.stride,
+      intrapredictor, bw,
+      xd->mi[0]->mbmi.interintra_mode, bw, bh,
+      xd->up_available, xd->left_available, 0, 0);
+  combine_interintra(xd->mi[0]->mbmi.interintra_mode,
+#if CONFIG_MASKED_INTERINTRA
+                     xd->mi[0]->mbmi.use_masked_interintra,
+                     xd->mi[0]->mbmi.interintra_mask_index,
+                     bsize,
+#endif
+                     xd->plane[0].dst.buf, xd->plane[0].dst.stride,
+                     ypred, ystride, intrapredictor, bw, bw, bh);
+}
+
+void vp9_build_interintra_predictors_sbuv(MACROBLOCKD *xd,
+                                          uint8_t *upred,
+                                          uint8_t *vpred,
+                                          int ustride, int vstride,
+                                          BLOCK_SIZE bsize) {
+  int bwl = b_width_log2(bsize), bw = 2 << bwl;
+  int bhl = b_height_log2(bsize), bh = 2 << bhl;
+  uint8_t uintrapredictor[1024];
+  uint8_t vintrapredictor[1024];
+  build_intra_predictors_for_interintra(
+      xd, xd->plane[1].dst.buf, xd->plane[1].dst.stride,
+      uintrapredictor, bw,
+      xd->mi[0]->mbmi.interintra_uv_mode, bw, bh,
+      xd->up_available, xd->left_available, 0, 1);
+  build_intra_predictors_for_interintra(
+      xd, xd->plane[2].dst.buf, xd->plane[1].dst.stride,
+      vintrapredictor, bw,
+      xd->mi[0]->mbmi.interintra_uv_mode, bw, bh,
+      xd->up_available, xd->left_available, 0, 2);
+  combine_interintra(xd->mi[0]->mbmi.interintra_uv_mode,
+#if CONFIG_MASKED_INTERINTRA
+                     xd->mi[0]->mbmi.use_masked_interintra,
+                     xd->mi[0]->mbmi.interintra_uv_mask_index,
+                     bsize,
+#endif
+                     xd->plane[1].dst.buf, xd->plane[1].dst.stride,
+                     upred, ustride, uintrapredictor, bw, bw, bh);
+  combine_interintra(xd->mi[0]->mbmi.interintra_uv_mode,
+#if CONFIG_MASKED_INTERINTRA
+                     xd->mi[0]->mbmi.use_masked_interintra,
+                     xd->mi[0]->mbmi.interintra_uv_mask_index,
+                     bsize,
+#endif
+                     xd->plane[2].dst.buf, xd->plane[2].dst.stride,
+                     vpred, vstride, vintrapredictor, bw, bw, bh);
+}
+
+void vp9_build_interintra_predictors(MACROBLOCKD *xd,
+                                     uint8_t *ypred,
+                                     uint8_t *upred,
+                                     uint8_t *vpred,
+                                     int ystride, int ustride, int vstride,
+                                     BLOCK_SIZE bsize) {
+  vp9_build_interintra_predictors_sby(xd, ypred, ystride, bsize);
+  vp9_build_interintra_predictors_sbuv(xd, upred, vpred,
+                                       ustride, vstride, bsize);
+}
+#endif

vp9/common/vp9_reconintra.h

@@ -20,9 +20,37 @@ extern "C" {
 
 void vp9_predict_intra_block(const MACROBLOCKD *xd, int block_idx, int bwl_in,
                              TX_SIZE tx_size, PREDICTION_MODE mode,
+#if CONFIG_FILTERINTRA
+                             int filterbit,
+#endif
                              const uint8_t *ref, int ref_stride,
                              uint8_t *dst, int dst_stride,
                              int aoff, int loff, int plane);
+#if CONFIG_INTERINTRA
+void vp9_build_interintra_predictors(MACROBLOCKD *xd,
+                                     uint8_t *ypred,
+                                     uint8_t *upred,
+                                     uint8_t *vpred,
+                                     int ystride,
+                                     int ustride,
+                                     int vstride,
+                                     BLOCK_SIZE bsize);
+void vp9_build_interintra_predictors_sby(MACROBLOCKD *xd,
+                                         uint8_t *ypred,
+                                         int ystride,
+                                         BLOCK_SIZE bsize);
+void vp9_build_interintra_predictors_sbuv(MACROBLOCKD *xd,
+                                          uint8_t *upred,
+                                          uint8_t *vpred,
+                                          int ustride, int vstride,
+                                          BLOCK_SIZE bsize);
+#if CONFIG_MASKED_INTERINTRA
+void vp9_generate_masked_weight_interintra(int mask_index,
+                                           BLOCK_SIZE sb_type,
+                                           int h, int w,
+                                           uint8_t *mask, int stride);
+#endif
+#endif
 #ifdef __cplusplus
 }  // extern "C"
 #endif

vp9/common/vp9_rtcd_defs.pl

@@ -506,6 +506,125 @@ specialize qw/vp9_sub_pixel_variance8x8/, "$sse2_x86inc", "$ssse3_x86inc";
 add_proto qw/unsigned int vp9_sub_pixel_avg_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
 specialize qw/vp9_sub_pixel_avg_variance8x8/, "$sse2_x86inc", "$ssse3_x86inc";
 
+if ((vpx_config("CONFIG_MASKED_INTERINTER") eq "yes") || ((vpx_config("CONFIG_INTERINTRA") eq "yes") && (vpx_config("CONFIG_MASKED_INTERINTRA") eq "yes"))) {
+add_proto qw/unsigned int vp9_masked_variance32x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance32x16/;
+
+add_proto qw/unsigned int vp9_masked_variance16x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masdctked_variance16x32/;
+
+add_proto qw/unsigned int vp9_masked_variance64x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance64x32/;
+
+add_proto qw/unsigned int vp9_masked_variance32x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance32x64/;
+
+add_proto qw/unsigned int vp9_masked_variance32x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance32x32/;
+
+add_proto qw/unsigned int vp9_masked_variance64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance64x64/;
+
+add_proto qw/unsigned int vp9_masked_variance16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance16x16/;
+
+add_proto qw/unsigned int vp9_masked_variance16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance16x8/;
+
+add_proto qw/unsigned int vp9_masked_variance8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance8x16/;
+
+add_proto qw/unsigned int vp9_masked_variance8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance8x8/;
+
+add_proto qw/unsigned int vp9_masked_variance8x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance8x4/;
+
+add_proto qw/unsigned int vp9_masked_variance4x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance4x8/;
+
+add_proto qw/unsigned int vp9_masked_variance4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_variance4x4/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance64x64/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance32x64/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance64x32/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance32x16/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance16x32/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance32x32/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance16x16/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance8x16/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance16x8/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance8x8/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance8x4/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance4x8/;
+
+add_proto qw/unsigned int vp9_masked_sub_pixel_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride, unsigned int *sse";
+specialize qw/vp9_masked_sub_pixel_variance4x4/;
+
+add_proto qw/unsigned int vp9_masked_sad64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad64x64/;
+
+add_proto qw/unsigned int vp9_masked_sad32x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad32x64/;
+
+add_proto qw/unsigned int vp9_masked_sad64x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad64x32/;
+
+add_proto qw/unsigned int vp9_masked_sad32x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad32x16/;
+
+add_proto qw/unsigned int vp9_masked_sad16x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad16x32/;
+
+add_proto qw/unsigned int vp9_masked_sad32x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad32x32/;
+
+add_proto qw/unsigned int vp9_masked_sad16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad16x16/;
+
+add_proto qw/unsigned int vp9_masked_sad16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad16x8/;
+
+add_proto qw/unsigned int vp9_masked_sad8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad8x16/;
+
+add_proto qw/unsigned int vp9_masked_sad8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad8x8/;
+
+add_proto qw/unsigned int vp9_masked_sad8x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad8x4/;
+
+add_proto qw/unsigned int vp9_masked_sad4x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad4x8/;
+
+add_proto qw/unsigned int vp9_masked_sad4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *mask, int mask_stride";
+specialize qw/vp9_masked_sad4x4/;
+}
+
 # TODO(jingning): need to convert 8x4/4x8 functions into mmx/sse form
 add_proto qw/unsigned int vp9_sub_pixel_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
 specialize qw/vp9_sub_pixel_variance8x4/, "$sse2_x86inc", "$ssse3_x86inc";

vp9/common/vp9_scale.h

@@ -46,8 +46,8 @@ static INLINE int vp9_is_valid_scale(const struct scale_factors *sf) {
 }
 
 static INLINE int vp9_is_scaled(const struct scale_factors *sf) {
-  return sf->x_scale_fp != REF_NO_SCALE ||
-         sf->y_scale_fp != REF_NO_SCALE;
+  return vp9_is_valid_scale(sf) &&
+         (sf->x_scale_fp != REF_NO_SCALE || sf->y_scale_fp != REF_NO_SCALE);
 }
 
 #ifdef __cplusplus

vp9/decoder/vp9_decodeframe.c

@@ -254,11 +254,24 @@ static void predict_and_reconstruct_intra_block(int plane, int block,
                                             : mi->mbmi.uv_mode;
   int x, y;
   uint8_t *dst;
+#if CONFIG_FILTERINTRA
+  int fbit;
+  if (plane == 0)
+    if (mi->mbmi.sb_type < BLOCK_8X8)
+      fbit = mi->b_filter_info[block];
+    else
+      fbit = is_filter_enabled(tx_size) ? mi->mbmi.filterbit : 0;
+  else
+    fbit = is_filter_enabled(tx_size) ? mi->mbmi.uv_filterbit : 0;
+#endif
   txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y);
   dst = &pd->dst.buf[4 * y * pd->dst.stride + 4 * x];
 
   vp9_predict_intra_block(xd, block >> (tx_size << 1),
                           b_width_log2(plane_bsize), tx_size, mode,
+#if CONFIG_FILTERINTRA
+                          fbit,
+#endif
                           dst, pd->dst.stride, dst, pd->dst.stride,
                           x, y, plane);
 
@@ -322,6 +335,84 @@ static MB_MODE_INFO *set_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd,
   return &xd->mi[0]->mbmi;
 }
 
+#if CONFIG_SUPERTX
+static void set_offsets_extend(VP9_COMMON *const cm, MACROBLOCKD *const xd,
+                               const TileInfo *const tile,
+                               BLOCK_SIZE top_bsize,
+                               int mi_row, int mi_col,
+                               int mi_row_ori, int mi_col_ori) {
+  const int bw = num_8x8_blocks_wide_lookup[top_bsize];
+  const int bh = num_8x8_blocks_high_lookup[top_bsize];
+  const int offset = mi_row * cm->mi_stride + mi_col;
+
+  xd->mi = cm->mi_grid_visible + offset;
+  xd->mi[0] = &cm->mi[offset];
+
+  set_mi_row_col(xd, tile, mi_row_ori, bh, mi_col_ori, bw,
+                 cm->mi_rows, cm->mi_cols);
+}
+
+static void set_mb_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd,
+                           const TileInfo *const tile,
+                           BLOCK_SIZE bsize, int mi_row, int mi_col) {
+  const int bw = num_8x8_blocks_wide_lookup[bsize];
+  const int bh = num_8x8_blocks_high_lookup[bsize];
+  const int x_mis = MIN(bw, cm->mi_cols - mi_col);
+  const int y_mis = MIN(bh, cm->mi_rows - mi_row);
+  const int offset = mi_row * cm->mi_stride + mi_col;
+  int x, y;
+
+  xd->mi = cm->mi_grid_visible + offset;
+  xd->mi[0] = &cm->mi[offset];
+  xd->mi[0]->mbmi.sb_type = bsize;
+  for (y = 0; y < y_mis; ++y)
+    for (x = !y; x < x_mis; ++x)
+      xd->mi[y * cm->mi_stride + x] = xd->mi[0];
+
+  set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
+}
+
+static void set_offsets_topblock(VP9_COMMON *const cm, MACROBLOCKD *const xd,
+                                 const TileInfo *const tile,
+                                 BLOCK_SIZE bsize, int mi_row, int mi_col) {
+  const int bw = num_8x8_blocks_wide_lookup[bsize];
+  const int bh = num_8x8_blocks_high_lookup[bsize];
+  const int offset = mi_row * cm->mi_stride + mi_col;
+
+  xd->mi = cm->mi_grid_visible + offset;
+  xd->mi[0] = &cm->mi[offset];
+
+  set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
+
+  vp9_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
+}
+
+static void set_param_topblock(VP9_COMMON *const cm,  MACROBLOCKD *const xd,
+                              BLOCK_SIZE bsize, int mi_row, int mi_col,
+#if CONFIG_EXT_TX
+                              int txfm,
+#endif
+                              int skip) {
+  const int bw = num_8x8_blocks_wide_lookup[bsize];
+  const int bh = num_8x8_blocks_high_lookup[bsize];
+  const int x_mis = MIN(bw, cm->mi_cols - mi_col);
+  const int y_mis = MIN(bh, cm->mi_rows - mi_row);
+  const int offset = mi_row * cm->mi_stride + mi_col;
+  int x, y;
+
+  xd->mi = cm->mi_grid_visible + offset;
+  xd->mi[0] = &cm->mi[offset];
+
+  for (y = 0; y < y_mis; ++y)
+    for (x = 0; x < x_mis; ++x) {
+      xd->mi[y * cm->mi_stride + x]->mbmi.skip = skip;
+#if CONFIG_EXT_TX
+      xd->mi[y * cm->mi_stride + x]->mbmi.ext_txfrm = txfm;
+#endif
+    }
+}
+#endif
+
 static void set_ref(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                     int idx, int mi_row, int mi_col) {
   MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
@@ -335,14 +426,246 @@ static void set_ref(VP9_COMMON *const cm, MACROBLOCKD *const xd,
   xd->corrupted |= ref_buffer->buf->corrupted;
 }
 
+#if CONFIG_SUPERTX
+static void dec_predict_b_extend(VP9_COMMON *const cm, MACROBLOCKD *const xd,
+                                 const TileInfo *const tile,
+                                 int mi_row, int mi_col,
+                                 int mi_row_ori, int mi_col_ori,
+                                 BLOCK_SIZE top_bsize) {
+  set_offsets_extend(cm, xd, tile, top_bsize, mi_row, mi_col,
+                     mi_row_ori, mi_col_ori);
+
+  set_ref(cm, xd, 0, mi_row_ori, mi_col_ori);
+  if (has_second_ref(&xd->mi[0]->mbmi))
+    set_ref(cm, xd, 1, mi_row_ori, mi_col_ori);
+  xd->mi[0]->mbmi.tx_size = b_width_log2(top_bsize);
+#if !CONFIG_MASKED_INTERINTER
+  vp9_dec_build_inter_predictors_sb(xd, mi_row_ori, mi_col_ori, top_bsize);
+#else
+  vp9_dec_build_inter_predictors_sb_extend(xd, mi_row, mi_col,
+                                           mi_row_ori, mi_col_ori, top_bsize);
+#endif
+}
+
+static void dec_predict_b_sub8x8_extend(VP9_COMMON *const cm,
+                                        MACROBLOCKD *const xd,
+                                        const TileInfo *const tile,
+                                        int mi_row, int mi_col,
+                                        int mi_row_ori, int mi_col_ori,
+                                        BLOCK_SIZE top_bsize,
+                                        PARTITION_TYPE partition) {
+  set_offsets_extend(cm, xd, tile, top_bsize, mi_row, mi_col,
+                     mi_row_ori, mi_col_ori);
+
+  set_ref(cm, xd, 0, mi_row_ori, mi_col_ori);
+  if (has_second_ref(&xd->mi[0]->mbmi))
+    set_ref(cm, xd, 1, mi_row_ori, mi_col_ori);
+  xd->mi[0]->mbmi.tx_size = b_width_log2(top_bsize);
+  vp9_dec_build_inter_predictors_sby_sub8x8_extend(xd, mi_row, mi_col,
+                                                   mi_row_ori, mi_col_ori,
+                                                   top_bsize, partition);
+  vp9_dec_build_inter_predictors_sbuv_sub8x8_extend(xd,
+#if CONFIG_MASKED_INTERINTER
+                                                    mi_row, mi_col,
+#endif
+                                                    mi_row_ori, mi_col_ori,
+                                                    top_bsize);
+}
+
+static void dec_predict_sb_complex(VP9_COMMON *const cm, MACROBLOCKD *const xd,
+                                   const TileInfo *const tile,
+                                   int mi_row, int mi_col,
+                                   int mi_row_ori, int mi_col_ori,
+                                   BLOCK_SIZE bsize, BLOCK_SIZE top_bsize,
+                                   uint8_t *dst_buf[3], int dst_stride[3]) {
+  const int bsl = b_width_log2(bsize), hbs = (1 << bsl) / 4;
+  PARTITION_TYPE partition;
+  BLOCK_SIZE subsize;
+  MB_MODE_INFO *mbmi;
+  int i, offset = mi_row * cm->mi_stride + mi_col;
+
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, tmp_buf1, MAX_MB_PLANE * 32 * 32);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, tmp_buf2, MAX_MB_PLANE * 32 * 32);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t, tmp_buf3, MAX_MB_PLANE * 32 * 32);
+  uint8_t *dst_buf1[3] = {tmp_buf1, tmp_buf1 + 32 * 32, tmp_buf1 + 2 * 32 * 32};
+  uint8_t *dst_buf2[3] = {tmp_buf2, tmp_buf2 + 32 * 32, tmp_buf2 + 2 * 32 * 32};
+  uint8_t *dst_buf3[3] = {tmp_buf3, tmp_buf3 + 32 * 32, tmp_buf3 + 2 * 32 * 32};
+  int dst_stride1[3] = {32, 32, 32};
+  int dst_stride2[3] = {32, 32, 32};
+  int dst_stride3[3] = {32, 32, 32};
+
+  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
+    return;
+
+  xd->mi = cm->mi_grid_visible + offset;
+  xd->mi[0] = &cm->mi[offset];
+  mbmi = &xd->mi[0]->mbmi;
+  partition = partition_lookup[bsl][mbmi->sb_type];
+  subsize = get_subsize(bsize, partition);
+
+  for (i = 0; i < MAX_MB_PLANE; i++) {
+    xd->plane[i].dst.buf = dst_buf[i];
+    xd->plane[i].dst.stride = dst_stride[i];
+  }
+
+  switch (partition) {
+    case PARTITION_NONE:
+      assert(bsize < top_bsize);
+      dec_predict_b_extend(cm, xd, tile, mi_row, mi_col, mi_row_ori, mi_col_ori,
+                           top_bsize);
+      break;
+    case PARTITION_HORZ:
+      if (bsize > BLOCK_8X8) {
+        dec_predict_b_extend(cm, xd, tile, mi_row, mi_col, mi_row_ori,
+                             mi_col_ori, top_bsize);
+      } else {
+        dec_predict_b_sub8x8_extend(cm, xd, tile, mi_row, mi_col,
+                                    mi_row_ori, mi_col_ori,
+                                    top_bsize, partition);
+      }
+      if (mi_row + hbs < cm->mi_rows && bsize > BLOCK_8X8) {
+        for (i = 0; i < MAX_MB_PLANE; i++) {
+          xd->plane[i].dst.buf = tmp_buf1 + i * 32 * 32;
+          xd->plane[i].dst.stride = 32;
+        }
+        dec_predict_b_extend(cm, xd, tile, mi_row + hbs, mi_col,
+                             mi_row_ori, mi_col_ori, top_bsize);
+        for (i = 0; i < MAX_MB_PLANE; i++) {
+          xd->plane[i].dst.buf = dst_buf[i];
+          xd->plane[i].dst.stride = dst_stride[i];
+          vp9_build_masked_inter_predictor_complex(dst_buf[i], dst_stride[i],
+                                                   dst_buf1[i], dst_stride1[i],
+                                                   i,
+                                                   mi_row, mi_col,
+                                                   mi_row_ori, mi_col_ori,
+                                                   bsize, top_bsize,
+                                                   PARTITION_HORZ);
+        }
+      }
+      break;
+    case PARTITION_VERT:
+      if (bsize > BLOCK_8X8) {
+        dec_predict_b_extend(cm, xd, tile, mi_row, mi_col, mi_row_ori,
+                             mi_col_ori, top_bsize);
+      } else {
+        dec_predict_b_sub8x8_extend(cm, xd, tile, mi_row, mi_col,
+                                    mi_row_ori, mi_col_ori,
+                                    top_bsize, partition);
+      }
+      if (mi_col + hbs < cm->mi_cols && bsize > BLOCK_8X8) {
+        for (i = 0; i < MAX_MB_PLANE; i++) {
+          xd->plane[i].dst.buf = tmp_buf1 + i * 32 * 32;
+          xd->plane[i].dst.stride = 32;
+        }
+        dec_predict_b_extend(cm, xd, tile, mi_row, mi_col + hbs, mi_row_ori,
+                             mi_col_ori, top_bsize);
+        for (i = 0; i < MAX_MB_PLANE; i++) {
+          xd->plane[i].dst.buf = dst_buf[i];
+          xd->plane[i].dst.stride = dst_stride[i];
+          vp9_build_masked_inter_predictor_complex(dst_buf[i], dst_stride[i],
+                                                   dst_buf1[i], dst_stride1[i],
+                                                   i,
+                                                   mi_row, mi_col,
+                                                   mi_row_ori, mi_col_ori,
+                                                   bsize, top_bsize,
+                                                   PARTITION_VERT);
+        }
+      }
+      break;
+    case PARTITION_SPLIT:
+      if (bsize == BLOCK_8X8) {
+        dec_predict_b_sub8x8_extend(cm, xd, tile, mi_row, mi_col,
+                                    mi_row_ori, mi_col_ori,
+                                    top_bsize, partition);
+      } else {
+        dec_predict_sb_complex(cm, xd, tile, mi_row, mi_col,
+                               mi_row_ori, mi_col_ori, subsize, top_bsize,
+                               dst_buf, dst_stride);
+        if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols)
+          dec_predict_sb_complex(cm, xd, tile, mi_row, mi_col + hbs,
+                                 mi_row_ori, mi_col_ori, subsize, top_bsize,
+                                 dst_buf1, dst_stride1);
+        if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols)
+          dec_predict_sb_complex(cm, xd, tile, mi_row + hbs, mi_col,
+                                 mi_row_ori, mi_col_ori, subsize, top_bsize,
+                                 dst_buf2, dst_stride2);
+        if (mi_row + hbs < cm->mi_rows && mi_col + hbs < cm->mi_cols)
+          dec_predict_sb_complex(cm, xd, tile, mi_row + hbs, mi_col + hbs,
+                                 mi_row_ori, mi_col_ori, subsize, top_bsize,
+                                 dst_buf3, dst_stride3);
+        for (i = 0; i < MAX_MB_PLANE; i++) {
+          if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols) {
+            vp9_build_masked_inter_predictor_complex(dst_buf[i], dst_stride[i],
+                                                     dst_buf1[i],
+                                                     dst_stride1[i],
+                                                     i, mi_row, mi_col,
+                                                     mi_row_ori, mi_col_ori,
+                                                     bsize, top_bsize,
+                                                     PARTITION_VERT);
+            if (mi_row + hbs < cm->mi_rows) {
+              vp9_build_masked_inter_predictor_complex(dst_buf2[i],
+                                                       dst_stride2[i],
+                                                       dst_buf3[i],
+                                                       dst_stride3[i],
+                                                       i, mi_row, mi_col,
+                                                       mi_row_ori, mi_col_ori,
+                                                       bsize, top_bsize,
+                                                       PARTITION_VERT);
+              vp9_build_masked_inter_predictor_complex(dst_buf[i],
+                                                       dst_stride[i],
+                                                       dst_buf2[i],
+                                                       dst_stride2[i],
+                                                       i, mi_row, mi_col,
+                                                       mi_row_ori, mi_col_ori,
+                                                       bsize, top_bsize,
+                                                       PARTITION_HORZ);
+            }
+          } else if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols) {
+            vp9_build_masked_inter_predictor_complex(dst_buf[i],
+                                                     dst_stride[i],
+                                                     dst_buf2[i],
+                                                     dst_stride2[i],
+                                                     i, mi_row, mi_col,
+                                                     mi_row_ori, mi_col_ori,
+                                                     bsize, top_bsize,
+                                                     PARTITION_HORZ);
+          }
+        }
+      }
+      break;
+    default:
+      assert(0);
+  }
+}
+#endif
+
 static void decode_block(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                          const TileInfo *const tile,
+#if CONFIG_SUPERTX
+                         int supertx_enabled,
+#endif
                          int mi_row, int mi_col,
                          vp9_reader *r, BLOCK_SIZE bsize) {
   const int less8x8 = bsize < BLOCK_8X8;
+#if !CONFIG_SUPERTX
   MB_MODE_INFO *mbmi = set_offsets(cm, xd, tile, bsize, mi_row, mi_col);
-  vp9_read_mode_info(cm, xd, tile, mi_row, mi_col, r);
+#else
+  MB_MODE_INFO *mbmi;
+  if (!supertx_enabled) {
+    mbmi = set_offsets(cm, xd, tile, bsize, mi_row, mi_col);
+  } else {
+    set_mb_offsets(cm, xd, tile, bsize, mi_row, mi_col);
+  }
+#endif
+  vp9_read_mode_info(cm, xd, tile,
+#if CONFIG_SUPERTX
+                     supertx_enabled,
+#endif
+                     mi_row, mi_col, r);
 
+#if CONFIG_SUPERTX
+  if (!supertx_enabled) {
+#endif
   if (less8x8)
     bsize = BLOCK_8X8;
 
@@ -376,6 +699,9 @@ static void decode_block(VP9_COMMON *const cm, MACROBLOCKD *const xd,
         mbmi->skip = 1;  // skip loopfilter
     }
   }
+#if CONFIG_SUPERTX
+  }
+#endif
 
   xd->corrupted |= vp9_reader_has_error(r);
 }
@@ -406,45 +732,161 @@ static PARTITION_TYPE read_partition(VP9_COMMON *cm, MACROBLOCKD *xd, int hbs,
 
 static void decode_partition(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                              const TileInfo *const tile,
+#if CONFIG_SUPERTX
+                             int read_token, int supertx_enabled,
+#endif
                              int mi_row, int mi_col,
                              vp9_reader* r, BLOCK_SIZE bsize) {
   const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2;
   PARTITION_TYPE partition;
   BLOCK_SIZE subsize;
+#if CONFIG_SUPERTX
+  int skip = 0;
+#if CONFIG_EXT_TX
+  int txfm = 0;
+#endif
+#endif
 
   if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
     return;
 
   partition = read_partition(cm, xd, hbs, mi_row, mi_col, bsize, r);
   subsize = get_subsize(bsize, partition);
+#if CONFIG_SUPERTX
+  if (cm->frame_type != KEY_FRAME &&
+      partition != PARTITION_NONE &&
+      bsize <= BLOCK_32X32 &&
+      !supertx_enabled) {
+    TX_SIZE supertx_size = b_width_log2(bsize);
+    if (partition == PARTITION_SPLIT) {
+      supertx_enabled = vp9_read(r, cm->fc.supertxsplit_prob[supertx_size]);
+      cm->counts.supertxsplit[supertx_size][supertx_enabled]++;
+    } else {
+      supertx_enabled = vp9_read(r, cm->fc.supertx_prob[supertx_size]);
+      cm->counts.supertx[supertx_size][supertx_enabled]++;
+    }
+  }
+  if (supertx_enabled && read_token) {
+    int offset = mi_row * cm->mi_stride + mi_col;
+    xd->mi = cm->mi_grid_visible + offset;
+    xd->mi[0] = &cm->mi[offset];
+    set_mi_row_col(xd, tile, mi_row, num_8x8_blocks_high_lookup[bsize],
+                   mi_col, num_8x8_blocks_wide_lookup[bsize],
+                   cm->mi_rows, cm->mi_cols);
+    set_skip_context(xd, mi_row, mi_col);
+    // Here we assume mbmi->segment_id = 0
+    skip = read_skip(cm, xd, 0, r);
+    if (skip)
+      reset_skip_context(xd, bsize);
+#if CONFIG_EXT_TX
+    if (bsize <= BLOCK_16X16 && !skip) {
+      txfm = vp9_read(r, cm->fc.ext_tx_prob);
+      if (!cm->frame_parallel_decoding_mode)
+        ++cm->counts.ext_tx[txfm];
+    }
+#endif
+  }
+#endif
   if (subsize < BLOCK_8X8) {
-    decode_block(cm, xd, tile, mi_row, mi_col, r, subsize);
+    decode_block(cm, xd, tile,
+#if CONFIG_SUPERTX
+                 supertx_enabled,
+#endif
+                 mi_row, mi_col, r, subsize);
   } else {
     switch (partition) {
       case PARTITION_NONE:
-        decode_block(cm, xd, tile, mi_row, mi_col, r, subsize);
+        decode_block(cm, xd, tile,
+#if CONFIG_SUPERTX
+                     supertx_enabled,
+#endif
+                     mi_row, mi_col, r, subsize);
         break;
       case PARTITION_HORZ:
-        decode_block(cm, xd, tile, mi_row, mi_col, r, subsize);
+        decode_block(cm, xd, tile,
+#if CONFIG_SUPERTX
+                     supertx_enabled,
+#endif
+                     mi_row, mi_col, r, subsize);
         if (mi_row + hbs < cm->mi_rows)
-          decode_block(cm, xd, tile, mi_row + hbs, mi_col, r, subsize);
+          decode_block(cm, xd, tile,
+#if CONFIG_SUPERTX
+                       supertx_enabled,
+#endif
+                       mi_row + hbs, mi_col, r, subsize);
         break;
       case PARTITION_VERT:
-        decode_block(cm, xd, tile, mi_row, mi_col, r, subsize);
+        decode_block(cm, xd, tile,
+#if CONFIG_SUPERTX
+                     supertx_enabled,
+#endif
+                     mi_row, mi_col, r, subsize);
         if (mi_col + hbs < cm->mi_cols)
-          decode_block(cm, xd, tile, mi_row, mi_col + hbs, r, subsize);
+          decode_block(cm, xd, tile,
+#if CONFIG_SUPERTX
+                       supertx_enabled,
+#endif
+                       mi_row, mi_col + hbs, r, subsize);
         break;
       case PARTITION_SPLIT:
-        decode_partition(cm, xd, tile, mi_row,       mi_col,       r, subsize);
-        decode_partition(cm, xd, tile, mi_row,       mi_col + hbs, r, subsize);
-        decode_partition(cm, xd, tile, mi_row + hbs, mi_col,       r, subsize);
-        decode_partition(cm, xd, tile, mi_row + hbs, mi_col + hbs, r, subsize);
+        decode_partition(cm, xd, tile,
+#if CONFIG_SUPERTX
+                         !supertx_enabled, supertx_enabled,
+#endif
+                         mi_row,       mi_col,       r, subsize);
+        decode_partition(cm, xd, tile,
+#if CONFIG_SUPERTX
+                         !supertx_enabled, supertx_enabled,
+#endif
+                         mi_row,       mi_col + hbs, r, subsize);
+        decode_partition(cm, xd, tile,
+#if CONFIG_SUPERTX
+                         !supertx_enabled, supertx_enabled,
+#endif
+                         mi_row + hbs, mi_col,       r, subsize);
+        decode_partition(cm, xd, tile,
+#if CONFIG_SUPERTX
+                         !supertx_enabled, supertx_enabled,
+#endif
+                         mi_row + hbs, mi_col + hbs, r, subsize);
         break;
       default:
         assert(0 && "Invalid partition type");
     }
   }
 
