Update aom to commit id f5bdeac22930ff4c6b219be49c843db35970b918

39 files changed, 1850 insertions, 2215 deletions

diff --git a/third_party/aom/aom_dsp/aom_convolve.c b/third_party/aom/aom_dsp/aom_convolve.c
index 4dac6aacc..c903ea52d 100644
--- a/third_party/aom/aom_dsp/aom_convolve.c
+++ b/third_party/aom/aom_dsp/aom_convolve.c
@@ -41,6 +41,29 @@ static void convolve_horiz(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
+static void convolve_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                   uint8_t *dst, ptrdiff_t dst_stride,
+                                   const InterpKernel *x_filters, int x0_qn,
+                                   int x_step_qn, int w, int h) {
+  int x, y;
+  src -= SUBPEL_TAPS / 2 - 1;
+  for (y = 0; y < h; ++y) {
+    int x_qn = x0_qn;
+    for (x = 0; x < w; ++x) {
+      const uint8_t *const src_x = &src[x_qn >> SCALE_SUBPEL_BITS];  // q8
+      const int x_filter_idx = (x_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
+      assert(x_filter_idx < SUBPEL_SHIFTS);
+      const int16_t *const x_filter = x_filters[x_filter_idx];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
+      dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
+      x_qn += x_step_qn;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
 static void convolve_avg_horiz(const uint8_t *src, ptrdiff_t src_stride,
                                uint8_t *dst, ptrdiff_t dst_stride,
                                const InterpKernel *x_filters, int x0_q4,
@@ -63,6 +86,30 @@ static void convolve_avg_horiz(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
+static void convolve_avg_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                       uint8_t *dst, ptrdiff_t dst_stride,
+                                       const InterpKernel *x_filters, int x0_qn,
+                                       int x_step_qn, int w, int h) {
+  int x, y;
+  src -= SUBPEL_TAPS / 2 - 1;
+  for (y = 0; y < h; ++y) {
+    int x_qn = x0_qn;
+    for (x = 0; x < w; ++x) {
+      const uint8_t *const src_x = &src[x_qn >> SCALE_SUBPEL_BITS];
+      const int x_filter_idx = (x_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
+      assert(x_filter_idx < SUBPEL_SHIFTS);
+      const int16_t *const x_filter = x_filters[x_filter_idx];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
+      dst[x] = ROUND_POWER_OF_TWO(
+          dst[x] + clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
+      x_qn += x_step_qn;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
 static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
                           uint8_t *dst, ptrdiff_t dst_stride,
                           const InterpKernel *y_filters, int y0_q4,
@@ -86,6 +133,31 @@ static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
+static void convolve_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                  uint8_t *dst, ptrdiff_t dst_stride,
+                                  const InterpKernel *y_filters, int y0_qn,
+                                  int y_step_qn, int w, int h) {
+  int x, y;
+  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+
+  for (x = 0; x < w; ++x) {
+    int y_qn = y0_qn;
+    for (y = 0; y < h; ++y) {
+      const unsigned char *src_y =
+          &src[(y_qn >> SCALE_SUBPEL_BITS) * src_stride];
+      const int16_t *const y_filter =
+          y_filters[(y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_y[k * src_stride] * y_filter[k];
+      dst[y * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
+      y_qn += y_step_qn;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
 static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
                               uint8_t *dst, ptrdiff_t dst_stride,
                               const InterpKernel *y_filters, int y0_q4,
@@ -112,6 +184,34 @@ static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
+static void convolve_avg_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                      uint8_t *dst, ptrdiff_t dst_stride,
+                                      const InterpKernel *y_filters, int y0_qn,
+                                      int y_step_qn, int w, int h) {
+  int x, y;
+  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+
+  for (x = 0; x < w; ++x) {
+    int y_qn = y0_qn;
+    for (y = 0; y < h; ++y) {
+      const unsigned char *src_y =
+          &src[(y_qn >> SCALE_SUBPEL_BITS) * src_stride];
+      const int16_t *const y_filter =
+          y_filters[(y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_y[k * src_stride] * y_filter[k];
+      dst[y * dst_stride] = ROUND_POWER_OF_TWO(
+          dst[y * dst_stride] +
+              clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)),
+          1);
+      y_qn += y_step_qn;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
 static void convolve(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                      ptrdiff_t dst_stride, const InterpKernel *const x_filters,
                      int x0_q4, int x_step_q4,
@@ -146,6 +246,41 @@ static void convolve(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                 dst_stride, y_filters, y0_q4, y_step_q4, w, h);
 }
 
+static void convolve_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                             uint8_t *dst, ptrdiff_t dst_stride,
+                             const InterpKernel *const x_filters, int x0_qn,
+                             int x_step_qn, const InterpKernel *const y_filters,
+                             int y0_qn, int y_step_qn, int w, int h) {
+  // TODO(afergs): Update comment here
+  // 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_qn = 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[MAX_EXT_SIZE * MAX_SB_SIZE];
+  int intermediate_height =
+      (((h - 1) * y_step_qn + y0_qn) >> SCALE_SUBPEL_BITS) + SUBPEL_TAPS;
+
+  assert(w <= MAX_SB_SIZE);
+  assert(h <= MAX_SB_SIZE);
+
+  assert(y_step_qn <= SCALE_SUBPEL_BITS * 2);
+  assert(x_step_qn <= SCALE_SUBPEL_BITS * 2);
+
+  convolve_horiz_scale_c(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride,
+                         temp, MAX_SB_SIZE, x_filters, x0_qn, x_step_qn, w,
+                         intermediate_height);
+  convolve_vert_scale_c(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1), MAX_SB_SIZE,
+                        dst, dst_stride, y_filters, y0_qn, y_step_qn, w, h);
+}
+
 static const InterpKernel *get_filter_base(const int16_t *filter) {
   // NOTE: This assumes that the filter table is 256-byte aligned.
   // TODO(agrange) Modify to make independent of table alignment.
@@ -171,6 +306,21 @@ void aom_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
                  w, h);
 }
 
+void aom_convolve8_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                 uint8_t *dst, ptrdiff_t dst_stride,
+                                 const int16_t *filter_x, int subpel_x,
+                                 int x_step_qn, const int16_t *filter_y,
+                                 int subpel_y, int y_step_qn, int w, int h) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+
+  (void)subpel_y;
+  (void)filter_y;
+  (void)y_step_qn;
+
+  convolve_horiz_scale_c(src, src_stride, dst, dst_stride, filters_x, subpel_x,
+                         x_step_qn, w, h);
+}
+
 void aom_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
                                uint8_t *dst, ptrdiff_t dst_stride,
                                const int16_t *filter_x, int x_step_q4,
@@ -186,6 +336,22 @@ void aom_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
                      x_step_q4, w, h);
 }
 
+void aom_convolve8_avg_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                     uint8_t *dst, ptrdiff_t dst_stride,
+                                     const int16_t *filter_x, int subpel_x,
+                                     int x_step_qn, const int16_t *filter_y,
+                                     int subpel_y, int y_step_qn, int w,
+                                     int h) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+
+  (void)subpel_y;
+  (void)filter_y;
+  (void)y_step_qn;
+
+  convolve_avg_horiz_scale_c(src, src_stride, dst, dst_stride, filters_x,
+                             subpel_x, x_step_qn, w, h);
+}
+
 void aom_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
                           uint8_t *dst, ptrdiff_t dst_stride,
                           const int16_t *filter_x, int x_step_q4,
@@ -201,6 +367,21 @@ void aom_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
                 w, h);
 }
 
+void aom_convolve8_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                uint8_t *dst, ptrdiff_t dst_stride,
+                                const int16_t *filter_x, int subpel_x,
+                                int x_step_qn, const int16_t *filter_y,
+                                int subpel_y, int y_step_qn, int w, int h) {
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+
+  (void)subpel_x;
+  (void)filter_x;
+  (void)x_step_qn;
+
+  convolve_vert_scale_c(src, src_stride, dst, dst_stride, filters_y, subpel_y,
+                        y_step_qn, w, h);
+}
+
 void aom_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
                               uint8_t *dst, ptrdiff_t dst_stride,
                               const int16_t *filter_x, int x_step_q4,
@@ -216,6 +397,21 @@ void aom_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
                     y_step_q4, w, h);
 }
 
+void aom_convolve8_avg_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                    uint8_t *dst, ptrdiff_t dst_stride,
+                                    const int16_t *filter_x, int subpel_x,
+                                    int x_step_qn, const int16_t *filter_y,
+                                    int subpel_y, int y_step_qn, int w, int h) {
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+
+  (void)subpel_x;
+  (void)filter_x;
+  (void)x_step_qn;
+
+  convolve_avg_vert_scale_c(src, src_stride, dst, dst_stride, filters_y,
+                            subpel_y, y_step_qn, w, h);
+}
+
 void aom_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                      ptrdiff_t dst_stride, const int16_t *filter_x,
                      int x_step_q4, const int16_t *filter_y, int y_step_q4,
@@ -230,6 +426,19 @@ void aom_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
            filters_y, y0_q4, y_step_q4, w, h);
 }
 
+void aom_convolve8_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                           uint8_t *dst, ptrdiff_t dst_stride,
+                           const int16_t *filter_x, int subpel_x, int x_step_qn,
+                           const int16_t *filter_y, int subpel_y, int y_step_qn,
+                           int w, int h) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+
+  convolve_scale_c(src, src_stride, dst, dst_stride, filters_x, subpel_x,
+                   x_step_qn, filters_y, subpel_y, y_step_qn, w, h);
+}
+
 void aom_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                          ptrdiff_t dst_stride, const int16_t *filter_x,
                          int x_step_q4, const int16_t *filter_y, int y_step_q4,
@@ -245,6 +454,22 @@ void aom_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                      h);
 }
 