+#if CONFIG_SUPERTX
+  if (supertx_enabled && read_token) {
+    uint8_t *dst_buf[3];
+    int dst_stride[3], i;
+
+    vp9_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
+    for (i = 0; i < MAX_MB_PLANE; i++) {
+      dst_buf[i] = xd->plane[i].dst.buf;
+      dst_stride[i] = xd->plane[i].dst.stride;
+    }
+    dec_predict_sb_complex(cm, xd, tile, mi_row, mi_col, mi_row, mi_col,
+                           bsize, bsize, dst_buf, dst_stride);
+
+    if (!skip) {
+      int eobtotal = 0;
+      struct inter_args arg = { cm, xd, r, &eobtotal };
+      set_offsets_topblock(cm, xd, tile, bsize, mi_row, mi_col);
+#if CONFIG_EXT_TX
+      xd->mi[0]->mbmi.ext_txfrm = txfm;
+#endif
+      vp9_foreach_transformed_block(xd, bsize, reconstruct_inter_block, &arg);
+      if (!(subsize < BLOCK_8X8) && eobtotal == 0)
+        skip = 1;
+    }
+    set_param_topblock(cm, xd, bsize, mi_row, mi_col,
+#if CONFIG_EXT_TX
+                       txfm,
+#endif
+                       skip);
+  }
+#endif
+
   // update partition context
   if (bsize >= BLOCK_8X8 &&
       (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT))
@@ -685,6 +1127,10 @@ static void setup_tile_info(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
   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)
@@ -838,7 +1284,11 @@ static const uint8_t *decode_tiles(VP9Decoder *pbi,
         vp9_zero(tile_data->xd.left_seg_context);
         for (mi_col = tile.mi_col_start; mi_col < tile.mi_col_end;
              mi_col += MI_BLOCK_SIZE) {
-          decode_partition(tile_data->cm, &tile_data->xd, &tile, mi_row, mi_col,
+          decode_partition(tile_data->cm, &tile_data->xd, &tile,
+#if CONFIG_SUPERTX
+                           1, 0,
+#endif
+                           mi_row, mi_col,
                            &tile_data->bit_reader, BLOCK_64X64);
         }
       }
@@ -892,6 +1342,9 @@ static int tile_worker_hook(void *arg1, void *arg2) {
     for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
          mi_col += MI_BLOCK_SIZE) {
       decode_partition(tile_data->cm, &tile_data->xd, tile,
+#if CONFIG_SUPERTX
+                       1, 0,
+#endif
                        mi_row, mi_col, &tile_data->bit_reader, BLOCK_64X64);
     }
   }
@@ -1077,7 +1530,7 @@ static size_t read_uncompressed_header(VP9Decoder *pbi,
     // Show an existing frame directly.
     const int frame_to_show = cm->ref_frame_map[vp9_rb_read_literal(rb, 3)];
 
-    if (cm->frame_bufs[frame_to_show].ref_count < 1)
+    if (frame_to_show < 0 || cm->frame_bufs[frame_to_show].ref_count < 1)
       vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
                          "Buffer %d does not contain a decoded frame",
                          frame_to_show);
@@ -1242,6 +1695,62 @@ static int read_compressed_header(VP9Decoder *pbi, const uint8_t *data,
         vp9_diff_update_prob(&r, &fc->partition_prob[j][i]);
 
     read_mv_probs(nmvc, cm->allow_high_precision_mv, &r);
+
+#if CONFIG_EXT_TX
+    vp9_diff_update_prob(&r, &fc->ext_tx_prob);
+#endif
+
+#if CONFIG_MASKED_INTERINTER
+    if (cm->reference_mode != SINGLE_REFERENCE) {
+      cm->use_masked_interinter = vp9_read_bit(&r);
+      if (cm->use_masked_interinter) {
+        for (i = 0; i < BLOCK_SIZES; i++) {
+          if (get_mask_bits(i))
+            vp9_diff_update_prob(&r, &fc->masked_interinter_prob[i]);
+        }
+      }
+    } else {
+      cm->use_masked_interinter = 0;
+    }
+#endif
+
+#if CONFIG_INTERINTRA
+    if (cm->reference_mode != COMPOUND_REFERENCE) {
+      cm->use_interintra = vp9_read_bit(&r);
+      if (cm->use_interintra) {
+        for (i = 0; i < BLOCK_SIZES; i++) {
+          if (is_interintra_allowed(i)) {
+            vp9_diff_update_prob(&r, &fc->interintra_prob[i]);
+          }
+        }
+#if CONFIG_MASKED_INTERINTRA
+        cm->use_masked_interintra = vp9_read_bit(&r);
+        if (cm->use_masked_interintra) {
+          for (i = 0; i < BLOCK_SIZES; i++) {
+            if (is_interintra_allowed(i) && get_mask_bits_interintra(i))
+              vp9_diff_update_prob(&r, &fc->masked_interintra_prob[i]);
+          }
+        }
+      } else {
+        cm->use_masked_interintra = 0;
+#endif
+      }
+    } else {
+      cm->use_interintra = 0;
+#if CONFIG_MASKED_INTERINTRA
+      cm->use_masked_interintra = 0;
+#endif
+    }
+#endif
+
+#if CONFIG_COPY_CODING
+    for (j = 0; j < COPY_MODE_CONTEXTS; j++) {
+      for (i = 0; i < 1; i++)
+        vp9_diff_update_prob(&r, &fc->copy_mode_probs_l2[j][i]);
+      for (i = 0; i < 2; i++)
+        vp9_diff_update_prob(&r, &fc->copy_mode_probs[j][i]);
+    }
+#endif
   }
 
   return vp9_reader_has_error(&r);
@@ -1293,6 +1802,10 @@ static void debug_check_frame_counts(const VP9_COMMON *const cm) {
   assert(!memcmp(&cm->counts.tx, &zero_counts.tx, sizeof(cm->counts.tx)));
   assert(!memcmp(cm->counts.skip, zero_counts.skip, sizeof(cm->counts.skip)));
   assert(!memcmp(&cm->counts.mv, &zero_counts.mv, sizeof(cm->counts.mv)));
+#if CONFIG_EXT_TX
+  assert(!memcmp(cm->counts.ext_tx, zero_counts.ext_tx,
+                 sizeof(cm->counts.ext_tx)));
+#endif
 }
 #endif  // NDEBUG
 

vp9/decoder/vp9_decodemv.c

@@ -54,6 +54,36 @@ static PREDICTION_MODE read_inter_mode(VP9_COMMON *cm, vp9_reader *r, int ctx) {
   return NEARESTMV + mode;
 }
 
+#if CONFIG_COPY_CODING
+static COPY_MODE read_copy_mode(VP9_COMMON *cm, vp9_reader *r,
+                                int num_candidate, int ctx) {
+  COPY_MODE mode;
+
+  switch (num_candidate) {
+    case 0:
+      assert(0);
+      break;
+    case 1:
+      mode = REF0;
+      break;
+    case 2:
+      mode = REF0 + vp9_read_tree(r, vp9_copy_mode_tree_l2,
+                                  cm->fc.copy_mode_probs_l2[ctx]);
+      if (!cm->frame_parallel_decoding_mode)
+          ++cm->counts.copy_mode_l2[ctx][mode - REF0];
+      break;
+    default:
+      mode = REF0 + vp9_read_tree(r, vp9_copy_mode_tree,
+                                  cm->fc.copy_mode_probs[ctx]);
+      if (!cm->frame_parallel_decoding_mode)
+          ++cm->counts.copy_mode[ctx][mode - REF0];
+      break;
+  }
+
+  return mode;
+}
+#endif
+
 static int read_segment_id(vp9_reader *r, const struct segmentation *seg) {
   return vp9_read_tree(r, vp9_segment_tree, seg->tree_probs);
 }
@@ -144,7 +174,11 @@ static int read_inter_segment_id(VP9_COMMON *const cm, MACROBLOCKD *const xd,
   return segment_id;
 }
 
+#if CONFIG_SUPERTX
+int read_skip(VP9_COMMON *cm, const MACROBLOCKD *xd,
+#else
 static int read_skip(VP9_COMMON *cm, const MACROBLOCKD *xd,
+#endif
                      int segment_id, vp9_reader *r) {
   if (vp9_segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
     return 1;
@@ -175,29 +209,85 @@ static void read_intra_frame_mode_info(VP9_COMMON *const cm,
 
   switch (bsize) {
     case BLOCK_4X4:
+#if !CONFIG_FILTERINTRA
       for (i = 0; i < 4; ++i)
+#else
+      for (i = 0; i < 4; ++i) {
+#endif
         mi->bmi[i].as_mode =
             read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, i));
+#if CONFIG_FILTERINTRA
+        if (is_filter_allowed(mi->bmi[i].as_mode))
+          mi->b_filter_info[i] =
+              vp9_read(r, cm->fc.filterintra_prob[0][mi->bmi[i].as_mode]);
+        else
+          mi->b_filter_info[i] = 0;
+      }
+      mbmi->filterbit = mi->b_filter_info[3];
+#endif
       mbmi->mode = mi->bmi[3].as_mode;
       break;
     case BLOCK_4X8:
       mi->bmi[0].as_mode = mi->bmi[2].as_mode =
           read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 0));
+#if CONFIG_FILTERINTRA
+      if (is_filter_allowed(mi->bmi[0].as_mode))
+        mi->b_filter_info[0] = mi->b_filter_info[2] =
+            vp9_read(r, cm->fc.filterintra_prob[0][mi->bmi[0].as_mode]);
+      else
+        mi->b_filter_info[0] = mi->b_filter_info[2] = 0;
+#endif
       mi->bmi[1].as_mode = mi->bmi[3].as_mode = mbmi->mode =
           read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 1));
+#if CONFIG_FILTERINTRA
+      if (is_filter_allowed(mi->bmi[1].as_mode))
+        mi->b_filter_info[1] = mi->b_filter_info[3] = mbmi->filterbit =
+            vp9_read(r, cm->fc.filterintra_prob[0][mi->bmi[1].as_mode]);
+      else
+        mi->b_filter_info[1] = mi->b_filter_info[3] = mbmi->filterbit = 0;
+#endif
       break;
     case BLOCK_8X4:
       mi->bmi[0].as_mode = mi->bmi[1].as_mode =
           read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 0));
+#if CONFIG_FILTERINTRA
+      if (is_filter_allowed(mi->bmi[0].as_mode))
+        mi->b_filter_info[0] = mi->b_filter_info[1] =
+            vp9_read(r, cm->fc.filterintra_prob[0][mi->bmi[0].as_mode]);
+      else
+        mi->b_filter_info[0] = mi->b_filter_info[1] = 0;
+#endif
       mi->bmi[2].as_mode = mi->bmi[3].as_mode = mbmi->mode =
           read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 2));
+#if CONFIG_FILTERINTRA
+      if (is_filter_allowed(mi->bmi[2].as_mode))
+        mi->b_filter_info[2] = mi->b_filter_info[3] = mbmi->filterbit =
+            vp9_read(r, cm->fc.filterintra_prob[0][mi->bmi[2].as_mode]);
+      else
+        mi->b_filter_info[2] = mi->b_filter_info[3] = mbmi->filterbit = 0;
+#endif
       break;
     default:
       mbmi->mode = read_intra_mode(r,
                                    get_y_mode_probs(mi, above_mi, left_mi, 0));
+#if CONFIG_FILTERINTRA
+      if (is_filter_enabled(mbmi->tx_size) && is_filter_allowed(mbmi->mode))
+        mbmi->filterbit = vp9_read(r,
+                            cm->fc.filterintra_prob[mbmi->tx_size][mbmi->mode]);
+      else
+        mbmi->filterbit = 0;
+#endif
   }
 
   mbmi->uv_mode = read_intra_mode(r, vp9_kf_uv_mode_prob[mbmi->mode]);
+#if CONFIG_FILTERINTRA
+  if (is_filter_enabled(get_uv_tx_size(mbmi)) &&
+      is_filter_allowed(mbmi->uv_mode))
+    mbmi->uv_filterbit = vp9_read(r,
+        cm->fc.filterintra_prob[get_uv_tx_size(mbmi)][mbmi->uv_mode]);
+  else
+    mbmi->uv_filterbit = 0;
+#endif
 }
 
 static int read_mv_component(vp9_reader *r,
@@ -335,25 +425,97 @@ static void read_intra_block_mode_info(VP9_COMMON *const cm, MODE_INFO *mi,
 
   switch (bsize) {
     case BLOCK_4X4:
+#if !CONFIG_FILTERINTRA
       for (i = 0; i < 4; ++i)
+#else
+      for (i = 0; i < 4; ++i) {
+#endif
         mi->bmi[i].as_mode = read_intra_mode_y(cm, r, 0);
+#if CONFIG_FILTERINTRA
+        if (is_filter_allowed(mi->bmi[i].as_mode)) {
+          mi->b_filter_info[i] =
+              vp9_read(r, cm->fc.filterintra_prob[0][mi->bmi[i].as_mode]);
+          cm->counts.filterintra[0][mi->bmi[i].as_mode]
+                                   [mi->b_filter_info[i]]++;
+        } else {
+          mi->b_filter_info[i] = 0;
+        }
+      }
+      mbmi->filterbit = mi->b_filter_info[3];
+#endif
       mbmi->mode = mi->bmi[3].as_mode;
       break;
     case BLOCK_4X8:
       mi->bmi[0].as_mode = mi->bmi[2].as_mode = read_intra_mode_y(cm, r, 0);
+#if CONFIG_FILTERINTRA
+      if (is_filter_allowed(mi->bmi[0].as_mode)) {
+        mi->b_filter_info[0] = mi->b_filter_info[2] =
+            vp9_read(r, cm->fc.filterintra_prob[0][mi->bmi[0].as_mode]);
+        cm->counts.filterintra[0][mi->bmi[0].as_mode][mi->b_filter_info[0]]++;
+      } else {
+        mi->b_filter_info[0] = mi->b_filter_info[2] = 0;
+      }
+#endif
       mi->bmi[1].as_mode = mi->bmi[3].as_mode = mbmi->mode =
           read_intra_mode_y(cm, r, 0);
+#if CONFIG_FILTERINTRA
+      if (is_filter_allowed(mi->bmi[1].as_mode)) {
+        mi->b_filter_info[1] = mi->b_filter_info[3] = mbmi->filterbit =
+            vp9_read(r, cm->fc.filterintra_prob[0][mi->bmi[1].as_mode]);
+        cm->counts.filterintra[0][mi->bmi[1].as_mode][mi->b_filter_info[1]]++;
+      } else {
+        mi->b_filter_info[1] = mi->b_filter_info[3] = mbmi->filterbit = 0;
+      }
+#endif
       break;
     case BLOCK_8X4:
       mi->bmi[0].as_mode = mi->bmi[1].as_mode = read_intra_mode_y(cm, r, 0);
+#if CONFIG_FILTERINTRA
+      if (is_filter_allowed(mi->bmi[0].as_mode)) {
+        mi->b_filter_info[0] = mi->b_filter_info[1] =
+            vp9_read(r, cm->fc.filterintra_prob[0][mi->bmi[0].as_mode]);
+        cm->counts.filterintra[0][mi->bmi[0].as_mode][mi->b_filter_info[0]]++;
+      } else {
+        mi->b_filter_info[0] = mi->b_filter_info[1] = 0;
+      }
+#endif
       mi->bmi[2].as_mode = mi->bmi[3].as_mode = mbmi->mode =
           read_intra_mode_y(cm, r, 0);
+#if CONFIG_FILTERINTRA
+      if (is_filter_allowed(mi->bmi[2].as_mode)) {
+        mi->b_filter_info[2] = mi->b_filter_info[3] = mbmi->filterbit =
+            vp9_read(r, cm->fc.filterintra_prob[0][mi->bmi[2].as_mode]);
+        cm->counts.filterintra[0][mi->bmi[2].as_mode][mi->b_filter_info[2]]++;
+      } else {
+        mi->b_filter_info[2] = mi->b_filter_info[3] = mbmi->filterbit = 0;
+      }
+#endif
       break;
     default:
       mbmi->mode = read_intra_mode_y(cm, r, size_group_lookup[bsize]);
+#if CONFIG_FILTERINTRA
+      if (is_filter_allowed(mbmi->mode) && is_filter_enabled(mbmi->tx_size)) {
+        mbmi->filterbit = vp9_read(r,
+            cm->fc.filterintra_prob[mbmi->tx_size][mbmi->mode]);
+        cm->counts.filterintra[mbmi->tx_size][mbmi->mode][mbmi->filterbit]++;
+      } else {
+        mbmi->filterbit = 0;
+      }
+#endif
   }
 
   mbmi->uv_mode = read_intra_mode_uv(cm, r, mbmi->mode);
+#if CONFIG_FILTERINTRA
+  if (is_filter_allowed(mbmi->uv_mode) &&
+      is_filter_enabled(get_uv_tx_size(mbmi))) {
+    mbmi->uv_filterbit = vp9_read(r,
+        cm->fc.filterintra_prob[get_uv_tx_size(mbmi)][mbmi->uv_mode]);
+    cm->counts.filterintra[get_uv_tx_size(mbmi)]
+                           [mbmi->uv_mode][mbmi->uv_filterbit]++;
+  } else {
+    mbmi->uv_filterbit = 0;
+  }
+#endif
 }
 
 static INLINE int is_mv_valid(const MV *mv) {
@@ -422,6 +584,9 @@ static void read_inter_block_mode_info(VP9_COMMON *const cm,
                                        MACROBLOCKD *const xd,
                                        const TileInfo *const tile,
                                        MODE_INFO *const mi,
+#if CONFIG_SUPERTX && CONFIG_EXT_TX
+                                       int supertx_enabled,
+#endif
                                        int mi_row, int mi_col, vp9_reader *r) {
   MB_MODE_INFO *const mbmi = &mi->mbmi;
   const BLOCK_SIZE bsize = mbmi->sb_type;
@@ -464,6 +629,37 @@ static void read_inter_block_mode_info(VP9_COMMON *const cm,
                       ? read_switchable_interp_filter(cm, xd, r)
                       : cm->interp_filter;
 
+#if CONFIG_INTERINTRA
+    if ((cm->use_interintra) &&
+        is_interintra_allowed(bsize) &&
+        is_inter_mode(mbmi->mode) &&
+        (mbmi->ref_frame[1] <= INTRA_FRAME)) {
+      mbmi->ref_frame[1] = vp9_read(r, cm->fc.interintra_prob[bsize]) ?
+                           INTRA_FRAME : NONE;
+      cm->counts.interintra[bsize][mbmi->ref_frame[1] == INTRA_FRAME]++;
+#if CONFIG_MASKED_INTERINTRA
+      mbmi->use_masked_interintra = 0;
+#endif
+      if (mbmi->ref_frame[1] == INTRA_FRAME) {
+        mbmi->interintra_mode =
+            read_intra_mode_y(cm, r, size_group_lookup[bsize]);
+        mbmi->interintra_uv_mode = mbmi->interintra_mode;
+#if CONFIG_MASKED_INTERINTRA
+        if (cm->use_masked_interintra && get_mask_bits_interintra(bsize)) {
+          mbmi->use_masked_interintra = vp9_read(r,
+                                          cm->fc.masked_interintra_prob[bsize]);
+          cm->counts.masked_interintra[bsize][mbmi->use_masked_interintra]++;
+          if (mbmi->use_masked_interintra) {
+            mbmi->interintra_mask_index = vp9_read_literal(r,
+                                               get_mask_bits_interintra(bsize));
+            mbmi->interintra_uv_mask_index = mbmi->interintra_mask_index;
+          }
+        }
+#endif
+      }
+    }
+#endif
+
   if (bsize < BLOCK_8X8) {
     const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];  // 1 or 2
     const int num_4x4_h = num_4x4_blocks_high_lookup[bsize];  // 1 or 2
@@ -508,35 +704,160 @@ static void read_inter_block_mode_info(VP9_COMMON *const cm,
     xd->corrupted |= !assign_mv(cm, mbmi->mode, mbmi->mv, nearestmv,
                                 nearestmv, nearmv, is_compound, allow_hp, r);
   }
+#if CONFIG_MASKED_INTERINTER
+  mbmi->use_masked_interinter = 0;
+  if (cm->use_masked_interinter &&
+      cm->reference_mode != SINGLE_REFERENCE &&
+      is_inter_mode(mbmi->mode) &&
+      get_mask_bits(bsize) &&
+      mbmi->ref_frame[1] > INTRA_FRAME) {
+    mbmi->use_masked_interinter =
+        vp9_read(r, cm->fc.masked_interinter_prob[bsize]);
+    cm->counts.masked_interinter[bsize][mbmi->use_masked_interinter]++;
+    if (mbmi->use_masked_interinter) {
+      mbmi->mask_index = vp9_read_literal(r, get_mask_bits(bsize));
+    }
+  }
+#endif
 }
 
 static void read_inter_frame_mode_info(VP9_COMMON *const cm,
                                        MACROBLOCKD *const xd,
                                        const TileInfo *const tile,
+#if CONFIG_SUPERTX
+                                       int supertx_enabled,
+#endif
                                        int mi_row, int mi_col, vp9_reader *r) {
   MODE_INFO *const mi = xd->mi[0];
   MB_MODE_INFO *const mbmi = &mi->mbmi;
   int inter_block;
+#if CONFIG_COPY_CODING
+  int num_candidate = 0;
+  MB_MODE_INFO *inter_ref_list[18] = {NULL};
+#endif
 
   mbmi->mv[0].as_int = 0;
   mbmi->mv[1].as_int = 0;
+
+#if CONFIG_COPY_CODING
+  if (mbmi->sb_type >= BLOCK_8X8)
+    num_candidate = vp9_construct_ref_inter_list(cm, xd, mbmi->sb_type,
+                                               mi_row, mi_col, inter_ref_list);
+  if (mbmi->sb_type >= BLOCK_8X8 && num_candidate > 0) {
+    int ctx = vp9_get_copy_mode_context(xd);
+    int is_copy = vp9_read(r, cm->fc.copy_noref_prob[ctx][mbmi->sb_type]);
+
+    ++cm->counts.copy_noref[ctx][mbmi->sb_type][is_copy];
+    if (!is_copy) {
+      mbmi->copy_mode = NOREF;
+    } else {
+      mbmi->copy_mode = read_copy_mode(cm, r, num_candidate, ctx);
+    }
+  } else {
+    mbmi->copy_mode = NOREF;
+  }
+  if (mbmi->copy_mode != NOREF) {
+    BLOCK_SIZE bsize_backup = mbmi->sb_type;
+    int skip_backup = mbmi->skip;
+    COPY_MODE copy_mode_backup = mbmi->copy_mode;
+#if CONFIG_SUPERTX
+    TX_SIZE tx_size_backup = mbmi->tx_size;
+#endif
+#if CONFIG_EXT_TX
+    EXT_TX_TYPE ext_txfrm_backup = mbmi->ext_txfrm;
+#endif
+
+    inter_block = 1;
+    *mbmi = *inter_ref_list[mbmi->copy_mode - REF0];
+#if CONFIG_MASKED_INTERINTER
+    mbmi->use_masked_interinter = 0;
+#endif
+#if CONFIG_INTERINTRA
+    if (mbmi->ref_frame[1] == INTRA_FRAME)
+      mbmi->ref_frame[1] = NONE;
+#endif
+#if CONFIG_SUPERTX
+    mbmi->tx_size = tx_size_backup;
+#endif
+#if CONFIG_EXT_TX
+    mbmi->ext_txfrm = ext_txfrm_backup;
+#endif
+    mbmi->sb_type = bsize_backup;
+    mbmi->mode = NEARESTMV;
+    mbmi->skip = skip_backup;
+    mbmi->copy_mode = copy_mode_backup;
+  }
+#endif
+
+#if CONFIG_SUPERTX
+  if (!supertx_enabled) {
+#endif
   mbmi->segment_id = read_inter_segment_id(cm, xd, mi_row, mi_col, r);
   mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r);
+
+#if CONFIG_COPY_CODING
+  if (mbmi->copy_mode == NOREF)
+#endif
   inter_block = read_is_inter_block(cm, xd, mbmi->segment_id, r);
   mbmi->tx_size = read_tx_size(cm, xd, cm->tx_mode, mbmi->sb_type,
                                !mbmi->skip || !inter_block, r);
+#if CONFIG_EXT_TX
+  if (inter_block &&
+      mbmi->tx_size <= TX_16X16 &&
+      mbmi->sb_type >= BLOCK_8X8 &&
+#if CONFIG_SUPERTX
+      !supertx_enabled &&
+#endif
+      !vp9_segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP) &&
+      !mbmi->skip) {
+    mbmi->ext_txfrm = vp9_read(r, cm->fc.ext_tx_prob);
+    if (!cm->frame_parallel_decoding_mode)
+      ++cm->counts.ext_tx[mbmi->ext_txfrm];
+  } else {
+    mbmi->ext_txfrm = NORM;
+  }
+#endif
+#if CONFIG_SUPERTX
+  } else {
+    const int ctx = vp9_get_intra_inter_context(xd);
+    mbmi->segment_id = 0;
+    inter_block = 1;
+    if (!cm->frame_parallel_decoding_mode)
+#if CONFIG_COPY_CODING
+      if (mbmi->copy_mode == NOREF)
+#endif
+      ++cm->counts.intra_inter[ctx][1];
+  }
+#endif
 
+#if CONFIG_COPY_CODING
+  if (mbmi->copy_mode == NOREF) {
+#endif
   if (inter_block)
-    read_inter_block_mode_info(cm, xd, tile, mi, mi_row, mi_col, r);
+    read_inter_block_mode_info(cm, xd, tile, mi,
+#if CONFIG_SUPERTX && CONFIG_EXT_TX
+                               supertx_enabled,
+#endif
+                               mi_row, mi_col, r);
   else
     read_intra_block_mode_info(cm, mi, r);
+#if CONFIG_COPY_CODING
+  }
+#endif
 }
 
 void vp9_read_mode_info(VP9_COMMON *cm, MACROBLOCKD *xd,
                         const TileInfo *const tile,
+#if CONFIG_SUPERTX
+                        int supertx_enabled,
+#endif
                         int mi_row, int mi_col, vp9_reader *r) {
   if (frame_is_intra_only(cm))
     read_intra_frame_mode_info(cm, xd, mi_row, mi_col, r);
   else
-    read_inter_frame_mode_info(cm, xd, tile, mi_row, mi_col, r);
+    read_inter_frame_mode_info(cm, xd, tile,
+#if CONFIG_SUPERTX
+                               supertx_enabled,
+#endif
+                               mi_row, mi_col, r);
 }

vp9/decoder/vp9_decodemv.h

@@ -21,8 +21,16 @@ struct TileInfo;
 
 void vp9_read_mode_info(VP9_COMMON *cm, MACROBLOCKD *xd,
                         const struct TileInfo *const tile,
+#if CONFIG_SUPERTX
+                        int supertx_enabled,
+#endif
                         int mi_row, int mi_col, vp9_reader *r);
 
+#if CONFIG_SUPERTX
+int read_skip(VP9_COMMON *cm, const MACROBLOCKD *xd,
+              int segment_id, vp9_reader *r);
+#endif
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif

vp9/decoder/vp9_decoder.c

@@ -37,7 +37,6 @@ static void initialize_dec() {
 
   if (!init_done) {
     vp9_init_neighbors();
-    vp9_init_quant_tables();
     init_done = 1;
   }
 }
@@ -244,8 +243,8 @@ int vp9_receive_compressed_data(VP9Decoder *pbi,
 
   // Check if the previous frame was a frame without any references to it.
   // Release frame buffer if not decoding in frame parallel mode.
-  if (!pbi->frame_parallel_decode && cm->new_fb_idx >= 0 &&
-         cm->frame_bufs[cm->new_fb_idx].ref_count == 0)
+  if (!pbi->frame_parallel_decode && cm->new_fb_idx >= 0
+      && cm->frame_bufs[cm->new_fb_idx].ref_count == 0)
     cm->release_fb_cb(cm->cb_priv,
                       &cm->frame_bufs[cm->new_fb_idx].raw_frame_buffer);
   cm->new_fb_idx = get_free_fb(cm);
@@ -260,10 +259,10 @@ int vp9_receive_compressed_data(VP9Decoder *pbi,
     // TODO(jkoleszar): Error concealment is undefined and non-normative
     // at this point, but if it becomes so, [0] may not always be the correct
     // thing to do here.
-    if (cm->frame_refs[0].idx != INT_MAX)
+    if (cm->frame_refs[0].idx != INT_MAX && cm->frame_refs[0].buf != NULL)
       cm->frame_refs[0].buf->corrupted = 1;
 
-    if (cm->frame_bufs[cm->new_fb_idx].ref_count > 0)
+    if (cm->new_fb_idx > 0 && cm->frame_bufs[cm->new_fb_idx].ref_count > 0)
       cm->frame_bufs[cm->new_fb_idx].ref_count--;
 
     return -1;

vp9/decoder/vp9_dthread.h

@@ -40,6 +40,23 @@ typedef struct VP9LfSyncData {
   int sync_range;
 } VP9LfSync;
 
+// WorkerData for the FrameWorker thread. It contains all the information of
+// the worker and decode structures for decoding a frame.
+typedef struct FrameWorkerData {
+  struct VP9Decoder *pbi;
+  const uint8_t *data;
+  const uint8_t *data_end;
+  size_t data_size;
+  void *user_priv;
+  int result;
+  int worker_id;
+
+  // scratch_buffer is used in frame parallel mode only.
+  // It is used to make a copy of the compressed data.
+  uint8_t *scratch_buffer;
+  size_t scratch_buffer_size;
+} FrameWorkerData;
+
 // Allocate memory for loopfilter row synchronization.
 void vp9_loop_filter_alloc(struct VP9Common *cm, VP9LfSync *lf_sync,
                            int rows, int width);

vp9/decoder/vp9_read_bit_buffer.c

@@ -10,7 +10,7 @@
 #include "vp9/decoder/vp9_read_bit_buffer.h"
 
 size_t vp9_rb_bytes_read(struct vp9_read_bit_buffer *rb) {
-  return rb->bit_offset / CHAR_BIT + (rb->bit_offset % CHAR_BIT > 0);
+  return (rb->bit_offset + CHAR_BIT - 1) / CHAR_BIT;
 }
 
 int vp9_rb_read_bit(struct vp9_read_bit_buffer *rb) {

vp9/encoder/vp9_bitstream.c

@@ -38,12 +38,32 @@ static struct vp9_token intra_mode_encodings[INTRA_MODES];
 static struct vp9_token switchable_interp_encodings[SWITCHABLE_FILTERS];
 static struct vp9_token partition_encodings[PARTITION_TYPES];
 static struct vp9_token inter_mode_encodings[INTER_MODES];
+#if CONFIG_COPY_CODING
+static struct vp9_token copy_mode_encodings_l2[2];
+static struct vp9_token copy_mode_encodings[COPY_MODE_COUNT - 1];
+#endif
+
+#if CONFIG_SUPERTX
+static int vp9_check_supertx(VP9_COMMON *cm, int mi_row, int mi_col,
+                             BLOCK_SIZE bsize) {
+  MODE_INFO **mi;
+
+  mi = cm->mi_grid_visible + (mi_row * cm->mi_stride + mi_col);
+
+  return mi[0]->mbmi.tx_size == bsize_to_tx_size(bsize) &&
+         mi[0]->mbmi.sb_type < bsize;
+}
+#endif
 
 void vp9_entropy_mode_init() {
   vp9_tokens_from_tree(intra_mode_encodings, vp9_intra_mode_tree);
   vp9_tokens_from_tree(switchable_interp_encodings, vp9_switchable_interp_tree);
   vp9_tokens_from_tree(partition_encodings, vp9_partition_tree);
   vp9_tokens_from_tree(inter_mode_encodings, vp9_inter_mode_tree);
+#if CONFIG_COPY_CODING
+  vp9_tokens_from_tree(copy_mode_encodings_l2, vp9_copy_mode_tree_l2);
+  vp9_tokens_from_tree(copy_mode_encodings, vp9_copy_mode_tree);
+#endif
 }
 
 static void write_intra_mode(vp9_writer *w, PREDICTION_MODE mode,
@@ -58,6 +78,21 @@ static void write_inter_mode(vp9_writer *w, PREDICTION_MODE mode,
                   &inter_mode_encodings[INTER_OFFSET(mode)]);
 }
 
+#if CONFIG_COPY_CODING
+static void write_copy_mode(VP9_COMMON *cm, vp9_writer *w, COPY_MODE mode,
+                            int inter_ref_count, int copy_mode_context) {
+  if (inter_ref_count == 2) {
+    vp9_write_token(w, vp9_copy_mode_tree_l2,
+                    cm->fc.copy_mode_probs_l2[copy_mode_context],
+                    &copy_mode_encodings_l2[mode - REF0]);
+  } else if (inter_ref_count > 2) {
+    vp9_write_token(w, vp9_copy_mode_tree,
+                    cm->fc.copy_mode_probs[copy_mode_context],
+                    &copy_mode_encodings[mode - REF0]);
+  }
+}
+#endif
+
 static void encode_unsigned_max(struct vp9_write_bit_buffer *wb,
                                 int data, int max) {
   vp9_wb_write_literal(wb, data, get_unsigned_bits(max));
@@ -225,6 +260,9 @@ static void write_ref_frames(const VP9_COMP *cpi, vp9_writer *w) {
 }
 
 static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
+#if CONFIG_SUPERTX
+                                int supertx_enabled,
+#endif
                                 vp9_writer *w) {
   VP9_COMMON *const cm = &cpi->common;
   const nmv_context *nmvc = &cm->fc.nmvc;
@@ -239,7 +277,19 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
   const int is_inter = is_inter_block(mbmi);
   const int is_compound = has_second_ref(mbmi);
   int skip, ref;
+#if CONFIG_COPY_CODING
+  int copy_mode_context = vp9_get_copy_mode_context(xd);
+#endif
 
+#if CONFIG_COPY_CODING
+  if (bsize >= BLOCK_8X8 && mbmi->inter_ref_count > 0) {
+      vp9_write(w, mbmi->copy_mode != NOREF,
+                cm->fc.copy_noref_prob[copy_mode_context][bsize]);
+      if (mbmi->copy_mode != NOREF)
+        write_copy_mode(cm, w, mbmi->copy_mode, mbmi->inter_ref_count,
+                        copy_mode_context);
+  }
+#endif
   if (seg->update_map) {
     if (seg->temporal_update) {
       const int pred_flag = mbmi->seg_id_predicted;
@@ -252,20 +302,57 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
     }
   }
 
+#if CONFIG_SUPERTX
+  if (!supertx_enabled)
+#endif
   skip = write_skip(cpi, segment_id, mi, w);
+#if CONFIG_SUPERTX
+  else
+    skip = mbmi->skip;
+#endif
 
+#if CONFIG_SUPERTX
+  if (!supertx_enabled) {
+#endif
+#if CONFIG_COPY_CODING
+  if (mbmi->copy_mode == NOREF)
+#endif
   if (!vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME))
     vp9_write(w, is_inter, vp9_get_intra_inter_prob(cm, xd));
+#if CONFIG_SUPERTX
+  }
+#endif
 
   if (bsize >= BLOCK_8X8 && cm->tx_mode == TX_MODE_SELECT &&
+#if CONFIG_SUPERTX
+      !supertx_enabled &&
+#endif
       !(is_inter &&
         (skip || vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP)))) {
     write_selected_tx_size(cpi, mbmi->tx_size, bsize, w);
   }
+#if CONFIG_EXT_TX
+    if (is_inter &&
+        mbmi->tx_size <= TX_16X16 &&
+        bsize >= BLOCK_8X8 &&
+#if CONFIG_SUPERTX
+        !supertx_enabled &&
+#endif
+        !mbmi->skip &&
+        !vp9_segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
+      vp9_write(w, mbmi->ext_txfrm, cm->fc.ext_tx_prob);
+    }
+#endif
 
   if (!is_inter) {
     if (bsize >= BLOCK_8X8) {
       write_intra_mode(w, mode, cm->fc.y_mode_prob[size_group_lookup[bsize]]);
+#if CONFIG_FILTERINTRA
+      if (is_filter_allowed(mode) && is_filter_enabled(mbmi->tx_size)) {
+        vp9_write(w, mbmi->filterbit,
+                  cm->fc.filterintra_prob[mbmi->tx_size][mode]);
+      }
+#endif
     } else {
       int idx, idy;
       const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];
@@ -274,11 +361,28 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
         for (idx = 0; idx < 2; idx += num_4x4_w) {
           const PREDICTION_MODE b_mode = mi->bmi[idy * 2 + idx].as_mode;
           write_intra_mode(w, b_mode, cm->fc.y_mode_prob[0]);
+#if CONFIG_FILTERINTRA
+          if (is_filter_allowed(b_mode)) {
+            vp9_write(w, mi->b_filter_info[idy * 2 + idx],
+                      cm->fc.filterintra_prob[0][b_mode]);
+          }
+#endif
         }
       }
     }
     write_intra_mode(w, mbmi->uv_mode, cm->fc.uv_mode_prob[mode]);
+#if CONFIG_FILTERINTRA
+    if (is_filter_allowed(mbmi->uv_mode) &&
+        is_filter_enabled(get_uv_tx_size(mbmi))) {
+      vp9_write(w, mbmi->uv_filterbit,
+                cm->fc.filterintra_prob[get_uv_tx_size(mbmi)][mbmi->uv_mode]);
+    }
+#endif
+#if !CONFIG_COPY_CODING
   } else {
+#else
+  } else if (mbmi->copy_mode == NOREF) {
+#endif
     const int mode_ctx = mbmi->mode_context[mbmi->ref_frame[0]];
     const vp9_prob *const inter_probs = cm->fc.inter_mode_probs[mode_ctx];
     write_ref_frames(cpi, w);
@@ -300,6 +404,32 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
       assert(mbmi->interp_filter == cm->interp_filter);
     }
 
+#if CONFIG_INTERINTRA
+    if ((cm->use_interintra) &&
+        cpi->common.reference_mode != COMPOUND_REFERENCE &&
+        is_interintra_allowed(bsize) &&
+        is_inter_mode(mode) &&
+        (mbmi->ref_frame[1] <= INTRA_FRAME)) {
+        vp9_write(w, mbmi->ref_frame[1] == INTRA_FRAME,
+                  cm->fc.interintra_prob[bsize]);
+        if (mbmi->ref_frame[1] == INTRA_FRAME) {
+          write_intra_mode(w, mbmi->interintra_mode,
+                           cm->fc.y_mode_prob[size_group_lookup[bsize]]);
+#if CONFIG_MASKED_INTERINTRA
+        if (get_mask_bits_interintra(bsize) &&
+            cm->use_masked_interintra) {
+          vp9_write(w, mbmi->use_masked_interintra,
+                    cm->fc.masked_interintra_prob[bsize]);
+          if (mbmi->use_masked_interintra) {
+            vp9_write_literal(w, mbmi->interintra_mask_index,
+                              get_mask_bits_interintra(bsize));
+          }
+        }
+#endif
+      }
+    }
+#endif
+
     if (bsize < BLOCK_8X8) {
       const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];
       const int num_4x4_h = num_4x4_blocks_high_lookup[bsize];
@@ -326,6 +456,18 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
                         allow_hp);
       }
     }
+#if CONFIG_MASKED_INTERINTER
+  if (cm->use_masked_interinter &&
+      cm->reference_mode != SINGLE_REFERENCE &&
+      is_inter_mode(mode) &&
+      get_mask_bits(mbmi->sb_type) &&
+      mbmi->ref_frame[1] > INTRA_FRAME) {
+    vp9_write(w, mbmi->use_masked_interinter,
+              cm->fc.masked_interinter_prob[bsize]);
+    if (mbmi->use_masked_interinter)
+      vp9_write_literal(w, mbmi->mask_index, get_mask_bits(mbmi->sb_type));
+  }
+#endif
   }
 }
 
@@ -350,6 +492,11 @@ static void write_mb_modes_kf(const VP9_COMP *cpi, MODE_INFO **mi_8x8,
 
   if (bsize >= BLOCK_8X8) {
     write_intra_mode(w, mbmi->mode, get_y_mode_probs(mi, above_mi, left_mi, 0));
+#if CONFIG_FILTERINTRA
+    if (is_filter_allowed(mbmi->mode) && is_filter_enabled(mbmi->tx_size))
+      vp9_write(w, mbmi->filterbit,
+                cm->fc.filterintra_prob[mbmi->tx_size][mbmi->mode]);
+#endif
   } else {
     const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];
     const int num_4x4_h = num_4x4_blocks_high_lookup[bsize];
@@ -360,15 +507,29 @@ static void write_mb_modes_kf(const VP9_COMP *cpi, MODE_INFO **mi_8x8,
         const int block = idy * 2 + idx;
         write_intra_mode(w, mi->bmi[block].as_mode,
                          get_y_mode_probs(mi, above_mi, left_mi, block));
+#if CONFIG_FILTERINTRA
+        if (is_filter_allowed(mi->bmi[block].as_mode))
+          vp9_write(w, mi->b_filter_info[block],
+                    cm->fc.filterintra_prob[0][mi->bmi[block].as_mode]);
+#endif
       }
     }
   }
 
   write_intra_mode(w, mbmi->uv_mode, vp9_kf_uv_mode_prob[mbmi->mode]);
+#if CONFIG_FILTERINTRA
+  if (is_filter_allowed(mbmi->uv_mode) &&
+      is_filter_enabled(get_uv_tx_size(mbmi)))
+    vp9_write(w, mbmi->uv_filterbit,
+              cm->fc.filterintra_prob[get_uv_tx_size(mbmi)][mbmi->uv_mode]);
+#endif
 }
 
 static void write_modes_b(VP9_COMP *cpi, const TileInfo *const tile,
                           vp9_writer *w, TOKENEXTRA **tok, TOKENEXTRA *tok_end,
+#if CONFIG_SUPERTX
+                          int supertx_enabled,
+#endif
                           int mi_row, int mi_col) {
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &cpi->mb.e_mbd;
@@ -384,11 +545,21 @@ static void write_modes_b(VP9_COMP *cpi, const TileInfo *const tile,
   if (frame_is_intra_only(cm)) {
     write_mb_modes_kf(cpi, xd->mi, w);
   } else {
+#if CONFIG_SUPERTX
+    pack_inter_mode_mvs(cpi, m, supertx_enabled, w);
+#else
     pack_inter_mode_mvs(cpi, m, w);
+#endif
   }
 
+#if CONFIG_SUPERTX
+  if (!supertx_enabled) {
+#endif
   assert(*tok < tok_end);
   pack_mb_tokens(w, tok, tok_end);
+#if CONFIG_SUPERTX
+  }
+#endif
 }
 
 static void write_partition(VP9_COMMON *cm, MACROBLOCKD *xd,
@@ -415,6 +586,9 @@ static void write_partition(VP9_COMMON *cm, MACROBLOCKD *xd,
 static void write_modes_sb(VP9_COMP *cpi,
                            const TileInfo *const tile,
                            vp9_writer *w, TOKENEXTRA **tok, TOKENEXTRA *tok_end,
+#if CONFIG_SUPERTX
+                           int pack_token, int supertx_enabled,
+#endif
                            int mi_row, int mi_col, BLOCK_SIZE bsize) {
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &cpi->mb.e_mbd;
@@ -431,36 +605,105 @@ static void write_modes_sb(VP9_COMP *cpi,
   partition = partition_lookup[bsl][m->mbmi.sb_type];
   write_partition(cm, xd, bs, mi_row, mi_col, partition, bsize, w);
   subsize = get_subsize(bsize, partition);
+#if CONFIG_SUPERTX
+  xd->mi = cm->mi_grid_visible + (mi_row * cm->mi_stride + mi_col);
+  set_mi_row_col(xd, tile,
+                 mi_row, num_8x8_blocks_high_lookup[bsize],
+                 mi_col, num_8x8_blocks_wide_lookup[bsize],
+                 cm->mi_rows, cm->mi_cols);
+  if (!supertx_enabled && cm->frame_type != KEY_FRAME &&
+      partition != PARTITION_NONE && bsize <= BLOCK_32X32) {
+    TX_SIZE supertx_size = bsize_to_tx_size(bsize);  // b_width_log2(bsize);
+    vp9_prob prob = partition == PARTITION_SPLIT ?
+                    cm->fc.supertxsplit_prob[supertx_size] :
+                    cm->fc.supertx_prob[supertx_size];
+    supertx_enabled = (xd->mi[0]->mbmi.tx_size == supertx_size);
+    vp9_write(w, supertx_enabled, prob);
+    if (supertx_enabled) {
+      vp9_write(w, xd->mi[0]->mbmi.skip, vp9_get_skip_prob(cm, xd));
+#if CONFIG_EXT_TX
+      if (supertx_size <= TX_16X16 && !xd->mi[0]->mbmi.skip)
+        vp9_write(w, xd->mi[0]->mbmi.ext_txfrm, cm->fc.ext_tx_prob);
+#endif
+    }
+  }
+#endif
   if (subsize < BLOCK_8X8) {
-    write_modes_b(cpi, tile, w, tok, tok_end, mi_row, mi_col);
+    write_modes_b(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                  supertx_enabled,
+#endif
+                  mi_row, mi_col);
   } else {
     switch (partition) {
       case PARTITION_NONE:
-        write_modes_b(cpi, tile, w, tok, tok_end, mi_row, mi_col);
+        write_modes_b(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                      supertx_enabled,
+#endif
+                      mi_row, mi_col);
         break;
       case PARTITION_HORZ:
-        write_modes_b(cpi, tile, w, tok, tok_end, mi_row, mi_col);
+        write_modes_b(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                      supertx_enabled,
+#endif
+                      mi_row, mi_col);
         if (mi_row + bs < cm->mi_rows)
-          write_modes_b(cpi, tile, w, tok, tok_end, mi_row + bs, mi_col);
+          write_modes_b(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                        supertx_enabled,
+#endif
+                        mi_row + bs, mi_col);
         break;
       case PARTITION_VERT:
-        write_modes_b(cpi, tile, w, tok, tok_end, mi_row, mi_col);
+        write_modes_b(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                      supertx_enabled,
+#endif
+                      mi_row, mi_col);
         if (mi_col + bs < cm->mi_cols)
-          write_modes_b(cpi, tile, w, tok, tok_end, mi_row, mi_col + bs);
+          write_modes_b(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                        supertx_enabled,
+#endif
+                        mi_row, mi_col + bs);
         break;
       case PARTITION_SPLIT:
-        write_modes_sb(cpi, tile, w, tok, tok_end, mi_row, mi_col, subsize);
-        write_modes_sb(cpi, tile, w, tok, tok_end, mi_row, mi_col + bs,
+        write_modes_sb(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                       !supertx_enabled, supertx_enabled,
+#endif
+                       mi_row, mi_col, subsize);
+        write_modes_sb(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                       !supertx_enabled, supertx_enabled,
+#endif
+                       mi_row, mi_col + bs,
                        subsize);
-        write_modes_sb(cpi, tile, w, tok, tok_end, mi_row + bs, mi_col,
+        write_modes_sb(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                       !supertx_enabled, supertx_enabled,
+#endif
+                       mi_row + bs, mi_col,
                        subsize);
-        write_modes_sb(cpi, tile, w, tok, tok_end, mi_row + bs, mi_col + bs,
+        write_modes_sb(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                       !supertx_enabled, supertx_enabled,
+#endif
+                       mi_row + bs, mi_col + bs,
                        subsize);
         break;
       default:
         assert(0);
     }
   }
+#if CONFIG_SUPERTX
+  if (partition != PARTITION_NONE && supertx_enabled && pack_token) {
+    assert(*tok < tok_end);
+    pack_mb_tokens(w, tok, tok_end);
+  }
+#endif
 
   // update partition context
   if (bsize >= BLOCK_8X8 &&
@@ -478,7 +721,11 @@ static void write_modes(VP9_COMP *cpi,
     vp9_zero(cpi->mb.e_mbd.left_seg_context);
     for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
          mi_col += MI_BLOCK_SIZE)
-      write_modes_sb(cpi, tile, w, tok, tok_end, mi_row, mi_col,
+      write_modes_sb(cpi, tile, w, tok, tok_end,
+#if CONFIG_SUPERTX
+                     1, 0,
+#endif
+                     mi_row, mi_col,
                      BLOCK_64X64);
   }
 }
@@ -890,14 +1137,8 @@ static void write_tile_info(VP9_COMMON *cm, struct vp9_write_bit_buffer *wb) {
 }
 
 static int get_refresh_mask(VP9_COMP *cpi) {
-    // Should the GF or ARF be updated using the transmitted frame or buffer
-#if CONFIG_MULTIPLE_ARF
-    if (!cpi->multi_arf_enabled && cpi->refresh_golden_frame &&
-        !cpi->refresh_alt_ref_frame) {
-#else
-    if (cpi->refresh_golden_frame && !cpi->refresh_alt_ref_frame &&
-        !cpi->use_svc) {
-#endif
+    if (!cpi->multi_arf_allowed && cpi->refresh_golden_frame &&
+        cpi->rc.is_src_frame_alt_ref && !cpi->use_svc) {
       // Preserve the previously existing golden frame and update the frame in
       // the alt ref slot instead. This is highly specific to the use of
       // alt-ref as a forward reference, and this needs to be generalized as
@@ -910,15 +1151,10 @@ static int get_refresh_mask(VP9_COMP *cpi) {
              (cpi->refresh_golden_frame << cpi->alt_fb_idx);
     } else {
       int arf_idx = cpi->alt_fb_idx;
-#if CONFIG_MULTIPLE_ARF
-      // Determine which ARF buffer to use to encode this ARF frame.
-      if (cpi->multi_arf_enabled) {
-        int sn = cpi->sequence_number;
-        arf_idx = (cpi->frame_coding_order[sn] < 0) ?
-            cpi->arf_buffer_idx[sn + 1] :
-            cpi->arf_buffer_idx[sn];
+      if ((cpi->pass == 2) && cpi->multi_arf_allowed) {
+        const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+        arf_idx = gf_group->arf_update_idx[gf_group->index];
       }
-#endif
       return (cpi->refresh_last_frame << cpi->lst_fb_idx) |
              (cpi->refresh_golden_frame << cpi->gld_fb_idx) |
              (cpi->refresh_alt_ref_frame << arf_idx);
@@ -1187,6 +1423,104 @@ static size_t write_compressed_header(VP9_COMP *cpi, uint8_t *data) {
                        cm->counts.partition[i], PARTITION_TYPES, &header_bc);
 
     vp9_write_nmv_probs(cm, cm->allow_high_precision_mv, &header_bc);
+
+#if CONFIG_EXT_TX
+    vp9_cond_prob_diff_update(&header_bc, &fc->ext_tx_prob, cm->counts.ext_tx);
+#endif
+
+#if CONFIG_MASKED_INTERINTER
+    if (cm->reference_mode != SINGLE_REFERENCE) {
+      if (!cpi->dummy_packing && cm->use_masked_interinter) {
+        cm->use_masked_interinter = 0;
+        for (i = 0; i < BLOCK_SIZES; i++)
+          if (get_mask_bits(i) && (cm->counts.masked_interinter[i][1] > 0)) {
+            cm->use_masked_interinter = 1;
+            break;
+          }
+      }
+      vp9_write_bit(&header_bc, cm->use_masked_interinter);
+      if (cm->use_masked_interinter) {
+        for (i = 0; i < BLOCK_SIZES; i++)
+          if (get_mask_bits(i))
+            vp9_cond_prob_diff_update(&header_bc,
+                                      &fc->masked_interinter_prob[i],
+                                      cm->counts.masked_interinter[i]);
+      } else {
+        vp9_zero(cm->counts.masked_interinter);
+      }
+    } else {
+      if (!cpi->dummy_packing)
+        cm->use_masked_interinter = 0;
+      vp9_zero(cm->counts.masked_interinter);
+    }
+#endif
+
+#if CONFIG_INTERINTRA
+    if (cm->reference_mode != COMPOUND_REFERENCE) {
+      if (!cpi->dummy_packing && cm->use_interintra) {
+        cm->use_interintra = 0;
+        for (i = 0; i < BLOCK_SIZES; i++) {
+          if (is_interintra_allowed(i) && (cm->counts.interintra[i][1] > 0)) {
+            cm->use_interintra = 1;
+            break;
+          }
+        }
+      }
+      vp9_write_bit(&header_bc, cm->use_interintra);
+      if (cm->use_interintra) {
+        for (i = 0; i < BLOCK_SIZES; i++) {
+          if (is_interintra_allowed(i)) {
+            vp9_cond_prob_diff_update(&header_bc,
+                                      &fc->interintra_prob[i],
+                                      cm->counts.interintra[i]);
+          }
+        }
+#if CONFIG_MASKED_INTERINTRA
+        if (!cpi->dummy_packing && cm->use_masked_interintra) {
+          cm->use_masked_interintra = 0;
+          for (i = 0; i < BLOCK_SIZES; i++) {
+            if (is_interintra_allowed(i) && get_mask_bits_interintra(i) &&
+                (cm->counts.masked_interintra[i][1] > 0)) {
+              cm->use_masked_interintra = 1;
+              break;
+            }
+          }
+        }
+        vp9_write_bit(&header_bc, cm->use_masked_interintra);
+        if (cm->use_masked_interintra) {
+          for (i = 0; i < BLOCK_SIZES; i++) {
+            if (is_interintra_allowed(i) && get_mask_bits_interintra(i))
+              vp9_cond_prob_diff_update(&header_bc,
+                                        &fc->masked_interintra_prob[i],
+                                        cm->counts.masked_interintra[i]);
+          }
+        } else {
+          vp9_zero(cm->counts.masked_interintra);
+        }
+#endif
+      } else {
+        vp9_zero(cm->counts.interintra);
+      }
+    } else {
+      if (!cpi->dummy_packing)
+        cm->use_interintra = 0;
+      vp9_zero(cm->counts.interintra);
+#if CONFIG_MASKED_INTERINTRA
+      if (!cpi->dummy_packing)
+        cm->use_masked_interintra = 0;
+      vp9_zero(cm->counts.masked_interintra);
+#endif
+    }
+#endif
+
+#if CONFIG_COPY_CODING
+    for (i = 0; i < COPY_MODE_CONTEXTS; i++) {
+      prob_diff_update(vp9_copy_mode_tree_l2, cm->fc.copy_mode_probs_l2[i],
+                       cm->counts.copy_mode_l2[i], 2, &header_bc);
+      prob_diff_update(vp9_copy_mode_tree, cm->fc.copy_mode_probs[i],
+                       cm->counts.copy_mode[i], 3, &header_bc);
+    }
+#endif
   }
 
   vp9_stop_encode(&header_bc);

vp9/encoder/vp9_block.h

@@ -93,8 +93,6 @@ struct macroblock {
 
   int encode_breakout;
 
-  int in_active_map;
-
   // note that token_costs is the cost when eob node is skipped
   vp9_coeff_cost token_costs[TX_SIZES];
 

vp9/encoder/vp9_denoiser.c

@@ -8,10 +8,10 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#include <stdio.h>
-#include <stdint.h>
-#include "vp9/encoder/vp9_denoiser.h"
+#include <assert.h>
 #include "vpx_scale/yv12config.h"
+#include "vpx/vpx_integer.h"
+#include "vp9/encoder/vp9_denoiser.h"
 
 static const int widths[]  = {4, 4, 8, 8,  8, 16, 16, 16, 32, 32, 32, 64, 64};
 static const int heights[] = {4, 8, 4, 8, 16,  8, 16, 32, 16, 32, 64, 32, 64};
@@ -20,30 +20,180 @@ int vp9_denoiser_filter() {
   return 0;
 }
 
-void vp9_denoiser_denoise(VP9_DENOISER *denoiser,
-                          MACROBLOCK *mb, MODE_INFO **grid,
+static int update_running_avg(const uint8_t *mc_avg, int mc_avg_stride,
+                              uint8_t *avg, int avg_stride,
+                              const uint8_t *sig, int sig_stride,
+                              int increase_denoising, BLOCK_SIZE bs) {
+  int r, c;
+  int diff, adj, absdiff;
+  int shift_inc1 = 0, shift_inc2 = 1;
+  int adj_val[] = {3, 4, 6};
+  int total_adj = 0;
+
+  if (increase_denoising) {
+    shift_inc1 = 1;
+    shift_inc2 = 2;
+  }
+
+  for (r = 0; r < heights[bs]; ++r) {
+    for (c = 0; c < widths[bs]; ++c) {
+      diff = mc_avg[c] - sig[c];
+      absdiff = abs(diff);
+
+      if (absdiff <= 3 + shift_inc1) {
+        avg[c] = mc_avg[c];
+        total_adj += diff;
+      } else {
+        switch (absdiff) {
+          case 4: case 5: case 6: case 7:
+            adj = adj_val[0];
+            break;
+          case 8: case 9: case 10: case 11:
+          case 12: case 13: case 14: case 15:
+            adj = adj_val[1];
+            break;
+          default:
+            adj = adj_val[2];
+        }
+        if (diff > 0) {
+          avg[c] = MIN(UINT8_MAX, sig[c] + adj);
+          total_adj += adj;
+        } else {
+          avg[c] = MAX(0, sig[c] - adj);
+          total_adj -= adj;
+        }
+      }
+    }
+    sig += sig_stride;
+    avg += avg_stride;
+    mc_avg += mc_avg_stride;
+  }
+  return total_adj;
+}
+
+static uint8_t *block_start(uint8_t *framebuf, int stride,
+                            int mi_row, int mi_col) {
+  return framebuf + (stride * mi_row * 8) + (mi_col * 8);
+}
+
+void copy_block(uint8_t *dest, int dest_stride,
+                uint8_t *src, int src_stride, BLOCK_SIZE bs) {
+  int r, c;
+  for (r = 0; r < heights[bs]; ++r) {
+    for (c = 0; c < widths[bs]; ++c) {
+      dest[c] = src[c];
+    }
+    dest += dest_stride;
+    src += src_stride;
+  }
+}
+
+void vp9_denoiser_denoise(VP9_DENOISER *denoiser, MACROBLOCK *mb,
                           int mi_row, int mi_col, BLOCK_SIZE bs) {
-  return;
+  int decision = COPY_BLOCK;
+
+  YV12_BUFFER_CONFIG avg = denoiser->running_avg_y[INTRA_FRAME];
+  YV12_BUFFER_CONFIG mc_avg = denoiser->mc_running_avg_y;
+  uint8_t *avg_start = block_start(avg.y_buffer, avg.y_stride, mi_row, mi_col);
+  uint8_t *mc_avg_start = block_start(mc_avg.y_buffer, mc_avg.y_stride,
+                                          mi_row, mi_col);
+  struct buf_2d src = mb->plane[0].src;
+
+
+  update_running_avg(mc_avg_start, mc_avg.y_stride, avg_start, avg.y_stride,
+                     mb->plane[0].src.buf, mb->plane[0].src.stride, 0, bs);
+
+  if (decision == FILTER_BLOCK) {
+    // TODO(tkopp)
+  }
+  if (decision == COPY_BLOCK) {
+    copy_block(avg_start, avg.y_stride, src.buf, src.stride, bs);
+  }
+}
+
+static void copy_frame(YV12_BUFFER_CONFIG dest, const YV12_BUFFER_CONFIG src) {
+  int r, c;
+  const uint8_t *srcbuf = src.y_buffer;
+  uint8_t *destbuf = dest.y_buffer;
+  assert(dest.y_width == src.y_width);
+  assert(dest.y_height == src.y_height);
+
+  for (r = 0; r < dest.y_height; ++r) {
+    for (c = 0; c < dest.y_width; ++c) {
+      destbuf[c] = srcbuf[c];
+    }
+    destbuf += dest.y_stride;
+    srcbuf += src.y_stride;
+  }
 }
 
 void vp9_denoiser_update_frame_info(VP9_DENOISER *denoiser,
+                                    YV12_BUFFER_CONFIG src,
                                     FRAME_TYPE frame_type,
                                     int refresh_alt_ref_frame,
                                     int refresh_golden_frame,
                                     int refresh_last_frame) {
-  return;
+  if (frame_type == KEY_FRAME) {
+    int i;
+    copy_frame(denoiser->running_avg_y[LAST_FRAME], src);
+    for (i = 2; i < MAX_REF_FRAMES - 1; i++) {
+      copy_frame(denoiser->running_avg_y[i],
+                 denoiser->running_avg_y[LAST_FRAME]);
+    }
+  } else {  /* For non key frames */
+    if (refresh_alt_ref_frame) {
+      copy_frame(denoiser->running_avg_y[ALTREF_FRAME],
+                 denoiser->running_avg_y[INTRA_FRAME]);
+    }
+    if (refresh_golden_frame) {
+      copy_frame(denoiser->running_avg_y[GOLDEN_FRAME],
+                 denoiser->running_avg_y[INTRA_FRAME]);
+    }
+    if (refresh_last_frame) {
+      copy_frame(denoiser->running_avg_y[LAST_FRAME],
+                 denoiser->running_avg_y[INTRA_FRAME]);
+    }
+  }
 }
 
 void vp9_denoiser_update_frame_stats() {
-  return;
 }
 
 int vp9_denoiser_alloc(VP9_DENOISER *denoiser, int width, int height,
-                       int border) {
+                       int ssx, int ssy, int border) {
+  int i, fail;
+  assert(denoiser != NULL);
+
+  for (i = 0; i < MAX_REF_FRAMES; ++i) {
+    fail = vp9_alloc_frame_buffer(&denoiser->running_avg_y[i], width, height,
+                                  ssx, ssy, border);
+    if (fail) {
+      vp9_denoiser_free(denoiser);
+      return 1;
+    }
+  }
+
+  fail = vp9_alloc_frame_buffer(&denoiser->mc_running_avg_y, width, height,
+                                ssx, ssy, border);
+  if (fail) {
+    vp9_denoiser_free(denoiser);
+    return 1;
+  }
+
   return 0;
 }
 
 void vp9_denoiser_free(VP9_DENOISER *denoiser) {
+  int i;
+  if (denoiser == NULL) {
     return;
+  }
+  for (i = 0; i < MAX_REF_FRAMES; ++i) {
+    if (&denoiser->running_avg_y[i] != NULL) {
+      vp9_free_frame_buffer(&denoiser->running_avg_y[i]);
+    }
+  }
+  if (&denoiser->mc_running_avg_y != NULL) {
+    vp9_free_frame_buffer(&denoiser->mc_running_avg_y);
+  }
 }
-

vp9/encoder/vp9_denoiser.h

@@ -12,6 +12,7 @@
 #define VP9_ENCODER_DENOISER_H_
 
 #include "vp9/encoder/vp9_block.h"
+#include "vpx_scale/yv12config.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -23,24 +24,24 @@ enum vp9_denoiser_decision {
 };
 
 typedef struct vp9_denoiser {
-  struct buf_2d running_avg_y;
-  struct buf_2d mc_running_avg_y;
+  YV12_BUFFER_CONFIG running_avg_y[MAX_REF_FRAMES];
+  YV12_BUFFER_CONFIG mc_running_avg_y;
 } VP9_DENOISER;
 
 void vp9_denoiser_update_frame_info(VP9_DENOISER *denoiser,
+                                    YV12_BUFFER_CONFIG src,
                                     FRAME_TYPE frame_type,
                                     int refresh_alt_ref_frame,
                                     int refresh_golden_frame,
                                     int refresh_last_frame);
 
-void vp9_denoiser_denoise(VP9_DENOISER *denoiser,
-                          MACROBLOCK *mb, MODE_INFO **grid,
+void vp9_denoiser_denoise(VP9_DENOISER *denoiser, MACROBLOCK *mb,
                           int mi_row, int mi_col, BLOCK_SIZE bs);
 
 void vp9_denoiser_update_frame_stats();
 
 int vp9_denoiser_alloc(VP9_DENOISER *denoiser, int width, int height,
-                       int border);
+                       int ssx, int ssy, int border);
 
 void vp9_denoiser_free(VP9_DENOISER *denoiser);
 

vp9/encoder/vp9_encodemb.c

@@ -360,6 +360,9 @@ void vp9_xform_quant(MACROBLOCK *x, int plane, int block,
   const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
   int i, j;
   const int16_t *src_diff;
+#if CONFIG_EXT_TX
+  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+#endif
   txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
   src_diff = &p->src_diff[4 * (j * diff_stride + i)];
 