+void aom_convolve8_avg_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                               uint8_t *dst, ptrdiff_t dst_stride,
+                               const int16_t *filter_x, int subpel_x,
+                               int x_step_qn, const int16_t *filter_y,
+                               int subpel_y, int y_step_qn, int w, int h) {
+  /* Fixed size intermediate buffer places limits on parameters. */
+  DECLARE_ALIGNED(16, uint8_t, temp[MAX_SB_SIZE * MAX_SB_SIZE]);
+  assert(w <= MAX_SB_SIZE);
+  assert(h <= MAX_SB_SIZE);
+
+  aom_convolve8_scale_c(src, src_stride, temp, MAX_SB_SIZE, filter_x, subpel_x,
+                        x_step_qn, filter_y, subpel_y, y_step_qn, w, h);
+  aom_convolve_avg_c(temp, MAX_SB_SIZE, dst, dst_stride, NULL, 0, NULL, 0, w,
+                     h);
+}
+
 void aom_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                          ptrdiff_t dst_stride, const int16_t *filter_x,
                          int filter_x_stride, const int16_t *filter_y,
@@ -332,6 +557,7 @@ void aom_scaled_avg_2d_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                       filter_y, y_step_q4, w, h);
 }
 
+// TODO(afergs): Make sure this works too
 #if CONFIG_LOOP_RESTORATION
 static void convolve_add_src_horiz(const uint8_t *src, ptrdiff_t src_stride,
                                    uint8_t *dst, ptrdiff_t dst_stride,
@@ -569,6 +795,7 @@ void aom_convolve8_add_src_hip_c(const uint8_t *src, ptrdiff_t src_stride,
 }
 #endif  // CONFIG_LOOP_RESTORATION
 
+// TODO(afergs): Make sure this works too
 #if CONFIG_HIGHBITDEPTH
 static void highbd_convolve_horiz(const uint8_t *src8, ptrdiff_t src_stride,
                                   uint8_t *dst8, ptrdiff_t dst_stride,
diff --git a/third_party/aom/aom_dsp/aom_dsp.cmake b/third_party/aom/aom_dsp/aom_dsp.cmake
index 5a49ae817..3ce6761ca 100644
--- a/third_party/aom/aom_dsp/aom_dsp.cmake
+++ b/third_party/aom/aom_dsp/aom_dsp.cmake
@@ -287,6 +287,7 @@ if (CONFIG_AV1_ENCODER)
       "${AOM_ROOT}/aom_dsp/x86/fwd_dct32x32_impl_avx2.h"
       "${AOM_ROOT}/aom_dsp/x86/fwd_txfm_avx2.c"
       "${AOM_ROOT}/aom_dsp/x86/fwd_txfm_avx2.h"
+      "${AOM_ROOT}/aom_dsp/x86/highbd_quantize_intrin_avx2.c"
       "${AOM_ROOT}/aom_dsp/x86/sad4d_avx2.c"
       "${AOM_ROOT}/aom_dsp/x86/sad_avx2.c"
       "${AOM_ROOT}/aom_dsp/x86/sad_impl_avx2.c"
@@ -312,6 +313,7 @@ if (CONFIG_AV1_ENCODER)
         "${AOM_ROOT}/aom_dsp/x86/fwd_txfm_sse2.c"
         "${AOM_ROOT}/aom_dsp/x86/fwd_txfm_sse2.h"
         "${AOM_ROOT}/aom_dsp/x86/halfpix_variance_sse2.c"
+        "${AOM_ROOT}/aom_dsp/x86/highbd_quantize_intrin_sse2.c"
         "${AOM_ROOT}/aom_dsp/x86/variance_sse2.c"
         "${AOM_ROOT}/aom_dsp/x86/sum_squares_sse2.c")
 
@@ -330,10 +332,16 @@ if (CONFIG_AV1_ENCODER)
         "${AOM_ROOT}/aom_dsp/mips/variance_msa.c"
         "${AOM_ROOT}/aom_dsp/mips/sub_pixel_variance_msa.c")
 
+    if (CONFIG_EXT_INTER)
+      set(AOM_DSP_ENCODER_INTRIN_SSSE3
+          ${AOM_DSP_ENCODER_INTRIN_SSSE3}
+          "${AOM_ROOT}/aom_dsp/x86/masked_sad_intrin_ssse3.c"
+          "${AOM_ROOT}/aom_dsp/x86/masked_variance_intrin_ssse3.c")
+    endif ()
+
     if (CONFIG_HIGHBITDEPTH)
       set(AOM_DSP_ENCODER_INTRIN_SSE2
           ${AOM_DSP_ENCODER_INTRIN_SSE2}
-          "${AOM_ROOT}/aom_dsp/x86/highbd_quantize_intrin_sse2.c"
           "${AOM_ROOT}/aom_dsp/x86/highbd_subtract_sse2.c")
     endif ()
   endif ()
@@ -407,29 +415,38 @@ endif ()
 # has been created.
 function (setup_aom_dsp_targets)
   add_library(aom_dsp_common OBJECT ${AOM_DSP_COMMON_SOURCES})
-  set(AOM_LIB_TARGETS ${AOM_LIB_TARGETS} aom_dsp_common)
-  target_sources(aom PUBLIC $<TARGET_OBJECTS:aom_dsp_common>)
+  list(APPEND AOM_LIB_TARGETS aom_dsp_common)
+  create_dummy_source_file("aom_av1" "c" "dummy_source_file")
+  add_library(aom_dsp OBJECT "${dummy_source_file}")
+  target_sources(aom PRIVATE $<TARGET_OBJECTS:aom_dsp_common>)
+  list(APPEND AOM_LIB_TARGETS aom_dsp)
+
+  # Not all generators support libraries consisting only of object files. Add a
+  # dummy source file to the aom_dsp target.
+  add_dummy_source_file_to_target("aom_dsp" "c")
 
   if (CONFIG_AV1_DECODER)
     add_library(aom_dsp_decoder OBJECT ${AOM_DSP_DECODER_SOURCES})
     set(AOM_LIB_TARGETS ${AOM_LIB_TARGETS} aom_dsp_decoder)
-    target_sources(aom PUBLIC $<TARGET_OBJECTS:aom_dsp_decoder>)
+    target_sources(aom PRIVATE $<TARGET_OBJECTS:aom_dsp_decoder>)
   endif ()
 
   if (CONFIG_AV1_ENCODER)
     add_library(aom_dsp_encoder OBJECT ${AOM_DSP_ENCODER_SOURCES})
     set(AOM_LIB_TARGETS ${AOM_LIB_TARGETS} aom_dsp_encoder)
-    target_sources(aom PUBLIC $<TARGET_OBJECTS:aom_dsp_encoder>)
+    target_sources(aom PRIVATE $<TARGET_OBJECTS:aom_dsp_encoder>)
   endif ()
 
   if (HAVE_SSE2)
     add_asm_library("aom_dsp_common_sse2" "AOM_DSP_COMMON_ASM_SSE2" "aom")
     add_intrinsics_object_library("-msse2" "sse2" "aom_dsp_common"
-                                  "AOM_DSP_COMMON_INTRIN_SSE2")
+                                   "AOM_DSP_COMMON_INTRIN_SSE2" "aom")
+
     if (CONFIG_AV1_ENCODER)
-      add_asm_library("aom_dsp_encoder_sse2" "AOM_DSP_ENCODER_ASM_SSE2" "aom")
+      add_asm_library("aom_dsp_encoder_sse2" "AOM_DSP_ENCODER_ASM_SSE2"
+                      "aom")
       add_intrinsics_object_library("-msse2" "sse2" "aom_dsp_encoder"
-                                    "AOM_DSP_ENCODER_INTRIN_SSE2")
+                                    "AOM_DSP_ENCODER_INTRIN_SSE2" "aom")
     endif()
   endif ()
 
@@ -440,7 +457,7 @@ function (setup_aom_dsp_targets)
   if (HAVE_SSSE3)
     add_asm_library("aom_dsp_common_ssse3" "AOM_DSP_COMMON_ASM_SSSE3" "aom")
     add_intrinsics_object_library("-mssse3" "ssse3" "aom_dsp_common"
-                                  "AOM_DSP_COMMON_INTRIN_SSSE3")
+                                  "AOM_DSP_COMMON_INTRIN_SSSE3" "aom")
 
     if (CONFIG_AV1_ENCODER)
       if ("${AOM_TARGET_CPU}" STREQUAL "x86_64")
@@ -448,16 +465,20 @@ function (setup_aom_dsp_targets)
              ${AOM_DSP_ENCODER_ASM_SSSE3_X86_64})
       endif ()
       add_asm_library("aom_dsp_encoder_ssse3" "AOM_DSP_ENCODER_ASM_SSSE3" "aom")
+      if (AOM_DSP_ENCODER_INTRIN_SSSE3)
+        add_intrinsics_object_library("-mssse3" "ssse3" "aom_dsp_encoder"
+                                      "AOM_DSP_ENCODER_INTRIN_SSSE3" "aom")
+      endif ()
     endif ()
   endif ()
 
   if (HAVE_SSE4_1)
     add_intrinsics_object_library("-msse4.1" "sse4_1" "aom_dsp_common"
-                                  "AOM_DSP_COMMON_INTRIN_SSE4_1")
+                                  "AOM_DSP_COMMON_INTRIN_SSE4_1" "aom")
     if (CONFIG_AV1_ENCODER)
       if (AOM_DSP_ENCODER_INTRIN_SSE4_1)
         add_intrinsics_object_library("-msse4.1" "sse4_1" "aom_dsp_encoder"
-                                      "AOM_DSP_ENCODER_INTRIN_SSE4_1")
+                                      "AOM_DSP_ENCODER_INTRIN_SSE4_1" "aom")
       endif ()
       add_asm_library("aom_dsp_encoder_sse4_1" "AOM_DSP_ENCODER_ASM_SSE4_1"
                       "aom")
@@ -473,10 +494,10 @@ function (setup_aom_dsp_targets)
 
   if (HAVE_AVX2)
     add_intrinsics_object_library("-mavx2" "avx2" "aom_dsp_common"
-                                  "AOM_DSP_COMMON_INTRIN_AVX2")
+                                  "AOM_DSP_COMMON_INTRIN_AVX2" "aom")
     if (CONFIG_AV1_ENCODER)
       add_intrinsics_object_library("-mavx2" "avx2" "aom_dsp_encoder"
-                                    "AOM_DSP_ENCODER_INTRIN_AVX2")
+                                    "AOM_DSP_ENCODER_INTRIN_AVX2" "aom")
     endif ()
   endif ()
 
@@ -490,20 +511,21 @@ function (setup_aom_dsp_targets)
 
   if (HAVE_NEON)
     add_intrinsics_object_library("${AOM_NEON_INTRIN_FLAG}" "neon"
-                                  "aom_dsp_common" "AOM_DSP_COMMON_INTRIN_NEON")
+                                  "aom_dsp_common" "AOM_DSP_COMMON_INTRIN_NEON"
+                                  "aom")
   endif ()
 
   if (HAVE_DSPR2)
     add_intrinsics_object_library("" "dspr2" "aom_dsp_common"
-                                  "AOM_DSP_COMMON_INTRIN_DSPR2")
+                                  "AOM_DSP_COMMON_INTRIN_DSPR2" "aom")
   endif ()
 
   if (HAVE_MSA)
     add_intrinsics_object_library("" "msa" "aom_dsp_common"
-                                  "AOM_DSP_COMMON_INTRIN_MSA")
+                                  "AOM_DSP_COMMON_INTRIN_MSA" "aom")
     if (CONFIG_AV1_ENCODER)
       add_intrinsics_object_library("" "msa" "aom_dsp_encoder"
-                                    "AOM_DSP_ENCODER_INTRIN_MSA")
+                                    "AOM_DSP_ENCODER_INTRIN_MSA" "aom")
     endif ()
   endif ()
 
diff --git a/third_party/aom/aom_dsp/aom_dsp.mk b/third_party/aom/aom_dsp/aom_dsp.mk
index 6e2d5630e..f9d675ac0 100644
--- a/third_party/aom/aom_dsp/aom_dsp.mk
+++ b/third_party/aom/aom_dsp/aom_dsp.mk
@@ -290,9 +290,10 @@ DSP_SRCS-yes            += quantize.c
 DSP_SRCS-yes            += quantize.h
 
 DSP_SRCS-$(HAVE_SSE2)   += x86/quantize_sse2.c
-ifeq ($(CONFIG_HIGHBITDEPTH),yes)
+
 DSP_SRCS-$(HAVE_SSE2)   += x86/highbd_quantize_intrin_sse2.c
-endif
+DSP_SRCS-$(HAVE_AVX2)   += x86/highbd_quantize_intrin_avx2.c
+
 ifeq ($(ARCH_X86_64),yes)
 DSP_SRCS-$(HAVE_SSSE3)  += x86/quantize_ssse3_x86_64.asm
 DSP_SRCS-$(HAVE_AVX)    += x86/quantize_avx_x86_64.asm
diff --git a/third_party/aom/aom_dsp/aom_dsp_common.h b/third_party/aom/aom_dsp/aom_dsp_common.h
index 82f9a95e9..5b104321b 100644
--- a/third_party/aom/aom_dsp/aom_dsp_common.h
+++ b/third_party/aom/aom_dsp/aom_dsp_common.h
@@ -31,8 +31,6 @@ extern "C" {
 #define AOMMIN(x, y) (((x) < (y)) ? (x) : (y))
 #define AOMMAX(x, y) (((x) > (y)) ? (x) : (y))
 
-#define NELEMENTS(x) (sizeof((x)) / sizeof((x)[0]))
-
 #define IMPLIES(a, b) (!(a) || (b))  //  Logical 'a implies b' (or 'a -> b')
 
 #define IS_POWER_OF_TWO(x) (((x) & ((x)-1)) == 0)
@@ -54,16 +52,9 @@ extern "C" {
 #define UNLIKELY(v) (v)
 #endif
 
-#define AOM_SWAP(type, a, b) \
-  do {                       \
-    type c = (b);            \
-    b = a;                   \
-    a = c;                   \
-  } while (0)
-
 #if CONFIG_AOM_QM
 typedef uint16_t qm_val_t;
-#define AOM_QM_BITS 6
+#define AOM_QM_BITS 5
 #endif
 #if CONFIG_HIGHBITDEPTH
 // Note:
@@ -87,11 +78,14 @@ static INLINE int clamp(int value, int low, int high) {
   return value < low ? low : (value > high ? high : value);
 }
 
+static INLINE int64_t clamp64(int64_t value, int64_t low, int64_t high) {
+  return value < low ? low : (value > high ? high : value);
+}
+
 static INLINE double fclamp(double value, double low, double high) {
   return value < low ? low : (value > high ? high : value);
 }
 
-#if CONFIG_HIGHBITDEPTH
 static INLINE uint16_t clip_pixel_highbd(int val, int bd) {
   switch (bd) {
     case 8:
@@ -100,7 +94,6 @@ static INLINE uint16_t clip_pixel_highbd(int val, int bd) {
     case 12: return (uint16_t)clamp(val, 0, 4095);
   }
 }
-#endif  // CONFIG_HIGHBITDEPTH
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/third_party/aom/aom_dsp/aom_dsp_rtcd_defs.pl b/third_party/aom/aom_dsp/aom_dsp_rtcd_defs.pl
index 8047cbc09..0c0356870 100755
--- a/third_party/aom/aom_dsp/aom_dsp_rtcd_defs.pl
+++ b/third_party/aom/aom_dsp/aom_dsp_rtcd_defs.pl
@@ -40,12 +40,26 @@ foreach $w (@block_widths) {
     push @block_sizes, [$w, $h] if ($w <= 2*$h && $h <= 2*$w) ;
   }
 }
+if (aom_config("CONFIG_EXT_PARTITION_TYPES")) {
+  push @block_sizes, [4, 16];
+  push @block_sizes, [16, 4];
+  push @block_sizes, [8, 32];
+  push @block_sizes, [32, 8];
+}
 
 @tx_dims = (2, 4, 8, 16, 32);
 if (aom_config("CONFIG_TX64X64") eq "yes") {
   push @tx_dims, '64';
 }
 
+@tx_sizes = ();
+foreach $w (@tx_dims) {
+  push @tx_sizes, [$w, $w];
+  foreach $h (@tx_dims) {
+    push @tx_sizes, [$w, $h] if ($w >=4 && $h >=4 && ($w == 2*$h || $h == 2*$w));
+  }
+}
+
 @pred_names = qw/dc dc_top dc_left dc_128 v h d207e d63e d45e d117 d135 d153/;
 if (aom_config("CONFIG_ALT_INTRA") eq "yes") {
   push @pred_names, qw/paeth smooth/;
@@ -60,9 +74,8 @@ if (aom_config("CONFIG_ALT_INTRA") eq "yes") {
 # Intra prediction
 #
 
-foreach $dim (@tx_dims) {
-  $w = ${dim};
-  $h = ${dim};
+foreach (@tx_sizes) {
+  ($w, $h) = @$_;
   foreach $pred_name (@pred_names) {
     add_proto "void", "aom_${pred_name}_predictor_${w}x${h}",
               "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
@@ -142,20 +155,27 @@ if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
 #
 # Sub Pixel Filters
 #
-add_proto qw/void aom_convolve_copy/,       "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_convolve_avg/,        "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_convolve8/,           "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_convolve8_horiz/,     "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_convolve8_vert/,      "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_convolve8_avg/,       "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_convolve8_avg_vert/,  "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_scaled_2d/,           "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_scaled_horiz/,        "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_scaled_vert/,         "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_scaled_avg_2d/,       "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_scaled_avg_horiz/,    "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_scaled_avg_vert/,     "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve_copy/,             "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve_avg/,              "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8/,                 "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8_horiz/,           "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8_vert/,            "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8_avg/,             "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8_avg_horiz/,       "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8_avg_vert/,        "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_scaled_2d/,                 "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_scaled_horiz/,              "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_scaled_vert/,               "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_scaled_avg_2d/,             "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_scaled_avg_horiz/,          "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_scaled_avg_vert/,           "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+
+add_proto qw/void aom_convolve8_horiz_scale/,     "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int subpel_x, int x_step_q4, const int16_t *filter_y, int subpel_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8_vert_scale/,      "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int subpel_x, int x_step_q4, const int16_t *filter_y, int subpel_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8_avg_horiz_scale/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int subpel_x, int x_step_q4, const int16_t *filter_y, int subpel_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8_avg_vert_scale/,  "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int subpel_x, int x_step_q4, const int16_t *filter_y, int subpel_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8_scale/,           "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int subpel_x, int x_step_q4, const int16_t *filter_y, int subpel_y, int y_step_q4, int w, int h";
+add_proto qw/void aom_convolve8_avg_scale/,       "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int subpel_x, int x_step_q4, const int16_t *filter_y, int subpel_y, int y_step_q4, int w, int h";
 
 specialize qw/aom_convolve_copy       sse2      /;
 specialize qw/aom_convolve_avg        sse2      /;
@@ -334,24 +354,15 @@ if ((aom_config("CONFIG_AV1_ENCODER") eq "yes") || (aom_config("CONFIG_PVQ") eq
     add_proto qw/void aom_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_fdct8x8 sse2/, "$ssse3_x86_64";
 
-    add_proto qw/void aom_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
-    specialize qw/aom_fdct8x8_1 sse2/;
-
     add_proto qw/void aom_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_fdct16x16 sse2/;
 
-    add_proto qw/void aom_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
-    specialize qw/aom_fdct16x16_1 sse2 avx2/;
-
     add_proto qw/void aom_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_fdct32x32 sse2 avx2/;
 
     add_proto qw/void aom_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_fdct32x32_rd sse2 avx2/;
 