@@ -372,21 +375,45 @@ void vp9_xform_quant(MACROBLOCK *x, int plane, int block,
                            scan_order->iscan);
       break;
     case TX_16X16:
+#if CONFIG_EXT_TX
+      if (plane != 0 || mbmi->ext_txfrm == NORM) {
+#endif
       vp9_fdct16x16(src_diff, coeff, diff_stride);
+#if CONFIG_EXT_TX
+      } else {
+        vp9_fht16x16(src_diff, coeff, diff_stride, ADST_ADST);
+      }
+#endif
       vp9_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
                      p->quant, p->quant_shift, qcoeff, dqcoeff,
                      pd->dequant, p->zbin_extra, eob,
                      scan_order->scan, scan_order->iscan);
       break;
     case TX_8X8:
+#if CONFIG_EXT_TX
+      if (plane != 0 || mbmi->ext_txfrm == NORM) {
+#endif
       vp9_fdct8x8(src_diff, coeff, diff_stride);
+#if CONFIG_EXT_TX
+      } else {
+        vp9_fht8x8(src_diff, coeff, diff_stride, ADST_ADST);
+      }
+#endif
       vp9_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round,
                      p->quant, p->quant_shift, qcoeff, dqcoeff,
                      pd->dequant, p->zbin_extra, eob,
                      scan_order->scan, scan_order->iscan);
       break;
     case TX_4X4:
+#if CONFIG_EXT_TX
+      if (plane != 0 || mbmi->ext_txfrm == NORM) {
+#endif
       x->fwd_txm4x4(src_diff, coeff, diff_stride);
+#if CONFIG_EXT_TX
+      } else {
+        vp9_fht4x4(src_diff, coeff, diff_stride, ADST_ADST);
+      }
+#endif
       vp9_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round,
                      p->quant, p->quant_shift, qcoeff, dqcoeff,
                      pd->dequant, p->zbin_extra, eob,
@@ -409,6 +436,9 @@ static void encode_block(int plane, int block, BLOCK_SIZE plane_bsize,
   int i, j;
   uint8_t *dst;
   ENTROPY_CONTEXT *a, *l;
+#if CONFIG_EXT_TX
+  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+#endif
   txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
   dst = &pd->dst.buf[4 * j * pd->dst.stride + 4 * i];
   a = &ctx->ta[plane][i];
@@ -454,16 +484,43 @@ static void encode_block(int plane, int block, BLOCK_SIZE plane_bsize,
       vp9_idct32x32_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
       break;
     case TX_16X16:
+#if CONFIG_EXT_TX
+      if (plane != 0 || mbmi->ext_txfrm == NORM) {
+#endif
       vp9_idct16x16_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
+#if CONFIG_EXT_TX
+      } else {
+        vp9_iht16x16_add(ADST_ADST, dqcoeff, dst, pd->dst.stride,
+                         p->eobs[block]);
+      }
+#endif
       break;
     case TX_8X8:
+#if CONFIG_EXT_TX
+      if (plane != 0 || mbmi->ext_txfrm == NORM) {
+#endif
       vp9_idct8x8_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
+#if CONFIG_EXT_TX
+      } else {
+        vp9_iht8x8_add(ADST_ADST, dqcoeff, dst, pd->dst.stride,
+                       p->eobs[block]);
+      }
+#endif
       break;
     case TX_4X4:
+#if CONFIG_EXT_TX
+      if (plane != 0 || mbmi->ext_txfrm == NORM) {
+#endif
       // this is like vp9_short_idct4x4 but has a special case around eob<=1
       // which is significant (not just an optimization) for the lossless
       // case.
       x->itxm_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
+#if CONFIG_EXT_TX
+      } else {
+        vp9_iht4x4_add(ADST_ADST, dqcoeff, dst, pd->dst.stride,
+                       p->eobs[block]);
+      }
+#endif
       break;
     default:
       assert(0 && "Invalid transform size");
@@ -479,6 +536,10 @@ static void encode_block_pass1(int plane, int block, BLOCK_SIZE plane_bsize,
   int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   int i, j;
   uint8_t *dst;
+#if CONFIG_EXT_TX
+  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  mbmi->ext_txfrm = NORM;
+#endif
   txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
   dst = &pd->dst.buf[4 * j * pd->dst.stride + 4 * i];
 
@@ -517,6 +578,26 @@ void vp9_encode_sb(MACROBLOCK *x, BLOCK_SIZE bsize) {
   }
 }
 
+#if CONFIG_SUPERTX
+void vp9_encode_sb_supertx(MACROBLOCK *x, BLOCK_SIZE bsize) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+  struct optimize_ctx ctx;
+  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  struct encode_b_args arg = {x, &ctx, &mbmi->skip};
+  int plane;
+
+  for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
+    BLOCK_SIZE plane_size = bsize - 3 * (plane > 0);
+    const struct macroblockd_plane* const pd = &xd->plane[plane];
+    const TX_SIZE tx_size = plane ? get_uv_tx_size(mbmi) : mbmi->tx_size;
+    vp9_subtract_plane(x, bsize, plane);
+    vp9_get_entropy_contexts(bsize, tx_size, pd,
+                             ctx.ta[plane], ctx.tl[plane]);
+    encode_block(plane, 0, plane_size, bsize_to_tx_size(plane_size), &arg);
+  }
+}
+#endif
+
 static void encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
                                TX_SIZE tx_size, void *arg) {
   struct encode_b_args* const args = arg;
@@ -531,6 +612,9 @@ static void encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
   const scan_order *scan_order;
   TX_TYPE tx_type;
   PREDICTION_MODE mode;
+#if CONFIG_FILTERINTRA
+  int fbit = 0;
+#endif
   const int bwl = b_width_log2(plane_bsize);
   const int diff_stride = 4 * (1 << bwl);
   uint8_t *src, *dst;
@@ -544,11 +628,20 @@ static void encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
   src = &p->src.buf[4 * (j * src_stride + i)];
   src_diff = &p->src_diff[4 * (j * diff_stride + i)];
 
+#if CONFIG_FILTERINTRA
+      if (mbmi->sb_type < BLOCK_8X8 && plane == 0)
+        fbit = xd->mi[0]->b_filter_info[block];
+      else
+        fbit = plane == 0 ? mbmi->filterbit : mbmi->uv_filterbit;
+#endif
   switch (tx_size) {
     case TX_32X32:
       scan_order = &vp9_default_scan_orders[TX_32X32];
       mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
       vp9_predict_intra_block(xd, block >> 6, bwl, TX_32X32, mode,
+#if CONFIG_FILTERINTRA
+                              fbit,
+#endif
                               x->skip_encode ? src : dst,
                               x->skip_encode ? src_stride : dst_stride,
                               dst, dst_stride, i, j, plane);
@@ -569,6 +662,9 @@ static void encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
       scan_order = &vp9_scan_orders[TX_16X16][tx_type];
       mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
       vp9_predict_intra_block(xd, block >> 4, bwl, TX_16X16, mode,
+#if CONFIG_FILTERINTRA
+                              fbit,
+#endif
                               x->skip_encode ? src : dst,
                               x->skip_encode ? src_stride : dst_stride,
                               dst, dst_stride, i, j, plane);
@@ -589,6 +685,9 @@ static void encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
       scan_order = &vp9_scan_orders[TX_8X8][tx_type];
       mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
       vp9_predict_intra_block(xd, block >> 2, bwl, TX_8X8, mode,
+#if CONFIG_FILTERINTRA
+                              fbit,
+#endif
                               x->skip_encode ? src : dst,
                               x->skip_encode ? src_stride : dst_stride,
                               dst, dst_stride, i, j, plane);
@@ -609,6 +708,9 @@ static void encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
       scan_order = &vp9_scan_orders[TX_4X4][tx_type];
       mode = plane == 0 ? get_y_mode(xd->mi[0], block) : mbmi->uv_mode;
       vp9_predict_intra_block(xd, block, bwl, TX_4X4, mode,
+#if CONFIG_FILTERINTRA
+                              fbit,
+#endif
                               x->skip_encode ? src : dst,
                               x->skip_encode ? src_stride : dst_stride,
                               dst, dst_stride, i, j, plane);

vp9/encoder/vp9_encodemb.h

@@ -21,6 +21,9 @@ extern "C" {
 #endif
 
 void vp9_encode_sb(MACROBLOCK *x, BLOCK_SIZE bsize);
+#if CONFIG_SUPERTX
+void vp9_encode_sb_supertx(MACROBLOCK *x, BLOCK_SIZE bsize);
+#endif
 void vp9_encode_sby_pass1(MACROBLOCK *x, BLOCK_SIZE bsize);
 void vp9_xform_quant_fp(MACROBLOCK *x, int plane, int block,
                         BLOCK_SIZE plane_bsize, TX_SIZE tx_size);

vp9/encoder/vp9_encoder.c

@@ -64,6 +64,9 @@ void vp9_coef_tree_initialize();
 
 // #define OUTPUT_YUV_REC
 
+#ifdef OUTPUT_YUV_DENOISED
+FILE *yuv_denoised_file;
+#endif
 #ifdef OUTPUT_YUV_SRC
 FILE *yuv_file;
 #endif
@@ -103,7 +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) {
@@ -142,8 +145,6 @@ void vp9_initialize_enc() {
 
   if (!init_done) {
     vp9_init_neighbors();
-    vp9_init_quant_tables();
-
     vp9_coef_tree_initialize();
     vp9_tokenize_initialize();
     vp9_init_me_luts();
@@ -173,10 +174,8 @@ static void dealloc_compressor_data(VP9_COMP *cpi) {
   vp9_cyclic_refresh_free(cpi->cyclic_refresh);
   cpi->cyclic_refresh = NULL;
 
-  vpx_free(cpi->active_map);
-  cpi->active_map = NULL;
-
   vp9_free_frame_buffers(cm);
+  vp9_free_context_buffers(cm);
 
   vp9_free_frame_buffer(&cpi->last_frame_uf);
   vp9_free_frame_buffer(&cpi->scaled_source);
@@ -414,39 +413,46 @@ static void alloc_raw_frame_buffers(VP9_COMP *cpi) {
                        "Failed to allocate altref buffer");
 }
 
-void vp9_alloc_compressor_data(VP9_COMP *cpi) {
-  VP9_COMMON *cm = &cpi->common;
-
+static void alloc_ref_frame_buffers(VP9_COMP *cpi) {
+  VP9_COMMON *const cm = &cpi->common;
   if (vp9_alloc_frame_buffers(cm, cm->width, cm->height))
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate frame buffers");
+}
 
-  if (vp9_alloc_frame_buffer(&cpi->last_frame_uf,
+static void alloc_util_frame_buffers(VP9_COMP *cpi) {
+  VP9_COMMON *const cm = &cpi->common;
+  if (vp9_realloc_frame_buffer(&cpi->last_frame_uf,
                                cm->width, cm->height,
                                cm->subsampling_x, cm->subsampling_y,
-                             VP9_ENC_BORDER_IN_PIXELS))
+                               VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL))
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate last frame buffer");
 
-  if (vp9_alloc_frame_buffer(&cpi->scaled_source,
+  if (vp9_realloc_frame_buffer(&cpi->scaled_source,
                                cm->width, cm->height,
                                cm->subsampling_x, cm->subsampling_y,
-                             VP9_ENC_BORDER_IN_PIXELS))
+                               VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL))
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate scaled source buffer");
 
-  if (vp9_alloc_frame_buffer(&cpi->scaled_last_source,
+  if (vp9_realloc_frame_buffer(&cpi->scaled_last_source,
                                cm->width, cm->height,
                                cm->subsampling_x, cm->subsampling_y,
-                             VP9_ENC_BORDER_IN_PIXELS))
+                               VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL))
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate scaled last source buffer");
+}
+
+void vp9_alloc_compressor_data(VP9_COMP *cpi) {
+  VP9_COMMON *cm = &cpi->common;
+
+  vp9_alloc_context_buffers(cm, cm->width, cm->height);
 
   vpx_free(cpi->tok);
 
   {
     unsigned int tokens = get_token_alloc(cm->mb_rows, cm->mb_cols);
-
     CHECK_MEM_ERROR(cm, cpi->tok, vpx_calloc(tokens, sizeof(*cpi->tok)));
   }
 
@@ -456,41 +462,7 @@ void vp9_alloc_compressor_data(VP9_COMP *cpi) {
 static void update_frame_size(VP9_COMP *cpi) {
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &cpi->mb.e_mbd;
-
   vp9_update_frame_size(cm);
-
-  // Update size of buffers local to this frame
-  if (vp9_realloc_frame_buffer(&cpi->last_frame_uf,
-                               cm->width, cm->height,
-                               cm->subsampling_x, cm->subsampling_y,
-                               VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL))
-    vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
-                       "Failed to reallocate last frame buffer");
-
-  if (vp9_realloc_frame_buffer(&cpi->scaled_source,
-                               cm->width, cm->height,
-                               cm->subsampling_x, cm->subsampling_y,
-                               VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL))
-    vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
-                       "Failed to reallocate scaled source buffer");
-
-  if (vp9_realloc_frame_buffer(&cpi->scaled_last_source,
-                               cm->width, cm->height,
-                               cm->subsampling_x, cm->subsampling_y,
-                               VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL))
-    vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
-                       "Failed to reallocate scaled last source buffer");
-
-  {
-    int y_stride = cpi->scaled_source.y_stride;
-
-    if (cpi->sf.mv.search_method == NSTEP) {
-      vp9_init3smotion_compensation(&cpi->ss_cfg, y_stride);
-    } else if (cpi->sf.mv.search_method == DIAMOND) {
-      vp9_init_dsmotion_compensation(&cpi->ss_cfg, y_stride);
-    }
-  }
-
   init_macroblockd(cm, xd);
 }
 
@@ -518,6 +490,12 @@ static void set_tile_limits(VP9_COMP *cpi) {
   cm->log2_tile_rows = cpi->oxcf.tile_rows;
 }
 
+static void init_buffer_indices(VP9_COMP *cpi) {
+  cpi->lst_fb_idx = 0;
+  cpi->gld_fb_idx = 1;
+  cpi->alt_fb_idx = 2;
+}
+
 static void init_config(struct VP9_COMP *cpi, VP9EncoderConfig *oxcf) {
   VP9_COMMON *const cm = &cpi->common;
 
@@ -528,8 +506,6 @@ static void init_config(struct VP9_COMP *cpi, VP9EncoderConfig *oxcf) {
 
   cm->width = oxcf->width;
   cm->height = oxcf->height;
-  cm->subsampling_x = 0;
-  cm->subsampling_y = 0;
   vp9_alloc_compressor_data(cpi);
 
   // Spatial scalability.
@@ -548,10 +524,9 @@ static void init_config(struct VP9_COMP *cpi, VP9EncoderConfig *oxcf) {
   vp9_change_config(cpi, oxcf);
 
   cpi->static_mb_pct = 0;
+  cpi->ref_frame_flags = 0;
 
-  cpi->lst_fb_idx = 0;
-  cpi->gld_fb_idx = 1;
-  cpi->alt_fb_idx = 2;
+  init_buffer_indices(cpi);
 
   set_tile_limits(cpi);
 }
@@ -590,7 +565,6 @@ void vp9_change_config(struct VP9_COMP *cpi, const VP9EncoderConfig *oxcf) {
   cpi->pass = get_pass(cpi->oxcf.mode);
 
   rc->baseline_gf_interval = DEFAULT_GF_INTERVAL;
-  cpi->ref_frame_flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
 
   cpi->refresh_golden_frame = 0;
   cpi->refresh_last_frame = 1;
@@ -598,7 +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;
@@ -663,11 +637,7 @@ void vp9_change_config(struct VP9_COMP *cpi, const VP9EncoderConfig *oxcf) {
                                            (int)cpi->oxcf.target_bandwidth);
   }
 
-#if CONFIG_MULTIPLE_ARF
-  vp9_zero(cpi->alt_ref_source);
-#else
   cpi->alt_ref_source = NULL;
-#endif
   rc->is_src_frame_alt_ref = 0;
 
 #if 0
@@ -683,6 +653,12 @@ void vp9_change_config(struct VP9_COMP *cpi, const VP9EncoderConfig *oxcf) {
 
 #if CONFIG_DENOISING
   vp9_denoiser_alloc(&(cpi->denoiser), cm->width, cm->height,
+                     // TODO(tkopp) An unrelated bug causes
+                     // cm->subsampling_{x,y} to be uninitialized at this point
+                     // in execution. For now we assume YUV-420, which is x/y
+                     // subsampling of 1.
+                     1, 1,
+                     // cm->subsampling_x, cm->subsampling_y,
                      VP9_ENC_BORDER_IN_PIXELS);
 #endif
 }
@@ -781,10 +757,6 @@ VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf) {
   CHECK_MEM_ERROR(cm, cpi->coding_context.last_frame_seg_map_copy,
                   vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
 
-  CHECK_MEM_ERROR(cm, cpi->active_map, vpx_calloc(cm->MBs, 1));
-  vpx_memset(cpi->active_map, 1, cm->MBs);
-  cpi->active_map_enabled = 0;
-
   for (i = 0; i < (sizeof(cpi->mbgraph_stats) /
                    sizeof(cpi->mbgraph_stats[0])); i++) {
     CHECK_MEM_ERROR(cm, cpi->mbgraph_stats[i].mb_stats,
@@ -794,18 +766,23 @@ VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf) {
 
   cpi->refresh_alt_ref_frame = 0;
 
-#if CONFIG_MULTIPLE_ARF
-  // Turn multiple ARF usage on/off. This is a quick hack for the initial test
-  // version. It should eventually be set via the codec API.
-  cpi->multi_arf_enabled = 1;
-
-  if (cpi->multi_arf_enabled) {
-    cpi->sequence_number = 0;
-    cpi->frame_coding_order_period = 0;
-    vp9_zero(cpi->frame_coding_order);
-    vp9_zero(cpi->arf_buffer_idx);
+  // Note that at the moment multi_arf will not work with svc.
+  // For the current check in all the execution paths are defaulted to 0
+  // pending further tuning and testing. The code is left in place here
+  // as a place holder in regard to the required paths.
+  if (cpi->pass == 2) {
+    if (cpi->use_svc) {
+      cpi->multi_arf_allowed = 0;
+      cpi->multi_arf_enabled = 0;
+    } else {
+      // Disable by default for now.
+      cpi->multi_arf_allowed = 0;
+      cpi->multi_arf_enabled = 0;
+    }
+  } else {
+    cpi->multi_arf_allowed = 0;
+    cpi->multi_arf_enabled = 0;
   }
-#endif
 
   cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS;
 #if CONFIG_INTERNAL_STATS
@@ -860,6 +837,9 @@ VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf) {
   cpi->mb.nmvsadcost_hp[1] = &cpi->mb.nmvsadcosts_hp[1][MV_MAX];
   cal_nmvsadcosts_hp(cpi->mb.nmvsadcost_hp);
 
+#ifdef OUTPUT_YUV_DENOISED
+  yuv_denoised_file = fopen("denoised.yuv", "ab");
+#endif
 #ifdef OUTPUT_YUV_SRC
   yuv_file = fopen("bd.yuv", "ab");
 #endif
@@ -1005,6 +985,41 @@ VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf) {
       vp9_sub_pixel_avg_variance4x4,
       vp9_sad4x4x3, vp9_sad4x4x8, vp9_sad4x4x4d)
 
+#if ((CONFIG_MASKED_INTERINTRA && CONFIG_INTERINTRA) || \
+    CONFIG_MASKED_INTERINTER)
+#define MBFP(BT, MSDF, MVF, MSVF) \
+  cpi->fn_ptr[BT].msdf            = MSDF; \
+  cpi->fn_ptr[BT].mvf             = MVF; \
+  cpi->fn_ptr[BT].msvf            = MSVF;
+
+  MBFP(BLOCK_64X64, vp9_masked_sad64x64, vp9_masked_variance64x64,
+       vp9_masked_sub_pixel_variance64x64)
+  MBFP(BLOCK_64X32, vp9_masked_sad64x32, vp9_masked_variance64x32,
+         vp9_masked_sub_pixel_variance64x32)
+  MBFP(BLOCK_32X64, vp9_masked_sad32x64, vp9_masked_variance32x64,
+         vp9_masked_sub_pixel_variance32x64)
+  MBFP(BLOCK_32X32, vp9_masked_sad32x32, vp9_masked_variance32x32,
+       vp9_masked_sub_pixel_variance32x32)
+  MBFP(BLOCK_32X16, vp9_masked_sad32x16, vp9_masked_variance32x16,
+       vp9_masked_sub_pixel_variance32x16)
+  MBFP(BLOCK_16X32, vp9_masked_sad16x32, vp9_masked_variance16x32,
+       vp9_masked_sub_pixel_variance16x32)
+  MBFP(BLOCK_16X16, vp9_masked_sad16x16, vp9_masked_variance16x16,
+         vp9_masked_sub_pixel_variance16x16)
+  MBFP(BLOCK_16X8, vp9_masked_sad16x8, vp9_masked_variance16x8,
+         vp9_masked_sub_pixel_variance16x8)
+  MBFP(BLOCK_8X16, vp9_masked_sad8x16, vp9_masked_variance8x16,
+         vp9_masked_sub_pixel_variance8x16)
+  MBFP(BLOCK_8X8, vp9_masked_sad8x8, vp9_masked_variance8x8,
+       vp9_masked_sub_pixel_variance8x8)
+  MBFP(BLOCK_4X8, vp9_masked_sad4x8, vp9_masked_variance4x8,
+       vp9_masked_sub_pixel_variance4x8)
+  MBFP(BLOCK_8X4, vp9_masked_sad8x4, vp9_masked_variance8x4,
+       vp9_masked_sub_pixel_variance8x4)
+  MBFP(BLOCK_4X4, vp9_masked_sad4x4, vp9_masked_variance4x4,
+       vp9_masked_sub_pixel_variance4x4)
+#endif
+
   cpi->full_search_sad = vp9_full_search_sad;
   cpi->diamond_search_sad = vp9_diamond_search_sad;
   cpi->refining_search_sad = vp9_refining_search_sad;
@@ -1105,6 +1120,9 @@ void vp9_remove_compressor(VP9_COMP *cpi) {
   vp9_remove_common(&cpi->common);
   vpx_free(cpi);
 
+#ifdef OUTPUT_YUV_DENOISED
+  fclose(yuv_denoised_file);
+#endif
 #ifdef OUTPUT_YUV_SRC
   fclose(yuv_file);
 #endif
@@ -1286,13 +1304,13 @@ int vp9_update_entropy(VP9_COMP * cpi, int update) {
 }
 
 
-#ifdef OUTPUT_YUV_SRC
-void vp9_write_yuv_frame(YV12_BUFFER_CONFIG *s) {
+#if defined(OUTPUT_YUV_SRC) || defined(OUTPUT_YUV_DENOISED)
+void vp9_write_yuv_frame(YV12_BUFFER_CONFIG *s, FILE *f) {
   uint8_t *src = s->y_buffer;
   int h = s->y_height;
 
   do {
-    fwrite(src, s->y_width, 1,  yuv_file);
+    fwrite(src, s->y_width, 1, f);
     src += s->y_stride;
   } while (--h);
 
@@ -1300,7 +1318,7 @@ void vp9_write_yuv_frame(YV12_BUFFER_CONFIG *s) {
   h = s->uv_height;
 
   do {
-    fwrite(src, s->uv_width, 1,  yuv_file);
+    fwrite(src, s->uv_width, 1, f);
     src += s->uv_stride;
   } while (--h);
 
@@ -1308,7 +1326,7 @@ void vp9_write_yuv_frame(YV12_BUFFER_CONFIG *s) {
   h = s->uv_height;
 
   do {
-    fwrite(src, s->uv_width, 1, yuv_file);
+    fwrite(src, s->uv_width, 1, f);
     src += s->uv_stride;
   } while (--h);
 }
@@ -1509,14 +1527,8 @@ void vp9_update_reference_frames(VP9_COMP *cpi) {
                &cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
     ref_cnt_fb(cm->frame_bufs,
                &cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
-  }
-#if CONFIG_MULTIPLE_ARF
-  else if (!cpi->multi_arf_enabled && cpi->refresh_golden_frame &&
-      !cpi->refresh_alt_ref_frame) {
-#else
-  else if (cpi->refresh_golden_frame && !cpi->refresh_alt_ref_frame &&
-           !cpi->use_svc) {
-#endif
+  } else if (!cpi->multi_arf_allowed && cpi->refresh_golden_frame &&
+             cpi->rc.is_src_frame_alt_ref && !cpi->use_svc) {
     /* Preserve the previously existing golden frame and update the frame in
      * the alt ref slot instead. This is highly specific to the current use of
      * alt-ref as a forward reference, and this needs to be generalized as
@@ -1537,11 +1549,11 @@ void vp9_update_reference_frames(VP9_COMP *cpi) {
   } else { /* For non key/golden frames */
     if (cpi->refresh_alt_ref_frame) {
       int arf_idx = cpi->alt_fb_idx;
-#if CONFIG_MULTIPLE_ARF
-      if (cpi->multi_arf_enabled) {
-        arf_idx = cpi->arf_buffer_idx[cpi->sequence_number + 1];
+      if ((cpi->pass == 2) && cpi->multi_arf_allowed) {
+        const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+        arf_idx = gf_group->arf_update_idx[gf_group->index];
       }
-#endif
+
       ref_cnt_fb(cm->frame_bufs,
                  &cm->ref_frame_map[arf_idx], cm->new_fb_idx);
     }
@@ -1558,6 +1570,7 @@ void vp9_update_reference_frames(VP9_COMP *cpi) {
   }
 #if CONFIG_DENOISING
   vp9_denoiser_update_frame_info(&cpi->denoiser,
+                                *cpi->Source,
                                 cpi->common.frame_type,
                                 cpi->refresh_alt_ref_frame,
                                 cpi->refresh_golden_frame,
@@ -1593,13 +1606,15 @@ static void loopfilter_frame(VP9_COMP *cpi, VP9_COMMON *cm) {
 void vp9_scale_references(VP9_COMP *cpi) {
   VP9_COMMON *cm = &cpi->common;
   MV_REFERENCE_FRAME ref_frame;
+  const VP9_REFFRAME ref_mask[3] = {VP9_LAST_FLAG, VP9_GOLD_FLAG, VP9_ALT_FLAG};
 
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
     const int idx = cm->ref_frame_map[get_ref_frame_idx(cpi, ref_frame)];
     const YV12_BUFFER_CONFIG *const ref = &cm->frame_bufs[idx].buf;
 
-    if (ref->y_crop_width != cm->width ||
-        ref->y_crop_height != cm->height) {
+    // Need to convert from VP9_REFFRAME to index into ref_mask (subtract 1).
+    if ((cpi->ref_frame_flags & ref_mask[ref_frame - 1]) &&
+        (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height)) {
       const int new_fb = get_free_fb(cm);
       vp9_realloc_frame_buffer(&cm->frame_bufs[new_fb].buf,
                                cm->width, cm->height,
@@ -1987,6 +2002,45 @@ YV12_BUFFER_CONFIG *vp9_scale_if_required(VP9_COMMON *cm,
   }
 }
 
+#if CONFIG_MASKED_INTERINTER
+static void select_masked_interinter_mode(VP9_COMP *cpi) {
+  static const double threshold = 1/128.0;
+  VP9_COMMON *cm = &cpi->common;
+  int sum = cpi->masked_interinter_select_counts[1] +
+      cpi->masked_interinter_select_counts[0];
+  if (sum) {
+    double fraction = (double) cpi->masked_interinter_select_counts[1] / sum;
+    cm->use_masked_interinter = (fraction > threshold);
+  }
+}
+#endif
+
+#if CONFIG_INTERINTRA
+static void select_interintra_mode(VP9_COMP *cpi) {
+  static const double threshold = 0.007;
+  VP9_COMMON *cm = &cpi->common;
+  int sum = cpi->interintra_select_count[1] + cpi->interintra_select_count[0];
+  if (sum) {
+    double fraction = (double)cpi->interintra_select_count[1] / (double)sum;
+    cm->use_interintra = (fraction > threshold);
+  }
+}
+
+#if CONFIG_MASKED_INTERINTRA
+static void select_masked_interintra_mode(VP9_COMP *cpi) {
+  static const double threshold = 1/100.0;
+  VP9_COMMON *cm = &cpi->common;
+  int sum = cpi->masked_interintra_select_count[1] +
+      cpi->masked_interintra_select_count[0];
+  if (sum) {
+    double fraction = (double)cpi->masked_interintra_select_count[1] /
+                      (double)sum;
+    cm->use_masked_interintra = (fraction > threshold);
+  }
+}
+#endif
+#endif
+
 static void encode_frame_to_data_rate(VP9_COMP *cpi,
                                       size_t *size,
                                       uint8_t *dest,
@@ -2121,8 +2175,12 @@ static void encode_frame_to_data_rate(VP9_COMP *cpi,
   }
 #endif
 
+#ifdef OUTPUT_YUV_DENOISED
+  vp9_write_yuv_frame(&cpi->denoiser.running_avg_y[INTRA_FRAME],
+                      yuv_denoised_file);
+#endif
 #ifdef OUTPUT_YUV_SRC
-  vp9_write_yuv_frame(cpi->Source);
+  vp9_write_yuv_frame(cpi->Source, yuv_file);
 #endif
 
   set_speed_features(cpi);
@@ -2133,9 +2191,23 @@ 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 CONFIG_MASKED_INTERINTER
+  if (cm->current_video_frame == 0)
+    cm->use_masked_interinter = 0;
+#endif
+
+#if CONFIG_INTERINTRA
+  if (cm->current_video_frame == 0) {
+    cm->use_interintra = 1;
+#if CONFIG_MASKED_INTERINTRA
+    cm->use_masked_interintra = 1;
+#endif
+  }
+#endif
+
   if (cpi->sf.recode_loop == DISALLOW_RECODE) {
     encode_without_recode_loop(cpi, q);
   } else {
@@ -2197,6 +2269,18 @@ static void encode_frame_to_data_rate(VP9_COMP *cpi,
       vp9_adapt_mode_probs(cm);
       vp9_adapt_mv_probs(cm, cm->allow_high_precision_mv);
     }
+#if CONFIG_MASKED_INTERINTER
+    select_masked_interinter_mode(cpi);
+#endif
+#if CONFIG_INTERINTRA
+    select_interintra_mode(cpi);
+#if CONFIG_MASKED_INTERINTRA
+    if (cpi->common.use_interintra)
+      select_masked_interintra_mode(cpi);
+    else
+      cpi->common.use_masked_interintra = 0;
+#endif
+#endif
   }
 
   if (cpi->refresh_golden_frame == 1)
@@ -2221,31 +2305,8 @@ static void encode_frame_to_data_rate(VP9_COMP *cpi,
   if (cm->frame_type == KEY_FRAME) {
     // Tell the caller that the frame was coded as a key frame
     *frame_flags = cpi->frame_flags | FRAMEFLAGS_KEY;
-
-#if CONFIG_MULTIPLE_ARF
-    // Reset the sequence number.
-    if (cpi->multi_arf_enabled) {
-      cpi->sequence_number = 0;
-      cpi->frame_coding_order_period = cpi->new_frame_coding_order_period;
-      cpi->new_frame_coding_order_period = -1;
-    }
-#endif
   } else {
     *frame_flags = cpi->frame_flags & ~FRAMEFLAGS_KEY;
-
-#if CONFIG_MULTIPLE_ARF
-    /* Increment position in the coded frame sequence. */
-    if (cpi->multi_arf_enabled) {
-      ++cpi->sequence_number;
-      if (cpi->sequence_number >= cpi->frame_coding_order_period) {
-        cpi->sequence_number = 0;
-        cpi->frame_coding_order_period = cpi->new_frame_coding_order_period;
-        cpi->new_frame_coding_order_period = -1;
-      }
-      cpi->this_frame_weight = cpi->arf_weight[cpi->sequence_number];
-      assert(cpi->this_frame_weight >= 0);
-    }
-#endif
   }
 
   // Clear the one shot update flags for segmentation map and mode/ref loop
@@ -2299,6 +2360,16 @@ static void Pass2Encode(VP9_COMP *cpi, size_t *size,
   vp9_twopass_postencode_update(cpi);
 }
 