-    add_proto qw/void aom_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
-    specialize qw/aom_fdct32x32_1 sse2 avx2/;
-
     # High bit depth
     add_proto qw/void aom_highbd_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_highbd_fdct4x4 sse2/;
@@ -359,20 +370,15 @@ if ((aom_config("CONFIG_AV1_ENCODER") eq "yes") || (aom_config("CONFIG_PVQ") eq
     add_proto qw/void aom_highbd_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_highbd_fdct8x8 sse2/;
 
-    add_proto qw/void aom_highbd_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
-
     add_proto qw/void aom_highbd_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_highbd_fdct16x16 sse2/;
 
-    add_proto qw/void aom_highbd_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
-
     add_proto qw/void aom_highbd_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_highbd_fdct32x32 sse2/;
 
     add_proto qw/void aom_highbd_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_highbd_fdct32x32_rd sse2/;
 
-    add_proto qw/void aom_highbd_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
   } else {
     add_proto qw/void aom_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_fdct4x4 sse2 msa/;
@@ -383,47 +389,25 @@ if ((aom_config("CONFIG_AV1_ENCODER") eq "yes") || (aom_config("CONFIG_PVQ") eq
     add_proto qw/void aom_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_fdct8x8 sse2 neon msa/, "$ssse3_x86_64";
 
-    add_proto qw/void aom_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
-    specialize qw/aom_fdct8x8_1 sse2 neon msa/;
-
     add_proto qw/void aom_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_fdct16x16 sse2 msa/;
 
-    add_proto qw/void aom_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
-    specialize qw/aom_fdct16x16_1 sse2 avx2 msa/;
-
     add_proto qw/void aom_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_fdct32x32 sse2 avx2 msa/;
 
     add_proto qw/void aom_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
     specialize qw/aom_fdct32x32_rd sse2 avx2 msa/;
-
-    add_proto qw/void aom_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
-    specialize qw/aom_fdct32x32_1 sse2 avx2 msa/;
   }  # CONFIG_HIGHBITDEPTH
 }  # CONFIG_AV1_ENCODER
 
 #
 # Inverse transform
 if (aom_config("CONFIG_AV1") eq "yes") {
-if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
   add_proto qw/void aom_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
 
   add_proto qw/void aom_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
   specialize qw/aom_iwht4x4_16_add sse2/;
 
-  add_proto qw/void aom_highbd_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
-
-  add_proto qw/void aom_highbd_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
-
-  add_proto qw/void aom_highbd_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
-
-  add_proto qw/void aom_highbd_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
-
-  add_proto qw/void aom_highbd_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
-
-  add_proto qw/void aom_highbd_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
-
   add_proto qw/void aom_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
 
   add_proto qw/void aom_highbd_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
@@ -469,10 +453,8 @@ if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
 
     add_proto qw/void aom_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
     specialize qw/aom_idct32x32_1_add sse2 avx2/;
-
-    add_proto qw/void aom_highbd_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
-    specialize qw/aom_highbd_idct4x4_16_add sse2/;
   }
+if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
 } else {
   {
     add_proto qw/void aom_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
@@ -541,13 +523,12 @@ if (aom_config("CONFIG_AOM_QM") eq "yes") {
 
     add_proto qw/void aom_quantize_b_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
 
-    if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-      add_proto qw/void aom_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
+    add_proto qw/void aom_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
 
-      add_proto qw/void aom_highbd_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
+    add_proto qw/void aom_highbd_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
+
+    add_proto qw/void aom_highbd_quantize_b_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
 
-      add_proto qw/void aom_highbd_quantize_b_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
-    }  # CONFIG_HIGHBITDEPTH
   }  # CONFIG_AV1_ENCODER
 } else {
   if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
@@ -559,21 +540,23 @@ if (aom_config("CONFIG_AOM_QM") eq "yes") {
 
     add_proto qw/void aom_quantize_b_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
 
-    if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-      add_proto qw/void aom_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-      specialize qw/aom_highbd_quantize_b sse2/;
+    add_proto qw/void aom_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+    specialize qw/aom_highbd_quantize_b sse2 avx2/;
 
-      add_proto qw/void aom_highbd_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-      specialize qw/aom_highbd_quantize_b_32x32 sse2/;
+    add_proto qw/void aom_highbd_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+    specialize qw/aom_highbd_quantize_b_32x32 sse2/;
+
+    add_proto qw/void aom_highbd_quantize_b_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
 
-      add_proto qw/void aom_highbd_quantize_b_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-    }  # CONFIG_HIGHBITDEPTH
   }  # CONFIG_AV1_ENCODER
 } # CONFIG_AOM_QM
 if (aom_config("CONFIG_AV1") eq "yes") {
   #
   # Alpha blending with mask
   #
+  if (aom_config("CONFIG_CONVOLVE_ROUND") eq "yes") {
+    add_proto qw/void aom_blend_a64_d32_mask/, "int32_t *dst, uint32_t dst_stride, const int32_t *src0, uint32_t src0_stride, const int32_t *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int h, int w, int suby, int subx";
+  }
   add_proto qw/void aom_blend_a64_mask/, "uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int h, int w, int suby, int subx";
   add_proto qw/void aom_blend_a64_hmask/, "uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int h, int w";
   add_proto qw/void aom_blend_a64_vmask/, "uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int h, int w";
@@ -927,15 +910,15 @@ if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
 #
 # ...
 #
-add_proto qw/void aom_upsampled_pred/, "uint8_t *comp_pred, int width, int height, const uint8_t *ref, int ref_stride";
+add_proto qw/void aom_upsampled_pred/, "uint8_t *comp_pred, int width, int height, int subsample_x_q3, int subsample_y_q3, const uint8_t *ref, int ref_stride";
 specialize qw/aom_upsampled_pred sse2/;
-add_proto qw/void aom_comp_avg_upsampled_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, const uint8_t *ref, int ref_stride";
+add_proto qw/void aom_comp_avg_upsampled_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, int subsample_x_q3, int subsample_y_q3, const uint8_t *ref, int ref_stride";
 specialize qw/aom_comp_avg_upsampled_pred sse2/;
 
 if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-  add_proto qw/void aom_highbd_upsampled_pred/, "uint16_t *comp_pred, int width, int height, const uint8_t *ref8, int ref_stride";
+  add_proto qw/void aom_highbd_upsampled_pred/, "uint16_t *comp_pred, int width, int height, int subsample_x_q3, int subsample_y_q3, const uint8_t *ref8, int ref_stride, int bd";
   specialize qw/aom_highbd_upsampled_pred sse2/;
-  add_proto qw/void aom_highbd_comp_avg_upsampled_pred/, "uint16_t *comp_pred, const uint8_t *pred8, int width, int height, const uint8_t *ref8, int ref_stride";
+  add_proto qw/void aom_highbd_comp_avg_upsampled_pred/, "uint16_t *comp_pred, const uint8_t *pred8, int width, int height, int subsample_x_q3, int subsample_y_q3, const uint8_t *ref8, int ref_stride, int bd";
   specialize qw/aom_highbd_comp_avg_upsampled_pred sse2/;
 }
 
@@ -1005,6 +988,22 @@ specialize qw/aom_sub_pixel_avg_variance8x8        msa sse2 ssse3/;
 specialize qw/aom_sub_pixel_avg_variance8x4        msa sse2 ssse3/;
 specialize qw/aom_sub_pixel_avg_variance4x8        msa sse2 ssse3/;
 specialize qw/aom_sub_pixel_avg_variance4x4        msa sse2 ssse3/;
+
+if (aom_config("CONFIG_EXT_PARTITION_TYPES")) {
+  specialize qw/aom_variance4x16 sse2/;
+  specialize qw/aom_variance16x4 sse2/;
+  specialize qw/aom_variance8x32 sse2/;
+  specialize qw/aom_variance32x8 sse2/;
+  specialize qw/aom_sub_pixel_variance4x16 sse2 ssse3/;
+  specialize qw/aom_sub_pixel_variance16x4 sse2 ssse3/;
+  specialize qw/aom_sub_pixel_variance8x32 sse2 ssse3/;
+  specialize qw/aom_sub_pixel_variance32x8 sse2 ssse3/;
+  specialize qw/aom_sub_pixel_avg_variance4x16 sse2 ssse3/;
+  specialize qw/aom_sub_pixel_avg_variance16x4 sse2 ssse3/;
+  specialize qw/aom_sub_pixel_avg_variance8x32 sse2 ssse3/;
+  specialize qw/aom_sub_pixel_avg_variance32x8 sse2 ssse3/;
+}
+
 if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
   foreach $bd (8, 10, 12) {
     add_proto qw/unsigned int/, "aom_highbd_${bd}_variance2x2", "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
@@ -1021,6 +1020,8 @@ if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
       if ($w != 128 && $h != 128 && $w != 4 && $h != 4) {
         specialize "aom_highbd_${bd}_variance${w}x${h}", "sse2";
       }
+      # TODO(david.barker): When ext-partition-types is enabled, we currenly
+      # don't have vectorized 4x16 highbd variance functions
       if ($w == 4 && $h == 4) {
         specialize "aom_highbd_${bd}_variance${w}x${h}", "sse4_1";
       }
@@ -1496,10 +1497,10 @@ if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
 
 if (aom_config("CONFIG_EXT_INTER") eq "yes") {
   add_proto qw/void aom_comp_mask_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, const uint8_t *ref, int ref_stride, const uint8_t *mask, int mask_stride, int invert_mask";
-  add_proto qw/void aom_comp_mask_upsampled_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, const uint8_t *ref, int ref_stride, const uint8_t *mask, int mask_stride, int invert_mask";
+  add_proto qw/void aom_comp_mask_upsampled_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, int subsample_x_q3, int subsample_y_q3, const uint8_t *ref, int ref_stride, const uint8_t *mask, int mask_stride, int invert_mask";
   if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
     add_proto qw/void aom_highbd_comp_mask_pred/, "uint16_t *comp_pred, const uint8_t *pred8, int width, int height, const uint8_t *ref8, int ref_stride, const uint8_t *mask, int mask_stride, int invert_mask";
-    add_proto qw/void aom_highbd_comp_mask_upsampled_pred/, "uint16_t *comp_pred, const uint8_t *pred8, int width, int height, const uint8_t *ref8, int ref_stride, const uint8_t *mask, int mask_stride, int invert_mask";
+    add_proto qw/void aom_highbd_comp_mask_upsampled_pred/, "uint16_t *comp_pred, const uint8_t *pred8, int width, int height, int subsample_x_q3, int subsample_y_q3, const uint8_t *ref8, int ref_stride, const uint8_t *mask, int mask_stride, int invert_mask, int bd";
   }
 }
 
diff --git a/third_party/aom/aom_dsp/aom_filter.h b/third_party/aom/aom_dsp/aom_filter.h
index 04d113dd3..58e8bb284 100644
--- a/third_party/aom/aom_dsp/aom_filter.h
+++ b/third_party/aom/aom_dsp/aom_filter.h
@@ -25,6 +25,12 @@ extern "C" {
 #define SUBPEL_SHIFTS (1 << SUBPEL_BITS)
 #define SUBPEL_TAPS 8
 