+static void init_motion_estimation(VP9_COMP *cpi) {
+  int y_stride = cpi->scaled_source.y_stride;
+
+  if (cpi->sf.mv.search_method == NSTEP) {
+    vp9_init3smotion_compensation(&cpi->ss_cfg, y_stride);
+  } else if (cpi->sf.mv.search_method == DIAMOND) {
+    vp9_init_dsmotion_compensation(&cpi->ss_cfg, y_stride);
+  }
+}
+
 static void check_initial_width(VP9_COMP *cpi, int subsampling_x,
                                 int subsampling_y) {
   VP9_COMMON *const cm = &cpi->common;
@@ -2306,7 +2377,13 @@ static void check_initial_width(VP9_COMP *cpi, int subsampling_x,
   if (!cpi->initial_width) {
     cm->subsampling_x = subsampling_x;
     cm->subsampling_y = subsampling_y;
+
     alloc_raw_frame_buffers(cpi);
+    alloc_ref_frame_buffers(cpi);
+    alloc_util_frame_buffers(cpi);
+
+    init_motion_estimation(cpi);
+
     cpi->initial_width = cm->width;
     cpi->initial_height = cm->height;
   }
@@ -2321,11 +2398,22 @@ 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);
@@ -2353,13 +2441,6 @@ static int frame_is_reference(const VP9_COMP *cpi) {
          cm->seg.update_data;
 }
 
-#if CONFIG_MULTIPLE_ARF
-int is_next_frame_arf(VP9_COMP *cpi) {
-  // Negative entry in frame_coding_order indicates an ARF at this position.
-  return cpi->frame_coding_order[cpi->sequence_number + 1] < 0 ? 1 : 0;
-}
-#endif
-
 void adjust_frame_rate(VP9_COMP *cpi) {
   int64_t this_duration;
   int step = 0;
@@ -2398,6 +2479,46 @@ void adjust_frame_rate(VP9_COMP *cpi) {
   cpi->last_end_time_stamp_seen = cpi->source->ts_end;
 }
 
+// Returns 0 if this is not an alt ref else the offset of the source frame
+// used as the arf midpoint.
+static int get_arf_src_index(VP9_COMP *cpi) {
+  RATE_CONTROL *const rc = &cpi->rc;
+  int arf_src_index = 0;
+  if (is_altref_enabled(&cpi->oxcf)) {
+    if (cpi->pass == 2) {
+      const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+      if (gf_group->update_type[gf_group->index] == ARF_UPDATE) {
+        arf_src_index = gf_group->arf_src_offset[gf_group->index];
+      }
+    } else if (rc->source_alt_ref_pending) {
+      arf_src_index = rc->frames_till_gf_update_due;
+    }
+  }
+  return arf_src_index;
+}
+
+static void check_src_altref(VP9_COMP *cpi) {
+  RATE_CONTROL *const rc = &cpi->rc;
+
+  if (cpi->pass == 2) {
+    const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+    rc->is_src_frame_alt_ref =
+      (gf_group->update_type[gf_group->index] == OVERLAY_UPDATE);
+  } else {
+    rc->is_src_frame_alt_ref = cpi->alt_ref_source &&
+                               (cpi->source == cpi->alt_ref_source);
+  }
+
+  if (rc->is_src_frame_alt_ref) {
+    // Current frame is an ARF overlay frame.
+    cpi->alt_ref_source = NULL;
+
+    // Don't refresh the last buffer for an ARF overlay frame. It will
+    // become the GF so preserve last as an alternative prediction option.
+    cpi->refresh_last_frame = 0;
+  }
+}
+
 int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
                             size_t *size, uint8_t *dest,
                             int64_t *time_stamp, int64_t *time_end, int flush) {
@@ -2407,11 +2528,15 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
   struct vpx_usec_timer  cmptimer;
   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);
   }
 
@@ -2420,7 +2545,7 @@ 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;
@@ -2429,35 +2554,26 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
   cpi->refresh_golden_frame = 0;
   cpi->refresh_alt_ref_frame = 0;
 
-  // Should we code an alternate reference frame.
-  if (is_altref_enabled(&cpi->oxcf) && rc->source_alt_ref_pending) {
-    int frames_to_arf;
+  // Should we encode an arf frame.
+  arf_src_index = get_arf_src_index(cpi);
+  if (arf_src_index) {
+    assert(arf_src_index <= rc->frames_to_key);
 
-#if CONFIG_MULTIPLE_ARF
-    assert(!cpi->multi_arf_enabled ||
-           cpi->frame_coding_order[cpi->sequence_number] < 0);
-
-    if (cpi->multi_arf_enabled && (cpi->pass == 2))
-      frames_to_arf = (-cpi->frame_coding_order[cpi->sequence_number])
-          - cpi->next_frame_in_order;
+#ifdef CONFIG_SPATIAL_SVC
+    if (is_spatial_svc)
+      cpi->source = vp9_svc_lookahead_peek(cpi, cpi->lookahead,
+                                           arf_src_index, 1);
     else
 #endif
-      frames_to_arf = rc->frames_till_gf_update_due;
-
-    assert(frames_to_arf <= rc->frames_to_key);
-
-    if ((cpi->source = vp9_lookahead_peek(cpi->lookahead, frames_to_arf))) {
-#if CONFIG_MULTIPLE_ARF
-      cpi->alt_ref_source[cpi->arf_buffered] = cpi->source;
-#else
+      cpi->source = vp9_lookahead_peek(cpi->lookahead, arf_src_index);
+    if (cpi->source != NULL) {
       cpi->alt_ref_source = cpi->source;
-#endif
 
       if (cpi->oxcf.arnr_max_frames > 0) {
         // Produce the filtered ARF frame.
         // TODO(agrange) merge these two functions.
-        vp9_configure_arnr_filter(cpi, frames_to_arf, rc->gfu_boost);
-        vp9_temporal_filter_prepare(cpi, frames_to_arf);
+        vp9_configure_arnr_filter(cpi, arf_src_index, rc->gfu_boost);
+        vp9_temporal_filter_prepare(cpi, arf_src_index);
         vp9_extend_frame_borders(&cpi->alt_ref_buffer);
         force_src_buffer = &cpi->alt_ref_buffer;
       }
@@ -2467,10 +2583,6 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
       cpi->refresh_golden_frame = 0;
       cpi->refresh_last_frame = 0;
       rc->is_src_frame_alt_ref = 0;
-
-#if CONFIG_MULTIPLE_ARF
-      if (!cpi->multi_arf_enabled)
-#endif
       rc->source_alt_ref_pending = 0;
     } else {
       rc->source_alt_ref_pending = 0;
@@ -2478,48 +2590,31 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
   }
 
   if (!cpi->source) {
-#if CONFIG_MULTIPLE_ARF
-    int i;
-#endif
-
     // 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;
     }
 
-    if ((cpi->source = vp9_lookahead_pop(cpi->lookahead, flush))) {
+    // Read in the source frame.
+#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;
 
-#if CONFIG_MULTIPLE_ARF
-      // Is this frame the ARF overlay.
-      rc->is_src_frame_alt_ref = 0;
-      for (i = 0; i < cpi->arf_buffered; ++i) {
-        if (cpi->source == cpi->alt_ref_source[i]) {
-          rc->is_src_frame_alt_ref = 1;
-          cpi->refresh_golden_frame = 1;
-          break;
-        }
-      }
-#else
-      rc->is_src_frame_alt_ref = cpi->alt_ref_source &&
-                                 (cpi->source == cpi->alt_ref_source);
-#endif
-      if (rc->is_src_frame_alt_ref) {
-        // Current frame is an ARF overlay frame.
-#if CONFIG_MULTIPLE_ARF
-        cpi->alt_ref_source[i] = NULL;
-#else
-        cpi->alt_ref_source = NULL;
-#endif
-        // Don't refresh the last buffer for an ARF overlay frame. It will
-        // become the GF so preserve last as an alternative prediction option.
-        cpi->refresh_last_frame = 0;
-      }
-#if CONFIG_MULTIPLE_ARF
-      ++cpi->next_frame_in_order;
-#endif
+      // Check to see if the frame should be encoded as an arf overlay.
+      check_src_altref(cpi);
     }
   }
 
@@ -2535,12 +2630,9 @@ 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;
 
-#if CONFIG_MULTIPLE_ARF
-    if (cm->frame_type != KEY_FRAME && cpi->pass == 2)
-      rc->source_alt_ref_pending = is_next_frame_arf(cpi);
-#endif
   } else {
     *size = 0;
     if (flush && cpi->pass == 1 && !cpi->twopass.first_pass_done) {
@@ -2578,16 +2670,14 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
   cm->frame_bufs[cm->new_fb_idx].ref_count--;
   cm->new_fb_idx = get_free_fb(cm);
 
-#if CONFIG_MULTIPLE_ARF
-  /* Set up the correct ARF frame. */
-  if (cpi->refresh_alt_ref_frame) {
-    ++cpi->arf_buffered;
+  if (!cpi->use_svc && cpi->multi_arf_allowed) {
+    if (cm->frame_type == KEY_FRAME) {
+      init_buffer_indices(cpi);
+    } else if (cpi->pass == 2) {
+      const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+      cpi->alt_fb_idx = gf_group->arf_ref_idx[gf_group->index];
     }
-  if (cpi->multi_arf_enabled && (cm->frame_type != KEY_FRAME) &&
-      (cpi->pass == 2)) {
-    cpi->alt_fb_idx = cpi->arf_buffer_idx[cpi->sequence_number];
   }
-#endif
 
   cpi->frame_flags = *frame_flags;
 
@@ -2606,6 +2696,9 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
                            cm->subsampling_x, cm->subsampling_y,
                            VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL);
 
+  alloc_util_frame_buffers(cpi);
+  init_motion_estimation(cpi);
+
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
     const int idx = cm->ref_frame_map[get_ref_frame_idx(cpi, ref_frame)];
     YV12_BUFFER_CONFIG *const buf = &cm->frame_bufs[idx].buf;
@@ -2775,16 +2868,23 @@ int vp9_get_preview_raw_frame(VP9_COMP *cpi, YV12_BUFFER_CONFIG *dest,
 
 int vp9_set_active_map(VP9_COMP *cpi, unsigned char *map, int rows, int cols) {
   if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols) {
+    const int mi_rows = cpi->common.mi_rows;
+    const int mi_cols = cpi->common.mi_cols;
     if (map) {
-      vpx_memcpy(cpi->active_map, map, rows * cols);
-      cpi->active_map_enabled = 1;
-    } else {
-      cpi->active_map_enabled = 0;
+      int r, c;
+      for (r = 0; r < mi_rows; r++) {
+        for (c = 0; c < mi_cols; c++) {
+          cpi->segmentation_map[r * mi_cols + c] =
+              !map[(r >> 1) * cols + (c >> 1)];
+        }
+      }
+      vp9_enable_segfeature(&cpi->common.seg, 1, SEG_LVL_SKIP);
+      vp9_enable_segmentation(&cpi->common.seg);
+    } else {
+      vp9_disable_segmentation(&cpi->common.seg);
     }
-
     return 0;
   } else {
-    // cpi->active_map_enabled = 0;
     return -1;
   }
 }
@@ -2863,3 +2963,42 @@ 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);
+  }
+}

vp9/encoder/vp9_encoder.h

@@ -32,6 +32,7 @@
 #include "vp9/encoder/vp9_mcomp.h"
 #include "vp9/encoder/vp9_quantize.h"
 #include "vp9/encoder/vp9_ratectrl.h"
+#include "vp9/encoder/vp9_rdopt.h"
 #include "vp9/encoder/vp9_speed_features.h"
 #include "vp9/encoder/vp9_svc_layercontext.h"
 #include "vp9/encoder/vp9_tokenize.h"
@@ -46,9 +47,6 @@ extern "C" {
 
 #define DEFAULT_GF_INTERVAL         10
 
-#define MAX_MODES 30
-#define MAX_REFS  6
-
 typedef struct {
   int nmvjointcost[MV_JOINTS];
   int nmvcosts[2][MV_VALS];
@@ -66,57 +64,6 @@ typedef struct {
   FRAME_CONTEXT fc;
 } CODING_CONTEXT;
 
-// This enumerator type needs to be kept aligned with the mode order in
-// const MODE_DEFINITION vp9_mode_order[MAX_MODES] used in the rd code.
-typedef enum {
-  THR_NEARESTMV,
-  THR_NEARESTA,
-  THR_NEARESTG,
-
-  THR_DC,
-
-  THR_NEWMV,
-  THR_NEWA,
-  THR_NEWG,
-
-  THR_NEARMV,
-  THR_NEARA,
-  THR_COMP_NEARESTLA,
-  THR_COMP_NEARESTGA,
-
-  THR_TM,
-
-  THR_COMP_NEARLA,
-  THR_COMP_NEWLA,
-  THR_NEARG,
-  THR_COMP_NEARGA,
-  THR_COMP_NEWGA,
-
-  THR_ZEROMV,
-  THR_ZEROG,
-  THR_ZEROA,
-  THR_COMP_ZEROLA,
-  THR_COMP_ZEROGA,
-
-  THR_H_PRED,
-  THR_V_PRED,
-  THR_D135_PRED,
-  THR_D207_PRED,
-  THR_D153_PRED,
-  THR_D63_PRED,
-  THR_D117_PRED,
-  THR_D45_PRED,
-} THR_MODES;
-
-typedef enum {
-  THR_LAST,
-  THR_GOLD,
-  THR_ALTR,
-  THR_COMP_LA,
-  THR_COMP_GA,
-  THR_INTRA,
-} THR_MODES_SUB8X8;
-
 typedef enum {
   // encode_breakout is disabled.
   ENCODE_BREAKOUT_DISABLED = 0,
@@ -293,32 +240,6 @@ static INLINE int is_best_mode(MODE mode) {
   return mode == ONE_PASS_BEST || mode == TWO_PASS_SECOND_BEST;
 }
 
-typedef struct RD_OPT {
-  // Thresh_mult is used to set a threshold for the rd score. A higher value
-  // means that we will accept the best mode so far more often. This number
-  // is used in combination with the current block size, and thresh_freq_fact
-  // to pick a threshold.
-  int thresh_mult[MAX_MODES];
-  int thresh_mult_sub8x8[MAX_REFS];
-
-  int threshes[MAX_SEGMENTS][BLOCK_SIZES][MAX_MODES];
-  int thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
-
-  int64_t comp_pred_diff[REFERENCE_MODES];
-  int64_t prediction_type_threshes[MAX_REF_FRAMES][REFERENCE_MODES];
-  int64_t tx_select_diff[TX_MODES];
-  // FIXME(rbultje) can this overflow?
-  int tx_select_threshes[MAX_REF_FRAMES][TX_MODES];
-
-  int64_t filter_diff[SWITCHABLE_FILTER_CONTEXTS];
-  int64_t filter_threshes[MAX_REF_FRAMES][SWITCHABLE_FILTER_CONTEXTS];
-  int64_t filter_cache[SWITCHABLE_FILTER_CONTEXTS];
-  int64_t mask_filter;
-
-  int RDMULT;
-  int RDDIV;
-} RD_OPT;
-
 typedef struct VP9_COMP {
   QUANTS quants;
   MACROBLOCK mb;
@@ -326,11 +247,7 @@ typedef struct VP9_COMP {
   VP9EncoderConfig oxcf;
   struct lookahead_ctx    *lookahead;
   struct lookahead_entry  *source;
-#if CONFIG_MULTIPLE_ARF
-  struct lookahead_entry  *alt_ref_source[REF_FRAMES];
-#else
   struct lookahead_entry  *alt_ref_source;
-#endif
   struct lookahead_entry  *last_source;
 
   YV12_BUFFER_CONFIG *Source;
@@ -349,9 +266,6 @@ typedef struct VP9_COMP {
   int gld_fb_idx;
   int alt_fb_idx;
 
-#if CONFIG_MULTIPLE_ARF
-  int alt_ref_fb_idx[REF_FRAMES - 3];
-#endif
   int refresh_last_frame;
   int refresh_golden_frame;
   int refresh_alt_ref_frame;
@@ -369,13 +283,6 @@ typedef struct VP9_COMP {
   TOKENEXTRA *tok;
   unsigned int tok_count[4][1 << 6];
 
-#if CONFIG_MULTIPLE_ARF
-  // Position within a frame coding order (including any additional ARF frames).
-  unsigned int sequence_number;
-  // Next frame in naturally occurring order that has not yet been coded.
-  int next_frame_in_order;
-#endif
-
   // Ambient reconstruction err target for force key frames
   int ambient_err;
 
@@ -425,9 +332,6 @@ typedef struct VP9_COMP {
 
   unsigned char *complexity_map;
 
-  unsigned char *active_map;
-  unsigned int active_map_enabled;
-
   CYCLIC_REFRESH *cyclic_refresh;
 
   fractional_mv_step_fp *find_fractional_mv_step;
@@ -505,28 +409,32 @@ typedef struct VP9_COMP {
   int intra_uv_mode_cost[FRAME_TYPES][INTRA_MODES];
   int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];
   int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];
+#if CONFIG_COPY_CODING
+  int copy_mode_cost_l2[COPY_MODE_CONTEXTS][2];
+  int copy_mode_cost[COPY_MODE_CONTEXTS][COPY_MODE_COUNT - 1];
+#endif
 
   PICK_MODE_CONTEXT *leaf_tree;
   PC_TREE *pc_tree;
   PC_TREE *pc_root;
   int partition_cost[PARTITION_CONTEXTS][PARTITION_TYPES];
 
-#if CONFIG_MULTIPLE_ARF
-  // ARF tracking variables.
+  int multi_arf_allowed;
   int multi_arf_enabled;
-  unsigned int frame_coding_order_period;
-  unsigned int new_frame_coding_order_period;
-  int frame_coding_order[MAX_LAG_BUFFERS * 2];
-  int arf_buffer_idx[MAX_LAG_BUFFERS * 3 / 2];
-  int arf_weight[MAX_LAG_BUFFERS];
-  int arf_buffered;
-  int this_frame_weight;
-  int max_arf_level;
-#endif
 
 #if CONFIG_DENOISING
   VP9_DENOISER denoiser;
 #endif
+
+#if CONFIG_MASKED_INTERINTER
+  unsigned int masked_interinter_select_counts[2];
+#endif
+#if CONFIG_INTERINTRA
+  unsigned int interintra_select_count[2];
+#if CONFIG_MASKED_INTERINTRA
+  unsigned int masked_interintra_select_count[2];
+#endif
+#endif
 } VP9_COMP;
 
 void vp9_initialize_enc();
@@ -622,10 +530,14 @@ 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) {

vp9/encoder/vp9_firstpass.c

@@ -33,7 +33,6 @@
 #include "vp9/encoder/vp9_firstpass.h"
 #include "vp9/encoder/vp9_mcomp.h"
 #include "vp9/encoder/vp9_quantize.h"
-#include "vp9/encoder/vp9_ratectrl.h"
 #include "vp9/encoder/vp9_rdopt.h"
 #include "vp9/encoder/vp9_variance.h"
 
@@ -56,14 +55,7 @@
 #define DOUBLE_DIVIDE_CHECK(x) ((x) < 0 ? (x) - 0.000001 : (x) + 0.000001)
 
 #define MIN_KF_BOOST        300
-
-#if CONFIG_MULTIPLE_ARF
-// Set MIN_GF_INTERVAL to 1 for the full decomposition.
-#define MIN_GF_INTERVAL             2
-#else
 #define MIN_GF_INTERVAL     4
-#endif
-
 #define LONG_TERM_VBR_CORRECTION
 
 static void swap_yv12(YV12_BUFFER_CONFIG *a, YV12_BUFFER_CONFIG *b) {
@@ -497,6 +489,8 @@ void vp9_first_pass(VP9_COMP *cpi) {
                                         &cpi->scaled_source);
   }
 
+  vp9_setup_block_planes(&x->e_mbd, cm->subsampling_x, cm->subsampling_y);
+
   vp9_setup_src_planes(x, cpi->Source, 0, 0);
   vp9_setup_pre_planes(xd, 0, first_ref_buf, 0, 0, NULL);
   vp9_setup_dst_planes(xd->plane, new_yv12, 0, 0);
@@ -504,8 +498,6 @@ void vp9_first_pass(VP9_COMP *cpi) {
   xd->mi = cm->mi_grid_visible;
   xd->mi[0] = cm->mi;
 
-  vp9_setup_block_planes(&x->e_mbd, cm->subsampling_x, cm->subsampling_y);
-
   vp9_frame_init_quantizer(cpi);
 
   for (i = 0; i < MAX_MB_PLANE; ++i) {
@@ -615,7 +607,8 @@ void vp9_first_pass(VP9_COMP *cpi) {
                                                 &unscaled_last_source_buf_2d);
 
         // TODO(pengchong): Replace the hard-coded threshold
-        if (raw_motion_error > 25) {
+        if (raw_motion_error > 25 ||
+            (cpi->use_svc && cpi->svc.number_temporal_layers == 1)) {
           // Test last reference frame using the previous best mv as the
           // starting point (best reference) for the search.
           first_pass_motion_search(cpi, x, &best_ref_mv.as_mv, &mv.as_mv,
@@ -1221,144 +1214,6 @@ static int calc_arf_boost(VP9_COMP *cpi, int offset,
   return arf_boost;
 }
 
-#if CONFIG_MULTIPLE_ARF
-// Work out the frame coding order for a GF or an ARF group.
-// The current implementation codes frames in their natural order for a
-// GF group, and inserts additional ARFs into an ARF group using a
-// binary split approach.
-// NOTE: this function is currently implemented recursively.
-static void schedule_frames(VP9_COMP *cpi, const int start, const int end,
-                            const int arf_idx, const int gf_or_arf_group,
-                            const int level) {
-  int i, abs_end, half_range;
-  int *cfo = cpi->frame_coding_order;
-  int idx = cpi->new_frame_coding_order_period;
-
-  // If (end < 0) an ARF should be coded at position (-end).
-  assert(start >= 0);
-
-  // printf("start:%d end:%d\n", start, end);
-
-  // GF Group: code frames in logical order.
-  if (gf_or_arf_group == 0) {
-    assert(end >= start);
-    for (i = start; i <= end; ++i) {
-      cfo[idx] = i;
-      cpi->arf_buffer_idx[idx] = arf_idx;
-      cpi->arf_weight[idx] = -1;
-      ++idx;
-    }
-    cpi->new_frame_coding_order_period = idx;
-    return;
-  }
-
-  // ARF Group: Work out the ARF schedule and mark ARF frames as negative.
-  if (end < 0) {
-    // printf("start:%d end:%d\n", -end, -end);
-    // ARF frame is at the end of the range.
-    cfo[idx] = end;
-    // What ARF buffer does this ARF use as predictor.
-    cpi->arf_buffer_idx[idx] = (arf_idx > 2) ? (arf_idx - 1) : 2;
-    cpi->arf_weight[idx] = level;
-    ++idx;
-    abs_end = -end;
-  } else {
-    abs_end = end;
-  }
-
-  half_range = (abs_end - start) >> 1;
-
-  // ARFs may not be adjacent, they must be separated by at least
-  // MIN_GF_INTERVAL non-ARF frames.
-  if ((start + MIN_GF_INTERVAL) >= (abs_end - MIN_GF_INTERVAL)) {
-    // printf("start:%d end:%d\n", start, abs_end);
-    // Update the coding order and active ARF.
-    for (i = start; i <= abs_end; ++i) {
-      cfo[idx] = i;
-      cpi->arf_buffer_idx[idx] = arf_idx;
-      cpi->arf_weight[idx] = -1;
-      ++idx;
-    }
-    cpi->new_frame_coding_order_period = idx;
-  } else {
-    // Place a new ARF at the mid-point of the range.
-    cpi->new_frame_coding_order_period = idx;
-    schedule_frames(cpi, start, -(start + half_range), arf_idx + 1,
-                    gf_or_arf_group, level + 1);
-    schedule_frames(cpi, start + half_range + 1, abs_end, arf_idx,
-                    gf_or_arf_group, level + 1);
-  }
-}
-
-#define FIXED_ARF_GROUP_SIZE 16
-
-void define_fixed_arf_period(VP9_COMP *cpi) {
-  int i;
-  int max_level = INT_MIN;
-
-  assert(cpi->multi_arf_enabled);
-  assert(cpi->oxcf.lag_in_frames >= FIXED_ARF_GROUP_SIZE);
-
-  // Save the weight of the last frame in the sequence before next
-  // sequence pattern overwrites it.
-  cpi->this_frame_weight = cpi->arf_weight[cpi->sequence_number];
-  assert(cpi->this_frame_weight >= 0);
-
-  cpi->twopass.gf_zeromotion_pct = 0;
-
-  // Initialize frame coding order variables.
-  cpi->new_frame_coding_order_period = 0;
-  cpi->next_frame_in_order = 0;
-  cpi->arf_buffered = 0;
-  vp9_zero(cpi->frame_coding_order);
-  vp9_zero(cpi->arf_buffer_idx);
-  vpx_memset(cpi->arf_weight, -1, sizeof(cpi->arf_weight));
-
-  if (cpi->rc.frames_to_key <= (FIXED_ARF_GROUP_SIZE + 8)) {
-    // Setup a GF group close to the keyframe.
-    cpi->rc.source_alt_ref_pending = 0;
-    cpi->rc.baseline_gf_interval = cpi->rc.frames_to_key;
-    schedule_frames(cpi, 0, (cpi->rc.baseline_gf_interval - 1), 2, 0, 0);
-  } else {
-    // Setup a fixed period ARF group.
-    cpi->rc.source_alt_ref_pending = 1;
-    cpi->rc.baseline_gf_interval = FIXED_ARF_GROUP_SIZE;
-    schedule_frames(cpi, 0, -(cpi->rc.baseline_gf_interval - 1), 2, 1, 0);
-  }
-
-  // Replace level indicator of -1 with correct level.
-  for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
-    if (cpi->arf_weight[i] > max_level) {
-      max_level = cpi->arf_weight[i];
-    }
-  }
-  ++max_level;
-  for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
-    if (cpi->arf_weight[i] == -1) {
-      cpi->arf_weight[i] = max_level;
-    }
-  }
-  cpi->max_arf_level = max_level;
-#if 0
-  printf("\nSchedule: ");
-  for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
-    printf("%4d ", cpi->frame_coding_order[i]);
-  }
-  printf("\n");
-  printf("ARFref:   ");
-  for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
-    printf("%4d ", cpi->arf_buffer_idx[i]);
-  }
-  printf("\n");
-  printf("Weight:   ");
-  for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
-    printf("%4d ", cpi->arf_weight[i]);
-  }
-  printf("\n");
-#endif
-}
-#endif
-
 // Calculate a section intra ratio used in setting max loop filter.
 static int calculate_section_intra_ratio(const FIRSTPASS_STATS *begin,
                                          const FIRSTPASS_STATS *end,
@@ -1428,6 +1283,18 @@ static int calculate_boost_bits(int frame_count,
   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;
@@ -1435,42 +1302,85 @@ static void allocate_gf_group_bits(VP9_COMP *cpi, int64_t gf_group_bits,
   TWO_PASS *twopass = &cpi->twopass;
   FIRSTPASS_STATS frame_stats;
   int i;
-  int group_frame_index = 1;
+  int frame_index = 1;
   int target_frame_size;
   int key_frame;
   const int max_bits = frame_max_bits(&cpi->rc, &cpi->oxcf);
   int64_t total_group_bits = gf_group_bits;
   double modified_err = 0.0;
   double err_fraction;
+  int mid_boost_bits = 0;
+  int mid_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.
-  // NOTE: We dont bother to check for the special case of ARF overlay
-  // frames here, as there is clamping code for this in the function
-  // vp9_rc_clamp_pframe_target_size(), which applies to one and two pass
-  // encodes.
   if (!key_frame) {
-    twopass->gf_group_bit_allocation[0] = gf_arf_bits;
+    if (rc->source_alt_ref_active) {
+      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] = 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] = arf_buffer_indices[0];
+      twopass->gf_group.arf_ref_idx[0] = arf_buffer_indices[0];
+    }
 
     // Step over the golden frame / overlay frame
     if (EOF == input_stats(twopass, &frame_stats))
       return;
   }
 
-  // Store the bits to spend on the ARF if there is one.
-  if (rc->source_alt_ref_pending) {
-    twopass->gf_group_bit_allocation[group_frame_index++] = gf_arf_bits;
-  }
-
-  // Deduct the boost bits for arf or gf if it is not a key frame.
+  // Deduct the boost bits for arf (or gf if it is not a key frame)
+  // from the group total.
   if (rc->source_alt_ref_pending || !key_frame)
     total_group_bits -= gf_arf_bits;
 
+  // Store the bits to spend on the ARF if there is one.
+  if (rc->source_alt_ref_pending) {
+    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] = arf_buffer_indices[0];
+    twopass->gf_group.arf_ref_idx[frame_index] = arf_buffer_indices[0];
+    ++frame_index;
+
+    if (cpi->multi_arf_enabled) {
+      // Set aside a slot for a level 1 arf.
+      twopass->gf_group.update_type[frame_index] = ARF_UPDATE;
+      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] = arf_buffer_indices[1];
+      twopass->gf_group.arf_ref_idx[frame_index] = arf_buffer_indices[0];
+      ++frame_index;
+    }
+  }
+
+  // Define middle frame
+  mid_frame_idx = frame_index + (rc->baseline_gf_interval >> 1) - 1;
+
   // 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;
 
@@ -1482,10 +1392,48 @@ static void allocate_gf_group_bits(VP9_COMP *cpi, int64_t gf_group_bits,
       err_fraction = 0.0;
 
     target_frame_size = (int)((double)total_group_bits * err_fraction);
+
+    if (rc->source_alt_ref_pending && cpi->multi_arf_enabled) {
+      mid_boost_bits += (target_frame_size >> 4);
+      target_frame_size -= (target_frame_size >> 4);
+
+      if (frame_index <= mid_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));
 
-    twopass->gf_group_bit_allocation[group_frame_index++] = target_frame_size;
+    twopass->gf_group.update_type[frame_index] = LF_UPDATE;
+    twopass->gf_group.rf_level[frame_index] = INTER_NORMAL;
+
+    twopass->gf_group.bit_allocation[frame_index] = target_frame_size;
+    ++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[mid_frame_idx] + mid_boost_bits;
+      twopass->gf_group.update_type[mid_frame_idx] = OVERLAY_UPDATE;
+      twopass->gf_group.bit_allocation[mid_frame_idx] = 0;
+    }
+  } else {
+    twopass->gf_group.update_type[frame_index] = GF_UPDATE;
+    twopass->gf_group.rf_level[frame_index] = GF_ARF_STD;
   }
 }
 
@@ -1528,8 +1476,7 @@ static void define_gf_group(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
   // Reset the GF group data structures unless this is a key
   // frame in which case it will already have been done.
   if (cpi->common.frame_type != KEY_FRAME) {
-    twopass->gf_group_index = 0;
-    vp9_zero(twopass->gf_group_bit_allocation);
+    vp9_zero(twopass->gf_group);
   }
 
   vp9_clear_system_state();
@@ -1651,24 +1598,14 @@ static void define_gf_group(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
     }
   }
 
-#if CONFIG_MULTIPLE_ARF
-  if (cpi->multi_arf_enabled) {
-    // Initialize frame coding order variables.
-    cpi->new_frame_coding_order_period = 0;
-    cpi->next_frame_in_order = 0;
-    cpi->arf_buffered = 0;
-    vp9_zero(cpi->frame_coding_order);
-    vp9_zero(cpi->arf_buffer_idx);
-    vpx_memset(cpi->arf_weight, -1, sizeof(cpi->arf_weight));
-  }
-#endif
-
   // Set the interval until the next gf.
   if (cpi->common.frame_type == KEY_FRAME || rc->source_alt_ref_active)
     rc->baseline_gf_interval = i - 1;
   else
     rc->baseline_gf_interval = i;
 
+  rc->frames_till_gf_update_due = rc->baseline_gf_interval;
+
   // Should we use the alternate reference frame.
   if (allow_alt_ref &&
       (i < cpi->oxcf.lag_in_frames) &&
@@ -1681,62 +1618,11 @@ static void define_gf_group(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
                                    &b_boost);
     rc->source_alt_ref_pending = 1;
 
-#if CONFIG_MULTIPLE_ARF
-    // Set the ARF schedule.
-    if (cpi->multi_arf_enabled) {
-      schedule_frames(cpi, 0, -(rc->baseline_gf_interval - 1), 2, 1, 0);
-    }
-#endif
   } else {
     rc->gfu_boost = (int)boost_score;
     rc->source_alt_ref_pending = 0;
-#if CONFIG_MULTIPLE_ARF
-    // Set the GF schedule.
-    if (cpi->multi_arf_enabled) {
-      schedule_frames(cpi, 0, rc->baseline_gf_interval - 1, 2, 0, 0);
-      assert(cpi->new_frame_coding_order_period ==
-             rc->baseline_gf_interval);
-    }
-#endif
   }
 