+#define SCALE_SUBPEL_BITS 10
+#define SCALE_SUBPEL_SHIFTS (1 << SCALE_SUBPEL_BITS)
+#define SCALE_SUBPEL_MASK (SCALE_SUBPEL_SHIFTS - 1)
+#define SCALE_EXTRA_BITS (SCALE_SUBPEL_BITS - SUBPEL_BITS)
+#define SCALE_EXTRA_OFF ((1 << SCALE_EXTRA_BITS) / 2)
+
 typedef int16_t InterpKernel[SUBPEL_TAPS];
 
 #define BIL_SUBPEL_BITS 3
diff --git a/third_party/aom/aom_dsp/bitreader.h b/third_party/aom/aom_dsp/bitreader.h
index 5bad70cb3..88bedccc2 100644
--- a/third_party/aom/aom_dsp/bitreader.h
+++ b/third_party/aom/aom_dsp/bitreader.h
@@ -194,9 +194,7 @@ static INLINE int aom_read_symbol_(aom_reader *r, aom_cdf_prob *cdf,
                                    int nsymbs ACCT_STR_PARAM) {
   int ret;
   ret = aom_read_cdf(r, cdf, nsymbs, ACCT_STR_NAME);
-#if CONFIG_EC_ADAPT
   update_cdf(cdf, ret, nsymbs);
-#endif
   return ret;
 }
 
diff --git a/third_party/aom/aom_dsp/bitwriter.h b/third_party/aom/aom_dsp/bitwriter.h
index 588e47bf3..68bc1c8f8 100644
--- a/third_party/aom/aom_dsp/bitwriter.h
+++ b/third_party/aom/aom_dsp/bitwriter.h
@@ -140,9 +140,7 @@ static INLINE void aom_write_cdf(aom_writer *w, int symb,
 static INLINE void aom_write_symbol(aom_writer *w, int symb, aom_cdf_prob *cdf,
                                     int nsymbs) {
   aom_write_cdf(w, symb, cdf, nsymbs);
-#if CONFIG_EC_ADAPT
   update_cdf(cdf, symb, nsymbs);
-#endif
 }
 
 static INLINE void aom_write_tree_as_cdf(aom_writer *w,
diff --git a/third_party/aom/aom_dsp/blend_a64_mask.c b/third_party/aom/aom_dsp/blend_a64_mask.c
index 3e15542c9..c35fa19f8 100644
--- a/third_party/aom/aom_dsp/blend_a64_mask.c
+++ b/third_party/aom/aom_dsp/blend_a64_mask.c
@@ -18,6 +18,69 @@
 
 #include "./aom_dsp_rtcd.h"
 
+#if CONFIG_CONVOLVE_ROUND
+// Blending with alpha mask. Mask values come from the range [0, 64],
+// as described for AOM_BLEND_A64 in aom_dsp/blend.h. src0 or src1 can
+// be the same as dst, or dst can be different from both sources.
+
+void aom_blend_a64_d32_mask_c(int32_t *dst, uint32_t dst_stride,
+                              const int32_t *src0, uint32_t src0_stride,
+                              const int32_t *src1, uint32_t src1_stride,
+                              const uint8_t *mask, uint32_t mask_stride, int h,
+                              int w, int subh, int subw) {
+  int i, j;
+
+  assert(IMPLIES(src0 == dst, src0_stride == dst_stride));
+  assert(IMPLIES(src1 == dst, src1_stride == dst_stride));
+
+  assert(h >= 1);
+  assert(w >= 1);
+  assert(IS_POWER_OF_TWO(h));
+  assert(IS_POWER_OF_TWO(w));
+
+  if (subw == 0 && subh == 0) {
+    for (i = 0; i < h; ++i) {
+      for (j = 0; j < w; ++j) {
+        const int m = mask[i * mask_stride + j];
+        dst[i * dst_stride + j] = AOM_BLEND_A64(m, src0[i * src0_stride + j],
+                                                src1[i * src1_stride + j]);
+      }
+    }
+  } else if (subw == 1 && subh == 1) {
+    for (i = 0; i < h; ++i) {
+      for (j = 0; j < w; ++j) {
+        const int m = ROUND_POWER_OF_TWO(
+            mask[(2 * i) * mask_stride + (2 * j)] +
+                mask[(2 * i + 1) * mask_stride + (2 * j)] +
+                mask[(2 * i) * mask_stride + (2 * j + 1)] +
+                mask[(2 * i + 1) * mask_stride + (2 * j + 1)],
+            2);
+        dst[i * dst_stride + j] = AOM_BLEND_A64(m, src0[i * src0_stride + j],
+                                                src1[i * src1_stride + j]);
+      }
+    }
+  } else if (subw == 1 && subh == 0) {
+    for (i = 0; i < h; ++i) {
+      for (j = 0; j < w; ++j) {
+        const int m = AOM_BLEND_AVG(mask[i * mask_stride + (2 * j)],
+                                    mask[i * mask_stride + (2 * j + 1)]);
+        dst[i * dst_stride + j] = AOM_BLEND_A64(m, src0[i * src0_stride + j],
+                                                src1[i * src1_stride + j]);
+      }
+    }
+  } else {
+    for (i = 0; i < h; ++i) {
+      for (j = 0; j < w; ++j) {
+        const int m = AOM_BLEND_AVG(mask[(2 * i) * mask_stride + j],
+                                    mask[(2 * i + 1) * mask_stride + j]);
+        dst[i * dst_stride + j] = AOM_BLEND_A64(m, src0[i * src0_stride + j],
+                                                src1[i * src1_stride + j]);
+      }
+    }
+  }
+}
+#endif  // CONFIG_CONVOLVE_ROUND
+
 // Blending with alpha mask. Mask values come from the range [0, 64],
 // as described for AOM_BLEND_A64 in aom_dsp/blend.h. src0 or src1 can
 // be the same as dst, or dst can be different from both sources.
diff --git a/third_party/aom/aom_dsp/fwd_txfm.c b/third_party/aom/aom_dsp/fwd_txfm.c
index 12ee02ba1..1ceef7782 100644
--- a/third_party/aom/aom_dsp/fwd_txfm.c
+++ b/third_party/aom/aom_dsp/fwd_txfm.c
@@ -172,15 +172,6 @@ void aom_fdct8x8_c(const int16_t *input, tran_low_t *final_output, int stride) {
   }
 }
 
-void aom_fdct8x8_1_c(const int16_t *input, tran_low_t *output, int stride) {
-  int r, c;
-  tran_low_t sum = 0;
-  for (r = 0; r < 8; ++r)
-    for (c = 0; c < 8; ++c) sum += input[r * stride + c];
-
-  output[0] = sum;
-}
-
 void aom_fdct16x16_c(const int16_t *input, tran_low_t *output, int stride) {
   // The 2D transform is done with two passes which are actually pretty
   // similar. In the first one, we transform the columns and transpose
@@ -361,15 +352,6 @@ void aom_fdct16x16_c(const int16_t *input, tran_low_t *output, int stride) {
   }
 }
 
-void aom_fdct16x16_1_c(const int16_t *input, tran_low_t *output, int stride) {
-  int r, c;
-  int sum = 0;
-  for (r = 0; r < 16; ++r)
-    for (c = 0; c < 16; ++c) sum += input[r * stride + c];
-
-  output[0] = (tran_low_t)(sum >> 1);
-}
-
 static INLINE tran_high_t dct_32_round(tran_high_t input) {
   tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
   // TODO(debargha, peter.derivaz): Find new bounds for this assert,
@@ -758,15 +740,6 @@ void aom_fdct32x32_rd_c(const int16_t *input, tran_low_t *out, int stride) {
   }
 }
 
-void aom_fdct32x32_1_c(const int16_t *input, tran_low_t *output, int stride) {
-  int r, c;
-  int sum = 0;
-  for (r = 0; r < 32; ++r)
-    for (c = 0; c < 32; ++c) sum += input[r * stride + c];
-
-  output[0] = (tran_low_t)(sum >> 3);
-}
-
 #if CONFIG_HIGHBITDEPTH
 void aom_highbd_fdct4x4_c(const int16_t *input, tran_low_t *output,
                           int stride) {
@@ -778,32 +751,17 @@ void aom_highbd_fdct8x8_c(const int16_t *input, tran_low_t *final_output,
   aom_fdct8x8_c(input, final_output, stride);
 }
 
-void aom_highbd_fdct8x8_1_c(const int16_t *input, tran_low_t *final_output,
-                            int stride) {
-  aom_fdct8x8_1_c(input, final_output, stride);
-}
-
 void aom_highbd_fdct16x16_c(const int16_t *input, tran_low_t *output,
                             int stride) {
   aom_fdct16x16_c(input, output, stride);
 }
 
-void aom_highbd_fdct16x16_1_c(const int16_t *input, tran_low_t *output,
-                              int stride) {
-  aom_fdct16x16_1_c(input, output, stride);
-}
-
 void aom_highbd_fdct32x32_c(const int16_t *input, tran_low_t *out, int stride) {
   aom_fdct32x32_c(input, out, stride);
 }
-
 void aom_highbd_fdct32x32_rd_c(const int16_t *input, tran_low_t *out,
                                int stride) {
   aom_fdct32x32_rd_c(input, out, stride);
 }
 
-void aom_highbd_fdct32x32_1_c(const int16_t *input, tran_low_t *out,
-                              int stride) {
-  aom_fdct32x32_1_c(input, out, stride);
-}
 #endif  // CONFIG_HIGHBITDEPTH
diff --git a/third_party/aom/aom_dsp/fwd_txfm.h b/third_party/aom/aom_dsp/fwd_txfm.h
index 579dbd06e..f4dc04ab4 100644
--- a/third_party/aom/aom_dsp/fwd_txfm.h
+++ b/third_party/aom/aom_dsp/fwd_txfm.h
@@ -20,10 +20,5 @@ static INLINE tran_high_t saturate_int16(tran_high_t value) {
   return result < INT16_MIN ? INT16_MIN : result;
 }
 
-static INLINE tran_high_t fdct_round_shift(tran_high_t input) {
-  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
-  return rv;
-}
-
 void aom_fdct32(const tran_high_t *input, tran_high_t *output, int round);
 #endif  // AOM_DSP_FWD_TXFM_H_
diff --git a/third_party/aom/aom_dsp/intrapred.c b/third_party/aom/aom_dsp/intrapred.c
index 370d0374b..b4d47ae89 100644
--- a/third_party/aom/aom_dsp/intrapred.c
+++ b/third_party/aom/aom_dsp/intrapred.c
@@ -23,13 +23,14 @@
 #define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
 #define AVG2(a, b) (((a) + (b) + 1) >> 1)
 
-static INLINE void d207e_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                   const uint8_t *above, const uint8_t *left) {
+static INLINE void d207e_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                   int bh, const uint8_t *above,
+                                   const uint8_t *left) {
   int r, c;
   (void)above;
 
-  for (r = 0; r < bs; ++r) {
-    for (c = 0; c < bs; ++c) {
+  for (r = 0; r < bh; ++r) {
+    for (c = 0; c < bw; ++c) {
       dst[c] = c & 1 ? AVG3(left[(c >> 1) + r], left[(c >> 1) + r + 1],
                             left[(c >> 1) + r + 2])
                      : AVG2(left[(c >> 1) + r], left[(c >> 1) + r + 1]);
@@ -38,12 +39,13 @@ static INLINE void d207e_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
   }
 }
 
-static INLINE void d63e_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                  const uint8_t *above, const uint8_t *left) {
+static INLINE void d63e_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                  int bh, const uint8_t *above,
+                                  const uint8_t *left) {
   int r, c;
   (void)left;
-  for (r = 0; r < bs; ++r) {
-    for (c = 0; c < bs; ++c) {
+  for (r = 0; r < bh; ++r) {
+    for (c = 0; c < bw; ++c) {
       dst[c] = r & 1 ? AVG3(above[(r >> 1) + c], above[(r >> 1) + c + 1],
                             above[(r >> 1) + c + 2])
                      : AVG2(above[(r >> 1) + c], above[(r >> 1) + c + 1]);
@@ -52,46 +54,49 @@ static INLINE void d63e_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
   }
 }
 
-static INLINE void d45e_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                  const uint8_t *above, const uint8_t *left) {
+static INLINE void d45e_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                  int bh, const uint8_t *above,
+                                  const uint8_t *left) {
   int r, c;
   (void)left;
-  for (r = 0; r < bs; ++r) {
-    for (c = 0; c < bs; ++c) {
+  for (r = 0; r < bh; ++r) {
+    for (c = 0; c < bw; ++c) {
       dst[c] = AVG3(above[r + c], above[r + c + 1],
-                    above[r + c + 1 + (r + c + 2 < bs * 2)]);
+                    above[r + c + 1 + (r + c + 2 < bw + bh)]);
     }
     dst += stride;
   }
 }
 
-static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                  const uint8_t *above, const uint8_t *left) {
+static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                  int bh, const uint8_t *above,
+                                  const uint8_t *left) {
   int r, c;
 
   // first row
-  for (c = 0; c < bs; c++) dst[c] = AVG2(above[c - 1], above[c]);
+  for (c = 0; c < bw; c++) dst[c] = AVG2(above[c - 1], above[c]);
   dst += stride;
 
   // second row
   dst[0] = AVG3(left[0], above[-1], above[0]);
-  for (c = 1; c < bs; c++) dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
+  for (c = 1; c < bw; c++) dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
   dst += stride;
 
   // the rest of first col
   dst[0] = AVG3(above[-1], left[0], left[1]);
-  for (r = 3; r < bs; ++r)
+  for (r = 3; r < bh; ++r)
     dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);
 
   // the rest of the block
-  for (r = 2; r < bs; ++r) {
-    for (c = 1; c < bs; c++) dst[c] = dst[-2 * stride + c - 1];
+  for (r = 2; r < bh; ++r) {
+    for (c = 1; c < bw; c++) dst[c] = dst[-2 * stride + c - 1];
     dst += stride;
   }
 }
 
-static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                  const uint8_t *above, const uint8_t *left) {
+static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                  int bh, const uint8_t *above,
+                                  const uint8_t *left) {
   int i;
 #if CONFIG_TX64X64
 #if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ > 7
@@ -111,64 +116,65 @@ static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
 #endif
 #endif  // CONFIG_TX64X64
 
-  // dst(bs, bs - 2)[0], i.e., border starting at bottom-left
-  for (i = 0; i < bs - 2; ++i) {
-    border[i] = AVG3(left[bs - 3 - i], left[bs - 2 - i], left[bs - 1 - i]);
+  // dst(bh, bh - 2)[0], i.e., border starting at bottom-left
+  for (i = 0; i < bh - 2; ++i) {
+    border[i] = AVG3(left[bh - 3 - i], left[bh - 2 - i], left[bh - 1 - i]);
   }
-  border[bs - 2] = AVG3(above[-1], left[0], left[1]);
-  border[bs - 1] = AVG3(left[0], above[-1], above[0]);
-  border[bs - 0] = AVG3(above[-1], above[0], above[1]);
+  border[bh - 2] = AVG3(above[-1], left[0], left[1]);
+  border[bh - 1] = AVG3(left[0], above[-1], above[0]);
+  border[bh - 0] = AVG3(above[-1], above[0], above[1]);
   // dst[0][2, size), i.e., remaining top border ascending
-  for (i = 0; i < bs - 2; ++i) {
-    border[bs + 1 + i] = AVG3(above[i], above[i + 1], above[i + 2]);
+  for (i = 0; i < bw - 2; ++i) {
+    border[bh + 1 + i] = AVG3(above[i], above[i + 1], above[i + 2]);
   }
 
-  for (i = 0; i < bs; ++i) {
-    memcpy(dst + i * stride, border + bs - 1 - i, bs);
+  for (i = 0; i < bh; ++i) {
+    memcpy(dst + i * stride, border + bh - 1 - i, bw);
   }
 }
 
-static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                  const uint8_t *above, const uint8_t *left) {
+static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                  int bh, const uint8_t *above,
+                                  const uint8_t *left) {
   int r, c;
   dst[0] = AVG2(above[-1], left[0]);
-  for (r = 1; r < bs; r++) dst[r * stride] = AVG2(left[r - 1], left[r]);
+  for (r = 1; r < bh; r++) dst[r * stride] = AVG2(left[r - 1], left[r]);
   dst++;
 
   dst[0] = AVG3(left[0], above[-1], above[0]);
   dst[stride] = AVG3(above[-1], left[0], left[1]);
-  for (r = 2; r < bs; r++)
+  for (r = 2; r < bh; r++)
     dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
   dst++;
 
-  for (c = 0; c < bs - 2; c++)
+  for (c = 0; c < bw - 2; c++)
     dst[c] = AVG3(above[c - 1], above[c], above[c + 1]);
   dst += stride;
 
-  for (r = 1; r < bs; ++r) {
-    for (c = 0; c < bs - 2; c++) dst[c] = dst[-stride + c - 2];
+  for (r = 1; r < bh; ++r) {
+    for (c = 0; c < bw - 2; c++) dst[c] = dst[-stride + c - 2];
     dst += stride;
   }
 }
 
-static INLINE void v_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+static INLINE void v_predictor(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
                                const uint8_t *above, const uint8_t *left) {
   int r;
   (void)left;
 
-  for (r = 0; r < bs; r++) {
-    memcpy(dst, above, bs);
+  for (r = 0; r < bh; r++) {
+    memcpy(dst, above, bw);
     dst += stride;
   }
 }
 
-static INLINE void h_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+static INLINE void h_predictor(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
                                const uint8_t *above, const uint8_t *left) {
   int r;
   (void)above;
 
-  for (r = 0; r < bs; r++) {
-    memset(dst, left[r], bs);
+  for (r = 0; r < bh; r++) {
+    memset(dst, left[r], bw);
     dst += stride;
   }
 }
@@ -189,13 +195,14 @@ static INLINE uint16_t paeth_predictor_single(uint16_t left, uint16_t top,
              : (p_top <= p_top_left) ? top : top_left;
 }
 
-static INLINE void paeth_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                   const uint8_t *above, const uint8_t *left) {
+static INLINE void paeth_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                   int bh, const uint8_t *above,
+                                   const uint8_t *left) {
   int r, c;
   const uint8_t ytop_left = above[-1];
 
-  for (r = 0; r < bs; r++) {
-    for (c = 0; c < bs; c++)
+  for (r = 0; r < bh; r++) {
+    for (c = 0; c < bw; c++)
       dst[c] = (uint8_t)paeth_predictor_single(left[r], above[c], ytop_left);
     dst += stride;
   }
@@ -236,32 +243,38 @@ static const uint8_t sm_weight_arrays[2 * MAX_BLOCK_DIM] = {
 };
 
 // Some basic checks on weights for smooth predictor.
-#define sm_weights_sanity_checks(weights, weights_scale, pred_scale) \
-  assert(weights[0] < weights_scale);                                \
-  assert(weights_scale - weights[bs - 1] < weights_scale);           \
+#define sm_weights_sanity_checks(weights_w, weights_h, weights_scale, \
+                                 pred_scale)                          \
+  assert(weights_w[0] < weights_scale);                               \
+  assert(weights_h[0] < weights_scale);                               \
+  assert(weights_scale - weights_w[bw - 1] < weights_scale);          \
+  assert(weights_scale - weights_h[bh - 1] < weights_scale);          \
   assert(pred_scale < 31)  // ensures no overflow when calculating predictor.
 