-#if CONFIG_MULTIPLE_ARF
-  if (cpi->multi_arf_enabled && (cpi->common.frame_type != KEY_FRAME)) {
-    int max_level = INT_MIN;
-    // Replace level indicator of -1 with correct level.
-    for (i = 0; i < cpi->frame_coding_order_period; ++i) {
-      if (cpi->arf_weight[i] > max_level) {
-        max_level = cpi->arf_weight[i];
-      }
-    }
-    ++max_level;
-    for (i = 0; i < cpi->frame_coding_order_period; ++i) {
-      if (cpi->arf_weight[i] == -1) {
-        cpi->arf_weight[i] = max_level;
-      }
-    }
-    cpi->max_arf_level = max_level;
-  }
-#if 0
-  if (cpi->multi_arf_enabled) {
-    printf("\nSchedule: ");
-    for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
-      printf("%4d ", cpi->frame_coding_order[i]);
-    }
-    printf("\n");
-    printf("ARFref:   ");
-    for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
-      printf("%4d ", cpi->arf_buffer_idx[i]);
-    }
-    printf("\n");
-    printf("Weight:   ");
-    for (i = 0; i < cpi->new_frame_coding_order_period; ++i) {
-      printf("%4d ", cpi->arf_weight[i]);
-    }
-    printf("\n");
-  }
-#endif
-#endif
   // Reset the file position.
   reset_fpf_position(twopass, start_pos);
 
@@ -1886,8 +1772,7 @@ static void find_next_key_frame(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
   cpi->common.frame_type = KEY_FRAME;
 
   // Reset the GF group data structures.
-  twopass->gf_group_index = 0;
-  vp9_zero(twopass->gf_group_bit_allocation);
+  vp9_zero(twopass->gf_group);
 
   // Is this a forced key frame by interval.
   rc->this_key_frame_forced = rc->next_key_frame_forced;
@@ -2078,7 +1963,9 @@ static void find_next_key_frame(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
   twopass->kf_group_bits -= kf_bits;
 
   // Save the bits to spend on the key frame.
-  twopass->gf_group_bit_allocation[0] = kf_bits;
+  twopass->gf_group.bit_allocation[0] = kf_bits;
+  twopass->gf_group.update_type[0] = KF_UPDATE;
+  twopass->gf_group.rf_level[0] = KF_STD;
 
   // Note the total error score of the kf group minus the key frame itself.
   twopass->kf_group_error_left = (int)(kf_group_err - kf_mod_err);
@@ -2106,6 +1993,44 @@ void vbr_rate_correction(int * this_frame_target,
   }
 }
 
+// Define the reference buffers that will be updated post encode.
+void configure_buffer_updates(VP9_COMP *cpi) {
+  TWO_PASS *const twopass = &cpi->twopass;
+
+  cpi->rc.is_src_frame_alt_ref = 0;
+  switch (twopass->gf_group.update_type[twopass->gf_group.index]) {
+    case KF_UPDATE:
+      cpi->refresh_last_frame = 1;
+      cpi->refresh_golden_frame = 1;
+      cpi->refresh_alt_ref_frame = 1;
+      break;
+    case LF_UPDATE:
+      cpi->refresh_last_frame = 1;
+      cpi->refresh_golden_frame = 0;
+      cpi->refresh_alt_ref_frame = 0;
+      break;
+    case GF_UPDATE:
+      cpi->refresh_last_frame = 1;
+      cpi->refresh_golden_frame = 1;
+      cpi->refresh_alt_ref_frame = 0;
+      break;
+    case OVERLAY_UPDATE:
+      cpi->refresh_last_frame = 0;
+      cpi->refresh_golden_frame = 1;
+      cpi->refresh_alt_ref_frame = 0;
+      cpi->rc.is_src_frame_alt_ref = 1;
+      break;
+    case ARF_UPDATE:
+      cpi->refresh_last_frame = 0;
+      cpi->refresh_golden_frame = 0;
+      cpi->refresh_alt_ref_frame = 1;
+      break;
+    default:
+      assert(0);
+  }
+}
+
+
 void vp9_rc_get_second_pass_params(VP9_COMP *cpi) {
   VP9_COMMON *const cm = &cpi->common;
   RATE_CONTROL *const rc = &cpi->rc;
@@ -2130,14 +2055,12 @@ void vp9_rc_get_second_pass_params(VP9_COMP *cpi) {
   if (!twopass->stats_in)
     return;
 
-  // Increment the gf group index.
-  ++twopass->gf_group_index;
-
   // If this is an arf frame then we dont want to read the stats file or
   // advance the input pointer as we already have what we need.
-  if (cpi->refresh_alt_ref_frame) {
+  if (twopass->gf_group.update_type[twopass->gf_group.index] == ARF_UPDATE) {
     int target_rate;
-    target_rate = twopass->gf_group_bit_allocation[twopass->gf_group_index];
+    configure_buffer_updates(cpi);
+    target_rate = twopass->gf_group.bit_allocation[twopass->gf_group.index];
     target_rate = vp9_rc_clamp_pframe_target_size(cpi, target_rate);
     rc->base_frame_target = target_rate;
 #ifdef LONG_TERM_VBR_CORRECTION
@@ -2201,15 +2124,7 @@ void vp9_rc_get_second_pass_params(VP9_COMP *cpi) {
 
   // Define a new GF/ARF group. (Should always enter here for key frames).
   if (rc->frames_till_gf_update_due == 0) {
-#if CONFIG_MULTIPLE_ARF
-    if (cpi->multi_arf_enabled) {
-      define_fixed_arf_period(cpi);
-    } else {
-#endif
     define_gf_group(cpi, &this_frame_copy);
-#if CONFIG_MULTIPLE_ARF
-    }
-#endif
 
     if (twopass->gf_zeromotion_pct > 995) {
       // As long as max_thresh for encode breakout is small enough, it is ok
@@ -2233,7 +2148,9 @@ void vp9_rc_get_second_pass_params(VP9_COMP *cpi) {
     }
   }
 
-  target_rate = twopass->gf_group_bit_allocation[twopass->gf_group_index];
+  configure_buffer_updates(cpi);
+
+  target_rate = twopass->gf_group.bit_allocation[twopass->gf_group.index];
   if (cpi->common.frame_type == KEY_FRAME)
     target_rate = vp9_rc_clamp_iframe_target_size(cpi, target_rate);
   else
@@ -2296,4 +2213,7 @@ void vp9_twopass_postencode_update(VP9_COMP *cpi) {
     twopass->kf_group_bits -= bits_used;
   }
   twopass->kf_group_bits = MAX(twopass->kf_group_bits, 0);
+
+  // Increment the gf group index ready for the next frame.
+  ++twopass->gf_group.index;
 }

vp9/encoder/vp9_firstpass.h

@@ -12,6 +12,7 @@
 #define VP9_ENCODER_VP9_FIRSTPASS_H_
 
 #include "vp9/encoder/vp9_lookahead.h"
+#include "vp9/encoder/vp9_ratectrl.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -39,6 +40,25 @@ typedef struct {
   int64_t spatial_layer_id;
 } FIRSTPASS_STATS;
 
+typedef enum {
+  KF_UPDATE = 0,
+  LF_UPDATE = 1,
+  GF_UPDATE = 2,
+  ARF_UPDATE = 3,
+  OVERLAY_UPDATE = 4,
+  FRAME_UPDATE_TYPES = 5
+} FRAME_UPDATE_TYPE;
+
+typedef struct {
+  unsigned char index;
+  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 {
   unsigned int section_intra_rating;
   unsigned int next_iiratio;
@@ -68,8 +88,7 @@ typedef struct {
 
   int active_worst_quality;
 
-  int gf_group_index;
-  int gf_group_bit_allocation[MAX_LAG_BUFFERS * 2];
+  GF_GROUP gf_group;
 } TWO_PASS;
 
 struct VP9_COMP;

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) {

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
  *

vp9/encoder/vp9_mbgraph.c

@@ -64,6 +64,9 @@ static unsigned int do_16x16_motion_iteration(VP9_COMP *cpi,
   xd->mi[0]->mbmi.mode = NEWMV;
   xd->mi[0]->mbmi.mv[0].as_mv = *dst_mv;
 
+#if CONFIG_INTERINTRA
+  xd->mi[0]->mbmi.ref_frame[1] = NONE;
+#endif
   vp9_build_inter_predictors_sby(xd, mb_row, mb_col, BLOCK_16X16);
 
   /* restore UMV window */
@@ -141,6 +144,9 @@ static int find_best_16x16_intra(VP9_COMP *cpi, PREDICTION_MODE *pbest_mode) {
 
     xd->mi[0]->mbmi.mode = mode;
     vp9_predict_intra_block(xd, 0, 2, TX_16X16, mode,
+#if CONFIG_FILTERINTRA
+                            0,
+#endif
                             x->plane[0].src.buf, x->plane[0].src.stride,
                             xd->plane[0].dst.buf, xd->plane[0].dst.stride,
                             0, 0, 0);

vp9/encoder/vp9_mcomp.c

@@ -1640,3 +1640,354 @@ int vp9_full_pixel_search(VP9_COMP *cpi, MACROBLOCK *x,
 
   return var;
 }
+
+#if ((CONFIG_MASKED_INTERINTRA && CONFIG_INTERINTRA) || \
+    CONFIG_MASKED_INTERINTER)
+/* returns subpixel variance error function */
+#define DIST(r, c) \
+    vfp->msvf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r), z, \
+              src_stride, mask, mask_stride, &sse)
+
+/* checks if (r, c) has better score than previous best */
+
+#define MVC(r, c)                                       \
+    (mvcost ?                                           \
+     ((mvjcost[((r) != rr) * 2 + ((c) != rc)] +         \
+       mvcost[0][((r) - rr)] + mvcost[1][((c) - rc)]) * \
+      error_per_bit + 4096) >> 13 : 0)
+
+#define CHECK_BETTER(v, r, c) \
+  if (c >= minc && c <= maxc && r >= minr && r <= maxr) {              \
+    thismse = (DIST(r, c));                                            \
+    if ((v = MVC(r, c) + thismse) < besterr) {                         \
+      besterr = v;                                                     \
+      br = r;                                                          \
+      bc = c;                                                          \
+      *distortion = thismse;                                           \
+      *sse1 = sse;                                                     \
+    }                                                                  \
+  } else {                                                             \
+    v = INT_MAX;                                                       \
+  }
+
+int vp9_find_best_masked_sub_pixel_tree(const MACROBLOCK *x,
+                                        uint8_t *mask, int mask_stride,
+                                        MV *bestmv, const MV *ref_mv,
+                                        int allow_hp,
+                                        int error_per_bit,
+                                        const vp9_variance_fn_ptr_t *vfp,
+                                        int forced_stop,
+                                        int iters_per_step,
+                                        int *mvjcost, int *mvcost[2],
+                                        int *distortion,
+                                        unsigned int *sse1, int is_second) {
+  const uint8_t *const z = x->plane[0].src.buf;
+  const int src_stride = x->plane[0].src.stride;
+  const MACROBLOCKD *xd = &x->e_mbd;
+  unsigned int besterr = INT_MAX;
+  unsigned int sse;
+  unsigned int whichdir;
+  int thismse;
+  unsigned int halfiters = iters_per_step;
+  unsigned int quarteriters = iters_per_step;
+  unsigned int eighthiters = iters_per_step;
+
+  const int y_stride = xd->plane[0].pre[is_second].stride;
+  const int offset = bestmv->row * y_stride + bestmv->col;
+  const uint8_t *const y = xd->plane[0].pre[is_second].buf;
+
+  int rr = ref_mv->row;
+  int rc = ref_mv->col;
+  int br = bestmv->row * 8;
+  int bc = bestmv->col * 8;
+  int hstep = 4;
+  const int minc = MAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);
+  const int maxc = MIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);
+  const int minr = MAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);
+  const int maxr = MIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);
+
+  int tr = br;
+  int tc = bc;
+
+  // central mv
+  bestmv->row *= 8;
+  bestmv->col *= 8;
+
+  // calculate central point error
+  besterr = vfp->mvf(y + offset, y_stride, z, src_stride, mask, mask_stride,
+                     sse1);
+  *distortion = besterr;
+  besterr += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost, error_per_bit);
+
+  // 1/2 pel
+  FIRST_LEVEL_CHECKS;
+  if (halfiters > 1) {
+    SECOND_LEVEL_CHECKS;
+  }
+  tr = br;
+  tc = bc;
+
+  // Note forced_stop: 0 - full, 1 - qtr only, 2 - half only
+  if (forced_stop != 2) {
+    hstep >>= 1;
+    FIRST_LEVEL_CHECKS;
+    if (quarteriters > 1) {
+      SECOND_LEVEL_CHECKS;
+    }
+    tr = br;
+    tc = bc;
+  }
+
+  if (allow_hp && vp9_use_mv_hp(ref_mv) && forced_stop == 0) {
+    hstep >>= 1;
+    FIRST_LEVEL_CHECKS;
+    if (eighthiters > 1) {
+      SECOND_LEVEL_CHECKS;
+    }
+    tr = br;
+    tc = bc;
+  }
+  // These lines insure static analysis doesn't warn that
+  // tr and tc aren't used after the above point.
+  (void) tr;
+  (void) tc;
+
+  bestmv->row = br;
+  bestmv->col = bc;
+
+  if ((abs(bestmv->col - ref_mv->col) > (MAX_FULL_PEL_VAL << 3)) ||
+      (abs(bestmv->row - ref_mv->row) > (MAX_FULL_PEL_VAL << 3)))
+    return INT_MAX;
+
+  return besterr;
+}
+
+#undef DIST
+#undef MVC
+#undef CHECK_BETTER
+
+int vp9_get_masked_mvpred_var(const MACROBLOCK *x,
+                              uint8_t *mask, int mask_stride,
+                              const MV *best_mv, const MV *center_mv,
+                              const vp9_variance_fn_ptr_t *vfp,
+                              int use_mvcost, int is_second) {
+  const MACROBLOCKD *const xd = &x->e_mbd;
+  const struct buf_2d *const what = &x->plane[0].src;
+  const struct buf_2d *const in_what = &xd->plane[0].pre[is_second];
+  const MV mv = {best_mv->row * 8, best_mv->col * 8};
+  unsigned int unused;
+
+  return vfp->mvf(what->buf, what->stride,
+                  get_buf_from_mv(in_what, best_mv), in_what->stride,
+                  mask, mask_stride, &unused) +
+      (use_mvcost ?  mv_err_cost(&mv, center_mv, x->nmvjointcost,
+                                 x->mvcost, x->errorperbit) : 0);
+}
+
+int vp9_masked_refining_search_sad_c(const MACROBLOCK *x,
+                                     uint8_t *mask, int mask_stride,
+                                     MV *ref_mv, int error_per_bit,
+                                     int search_range,
+                                     const vp9_variance_fn_ptr_t *fn_ptr,
+                                     const MV *center_mv, int is_second) {
+  const MV neighbors[4] = {{ -1, 0}, {0, -1}, {0, 1}, {1, 0}};
+  const MACROBLOCKD *const xd = &x->e_mbd;
+  const struct buf_2d *const what = &x->plane[0].src;
+  const struct buf_2d *const in_what = &xd->plane[0].pre[is_second];
+  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
+  unsigned int best_sad = fn_ptr->msdf(what->buf, what->stride,
+                                       get_buf_from_mv(in_what, ref_mv),
+                                       in_what->stride, mask, mask_stride) +
+      mvsad_err_cost(x, ref_mv, &fcenter_mv, error_per_bit);
+  int i, j;
+
+  for (i = 0; i < search_range; i++) {
+    int best_site = -1;
+
+    for (j = 0; j < 4; j++) {
+      const MV mv = {ref_mv->row + neighbors[j].row,
+                     ref_mv->col + neighbors[j].col};
+      if (is_mv_in(x, &mv)) {
+        unsigned int sad = fn_ptr->msdf(what->buf, what->stride,
+            get_buf_from_mv(in_what, &mv), in_what->stride, mask, mask_stride);
+        if (sad < best_sad) {
+          sad += mvsad_err_cost(x, &mv, &fcenter_mv, error_per_bit);
+          if (sad < best_sad) {
+            best_sad = sad;
+            best_site = j;
+          }
+        }
+      }
+    }
+
+    if (best_site == -1) {
+      break;
+    } else {
+      ref_mv->row += neighbors[best_site].row;
+      ref_mv->col += neighbors[best_site].col;
+    }
+  }
+  return best_sad;
+}
+
+int vp9_masked_diamond_search_sad_c(const MACROBLOCK *x,
+                             const search_site_config *cfg,
+                             uint8_t *mask, int mask_stride,
+                             MV *ref_mv, MV *best_mv,
+                             int search_param, int sad_per_bit, int *num00,
+                             const vp9_variance_fn_ptr_t *fn_ptr,
+                             const MV *center_mv, int is_second) {
+  const MACROBLOCKD *const xd = &x->e_mbd;
+  const struct buf_2d *const what = &x->plane[0].src;
+  const struct buf_2d *const in_what = &xd->plane[0].pre[is_second];
+  // search_param determines the length of the initial step and hence the number
+  // of iterations
+  // 0 = initial step (MAX_FIRST_STEP) pel : 1 = (MAX_FIRST_STEP/2) pel, 2 =
+  // (MAX_FIRST_STEP/4) pel... etc.
+  const search_site *const ss = &cfg->ss[search_param * cfg->searches_per_step];
+  const int tot_steps = (cfg->ss_count / cfg->searches_per_step) - search_param;
+  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
+  const uint8_t *best_address, *in_what_ref;
+  int best_sad = INT_MAX;
+  int best_site = 0;
+  int last_site = 0;
+  int i, j, step;
+
+  clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+  in_what_ref = get_buf_from_mv(in_what, ref_mv);
+  best_address = in_what_ref;
+  *num00 = 0;
+  *best_mv = *ref_mv;
+
+  // Check the starting position
+  best_sad = fn_ptr->msdf(what->buf, what->stride,
+                         best_address, in_what->stride,
+                         mask, mask_stride) +
+      mvsad_err_cost(x, best_mv, &fcenter_mv, sad_per_bit);
+
+  i = 1;
+
+  for (step = 0; step < tot_steps; step++) {
+    for (j = 0; j < cfg->searches_per_step; j++) {
+      const MV mv = {best_mv->row + ss[i].mv.row,
+                     best_mv->col + ss[i].mv.col};
+      if (is_mv_in(x, &mv)) {
+       int sad = fn_ptr->msdf(what->buf, what->stride,
+                              best_address + ss[i].offset, in_what->stride,
+                              mask, mask_stride);
+        if (sad < best_sad) {
+          sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
+          if (sad < best_sad) {
+            best_sad = sad;
+            best_site = i;
+          }
+        }
+      }
+
+      i++;
+    }
+
+    if (best_site != last_site) {
+      best_mv->row += ss[best_site].mv.row;
+      best_mv->col += ss[best_site].mv.col;
+      best_address += ss[best_site].offset;
+      last_site = best_site;
+#if defined(NEW_DIAMOND_SEARCH)
+      while (1) {
+        const MV this_mv = {best_mv->row + ss[best_site].mv.row,
+                            best_mv->col + ss[best_site].mv.col};
+        if (is_mv_in(x, &this_mv)) {
+          int sad = fn_ptr->msdf(what->buf, what->stride,
+                                 best_address + ss[best_site].offset,
+                                 in_what->stride, mask, mask_stride);
+          if (sad < best_sad) {
+            sad += mvsad_err_cost(x, &this_mv, &fcenter_mv, sad_per_bit);
+            if (sad < best_sad) {
+              best_sad = sad;
+              best_mv->row += ss[best_site].mv.row;
+              best_mv->col += ss[best_site].mv.col;
+              best_address += ss[best_site].offset;
+              continue;
+            }
+          }
+        }
+        break;
+      }
+#endif
+    } else if (best_address == in_what_ref) {
+      (*num00)++;
+    }
+  }
+  return best_sad;
+}
+
+int vp9_masked_full_pixel_diamond(const VP9_COMP *cpi, MACROBLOCK *x,
+                                  uint8_t *mask, int mask_stride,
+                                  MV *mvp_full, int step_param,
+                                  int sadpb, int further_steps, int do_refine,
+                                  const vp9_variance_fn_ptr_t *fn_ptr,
+                                  const MV *ref_mv, MV *dst_mv,
+                                  int is_second) {
+  MV temp_mv;
+  int thissme, n, num00 = 0;
+  int bestsme = vp9_masked_diamond_search_sad_c(x, &cpi->ss_cfg,
+                                                mask, mask_stride,
+                                                mvp_full, &temp_mv,
+                                                step_param, sadpb, &n,
+                                                fn_ptr, ref_mv, is_second);
+  if (bestsme < INT_MAX)
+    bestsme = vp9_get_masked_mvpred_var(x, mask, mask_stride, &temp_mv, ref_mv,
+                                        fn_ptr, 1, is_second);
+  *dst_mv = temp_mv;
+
+  // If there won't be more n-step search, check to see if refining search is
+  // needed.
+  if (n > further_steps)
+    do_refine = 0;
+
+  while (n < further_steps) {
+    ++n;
+
+    if (num00) {
+      num00--;
+    } else {
+      thissme = vp9_masked_diamond_search_sad_c(x, &cpi->ss_cfg,
+                                                mask, mask_stride,
+                                                mvp_full, &temp_mv,
+                                                step_param + n, sadpb, &num00,
+                                                fn_ptr, ref_mv, is_second);
+      if (thissme < INT_MAX)
+        thissme = vp9_get_masked_mvpred_var(x, mask, mask_stride,
+                                            &temp_mv, ref_mv, fn_ptr, 1,
+                                            is_second);
+
+      // check to see if refining search is needed.
+      if (num00 > further_steps - n)
+        do_refine = 0;
+
+      if (thissme < bestsme) {
+        bestsme = thissme;
+        *dst_mv = temp_mv;
+      }
+    }
+  }
+
+  // final 1-away diamond refining search
+  if (do_refine) {
+    const int search_range = 8;
+    MV best_mv = *dst_mv;
+    thissme = vp9_masked_refining_search_sad_c(x, mask, mask_stride,
+                                               &best_mv, sadpb, search_range,
+                                               fn_ptr, ref_mv, is_second);
+    if (thissme < INT_MAX)
+      thissme = vp9_get_masked_mvpred_var(x, mask, mask_stride,
+                                          &best_mv, ref_mv, fn_ptr, 1,
+                                          is_second);
+    if (thissme < bestsme) {
+      bestsme = thissme;
+      *dst_mv = best_mv;
+    }
+  }
+  return bestsme;
+}
+#endif

vp9/encoder/vp9_mcomp.h

@@ -153,6 +153,29 @@ int vp9_full_pixel_search(struct VP9_COMP *cpi, MACROBLOCK *x,
                           int step_param, int error_per_bit,
                           const MV *ref_mv, MV *tmp_mv,
                           int var_max, int rd);
+
+#if ((CONFIG_MASKED_INTERINTRA && CONFIG_INTERINTRA) || \
+    CONFIG_MASKED_INTERINTER)
+int vp9_find_best_masked_sub_pixel_tree(const MACROBLOCK *x,
+                                        uint8_t *mask, int mask_stride,
+                                        MV *bestmv, const MV *ref_mv,
+                                        int allow_hp,
+                                        int error_per_bit,
+                                        const vp9_variance_fn_ptr_t *vfp,
+                                        int forced_stop,
+                                        int iters_per_step,
+                                        int *mvjcost, int *mvcost[2],
+                                        int *distortion,
+                                        unsigned int *sse1, int is_second);
+int vp9_masked_full_pixel_diamond(const struct VP9_COMP *cpi, MACROBLOCK *x,
+                                  uint8_t *mask, int mask_stride,
+                                  MV *mvp_full, int step_param,
+                                  int sadpb, int further_steps, int do_refine,
+                                  const vp9_variance_fn_ptr_t *fn_ptr,
+                                  const MV *ref_mv, MV *dst_mv,
+                                  int is_second);
+#endif
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif

vp9/encoder/vp9_pickmode.c

@@ -23,9 +23,89 @@
 #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"
 
+static int mv_refs_rt(const VP9_COMMON *cm, const MACROBLOCKD *xd,
+                       const TileInfo *const tile,
+                       MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
+                       int_mv *mv_ref_list,
+                       int mi_row, int mi_col) {
+  const int *ref_sign_bias = cm->ref_frame_sign_bias;
+  int i, refmv_count = 0;
+
+  const POSITION *const mv_ref_search = mv_ref_blocks[mi->mbmi.sb_type];
+
+  int different_ref_found = 0;
+  int context_counter = 0;
+  int const_motion = 0;
+
+  // Blank the reference vector list
+  vpx_memset(mv_ref_list, 0, sizeof(*mv_ref_list) * MAX_MV_REF_CANDIDATES);
+
+  // The nearest 2 blocks are treated differently
+  // if the size < 8x8 we get the mv from the bmi substructure,
+  // and we also need to keep a mode count.
+  for (i = 0; i < 2; ++i) {
+    const POSITION *const mv_ref = &mv_ref_search[i];
+    if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
+      const MODE_INFO *const candidate_mi = xd->mi[mv_ref->col + mv_ref->row *
+                                                   xd->mi_stride];
+      const MB_MODE_INFO *const candidate = &candidate_mi->mbmi;
+      // Keep counts for entropy encoding.
+      context_counter += mode_2_counter[candidate->mode];
+      different_ref_found = 1;
+
+      if (candidate->ref_frame[0] == ref_frame)
+        ADD_MV_REF_LIST(get_sub_block_mv(candidate_mi, 0, mv_ref->col, -1));
+    }
+  }
+
+  const_motion = 1;
+
+  // Check the rest of the neighbors in much the same way
+  // as before except we don't need to keep track of sub blocks or
+  // mode counts.
+  for (; i < MVREF_NEIGHBOURS && !refmv_count; ++i) {
+    const POSITION *const mv_ref = &mv_ref_search[i];
+    if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
+      const MB_MODE_INFO *const candidate = &xd->mi[mv_ref->col + mv_ref->row *
+                                                    xd->mi_stride]->mbmi;
+      different_ref_found = 1;
+
+      if (candidate->ref_frame[0] == ref_frame)
+        ADD_MV_REF_LIST(candidate->mv[0]);
+    }
+  }
+
+  // Since we couldn't find 2 mvs from the same reference frame
+  // go back through the neighbors and find motion vectors from
+  // different reference frames.
+  if (different_ref_found && !refmv_count) {
+    for (i = 0; i < MVREF_NEIGHBOURS; ++i) {
+      const POSITION *mv_ref = &mv_ref_search[i];
+      if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
+        const MB_MODE_INFO *const candidate = &xd->mi[mv_ref->col + mv_ref->row
+                                              * xd->mi_stride]->mbmi;
+
+        // If the candidate is INTRA we don't want to consider its mv.
+        IF_DIFF_REF_FRAME_ADD_MV(candidate);
+      }
+    }
+  }
+
+ Done:
+
+  mi->mbmi.mode_context[ref_frame] = counter_to_context[context_counter];
+
+  // Clamp vectors
+  for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i)
+    clamp_mv_ref(&mv_ref_list[i].as_mv, xd);
+
+  return const_motion;
+}
+
 static void full_pixel_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
                                     BLOCK_SIZE bsize, int mi_row, int mi_col,
                                     int_mv *tmp_mv, int *rate_mv) {
@@ -172,15 +252,31 @@ static void model_rd_for_sb_y(VP9_COMP *cpi, BLOCK_SIZE bsize,
   else
     x->skip_txfm = 0;
 
-  // TODO(jingning) This is a temporary solution to account for frames with
-  // light changes. Need to customize the rate-distortion modeling for non-RD
-  // mode decision.
-  if ((sse >> 3) > var)
-    sse = var;
-  vp9_model_rd_from_var_lapndz(var + sse, 1 << num_pels_log2_lookup[bsize],
-                               ac_quant >> 3, &rate, &dist);
-  *out_rate_sum = rate;
+  vp9_model_rd_from_var_lapndz(sse - var, 1 << num_pels_log2_lookup[bsize],
+                               dc_quant >> 3, &rate, &dist);
+  *out_rate_sum = rate >> 1;
   *out_dist_sum = dist << 3;
+
+  vp9_model_rd_from_var_lapndz(var, 1 << num_pels_log2_lookup[bsize],
+                               ac_quant >> 3, &rate, &dist);
+  *out_rate_sum += rate;
+  *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.
@@ -228,6 +324,32 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
   int bsl = mi_width_log2_lookup[bsize];
   const int pred_filter_search = (((mi_row + mi_col) >> bsl) +
                                       get_chessboard_index(cm)) % 2;
+  int const_motion[MAX_REF_FRAMES] = { 0 };
+
+  // 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;
 
@@ -241,18 +363,36 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
   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;
 
   for (ref_frame = LAST_FRAME; ref_frame <= LAST_FRAME ; ++ref_frame) {
     x->pred_mv_sad[ref_frame] = INT_MAX;
     if (cpi->ref_frame_flags & flag_list[ref_frame]) {
-      vp9_setup_buffer_inter(cpi, x, tile,
-                             ref_frame, bsize, mi_row, mi_col,
-                             frame_mv[NEARESTMV], frame_mv[NEARMV], yv12_mb);
+      const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, ref_frame);
+      int_mv *const candidates = mbmi->ref_mvs[ref_frame];
+      const struct scale_factors *const sf = &cm->frame_refs[ref_frame - 1].sf;
+      vp9_setup_pred_block(xd, yv12_mb[ref_frame], yv12, mi_row, mi_col,
+                           sf, sf);
+
+      if (cm->coding_use_prev_mi)
+        vp9_find_mv_refs(cm, xd, tile, xd->mi[0], ref_frame,
+                         candidates, mi_row, mi_col);
+      else
+        const_motion[ref_frame] = mv_refs_rt(cm, xd, tile, xd->mi[0],
+                                             ref_frame, candidates,
+                                             mi_row, mi_col);
+
+      vp9_find_best_ref_mvs(xd, cm->allow_high_precision_mv, candidates,
+                            &frame_mv[NEARESTMV][ref_frame],
+                            &frame_mv[NEARMV][ref_frame]);
+
+      if (!vp9_is_scaled(sf) && bsize >= BLOCK_8X8)
+        vp9_mv_pred(cpi, x, yv12_mb[ref_frame][0].buf, yv12->y_stride,
+                    ref_frame, bsize);
     }
     frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
     frame_mv[ZEROMV][ref_frame].as_int = 0;
@@ -286,6 +426,10 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
     for (this_mode = NEARESTMV; this_mode <= NEWMV; ++this_mode) {
       int rate_mv = 0;
 
+      if (const_motion[ref_frame] &&
+          (this_mode == NEARMV || this_mode == ZEROMV))
+        continue;
+
       if (!(cpi->sf.inter_mode_mask[bsize] & (1 << this_mode)))
         continue;
 
@@ -324,6 +468,16 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
       // 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 +488,7 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
         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;
@@ -348,8 +503,24 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
             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];
@@ -370,17 +541,17 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
 
       // Skipping checking: test to see if this block can be reconstructed by
       // prediction only.
-      if (!x->in_active_map) {
-        x->skip = 1;
-      } else if (cpi->allow_encode_breakout && x->encode_breakout) {
+      if (cpi->allow_encode_breakout) {
         const BLOCK_SIZE uv_size = get_plane_block_size(bsize, &xd->plane[1]);
         unsigned int var = var_y, sse = sse_y;
         // Skipping threshold for ac.
         unsigned int thresh_ac;
         // Skipping threshold for dc.
         unsigned int thresh_dc;
+        if (x->encode_breakout > 0) {
           // Set a maximum for threshold to avoid big PSNR loss in low bit rate
-        // case. Use extreme low threshold for static frames to limit skipping.
+          // case. Use extreme low threshold for static frames to limit
+          // skipping.
           const unsigned int max_thresh = 36000;
           // The encode_breakout input
           const unsigned int min_thresh =
@@ -391,10 +562,14 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
           thresh_ac = clamp(thresh_ac, min_thresh, max_thresh);
 