 #define divide_round(value, bits) (((value) + (1 << ((bits)-1))) >> (bits))
 
-static INLINE void smooth_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                    const uint8_t *above, const uint8_t *left) {
-  const uint8_t below_pred = left[bs - 1];   // estimated by bottom-left pixel
-  const uint8_t right_pred = above[bs - 1];  // estimated by top-right pixel
-  const uint8_t *const sm_weights = sm_weight_arrays + bs;
+static INLINE void smooth_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                    int bh, const uint8_t *above,
+                                    const uint8_t *left) {
+  const uint8_t below_pred = left[bh - 1];   // estimated by bottom-left pixel
+  const uint8_t right_pred = above[bw - 1];  // estimated by top-right pixel
+  const uint8_t *const sm_weights_w = sm_weight_arrays + bw;
+  const uint8_t *const sm_weights_h = sm_weight_arrays + bh;
   // scale = 2 * 2^sm_weight_log2_scale
   const int log2_scale = 1 + sm_weight_log2_scale;
   const uint16_t scale = (1 << sm_weight_log2_scale);
-  sm_weights_sanity_checks(sm_weights, scale, log2_scale + sizeof(*dst));
+  sm_weights_sanity_checks(sm_weights_w, sm_weights_h, scale,
+                           log2_scale + sizeof(*dst));
   int r;
-  for (r = 0; r < bs; ++r) {
+  for (r = 0; r < bh; ++r) {
     int c;
-    for (c = 0; c < bs; ++c) {
+    for (c = 0; c < bw; ++c) {
       const uint8_t pixels[] = { above[c], below_pred, left[r], right_pred };
-      const uint8_t weights[] = { sm_weights[r], scale - sm_weights[r],
-                                  sm_weights[c], scale - sm_weights[c] };
+      const uint8_t weights[] = { sm_weights_h[r], scale - sm_weights_h[r],
+                                  sm_weights_w[c], scale - sm_weights_w[c] };
       uint32_t this_pred = 0;
       int i;
-      assert(scale >= sm_weights[r] && scale >= sm_weights[c]);
+      assert(scale >= sm_weights_h[r] && scale >= sm_weights_w[c]);
       for (i = 0; i < 4; ++i) {
         this_pred += weights[i] * pixels[i];
       }
@@ -272,20 +285,21 @@ static INLINE void smooth_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
 }
 
 #if CONFIG_SMOOTH_HV
-static INLINE void smooth_v_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                      const uint8_t *above,
+static INLINE void smooth_v_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                      int bh, const uint8_t *above,
                                       const uint8_t *left) {
-  const uint8_t below_pred = left[bs - 1];  // estimated by bottom-left pixel
-  const uint8_t *const sm_weights = sm_weight_arrays + bs;
+  const uint8_t below_pred = left[bh - 1];  // estimated by bottom-left pixel
+  const uint8_t *const sm_weights = sm_weight_arrays + bh;
   // scale = 2^sm_weight_log2_scale
   const int log2_scale = sm_weight_log2_scale;
   const uint16_t scale = (1 << sm_weight_log2_scale);
-  sm_weights_sanity_checks(sm_weights, scale, log2_scale + sizeof(*dst));
+  sm_weights_sanity_checks(sm_weights, sm_weights, scale,
+                           log2_scale + sizeof(*dst));
 
   int r;
-  for (r = 0; r < bs; r++) {
+  for (r = 0; r < bh; r++) {
     int c;
-    for (c = 0; c < bs; ++c) {
+    for (c = 0; c < bw; ++c) {
       const uint8_t pixels[] = { above[c], below_pred };
       const uint8_t weights[] = { sm_weights[r], scale - sm_weights[r] };
       uint32_t this_pred = 0;
@@ -300,20 +314,21 @@ static INLINE void smooth_v_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
   }
 }
 
-static INLINE void smooth_h_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                      const uint8_t *above,
+static INLINE void smooth_h_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                      int bh, const uint8_t *above,
                                       const uint8_t *left) {
-  const uint8_t right_pred = above[bs - 1];  // estimated by top-right pixel
-  const uint8_t *const sm_weights = sm_weight_arrays + bs;
+  const uint8_t right_pred = above[bw - 1];  // estimated by top-right pixel
+  const uint8_t *const sm_weights = sm_weight_arrays + bw;
   // scale = 2^sm_weight_log2_scale
   const int log2_scale = sm_weight_log2_scale;
   const uint16_t scale = (1 << sm_weight_log2_scale);
-  sm_weights_sanity_checks(sm_weights, scale, log2_scale + sizeof(*dst));
+  sm_weights_sanity_checks(sm_weights, sm_weights, scale,
+                           log2_scale + sizeof(*dst));
 
   int r;
-  for (r = 0; r < bs; r++) {
+  for (r = 0; r < bh; r++) {
     int c;
-    for (c = 0; c < bs; ++c) {
+    for (c = 0; c < bw; ++c) {
       const uint8_t pixels[] = { left[r], right_pred };
       const uint8_t weights[] = { sm_weights[c], scale - sm_weights[c] };
       uint32_t this_pred = 0;
@@ -331,74 +346,78 @@ static INLINE void smooth_h_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
 
 #else
 
-static INLINE void tm_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+static INLINE void tm_predictor(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
                                 const uint8_t *above, const uint8_t *left) {
   int r, c;
   int ytop_left = above[-1];
 
-  for (r = 0; r < bs; r++) {
-    for (c = 0; c < bs; c++)
+  for (r = 0; r < bh; r++) {
+    for (c = 0; c < bw; c++)
       dst[c] = clip_pixel(left[r] + above[c] - ytop_left);
     dst += stride;
   }
 }
 #endif  // CONFIG_ALT_INTRA
 
-static INLINE void dc_128_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                    const uint8_t *above, const uint8_t *left) {
+static INLINE void dc_128_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                    int bh, const uint8_t *above,
+                                    const uint8_t *left) {
   int r;
   (void)above;
   (void)left;
 
-  for (r = 0; r < bs; r++) {
-    memset(dst, 128, bs);
+  for (r = 0; r < bh; r++) {
+    memset(dst, 128, bw);
     dst += stride;
   }
 }
 
-static INLINE void dc_left_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                     const uint8_t *above,
+static INLINE void dc_left_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                     int bh, const uint8_t *above,
                                      const uint8_t *left) {
   int i, r, expected_dc, sum = 0;
   (void)above;
 
-  for (i = 0; i < bs; i++) sum += left[i];
-  expected_dc = (sum + (bs >> 1)) / bs;
+  for (i = 0; i < bh; i++) sum += left[i];
+  expected_dc = (sum + (bh >> 1)) / bh;
 
-  for (r = 0; r < bs; r++) {
-    memset(dst, expected_dc, bs);
+  for (r = 0; r < bh; r++) {
+    memset(dst, expected_dc, bw);
     dst += stride;
   }
 }
 
-static INLINE void dc_top_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
-                                    const uint8_t *above, const uint8_t *left) {
+static INLINE void dc_top_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
+                                    int bh, const uint8_t *above,
+                                    const uint8_t *left) {
   int i, r, expected_dc, sum = 0;
   (void)left;
 
-  for (i = 0; i < bs; i++) sum += above[i];
-  expected_dc = (sum + (bs >> 1)) / bs;
+  for (i = 0; i < bw; i++) sum += above[i];
+  expected_dc = (sum + (bw >> 1)) / bw;
 
-  for (r = 0; r < bs; r++) {
-    memset(dst, expected_dc, bs);
+  for (r = 0; r < bh; r++) {
+    memset(dst, expected_dc, bw);
     dst += stride;
   }
 }
 
-static INLINE void dc_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+static INLINE void dc_predictor(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
                                 const uint8_t *above, const uint8_t *left) {
   int i, r, expected_dc, sum = 0;
-  const int count = 2 * bs;
+  const int count = bw + bh;
 
-  for (i = 0; i < bs; i++) {
+  for (i = 0; i < bw; i++) {
     sum += above[i];
+  }
+  for (i = 0; i < bh; i++) {
     sum += left[i];
   }
 
   expected_dc = (sum + (count >> 1)) / count;
 
-  for (r = 0; r < bs; r++) {
-    memset(dst, expected_dc, bs);
+  for (r = 0; r < bh; r++) {
+    memset(dst, expected_dc, bw);
     dst += stride;
   }
 }
@@ -546,14 +565,14 @@ void aom_d153_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
 
 #if CONFIG_HIGHBITDEPTH
 static INLINE void highbd_d207e_predictor(uint16_t *dst, ptrdiff_t stride,
-                                          int bs, const uint16_t *above,
+                                          int bw, int bh, const uint16_t *above,
                                           const uint16_t *left, int bd) {
   int r, c;
   (void)above;
   (void)bd;
 