           // Adjust ac threshold according to partition size.
-        thresh_ac >>= 8 - (b_width_log2_lookup[bsize] +
-            b_height_log2_lookup[bsize]);
+          thresh_ac >>=
+              8 - (b_width_log2_lookup[bsize] + b_height_log2_lookup[bsize]);
 
           thresh_dc = (xd->plane[0].dequant[0] * xd->plane[0].dequant[0] >> 6);
+        } else {
+          thresh_ac = 0;
+          thresh_dc = 0;
+        }
 
         // Y skipping condition checking for ac and dc.
         if (var <= thresh_ac && (sse - var) <= thresh_dc) {
@@ -451,6 +626,16 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
         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)
@@ -458,6 +643,19 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
     }
   }
 
+  // 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;
@@ -469,18 +667,51 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
   // Perform intra prediction search, if the best SAD is above a certain
   // threshold.
   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) {
-      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);
+      bsize <= cpi->sf.max_intra_bsize) {
+    int i, j;
+    const int step   = 1 << mbmi->tx_size;
+    const int width  = num_4x4_blocks_wide_lookup[bsize];
+    const int height = num_4x4_blocks_high_lookup[bsize];
 
+    int rate2 = 0;
+    int64_t dist2 = 0;
+    const int dst_stride = pd->dst.stride;
+    const int src_stride = p->src.stride;
+    int block_idx = 0;
+
+    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;
+      }
+
+      for (j = 0; j < height; j += step) {
+        for (i = 0; i < width; i += step) {
+          vp9_predict_intra_block(xd, block_idx, b_width_log2(bsize),
+                                  mbmi->tx_size, this_mode,
+#if CONFIG_FILTERINTRA
+                                  0,
+#endif
+                                  &p->src.buf[4 * (j * dst_stride + i)],
+                                  src_stride,
+                                  &pd->dst.buf[4 * (j * dst_stride + i)],
+                                  dst_stride, i, j, 0);
           model_rd_for_sb_y(cpi, bsize, x, xd, &rate, &dist, &var_y, &sse_y);
+          rate2 += rate;
+          dist2 += dist;
+          ++block_idx;
+        }
+      }
+      rate = rate2;
+      dist = dist2;
+
       rate += cpi->mbmode_cost[this_mode];
       rate += intra_cost_penalty;
       this_rd = RDCOST(x->rdmult, x->rddiv, rate, dist);
 
+      if (cpi->sf.reuse_inter_pred_sby)
+        pd->dst = orig_dst;
+
       if (this_rd + intra_mode_cost < best_rd) {
         best_rd = this_rd;
         *returnrate = rate;
@@ -494,9 +725,9 @@ int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
       }
     }
   }
+
 #if CONFIG_DENOISING
-  vp9_denoiser_denoise(&cpi->denoiser, x, cpi->common.mi_grid_visible, mi_row,
-                       mi_col, bsize);
+  vp9_denoiser_denoise(&cpi->denoiser, x, mi_row, mi_col, bsize);
 #endif
 
   return INT64_MAX;

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,

vp9/encoder/vp9_ratectrl.c

@@ -186,6 +186,8 @@ static void update_buffer_level(VP9_COMP *cpi, int encoded_frame_size) {
 }
 
 void vp9_rc_init(const VP9EncoderConfig *oxcf, int pass, RATE_CONTROL *rc) {
+  int i;
+
   if (pass == 0 && oxcf->rc_mode == VPX_CBR) {
     rc->avg_frame_qindex[KEY_FRAME] = oxcf->worst_allowed_q;
     rc->avg_frame_qindex[INTER_FRAME] = oxcf->worst_allowed_q;
@@ -227,9 +229,9 @@ void vp9_rc_init(const VP9EncoderConfig *oxcf, int pass, RATE_CONTROL *rc) {
   rc->tot_q = 0.0;
   rc->avg_q = vp9_convert_qindex_to_q(oxcf->worst_allowed_q);
 
-  rc->rate_correction_factor = 1.0;
-  rc->key_frame_rate_correction_factor = 1.0;
-  rc->gf_rate_correction_factor = 1.0;
+  for (i = 0; i < RATE_FACTOR_LEVELS; ++i) {
+    rc->rate_correction_factors[i] = 1.0;
+  }
 }
 
 int vp9_rc_drop_frame(VP9_COMP *cpi) {
@@ -271,28 +273,40 @@ int vp9_rc_drop_frame(VP9_COMP *cpi) {
 }
 
 static double get_rate_correction_factor(const VP9_COMP *cpi) {
+  const RATE_CONTROL *const rc = &cpi->rc;
+
   if (cpi->common.frame_type == KEY_FRAME) {
-    return cpi->rc.key_frame_rate_correction_factor;
+    return rc->rate_correction_factors[KF_STD];
+  } else if (cpi->pass == 2) {
+    RATE_FACTOR_LEVEL rf_lvl =
+      cpi->twopass.gf_group.rf_level[cpi->twopass.gf_group.index];
+    return rc->rate_correction_factors[rf_lvl];
   } else {
     if ((cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame) &&
-        !cpi->rc.is_src_frame_alt_ref &&
+        !rc->is_src_frame_alt_ref &&
         !(cpi->use_svc && cpi->oxcf.rc_mode == VPX_CBR))
-      return cpi->rc.gf_rate_correction_factor;
+      return rc->rate_correction_factors[GF_ARF_STD];
     else
-      return cpi->rc.rate_correction_factor;
+      return rc->rate_correction_factors[INTER_NORMAL];
   }
 }
 
 static void set_rate_correction_factor(VP9_COMP *cpi, double factor) {
+  RATE_CONTROL *const rc = &cpi->rc;
+
   if (cpi->common.frame_type == KEY_FRAME) {
-    cpi->rc.key_frame_rate_correction_factor = factor;
+    rc->rate_correction_factors[KF_STD] = factor;
+  } else if (cpi->pass == 2) {
+    RATE_FACTOR_LEVEL rf_lvl =
+      cpi->twopass.gf_group.rf_level[cpi->twopass.gf_group.index];
+    rc->rate_correction_factors[rf_lvl] = factor;
   } else {
     if ((cpi->refresh_alt_ref_frame || cpi->refresh_golden_frame) &&
-        !cpi->rc.is_src_frame_alt_ref &&
+        !rc->is_src_frame_alt_ref &&
         !(cpi->use_svc && cpi->oxcf.rc_mode == VPX_CBR))
-      cpi->rc.gf_rate_correction_factor = factor;
+      rc->rate_correction_factors[GF_ARF_STD] = factor;
     else
-      cpi->rc.rate_correction_factor = factor;
+      rc->rate_correction_factors[INTER_NORMAL] = factor;
   }
 }
 
@@ -628,8 +642,8 @@ static int rc_pick_q_and_bounds_one_pass_vbr(const VP9_COMP *cpi,
 
   if (frame_is_intra_only(cm)) {
     active_best_quality = rc->best_quality;
-#if !CONFIG_MULTIPLE_ARF
-    // Handle the special case for key frames forced when we have75 reached
+
+    // Handle the special case for key frames forced when we have reached
     // the maximum key frame interval. Here force the Q to a range
     // based on the ambient Q to reduce the risk of popping.
     if (rc->this_key_frame_forced) {
@@ -660,13 +674,6 @@ static int rc_pick_q_and_bounds_one_pass_vbr(const VP9_COMP *cpi,
       active_best_quality += vp9_compute_qdelta(rc, q_val,
                                                 q_val * q_adj_factor);
     }
-#else
-    double current_q;
-    // Force the KF quantizer to be 30% of the active_worst_quality.
-    current_q = vp9_convert_qindex_to_q(active_worst_quality);
-    active_best_quality = active_worst_quality
-        + vp9_compute_qdelta(rc, current_q, current_q * 0.3);
-#endif
   } else if (!rc->is_src_frame_alt_ref &&
              (cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)) {
     // Use the lower of active_worst_quality and recent
@@ -768,23 +775,7 @@ static int rc_pick_q_and_bounds_one_pass_vbr(const VP9_COMP *cpi,
         q = *top_index;
     }
   }
-#if CONFIG_MULTIPLE_ARF
-  // Force the quantizer determined by the coding order pattern.
-  if (cpi->multi_arf_enabled && (cm->frame_type != KEY_FRAME) &&
-      cpi->oxcf.rc_mode != VPX_Q) {
-    double new_q;
-    double current_q = vp9_convert_qindex_to_q(active_worst_quality);
-    int level = cpi->this_frame_weight;
-    assert(level >= 0);
-    new_q = current_q * (1.0 - (0.2 * (cpi->max_arf_level - level)));
-    q = active_worst_quality +
-        vp9_compute_qdelta(rc, current_q, new_q);
 
-    *bottom_index = q;
-    *top_index    = q;
-    printf("frame:%d q:%d\n", cm->current_video_frame, q);
-  }
-#endif
   assert(*top_index <= rc->worst_quality &&
          *top_index >= rc->best_quality);
   assert(*bottom_index <= rc->worst_quality &&
@@ -805,7 +796,6 @@ static int rc_pick_q_and_bounds_two_pass(const VP9_COMP *cpi,
   int q;
 
   if (frame_is_intra_only(cm) || vp9_is_upper_layer_key_frame(cpi)) {
-#if !CONFIG_MULTIPLE_ARF
     // Handle the special case for key frames forced when we have75 reached
     // the maximum key frame interval. Here force the Q to a range
     // based on the ambient Q to reduce the risk of popping.
@@ -840,13 +830,6 @@ static int rc_pick_q_and_bounds_two_pass(const VP9_COMP *cpi,
       active_best_quality += vp9_compute_qdelta(rc, q_val,
                                                 q_val * q_adj_factor);
     }
-#else
-    double current_q;
-    // Force the KF quantizer to be 30% of the active_worst_quality.
-    current_q = vp9_convert_qindex_to_q(active_worst_quality);
-    active_best_quality = active_worst_quality
-        + vp9_compute_qdelta(rc, current_q, current_q * 0.3);
-#endif
   } else if (!rc->is_src_frame_alt_ref &&
              (cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)) {
     // Use the lower of active_worst_quality and recent
@@ -909,21 +892,20 @@ static int rc_pick_q_and_bounds_two_pass(const VP9_COMP *cpi,
   *bottom_index = active_best_quality;
 
 #if LIMIT_QRANGE_FOR_ALTREF_AND_KEY
-  {
-    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_golden_frame || 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;
   }
@@ -945,23 +927,7 @@ static int rc_pick_q_and_bounds_two_pass(const VP9_COMP *cpi,
         q = *top_index;
     }
   }
-#if CONFIG_MULTIPLE_ARF
-  // Force the quantizer determined by the coding order pattern.
-  if (cpi->multi_arf_enabled && (cm->frame_type != KEY_FRAME) &&
-      cpi->oxcf.rc_mode != VPX_Q) {
-    double new_q;
-    double current_q = vp9_convert_qindex_to_q(active_worst_quality);
-    int level = cpi->this_frame_weight;
-    assert(level >= 0);
-    new_q = current_q * (1.0 - (0.2 * (cpi->max_arf_level - level)));
-    q = active_worst_quality +
-        vp9_compute_qdelta(rc, current_q, new_q);
 
-    *bottom_index = q;
-    *top_index    = q;
-    printf("frame:%d q:%d\n", cm->current_video_frame, q);
-  }
-#endif
   assert(*top_index <= rc->worst_quality &&
          *top_index >= rc->best_quality);
   assert(*bottom_index <= rc->worst_quality &&
@@ -1026,10 +992,7 @@ static void update_alt_ref_frame_stats(VP9_COMP *cpi) {
   RATE_CONTROL *const rc = &cpi->rc;
   rc->frames_since_golden = 0;
 
-#if CONFIG_MULTIPLE_ARF
-  if (!cpi->multi_arf_enabled)
-#endif
-    // Clear the alternate reference update pending flag.
+  // Mark the alt ref as done (setting to 0 means no further alt refs pending).
   rc->source_alt_ref_pending = 0;
 
   // Set the alternate reference frame active flag
@@ -1044,8 +1007,13 @@ static void update_golden_frame_stats(VP9_COMP *cpi) {
     // this frame refreshes means next frames don't unless specified by user
     rc->frames_since_golden = 0;
 
-    if (!rc->source_alt_ref_pending)
+    if (cpi->pass == 2) {
+      if (!rc->source_alt_ref_pending &&
+          cpi->twopass.gf_group.rf_level[0] == GF_ARF_STD)
       rc->source_alt_ref_active = 0;
+    } else if (!rc->source_alt_ref_pending) {
+      rc->source_alt_ref_active = 0;
+    }
 
     // Decrement count down till next gf
     if (rc->frames_till_gf_update_due > 0)
@@ -1388,6 +1356,8 @@ void vp9_rc_set_gf_max_interval(const VP9EncoderConfig *const oxcf,
 
   // 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)

vp9/encoder/vp9_ratectrl.h

@@ -23,6 +23,15 @@ extern "C" {
 // Bits Per MB at different Q (Multiplied by 512)
 #define BPER_MB_NORMBITS    9
 
+typedef enum {
+  INTER_NORMAL = 0,
+  INTER_HIGH = 1,
+  GF_ARF_LOW = 2,
+  GF_ARF_STD = 3,
+  KF_STD = 4,
+  RATE_FACTOR_LEVELS = 5
+} RATE_FACTOR_LEVEL;
+
 typedef struct {
   // Rate targetting variables
   int base_frame_target;           // A baseline frame target before adjustment
@@ -37,9 +46,7 @@ typedef struct {
   int last_boost;
   int kf_boost;
 
-  double rate_correction_factor;
-  double key_frame_rate_correction_factor;
-  double gf_rate_correction_factor;
+  double rate_correction_factors[RATE_FACTOR_LEVELS];
 
   int frames_since_golden;
   int frames_till_gf_update_due;

vp9/encoder/vp9_rdopt.h

@@ -13,7 +13,10 @@
 
 #include <limits.h>
 
-#include "vp9/encoder/vp9_encoder.h"
+#include "vp9/common/vp9_blockd.h"
+
+#include "vp9/encoder/vp9_block.h"
+#include "vp9/encoder/vp9_context_tree.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -30,21 +33,125 @@ extern "C" {
 
 #define INVALID_MV 0x80008000
 
+#if !CONFIG_INTERINTRA
+#define MAX_MODES 30
+#else
+#define MAX_MODES 42
+#endif
+#define MAX_REFS  6
+
+// This enumerator type needs to be kept aligned with the mode order in
+// const MODE_DEFINITION vp9_mode_order[MAX_MODES] used in the rd code.
+typedef enum {
+  THR_NEARESTMV,
+  THR_NEARESTA,
+  THR_NEARESTG,
+
+  THR_DC,
+
+  THR_NEWMV,
+  THR_NEWA,
+  THR_NEWG,
+
+  THR_NEARMV,
+  THR_NEARA,
+  THR_COMP_NEARESTLA,
+  THR_COMP_NEARESTGA,
+
+  THR_TM,
+
+  THR_COMP_NEARLA,
+  THR_COMP_NEWLA,
+  THR_NEARG,
+  THR_COMP_NEARGA,
+  THR_COMP_NEWGA,
+
+  THR_ZEROMV,
+  THR_ZEROG,
+  THR_ZEROA,
+  THR_COMP_ZEROLA,
+  THR_COMP_ZEROGA,
+
+  THR_H_PRED,
+  THR_V_PRED,
+  THR_D135_PRED,
+  THR_D207_PRED,
+  THR_D153_PRED,
+  THR_D63_PRED,
+  THR_D117_PRED,
+  THR_D45_PRED,
+
+#if CONFIG_INTERINTRA
+  THR_COMP_INTERINTRA_ZEROL,
+  THR_COMP_INTERINTRA_NEARESTL,
+  THR_COMP_INTERINTRA_NEARL,
+  THR_COMP_INTERINTRA_NEWL,
+
+  THR_COMP_INTERINTRA_ZEROG,
+  THR_COMP_INTERINTRA_NEARESTG,
+  THR_COMP_INTERINTRA_NEARG,
+  THR_COMP_INTERINTRA_NEWG,
+
+  THR_COMP_INTERINTRA_ZEROA,
+  THR_COMP_INTERINTRA_NEARESTA,
+  THR_COMP_INTERINTRA_NEARA,
+  THR_COMP_INTERINTRA_NEWA,
+#endif
+} THR_MODES;
+
+typedef enum {
+  THR_LAST,
+  THR_GOLD,
+  THR_ALTR,
+  THR_COMP_LA,
+  THR_COMP_GA,
+  THR_INTRA,
+} THR_MODES_SUB8X8;
+
+typedef struct RD_OPT {
+  // Thresh_mult is used to set a threshold for the rd score. A higher value
+  // means that we will accept the best mode so far more often. This number
+  // is used in combination with the current block size, and thresh_freq_fact
+  // to pick a threshold.
+  int thresh_mult[MAX_MODES];
+  int thresh_mult_sub8x8[MAX_REFS];
+
+  int threshes[MAX_SEGMENTS][BLOCK_SIZES][MAX_MODES];
+  int thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
+
+  int64_t comp_pred_diff[REFERENCE_MODES];
+  int64_t prediction_type_threshes[MAX_REF_FRAMES][REFERENCE_MODES];
+  int64_t tx_select_diff[TX_MODES];
+  // FIXME(rbultje) can this overflow?
+  int tx_select_threshes[MAX_REF_FRAMES][TX_MODES];
+
+  int64_t filter_diff[SWITCHABLE_FILTER_CONTEXTS];
+  int64_t filter_threshes[MAX_REF_FRAMES][SWITCHABLE_FILTER_CONTEXTS];
+  int64_t filter_cache[SWITCHABLE_FILTER_CONTEXTS];
+  int64_t mask_filter;
+
+  int RDMULT;
+  int RDDIV;
+} RD_OPT;
+
+
 struct TileInfo;
+struct VP9_COMP;
+struct macroblock;
 
-int vp9_compute_rd_mult(const VP9_COMP *cpi, int qindex);
+int vp9_compute_rd_mult(const struct VP9_COMP *cpi, int qindex);
 
-void vp9_initialize_rd_consts(VP9_COMP *cpi);
+void vp9_initialize_rd_consts(struct VP9_COMP *cpi);
 
-void vp9_initialize_me_consts(VP9_COMP *cpi, int qindex);
+void vp9_initialize_me_consts(struct VP9_COMP *cpi, int qindex);
 
 void vp9_model_rd_from_var_lapndz(unsigned int var, unsigned int n,
                                   unsigned int qstep, int *rate,
                                   int64_t *dist);
 
-int vp9_get_switchable_rate(const VP9_COMP *cpi);
+int vp9_get_switchable_rate(const struct VP9_COMP *cpi);
 
-void vp9_setup_buffer_inter(VP9_COMP *cpi, MACROBLOCK *x,
+void vp9_setup_buffer_inter(struct VP9_COMP *cpi, struct macroblock *x,
                             const TileInfo *const tile,
                             MV_REFERENCE_FRAME ref_frame,
                             BLOCK_SIZE block_size,
@@ -53,26 +160,43 @@ void vp9_setup_buffer_inter(VP9_COMP *cpi, MACROBLOCK *x,
                             int_mv frame_near_mv[MAX_REF_FRAMES],
                             struct buf_2d yv12_mb[4][MAX_MB_PLANE]);
 
-const YV12_BUFFER_CONFIG *vp9_get_scaled_ref_frame(const VP9_COMP *cpi,
+const YV12_BUFFER_CONFIG *vp9_get_scaled_ref_frame(const struct VP9_COMP *cpi,
                                                    int ref_frame);
 
-void vp9_rd_pick_intra_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
+void vp9_rd_pick_intra_mode_sb(struct VP9_COMP *cpi, struct macroblock *x,
                                int *r, int64_t *d, BLOCK_SIZE bsize,
                                PICK_MODE_CONTEXT *ctx, int64_t best_rd);
 
-int64_t vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
+int64_t vp9_rd_pick_inter_mode_sb(struct VP9_COMP *cpi, struct macroblock *x,
                                   const struct TileInfo *const tile,
                                   int mi_row, int mi_col,
                                   int *returnrate,
+#if CONFIG_SUPERTX
+                                  int *returnrate_nocoef,
+#endif
+                                  int64_t *returndistortion,
+                                  BLOCK_SIZE bsize,
+                                  PICK_MODE_CONTEXT *ctx,
+                                  int64_t best_rd_so_far);
+
+int64_t vp9_rd_pick_inter_mode_sb_seg_skip(struct VP9_COMP *cpi,
+                                           struct macroblock *x,
+                                           const TileInfo *const tile,
+                                           int mi_row, int mi_col,
+                                           int *returnrate,
                                            int64_t *returndistortion,
                                            BLOCK_SIZE bsize,
                                            PICK_MODE_CONTEXT *ctx,
                                            int64_t best_rd_so_far);
 
-int64_t vp9_rd_pick_inter_mode_sub8x8(VP9_COMP *cpi, MACROBLOCK *x,
+int64_t vp9_rd_pick_inter_mode_sub8x8(struct VP9_COMP *cpi,
+                                      struct macroblock *x,
                                       const struct TileInfo *const tile,
                                       int mi_row, int mi_col,
                                       int *returnrate,
+#if CONFIG_SUPERTX
+                                      int *returnrate_nocoef,
+#endif
                                       int64_t *returndistortion,
                                       BLOCK_SIZE bsize,
                                       PICK_MODE_CONTEXT *ctx,
@@ -85,15 +209,40 @@ void vp9_get_entropy_contexts(BLOCK_SIZE bsize, TX_SIZE tx_size,
                               ENTROPY_CONTEXT t_above[16],
                               ENTROPY_CONTEXT t_left[16]);
 
-void vp9_set_rd_speed_thresholds(VP9_COMP *cpi);
+void vp9_set_rd_speed_thresholds(struct VP9_COMP *cpi);
 
-void vp9_set_rd_speed_thresholds_sub8x8(VP9_COMP *cpi);
+void vp9_set_rd_speed_thresholds_sub8x8(struct VP9_COMP *cpi);
 
 static INLINE int rd_less_than_thresh(int64_t best_rd, int thresh,
                                       int thresh_fact) {
     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);
+
+#if CONFIG_SUPERTX
+void txfm_rd_in_plane_supertx(MACROBLOCK *x,
+                              int *rate, int64_t *distortion,
+                              int *skippable, int64_t *sse,
+                              int64_t ref_best_rd, int plane,
+                              BLOCK_SIZE bsize, TX_SIZE tx_size,
+                              int use_fast_coef_casting);
+void txfm_rd_in_plane(MACROBLOCK *x,
+                      int *rate, int64_t *distortion,
+                      int *skippable, int64_t *sse,
+                      int64_t ref_best_rd, int plane,
+                      BLOCK_SIZE bsize, TX_SIZE tx_size,
+                      int use_fast_coef_casting);
+#endif
 #ifdef __cplusplus
 }  // extern "C"
 #endif

vp9/encoder/vp9_sad.c

@@ -131,3 +131,47 @@ sadMxN(4, 4)
 sadMxNxK(4, 4, 3)
 sadMxNxK(4, 4, 8)
 sadMxNx4D(4, 4)
+
+#if ((CONFIG_MASKED_INTERINTRA && CONFIG_INTERINTRA) || \
+     CONFIG_MASKED_INTERINTER)
+static INLINE unsigned int masked_sad(const uint8_t *a, int a_stride,
+                                      const uint8_t *b, int b_stride,
+                                      const uint8_t *m, int m_stride,
+                                      int width, int height) {
+  int y, x;
+  unsigned int sad = 0;
+
+  for (y = 0; y < height; y++) {
+    for (x = 0; x < width; x++)
+      sad += m[x] * abs(a[x] - b[x]);
+
+    a += a_stride;
+    b += b_stride;
+    m += m_stride;
+  }
+  sad = (sad + 31) >> 6;
+
+  return sad;
+}
+
+#define MASKSADMxN(m, n) \
+unsigned int vp9_masked_sad##m##x##n##_c(const uint8_t *src, int src_stride, \
+                                         const uint8_t *ref, int ref_stride, \
+                                         const uint8_t *msk, int msk_stride) { \
+  return masked_sad(src, src_stride, ref, ref_stride, msk, msk_stride, m, n); \
+}
+
+MASKSADMxN(64, 64)
+MASKSADMxN(64, 32)
+MASKSADMxN(32, 64)
+MASKSADMxN(32, 32)
+MASKSADMxN(32, 16)
+MASKSADMxN(16, 32)
+MASKSADMxN(16, 16)
+MASKSADMxN(16, 8)
+MASKSADMxN(8, 16)
+MASKSADMxN(8, 8)
+MASKSADMxN(8, 4)
+MASKSADMxN(4, 8)
+MASKSADMxN(4, 4)
+#endif

vp9/encoder/vp9_speed_features.c

@@ -84,16 +84,17 @@ static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm,
 
   if (speed >= 2) {
     if (MIN(cm->width, cm->height) >= 720) {
-      sf->lf_motion_threshold = LOW_MOITION_THRESHOLD;
+      sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
       sf->last_partitioning_redo_frequency = 3;
       sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
                                               : DISABLE_ALL_INTER_SPLIT;
     } else {
       sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY;
       sf->last_partitioning_redo_frequency = 2;
-      sf->lf_motion_threshold = NO_MOITION_THRESHOLD;
+      sf->lf_motion_threshold = NO_MOTION_THRESHOLD;
     }
-    sf->adaptive_pred_interp_filter = 2;
+
+    sf->adaptive_pred_interp_filter = 0;
     sf->reference_masking = 1;
     sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
                                  FLAG_SKIP_INTRA_BESTINTER |
@@ -114,7 +115,7 @@ static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm,
     else
       sf->disable_split_mask = DISABLE_ALL_INTER_SPLIT;
 
-    sf->lf_motion_threshold = LOW_MOITION_THRESHOLD;
+    sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
     sf->last_partitioning_redo_frequency = 3;
     sf->recode_loop = ALLOW_RECODE_KFMAXBW;
     sf->adaptive_rd_thresh = 3;
@@ -198,7 +199,7 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
     sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
     sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
     sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION;
-    sf->lf_motion_threshold = LOW_MOITION_THRESHOLD;
+    sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
     sf->adjust_partitioning_from_last_frame = 1;
     sf->last_partitioning_redo_frequency = 3;
     sf->use_lp32x32fdct = 1;
@@ -249,6 +250,8 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
   }
 
   if (speed >= 5) {
+    sf->auto_min_max_partition_size = (cm->frame_type == KEY_FRAME) ?
+        RELAXED_NEIGHBORING_MIN_MAX : STRICT_NEIGHBORING_MIN_MAX;
     sf->max_partition_size = BLOCK_32X32;
     sf->min_partition_size = BLOCK_8X8;
     sf->partition_check =
@@ -270,6 +273,10 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
     sf->source_var_thresh = 360;
 
     sf->tx_size_search_method = USE_TX_8X8;
+    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) {
@@ -335,6 +342,7 @@ void vp9_set_speed_features(VP9_COMP *cpi) {
   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;

vp9/encoder/vp9_speed_features.h

@@ -44,8 +44,8 @@ typedef enum {
 } SUBPEL_SEARCH_METHODS;
 
 typedef enum {
-  NO_MOITION_THRESHOLD = 0,
-  LOW_MOITION_THRESHOLD = 7
+  NO_MOTION_THRESHOLD = 0,
+  LOW_MOTION_THRESHOLD = 7
 } MOTION_THRESHOLD;
 
 typedef enum {
@@ -353,6 +353,11 @@ typedef struct SPEED_FEATURES {
 
   // 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;

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;
@@ -31,6 +32,7 @@ void vp9_init_layer_context(VP9_COMP *const cpi) {
   for (layer = 0; layer < layer_end; ++layer) {
     LAYER_CONTEXT *const lc = &svc->layer_context[layer];
     RATE_CONTROL *const lrc = &lc->rc;
+    int i;
     lc->current_video_frame_in_layer = 0;
     lrc->avg_frame_qindex[INTER_FRAME] = oxcf->worst_allowed_q;
     lrc->ni_av_qi = oxcf->worst_allowed_q;
@@ -42,8 +44,10 @@ void vp9_init_layer_context(VP9_COMP *const cpi) {
     lrc->ni_frames = 0;
     lrc->decimation_count = 0;
     lrc->decimation_factor = 0;
-    lrc->rate_correction_factor = 1.0;
-    lrc->key_frame_rate_correction_factor = 1.0;
+
+    for (i = 0; i < RATE_FACTOR_LEVELS; ++i) {
+      lrc->rate_correction_factors[i] = 1.0;
+    }
 
     if (svc->number_temporal_layers > 1) {
       lc->target_bandwidth = oxcf->ts_target_bitrate[layer];
@@ -206,3 +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;
+}

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

vp9/encoder/vp9_tokenize.c

@@ -334,3 +334,41 @@ void vp9_tokenize_sb(VP9_COMP *cpi, TOKENEXTRA **t, int dry_run,
     *t = t_backup;
   }
 }
+
+#if CONFIG_SUPERTX
+void vp9_tokenize_sb_supertx(VP9_COMP *cpi, TOKENEXTRA **t, int dry_run,
+                             BLOCK_SIZE bsize) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &cpi->mb.e_mbd;
+  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  TOKENEXTRA *t_backup = *t;
+  const int ctx = vp9_get_skip_context(xd);
+  const int skip_inc = !vp9_segfeature_active(&cm->seg, mbmi->segment_id,
+                                              SEG_LVL_SKIP);
+  struct tokenize_b_args arg = {cpi, xd, t};
+  int plane;
+  if (mbmi->skip) {
+    if (!dry_run)
+      cm->counts.skip[ctx][1] += skip_inc;
+    reset_skip_context(xd, bsize);
+    if (dry_run)
+      *t = t_backup;
+    return;
+  }
+
+  if (!dry_run) {
+    cm->counts.skip[ctx][0] += skip_inc;
+    for (plane = 0; plane < MAX_MB_PLANE; plane++) {
+      BLOCK_SIZE plane_size = plane ? (bsize - 3) : bsize;
+      tokenize_b(plane, 0, plane_size, b_width_log2(plane_size), &arg);
+    }
+  } else {
+    for (plane = 0; plane < MAX_MB_PLANE; plane++) {
+      BLOCK_SIZE plane_size = plane ? (bsize - 3) : bsize;
+      set_entropy_context_b(plane, 0, plane_size, b_width_log2(plane_size),
+                            &arg);
+    }
+    *t = t_backup;
+  }
+}
+#endif

vp9/encoder/vp9_tokenize.h

@@ -46,6 +46,10 @@ struct VP9_COMP;
 
 void vp9_tokenize_sb(struct VP9_COMP *cpi, TOKENEXTRA **t, int dry_run,
                      BLOCK_SIZE bsize);
+#if CONFIG_SUPERTX
+void vp9_tokenize_sb_supertx(struct VP9_COMP *cpi, TOKENEXTRA **t, int dry_run,
+                             BLOCK_SIZE bsize);
+#endif
 
 extern const int16_t *vp9_dct_value_cost_ptr;
 /* TODO: The Token field should be broken out into a separate char array to

vp9/encoder/vp9_variance.c

@@ -266,3 +266,98 @@ void vp9_comp_avg_pred(uint8_t *comp_pred, const uint8_t *pred, int width,
     ref += ref_stride;
   }
 }
+
+#if ((CONFIG_MASKED_INTERINTRA && CONFIG_INTERINTRA) || \
+    CONFIG_MASKED_INTERINTER)
+void masked_variance(const uint8_t *a, int  a_stride,
+                     const uint8_t *b, int  b_stride,
+                     const uint8_t *m, int  m_stride,
+                     int  w, int  h, unsigned int *sse, int *sum) {
+  int i, j;
+
+  *sum = 0;
+  *sse = 0;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      const int diff = (a[j] - b[j]) * (m[j]);
+      *sum += diff;
+      *sse += diff * diff;
+    }
+
+    a += a_stride;
+    b += b_stride;
+    m += m_stride;
+  }
+  *sum = (*sum >= 0) ? ((*sum + 31) >> 6) : -((-*sum + 31) >> 6);
+  *sse = (*sse + 2047) >> 12;
+}
+
+#define MASK_VAR(W, H) \
+unsigned int vp9_masked_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
+                                              const uint8_t *b, int b_stride, \
+                                              const uint8_t *m, int m_stride, \
+                                              unsigned int *sse) { \
+  int sum; \
+  masked_variance(a, a_stride, b, b_stride, m, m_stride, W, H, sse, &sum); \
+  return *sse - (((int64_t)sum * sum) / (W * H)); \
+}
+
+#define MASK_SUBPIX_VAR(W, H) \
+unsigned int vp9_masked_sub_pixel_variance##W##x##H##_c( \
+  const uint8_t *src, int  src_stride, \
+  int xoffset, int  yoffset, \
+  const uint8_t *dst, int dst_stride, \
+  const uint8_t *msk, int msk_stride, \
+  unsigned int *sse) { \
+  uint16_t fdata3[(H + 1) * W]; \
+  uint8_t temp2[H * W]; \
+\
+  var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, W, \
+                                    BILINEAR_FILTERS_2TAP(xoffset)); \
+  var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
+                                     BILINEAR_FILTERS_2TAP(yoffset)); \
+\
+  return vp9_masked_variance##W##x##H##_c(temp2, W, dst, dst_stride, \
+                                          msk, msk_stride, sse); \
+}
+
+MASK_VAR(4, 4)
+MASK_SUBPIX_VAR(4, 4)
+
+MASK_VAR(4, 8)
+MASK_SUBPIX_VAR(4, 8)
+
+MASK_VAR(8, 4)
+MASK_SUBPIX_VAR(8, 4)
+
+MASK_VAR(8, 8)
+MASK_SUBPIX_VAR(8, 8)
+
+MASK_VAR(8, 16)
+MASK_SUBPIX_VAR(8, 16)
+
+MASK_VAR(16, 8)
+MASK_SUBPIX_VAR(16, 8)
+
+MASK_VAR(16, 16)
+MASK_SUBPIX_VAR(16, 16)
+
+MASK_VAR(16, 32)
+MASK_SUBPIX_VAR(16, 32)
+
+MASK_VAR(32, 16)
+MASK_SUBPIX_VAR(32, 16)
+
+MASK_VAR(32, 32)
+MASK_SUBPIX_VAR(32, 32)
+
+MASK_VAR(32, 64)
+MASK_SUBPIX_VAR(32, 64)
+
+MASK_VAR(64, 32)
+MASK_SUBPIX_VAR(64, 32)
+
+MASK_VAR(64, 64)
+MASK_SUBPIX_VAR(64, 64)
+#endif

vp9/encoder/vp9_variance.h

@@ -67,6 +67,32 @@ typedef unsigned int (*vp9_subp_avg_variance_fn_t)(const uint8_t *src_ptr,
                                                    unsigned int *sse,
                                                    const uint8_t *second_pred);
 
+#if ((CONFIG_MASKED_INTERINTRA && CONFIG_INTERINTRA) || \
+    CONFIG_MASKED_INTERINTER)
+typedef unsigned int(*vp9_masked_sad_fn_t)(const uint8_t *src_ptr,
+                                           int source_stride,
+                                           const uint8_t *ref_ptr,
+                                           int ref_stride,
+                                           const uint8_t *msk_ptr,
+                                           int msk_stride);
+typedef unsigned int (*vp9_masked_variance_fn_t)(const uint8_t *src_ptr,
+                                                 int source_stride,
+                                                 const uint8_t *ref_ptr,
+                                                 int ref_stride,
+                                                 const uint8_t *msk_ptr,
+                                                 int msk_stride,
+                                                 unsigned int *sse);
+typedef unsigned int (*vp9_masked_subpixvariance_fn_t)(const uint8_t *src_ptr,
+                                                       int source_stride,
+                                                       int xoffset,
+                                                       int yoffset,
+                                                       const uint8_t *ref_ptr,
+                                                       int Refstride,
+                                                       const uint8_t *msk_ptr,
+                                                       int msk_stride,
+                                                       unsigned int *sse);
+#endif
+
 typedef struct vp9_variance_vtable {
   vp9_sad_fn_t               sdf;
   vp9_sad_avg_fn_t           sdaf;
@@ -76,6 +102,12 @@ typedef struct vp9_variance_vtable {
   vp9_sad_multi_fn_t         sdx3f;
   vp9_sad_multi_fn_t         sdx8f;
   vp9_sad_multi_d_fn_t       sdx4df;
+#if ((CONFIG_MASKED_INTERINTRA && CONFIG_INTERINTRA) || \
+    CONFIG_MASKED_INTERINTER)
+  vp9_masked_sad_fn_t            msdf;
+  vp9_masked_variance_fn_t       mvf;
+  vp9_masked_subpixvariance_fn_t msvf;
+#endif
 } vp9_variance_fn_ptr_t;
 
 void vp9_comp_avg_pred(uint8_t *comp_pred, const uint8_t *pred, int width,

vp9/vp9_cx_iface.c

@@ -88,7 +88,7 @@ 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;
+  vpx_codec_pkt_list_decl(128) pkt_list;
   unsigned int                 fixed_kf_cntr;
 };
 
@@ -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)
+  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;
 }

vp9/vp9_dx_iface.c

@@ -32,7 +32,6 @@ struct vpx_codec_alg_priv {
   vpx_codec_priv_t        base;
   vpx_codec_dec_cfg_t     cfg;
   vp9_stream_info_t       si;
-  struct VP9Decoder *pbi;
   int                     postproc_cfg_set;
   vp8_postproc_cfg_t      postproc_cfg;
   vpx_decrypt_cb          decrypt_cb;
@@ -42,6 +41,11 @@ struct vpx_codec_alg_priv {
   int                     frame_parallel_decode;  // frame-based threading.
   int                     last_show_frame;  // Index of last output frame.
 
+  VP9Worker               *frame_workers;
+  int                     num_frame_workers;
+  int                     next_submit_thread_id;
+  int                     next_output_thread_id;
+
   // External frame buffer info to save for VP9 common.
   void *ext_priv;  // Private data associated with the external frame buffers.
   vpx_get_frame_buffer_cb_fn_t get_ext_fb_cb;
@@ -85,11 +89,17 @@ static vpx_codec_err_t decoder_init(vpx_codec_ctx_t *ctx,
 }
 
 static vpx_codec_err_t decoder_destroy(vpx_codec_alg_priv_t *ctx) {
-  if (ctx->pbi) {
-    vp9_decoder_remove(ctx->pbi);
-    ctx->pbi = NULL;
+  if (ctx->frame_workers != NULL) {
+    int i;
+    for (i = 0; i < ctx->num_frame_workers; ++i) {
+      VP9Worker *const worker = &ctx->frame_workers[i];
+      FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
+      vp9_decoder_remove(worker_data->pbi);
+      vpx_free(worker_data);
+    }
   }
 
+  vpx_free(ctx->frame_workers);
   vpx_free(ctx);
 
   return VPX_CODEC_OK;
@@ -102,9 +112,6 @@ static vpx_codec_err_t decoder_peek_si_internal(const uint8_t *data,
                                                 void *decrypt_state) {
   uint8_t clear_buffer[9];
 
-  if (data_sz <= 8)
-    return VPX_CODEC_UNSUP_BITSTREAM;
-
   if (data + data_sz <= data)
     return VPX_CODEC_INVALID_PARAM;
 
@@ -125,12 +132,16 @@ static vpx_codec_err_t decoder_peek_si_internal(const uint8_t *data,
 
     if (frame_marker != VP9_FRAME_MARKER)
       return VPX_CODEC_UNSUP_BITSTREAM;
+
     if (version > 1) return VPX_CODEC_UNSUP_BITSTREAM;
 
     if (vp9_rb_read_bit(&rb)) {  // show an existing frame
       return VPX_CODEC_OK;
     }
 
+    if (data_sz <= 8)
+      return VPX_CODEC_UNSUP_BITSTREAM;
+
     si->is_kf = !vp9_rb_read_bit(&rb);
     if (si->is_kf) {
       const int sRGB = 7;
@@ -187,19 +198,28 @@ static vpx_codec_err_t decoder_get_si(vpx_codec_alg_priv_t *ctx,
   return VPX_CODEC_OK;
 }
 
+static void set_error_detail(vpx_codec_alg_priv_t *ctx,
+                             const char *const error) {
+  ctx->base.err_detail = error;
+}
+
 static vpx_codec_err_t update_error_state(vpx_codec_alg_priv_t *ctx,
                            const struct vpx_internal_error_info *error) {
   if (error->error_code)
-    ctx->base.err_detail = error->has_detail ? error->detail : NULL;
+    set_error_detail(ctx, error->has_detail ? error->detail : NULL);
 
   return error->error_code;
 }
 
 static void init_buffer_callbacks(vpx_codec_alg_priv_t *ctx) {
-  VP9_COMMON *const cm = &ctx->pbi->common;
+  int i;
+
+  for (i = 0; i < ctx->num_frame_workers; ++i) {
+    VP9Worker *const worker = &ctx->frame_workers[i];
+    FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
+    VP9_COMMON *const cm = &worker_data->pbi->common;
 
     cm->new_fb_idx = -1;
-
     if (ctx->get_ext_fb_cb != NULL && ctx->release_ext_fb_cb != NULL) {
       cm->get_fb_cb = ctx->get_ext_fb_cb;
       cm->release_fb_cb = ctx->release_ext_fb_cb;
@@ -214,6 +234,7 @@ static void init_buffer_callbacks(vpx_codec_alg_priv_t *ctx) {
 
       cm->cb_priv = &cm->int_frame_buffers;
     }
+  }
 }
 
 static void set_default_ppflags(vp8_postproc_cfg_t *cfg) {
@@ -231,15 +252,58 @@ static void set_ppflags(const vpx_codec_alg_priv_t *ctx,
   flags->noise_level = ctx->postproc_cfg.noise_level;
 }
 
-static void init_decoder(vpx_codec_alg_priv_t *ctx) {
-  ctx->pbi = vp9_decoder_create();
-  if (ctx->pbi == NULL)
-    return;
+static int frame_worker_hook(void *arg1, void *arg2) {
+  FrameWorkerData *const worker_data = (FrameWorkerData *)arg1;
+  const uint8_t *data = worker_data->data;
+  (void)arg2;
+  worker_data->result = vp9_receive_compressed_data(worker_data->pbi,
+                                                    worker_data->data_size,
+                                                    &data);
+  worker_data->data_end = data;
+  return !worker_data->result;
+}
+
+static vpx_codec_err_t init_decoder(vpx_codec_alg_priv_t *ctx) {
+  int i;
 
-  ctx->pbi->max_threads = ctx->cfg.threads;
-  ctx->pbi->inv_tile_order = ctx->invert_tile_order;
-  ctx->pbi->frame_parallel_decode = ctx->frame_parallel_decode;
   ctx->last_show_frame = -1;
+  ctx->next_submit_thread_id = 0;
+  ctx->next_output_thread_id = 0;
+  ctx->num_frame_workers =
+      (ctx->frame_parallel_decode == 1) ? ctx->cfg.threads: 1;
+
+  ctx->frame_workers = (VP9Worker *)
+      vpx_malloc(ctx->num_frame_workers * sizeof(*ctx->frame_workers));
+  if (ctx->frame_workers == NULL) {
+    set_error_detail(ctx, "Failed to allocate frame_workers");
+    return VPX_CODEC_MEM_ERROR;
+  }
+
+  for (i = 0; i < ctx->num_frame_workers; ++i) {
+    VP9Worker *const worker = &ctx->frame_workers[i];
+    FrameWorkerData *worker_data = NULL;
+    vp9_worker_init(worker);
+    worker->data1 = vpx_memalign(32, sizeof(FrameWorkerData));
+    if (worker->data1 == NULL) {
+      set_error_detail(ctx, "Failed to allocate worker_data");
+      return VPX_CODEC_MEM_ERROR;
+    }
+    worker_data = (FrameWorkerData *)worker->data1;
+    worker_data->pbi = vp9_decoder_create();
+    if (worker_data->pbi == NULL) {
+      set_error_detail(ctx, "Failed to allocate worker_data");
+      return VPX_CODEC_MEM_ERROR;
+    }
+
+    // If decoding in serial mode, FrameWorker thread could create tile worker
+    // thread or loopfilter thread.
+    worker_data->pbi->max_threads =
+        (ctx->frame_parallel_decode == 0) ? ctx->cfg.threads : 0;
+
+    worker_data->pbi->inv_tile_order = ctx->invert_tile_order;
+    worker_data->pbi->frame_parallel_decode = ctx->frame_parallel_decode;
+    worker->hook = (VP9WorkerHook)frame_worker_hook;
+  }
 
   // If postprocessing was enabled by the application and a
   // configuration has not been provided, default it.
@@ -248,14 +312,14 @@ static void init_decoder(vpx_codec_alg_priv_t *ctx) {
     set_default_ppflags(&ctx->postproc_cfg);
 
   init_buffer_callbacks(ctx);
+
+  return VPX_CODEC_OK;
 }
 
 static vpx_codec_err_t decode_one(vpx_codec_alg_priv_t *ctx,
                                   const uint8_t **data, unsigned int data_sz,
                                   void *user_priv, int64_t deadline) {
   vp9_ppflags_t flags = {0};
-  VP9_COMMON *cm = NULL;
-
   (void)deadline;
 
   // Determine the stream parameters. Note that we rely on peek_si to
@@ -272,22 +336,35 @@ static vpx_codec_err_t decode_one(vpx_codec_alg_priv_t *ctx,
       return VPX_CODEC_ERROR;
   }
 
-  // Initialize the decoder instance on the first frame
-  if (ctx->pbi == NULL) {
-    init_decoder(ctx);
-    if (ctx->pbi == NULL)
-      return VPX_CODEC_ERROR;
+  // Initialize the decoder workers on the first frame
+  if (ctx->frame_workers == NULL) {
+    const vpx_codec_err_t res = init_decoder(ctx);
+    if (res != VPX_CODEC_OK)
+      return res;
   }
 
+  if (!ctx->frame_parallel_decode) {
+    VP9Worker *const worker = ctx->frame_workers;
+    FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
+    worker_data->data = *data;
+    worker_data->data_size = data_sz;
+    worker_data->user_priv = user_priv;
+
     // Set these even if already initialized.  The caller may have changed the
     // decrypt config between frames.
-  ctx->pbi->decrypt_cb = ctx->decrypt_cb;
-  ctx->pbi->decrypt_state = ctx->decrypt_state;
+    worker_data->pbi->decrypt_cb = ctx->decrypt_cb;
+    worker_data->pbi->decrypt_state = ctx->decrypt_state;
 
-  cm = &ctx->pbi->common;
+    vp9_worker_execute(worker);
+    if (worker->had_error)
+      return update_error_state(ctx, &worker_data->pbi->common.error);
 
-  if (vp9_receive_compressed_data(ctx->pbi, data_sz, data))
-    return update_error_state(ctx, &cm->error);
+    // Update data pointer after decode.
+    *data = worker_data->data_end;
+  } else {
+    // TODO(hkuang): Implement frame parallel decode.
+    return VPX_CODEC_INCAPABLE;
+  }
 
   if (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC)
     set_ppflags(ctx, &flags);
@@ -306,10 +383,17 @@ static INLINE uint8_t read_marker(vpx_decrypt_cb decrypt_cb,
   return *data;
 }
 
-static void parse_superframe_index(const uint8_t *data, size_t data_sz,
+static vpx_codec_err_t parse_superframe_index(const uint8_t *data,
+                                              size_t data_sz,
                                               uint32_t sizes[8], int *count,
                                               vpx_decrypt_cb decrypt_cb,
                                               void *decrypt_state) {
+  // A chunk ending with a byte matching 0xc0 is an invalid chunk unless
+  // it is a super frame index. If the last byte of real video compression
+  // data is 0xc0 the encoder must add a 0 byte. If we have the marker but
+  // not the associated matching marker byte at the front of the index we have
+  // an invalid bitstream and need to return an error.
+
   uint8_t marker;
 
   assert(data_sz);
@@ -321,11 +405,23 @@ static void parse_superframe_index(const uint8_t *data, size_t data_sz,
     const uint32_t mag = ((marker >> 3) & 0x3) + 1;
     const size_t index_sz = 2 + mag * frames;
 
-    if (data_sz >= index_sz) {
-      uint8_t marker2 = read_marker(decrypt_cb, decrypt_state,
+    // This chunk is marked as having a superframe index but doesn't have
+    // enough data for it, thus it's an invalid superframe index.
+    if (data_sz < index_sz)
+      return VPX_CODEC_CORRUPT_FRAME;
+
+    {
+      const uint8_t marker2 = read_marker(decrypt_cb, decrypt_state,
                                           data + data_sz - index_sz);
 
-      if (marker == marker2) {
+      // This chunk is marked as having a superframe index but doesn't have
+      // the matching marker byte at the front of the index therefore it's an
+      // invalid chunk.
+      if (marker != marker2)
+        return VPX_CODEC_CORRUPT_FRAME;
+    }
+
+    {
       // Found a valid superframe index.
       uint32_t i, j;
       const uint8_t *x = &data[data_sz - index_sz + 1];
@@ -345,32 +441,9 @@ static void parse_superframe_index(const uint8_t *data, size_t data_sz,
           this_sz |= (*x++) << (j * 8);
         sizes[i] = this_sz;
       }
-
       *count = frames;
     }
   }
-  }
-}
-
-static vpx_codec_err_t decode_one_iter(vpx_codec_alg_priv_t *ctx,
-                                       const uint8_t **data_start_ptr,
-                                       const uint8_t *data_end,
-                                       uint32_t frame_size, void *user_priv,
-                                       long deadline) {
-  const vpx_codec_err_t res = decode_one(ctx, data_start_ptr, frame_size,
-                                         user_priv, deadline);
-  if (res != VPX_CODEC_OK)
-    return res;
-
-  // Account for suboptimal termination by the encoder.
-  while (*data_start_ptr < data_end) {
-    const uint8_t marker = read_marker(ctx->decrypt_cb, ctx->decrypt_state,
-                                       *data_start_ptr);
-    if (marker)
-      break;
-    (*data_start_ptr)++;
-  }
-
   return VPX_CODEC_OK;
 }
 
@@ -378,7 +451,7 @@ static vpx_codec_err_t decoder_decode(vpx_codec_alg_priv_t *ctx,
                                       const uint8_t *data, unsigned int data_sz,
                                       void *user_priv, long deadline) {
   const uint8_t *data_start = data;
-  const uint8_t *const data_end = data + data_sz;
+  const uint8_t * const data_end = data + data_sz;
   vpx_codec_err_t res;
   uint32_t frame_sizes[8];
   int frame_count;
@@ -386,32 +459,86 @@ static vpx_codec_err_t decoder_decode(vpx_codec_alg_priv_t *ctx,
   if (data == NULL || data_sz == 0)
     return VPX_CODEC_INVALID_PARAM;
 
-  parse_superframe_index(data, data_sz, frame_sizes, &frame_count,
+  res = parse_superframe_index(data, data_sz, frame_sizes, &frame_count,
                                ctx->decrypt_cb, ctx->decrypt_state);
+  if (res != VPX_CODEC_OK)
+    return res;
 
+  if (ctx->frame_parallel_decode) {
+    // Decode in frame parallel mode. When decoding in this mode, the frame
+    // passed to the decoder must be either a normal frame or a superframe with
+    // superframe index so the decoder could get each frame's start position
+    // in the superframe.
     if (frame_count > 0) {
       int i;
 
       for (i = 0; i < frame_count; ++i) {
+        const uint8_t *data_start_copy = data_start;
         const uint32_t frame_size = frame_sizes[i];
-      if (data_start < data ||
-          frame_size > (uint32_t)(data_end - data_start)) {
-        ctx->base.err_detail = "Invalid frame size in index";
+        vpx_codec_err_t res;
+        if (data_start < data
+            || frame_size > (uint32_t) (data_end - data_start)) {
+          set_error_detail(ctx, "Invalid frame size in index");
           return VPX_CODEC_CORRUPT_FRAME;
         }
 
-      res = decode_one_iter(ctx, &data_start, data_end, frame_size,
-                            user_priv, deadline);
+        res = decode_one(ctx, &data_start_copy, frame_size, user_priv,
+                         deadline);
         if (res != VPX_CODEC_OK)
           return res;
+
+        data_start += frame_size;
+      }
+    } else {
+      res = decode_one(ctx, &data_start, data_sz, user_priv, deadline);
+      if (res != VPX_CODEC_OK)
+        return res;
+
+      // Extra data detected after the frame.
+      if (data_start < data_end - 1) {
+        set_error_detail(ctx, "Fail to decode frame in parallel mode");
+        return VPX_CODEC_INCAPABLE;
+      }
+    }
+  } else {
+    // Decode in serial mode.
+    if (frame_count > 0) {
+      int i;
+
+      for (i = 0; i < frame_count; ++i) {
+        const uint8_t *data_start_copy = data_start;
+        const uint32_t frame_size = frame_sizes[i];
+        vpx_codec_err_t res;
+        if (data_start < data
+            || frame_size > (uint32_t) (data_end - data_start)) {
+          set_error_detail(ctx, "Invalid frame size in index");
+          return VPX_CODEC_CORRUPT_FRAME;
+        }
+
+        res = decode_one(ctx, &data_start_copy, frame_size, user_priv,
+                         deadline);
+        if (res != VPX_CODEC_OK)
+          return res;
+
+        data_start += frame_size;
       }
     } else {
       while (data_start < data_end) {
-      res = decode_one_iter(ctx, &data_start, data_end,
-                            (uint32_t)(data_end - data_start),
+        const uint32_t frame_size = (uint32_t) (data_end - data_start);
+        const vpx_codec_err_t res = decode_one(ctx, &data_start, frame_size,
                                                user_priv, deadline);
         if (res != VPX_CODEC_OK)
           return res;
+
+        // Account for suboptimal termination by the encoder.
+        while (data_start < data_end) {
+          const uint8_t marker = read_marker(ctx->decrypt_cb,
+                                             ctx->decrypt_state, data_start);
+          if (marker)
+            break;
+          ++data_start;
+        }
+      }
     }
   }
 
@@ -424,13 +551,15 @@ static vpx_image_t *decoder_get_frame(vpx_codec_alg_priv_t *ctx,
 
   // iter acts as a flip flop, so an image is only returned on the first
   // call to get_frame.
-  if (*iter == NULL && ctx->pbi != NULL) {
+  if (*iter == NULL && ctx->frame_workers != NULL) {
     YV12_BUFFER_CONFIG sd;
     vp9_ppflags_t flags = {0, 0, 0};
 
-    if (vp9_get_raw_frame(ctx->pbi, &sd, &flags) == 0) {
-      VP9_COMMON *cm = &ctx->pbi->common;
-      yuvconfig2image(&ctx->img, &sd, NULL);
+    VP9Worker *const worker = &ctx->frame_workers[ctx->next_output_thread_id];
+    FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
+    if (vp9_get_raw_frame(worker_data->pbi, &sd, &flags) == 0) {
+      VP9_COMMON *const cm = &worker_data->pbi->common;
+      yuvconfig2image(&ctx->img, &sd, worker_data->user_priv);
       ctx->img.fb_priv = cm->frame_bufs[cm->new_fb_idx].raw_frame_buffer.priv;
       img = &ctx->img;
       *iter = img;
@@ -442,7 +571,7 @@ static vpx_image_t *decoder_get_frame(vpx_codec_alg_priv_t *ctx,
               &cm->frame_bufs[ctx->last_show_frame].raw_frame_buffer);
         }
       }
-      ctx->last_show_frame = ctx->pbi->common.new_fb_idx;
+      ctx->last_show_frame = worker_data->pbi->common.new_fb_idx;
     }
   }
 
@@ -455,7 +584,7 @@ static vpx_codec_err_t decoder_set_fb_fn(
     vpx_release_frame_buffer_cb_fn_t cb_release, void *cb_priv) {
   if (cb_get == NULL || cb_release == NULL) {
     return VPX_CODEC_INVALID_PARAM;
-  } else if (ctx->pbi == NULL) {
+  } else if (ctx->frame_workers == NULL) {
     // If the decoder has already been initialized, do not accept changes to
     // the frame buffer functions.
     ctx->get_ext_fb_cb = cb_get;
@@ -471,12 +600,19 @@ static vpx_codec_err_t ctrl_set_reference(vpx_codec_alg_priv_t *ctx,
                                           va_list args) {
   vpx_ref_frame_t *const data = va_arg(args, vpx_ref_frame_t *);
 
+  // Only support this function in serial decode.
+  if (ctx->frame_parallel_decode) {
+    set_error_detail(ctx, "Not supported in frame parallel decode");
+    return VPX_CODEC_INCAPABLE;
+  }
+
   if (data) {
     vpx_ref_frame_t *const frame = (vpx_ref_frame_t *)data;
     YV12_BUFFER_CONFIG sd;
-
+    VP9Worker *const worker = ctx->frame_workers;
+    FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
     image2yuvconfig(&frame->img, &sd);
-    return vp9_set_reference_dec(&ctx->pbi->common,
+    return vp9_set_reference_dec(&worker_data->pbi->common,
                                  (VP9_REFFRAME)frame->frame_type, &sd);
   } else {
     return VPX_CODEC_INVALID_PARAM;
@@ -487,13 +623,19 @@ static vpx_codec_err_t ctrl_copy_reference(vpx_codec_alg_priv_t *ctx,
                                            va_list args) {
   vpx_ref_frame_t *data = va_arg(args, vpx_ref_frame_t *);
 
+  // Only support this function in serial decode.
+  if (ctx->frame_parallel_decode) {
+    set_error_detail(ctx, "Not supported in frame parallel decode");
+    return VPX_CODEC_INCAPABLE;
+  }
+
   if (data) {
-    vpx_ref_frame_t *frame = (vpx_ref_frame_t *)data;
+    vpx_ref_frame_t *frame = (vpx_ref_frame_t *) data;
     YV12_BUFFER_CONFIG sd;
-
+    VP9Worker *const worker = ctx->frame_workers;
+    FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
     image2yuvconfig(&frame->img, &sd);
-
-    return vp9_copy_reference_dec(ctx->pbi,
+    return vp9_copy_reference_dec(worker_data->pbi,
                                   (VP9_REFFRAME)frame->frame_type, &sd);
   } else {
     return VPX_CODEC_INVALID_PARAM;
@@ -504,11 +646,18 @@ static vpx_codec_err_t ctrl_get_reference(vpx_codec_alg_priv_t *ctx,
                                           va_list args) {
   vp9_ref_frame_t *data = va_arg(args, vp9_ref_frame_t *);
 
+  // Only support this function in serial decode.
+  if (ctx->frame_parallel_decode) {
+    set_error_detail(ctx, "Not supported in frame parallel decode");
+    return VPX_CODEC_INCAPABLE;
+  }
+
   if (data) {
     YV12_BUFFER_CONFIG* fb;
-
-    vp9_get_reference_dec(ctx->pbi, data->idx, &fb);
-    yuvconfig2image(&data->img, fb, NULL);
+    VP9Worker *const worker = ctx->frame_workers;
+    FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
+    vp9_get_reference_dec(worker_data->pbi, data->idx, &fb);
+    yuvconfig2image(&data->img, fb, worker_data->user_priv);
     return VPX_CODEC_OK;
   } else {
     return VPX_CODEC_INVALID_PARAM;
@@ -545,11 +694,20 @@ static vpx_codec_err_t ctrl_get_last_ref_updates(vpx_codec_alg_priv_t *ctx,
                                                  va_list args) {
   int *const update_info = va_arg(args, int *);
 
+  // Only support this function in serial decode.
+  if (ctx->frame_parallel_decode) {
+    set_error_detail(ctx, "Not supported in frame parallel decode");
+    return VPX_CODEC_INCAPABLE;
+  }
+
   if (update_info) {
-    if (ctx->pbi)
-      *update_info = ctx->pbi->refresh_frame_flags;
-    else
+    if (ctx->frame_workers) {
+      VP9Worker *const worker = ctx->frame_workers;
+      FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
+      *update_info = worker_data->pbi->refresh_frame_flags;
+    } else {
       return VPX_CODEC_ERROR;
+    }
     return VPX_CODEC_OK;
   } else {
     return VPX_CODEC_INVALID_PARAM;
@@ -561,11 +719,20 @@ static vpx_codec_err_t ctrl_get_frame_corrupted(vpx_codec_alg_priv_t *ctx,
                                                 va_list args) {
   int *corrupted = va_arg(args, int *);
 
+  // Only support this function in serial decode.
+  if (ctx->frame_parallel_decode) {
+    set_error_detail(ctx, "Not supported in frame parallel decode");
+    return VPX_CODEC_INCAPABLE;
+  }
+
   if (corrupted) {
-    if (ctx->pbi)
-      *corrupted = ctx->pbi->common.frame_to_show->corrupted;
-    else
+    if (ctx->frame_workers) {
+      VP9Worker *const worker = ctx->frame_workers;
+      FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
+      *corrupted = worker_data->pbi->common.frame_to_show->corrupted;
+    } else {
       return VPX_CODEC_ERROR;
+    }
     return VPX_CODEC_OK;
   } else {
     return VPX_CODEC_INVALID_PARAM;
@@ -576,9 +743,17 @@ static vpx_codec_err_t ctrl_get_display_size(vpx_codec_alg_priv_t *ctx,
                                              va_list args) {
   int *const display_size = va_arg(args, int *);
 
+  // Only support this function in serial decode.
+  if (ctx->frame_parallel_decode) {
+    set_error_detail(ctx, "Not supported in frame parallel decode");
+    return VPX_CODEC_INCAPABLE;
+  }
+
   if (display_size) {
-    if (ctx->pbi) {
-      const VP9_COMMON *const cm = &ctx->pbi->common;
+    if (ctx->frame_workers) {
+      VP9Worker *const worker = ctx->frame_workers;
+      FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
+      const VP9_COMMON *const cm = &worker_data->pbi->common;
       display_size[0] = cm->display_width;
       display_size[1] = cm->display_height;
     } else {

vp9/vp9cx.mk

@@ -105,11 +105,9 @@ VP9_CX_SRCS-$(HAVE_SSE3) += encoder/x86/vp9_sad_sse3.asm
 ifeq ($(CONFIG_USE_X86INC),yes)
 VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_dct_mmx.asm
 VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_error_sse2.asm
-VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_error_intrin_avx2.c
 VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_sad_sse2.asm
 VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_subtract_sse2.asm
 VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_variance_sse2.c
-VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_variance_avx2.c
 VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_subpel_variance.asm
 endif
 
@@ -124,7 +122,9 @@ VP9_CX_SRCS-$(ARCH_X86_64) += encoder/x86/vp9_ssim_opt_x86_64.asm
 VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_dct_sse2.c
 VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_dct32x32_sse2.c
 
-VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_dct_avx2.c
 VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_dct32x32_avx2.c
+VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_dct_avx2.c
+VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_error_intrin_avx2.c
+VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_variance_avx2.c
 
 VP9_CX_SRCS-yes := $(filter-out $(VP9_CX_SRCS_REMOVE-yes),$(VP9_CX_SRCS-yes))

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,
@@ -872,6 +829,7 @@ vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
             si->frame_within_gop);
   }
 
+  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 &&
@@ -879,14 +837,13 @@ vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
         svc_log(svc_ctx, SVC_LOG_DEBUG, "Skip encoding layer %d\n", si->layer);
         continue;
       }
-
-    if (rawimg != NULL) {
       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);
+  res = vpx_codec_encode(codec_ctx, rawimg, pts, (uint32_t)duration, 0,
+                         deadline);
   if (res != VPX_CODEC_OK) {
     return res;
   }
@@ -895,22 +852,19 @@ vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
   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);
+        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);
 
-          // save layer size in superframe index
-          superframe.sizes[superframe.count++] = frame_pkt_size;
-          superframe.magnitude |= frame_pkt_size;
+        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: {
@@ -918,19 +872,20 @@ vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
         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,
+                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->encode_frame_count, si->layer,
+                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[si->layer][i] += cx_pkt->data.psnr.psnr[i];
-            si->sse_sum[si->layer][i] += cx_pkt->data.psnr.sse[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: {
@@ -941,7 +896,7 @@ vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
           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;
+            return VPX_CODEC_MEM_ERROR;
           }
           si->rc_stats_buf = p;
           si->rc_stats_buf_size = new_size;
@@ -957,39 +912,7 @@ vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
       }
     }
   }
-    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 @@ 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);
     }

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