-  for (r = 0; r < bs; ++r) {
-    for (c = 0; c < bs; ++c) {
+  for (r = 0; r < bh; ++r) {
+    for (c = 0; c < bw; ++c) {
       dst[c] = c & 1 ? AVG3(left[(c >> 1) + r], left[(c >> 1) + r + 1],
                             left[(c >> 1) + r + 2])
                      : AVG2(left[(c >> 1) + r], left[(c >> 1) + r + 1]);
@@ -563,13 +582,13 @@ static INLINE void highbd_d207e_predictor(uint16_t *dst, ptrdiff_t stride,
 }
 
 static INLINE void highbd_d63e_predictor(uint16_t *dst, ptrdiff_t stride,
-                                         int bs, const uint16_t *above,
+                                         int bw, int bh, const uint16_t *above,
                                          const uint16_t *left, int bd) {
   int r, c;
   (void)left;
   (void)bd;
-  for (r = 0; r < bs; ++r) {
-    for (c = 0; c < bs; ++c) {
+  for (r = 0; r < bh; ++r) {
+    for (c = 0; c < bw; ++c) {
       dst[c] = r & 1 ? AVG3(above[(r >> 1) + c], above[(r >> 1) + c + 1],
                             above[(r >> 1) + c + 2])
                      : AVG2(above[(r >> 1) + c], above[(r >> 1) + c + 1]);
@@ -579,111 +598,111 @@ static INLINE void highbd_d63e_predictor(uint16_t *dst, ptrdiff_t stride,
 }
 
 static INLINE void highbd_d45e_predictor(uint16_t *dst, ptrdiff_t stride,
-                                         int bs, const uint16_t *above,
+                                         int bw, int bh, const uint16_t *above,
                                          const uint16_t *left, int bd) {
   int r, c;
   (void)left;
   (void)bd;
-  for (r = 0; r < bs; ++r) {
-    for (c = 0; c < bs; ++c) {
+  for (r = 0; r < bh; ++r) {
+    for (c = 0; c < bw; ++c) {
       dst[c] = AVG3(above[r + c], above[r + c + 1],
-                    above[r + c + 1 + (r + c + 2 < bs * 2)]);
+                    above[r + c + 1 + (r + c + 2 < bw + bh)]);
     }
     dst += stride;
   }
 }
 
 static INLINE void highbd_d117_predictor(uint16_t *dst, ptrdiff_t stride,
-                                         int bs, const uint16_t *above,
+                                         int bw, int bh, const uint16_t *above,
                                          const uint16_t *left, int bd) {
   int r, c;
   (void)bd;
 
   // first row
-  for (c = 0; c < bs; c++) dst[c] = AVG2(above[c - 1], above[c]);
+  for (c = 0; c < bw; c++) dst[c] = AVG2(above[c - 1], above[c]);
   dst += stride;
 
   // second row
   dst[0] = AVG3(left[0], above[-1], above[0]);
-  for (c = 1; c < bs; c++) dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
+  for (c = 1; c < bw; c++) dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
   dst += stride;
 
   // the rest of first col
   dst[0] = AVG3(above[-1], left[0], left[1]);
-  for (r = 3; r < bs; ++r)
+  for (r = 3; r < bh; ++r)
     dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);
 
   // the rest of the block
-  for (r = 2; r < bs; ++r) {
-    for (c = 1; c < bs; c++) dst[c] = dst[-2 * stride + c - 1];
+  for (r = 2; r < bh; ++r) {
+    for (c = 1; c < bw; c++) dst[c] = dst[-2 * stride + c - 1];
     dst += stride;
   }
 }
 
 static INLINE void highbd_d135_predictor(uint16_t *dst, ptrdiff_t stride,
-                                         int bs, const uint16_t *above,
+                                         int bw, int bh, const uint16_t *above,
                                          const uint16_t *left, int bd) {
   int r, c;
   (void)bd;
   dst[0] = AVG3(left[0], above[-1], above[0]);
-  for (c = 1; c < bs; c++) dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
+  for (c = 1; c < bw; c++) dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
 
   dst[stride] = AVG3(above[-1], left[0], left[1]);
-  for (r = 2; r < bs; ++r)
+  for (r = 2; r < bh; ++r)
     dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
 
   dst += stride;
-  for (r = 1; r < bs; ++r) {
-    for (c = 1; c < bs; c++) dst[c] = dst[-stride + c - 1];
+  for (r = 1; r < bh; ++r) {
+    for (c = 1; c < bw; c++) dst[c] = dst[-stride + c - 1];
     dst += stride;
   }
 }
 
 static INLINE void highbd_d153_predictor(uint16_t *dst, ptrdiff_t stride,
-                                         int bs, const uint16_t *above,
+                                         int bw, int bh, const uint16_t *above,
                                          const uint16_t *left, int bd) {
   int r, c;
   (void)bd;
   dst[0] = AVG2(above[-1], left[0]);
-  for (r = 1; r < bs; r++) dst[r * stride] = AVG2(left[r - 1], left[r]);
+  for (r = 1; r < bh; r++) dst[r * stride] = AVG2(left[r - 1], left[r]);
   dst++;
 
   dst[0] = AVG3(left[0], above[-1], above[0]);
   dst[stride] = AVG3(above[-1], left[0], left[1]);
-  for (r = 2; r < bs; r++)
+  for (r = 2; r < bh; r++)
     dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
   dst++;
 
-  for (c = 0; c < bs - 2; c++)
+  for (c = 0; c < bw - 2; c++)
     dst[c] = AVG3(above[c - 1], above[c], above[c + 1]);
   dst += stride;
 
-  for (r = 1; r < bs; ++r) {
-    for (c = 0; c < bs - 2; c++) dst[c] = dst[-stride + c - 2];
+  for (r = 1; r < bh; ++r) {
+    for (c = 0; c < bw - 2; c++) dst[c] = dst[-stride + c - 2];
     dst += stride;
   }
 }
 
-static INLINE void highbd_v_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
-                                      const uint16_t *above,
+static INLINE void highbd_v_predictor(uint16_t *dst, ptrdiff_t stride, int bw,
+                                      int bh, const uint16_t *above,
                                       const uint16_t *left, int bd) {
   int r;
   (void)left;
   (void)bd;
-  for (r = 0; r < bs; r++) {
-    memcpy(dst, above, bs * sizeof(uint16_t));
+  for (r = 0; r < bh; r++) {
+    memcpy(dst, above, bw * sizeof(uint16_t));
     dst += stride;
   }
 }
 
-static INLINE void highbd_h_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
-                                      const uint16_t *above,
+static INLINE void highbd_h_predictor(uint16_t *dst, ptrdiff_t stride, int bw,
+                                      int bh, const uint16_t *above,
                                       const uint16_t *left, int bd) {
   int r;
   (void)above;
   (void)bd;
-  for (r = 0; r < bs; r++) {
-    aom_memset16(dst, left[r], bs);
+  for (r = 0; r < bh; r++) {
+    aom_memset16(dst, left[r], bw);
     dst += stride;
   }
 }
@@ -777,39 +796,42 @@ void aom_highbd_d153_predictor_2x2_c(uint16_t *dst, ptrdiff_t stride,
 
 #if CONFIG_ALT_INTRA
 static INLINE void highbd_paeth_predictor(uint16_t *dst, ptrdiff_t stride,
-                                          int bs, const uint16_t *above,
+                                          int bw, int bh, const uint16_t *above,
                                           const uint16_t *left, int bd) {
   int r, c;
   const uint16_t ytop_left = above[-1];
   (void)bd;
 
-  for (r = 0; r < bs; r++) {
-    for (c = 0; c < bs; c++)
+  for (r = 0; r < bh; r++) {
+    for (c = 0; c < bw; c++)
       dst[c] = paeth_predictor_single(left[r], above[c], ytop_left);
     dst += stride;
   }
 }
 
 static INLINE void highbd_smooth_predictor(uint16_t *dst, ptrdiff_t stride,
-                                           int bs, const uint16_t *above,
+                                           int bw, int bh,
+                                           const uint16_t *above,
                                            const uint16_t *left, int bd) {
-  const uint16_t below_pred = left[bs - 1];   // estimated by bottom-left pixel
-  const uint16_t right_pred = above[bs - 1];  // estimated by top-right pixel
-  const uint8_t *const sm_weights = sm_weight_arrays + bs;
+  const uint16_t below_pred = left[bh - 1];   // estimated by bottom-left pixel
+  const uint16_t right_pred = above[bw - 1];  // estimated by top-right pixel
+  const uint8_t *const sm_weights_w = sm_weight_arrays + bw;
+  const uint8_t *const sm_weights_h = sm_weight_arrays + bh;
   // scale = 2 * 2^sm_weight_log2_scale
   const int log2_scale = 1 + sm_weight_log2_scale;
   const uint16_t scale = (1 << sm_weight_log2_scale);
-  sm_weights_sanity_checks(sm_weights, scale, log2_scale + sizeof(*dst));
+  sm_weights_sanity_checks(sm_weights_w, sm_weights_h, scale,
+                           log2_scale + sizeof(*dst));
   int r;
-  for (r = 0; r < bs; ++r) {
+  for (r = 0; r < bh; ++r) {
     int c;
-    for (c = 0; c < bs; ++c) {
+    for (c = 0; c < bw; ++c) {
       const uint16_t pixels[] = { above[c], below_pred, left[r], right_pred };
-      const uint8_t weights[] = { sm_weights[r], scale - sm_weights[r],
-                                  sm_weights[c], scale - sm_weights[c] };
+      const uint8_t weights[] = { sm_weights_h[r], scale - sm_weights_h[r],
+                                  sm_weights_w[c], scale - sm_weights_w[c] };
       uint32_t this_pred = 0;
       int i;
-      assert(scale >= sm_weights[r] && scale >= sm_weights[c]);
+      assert(scale >= sm_weights_h[r] && scale >= sm_weights_w[c]);
       for (i = 0; i < 4; ++i) {
         this_pred += weights[i] * pixels[i];
       }
@@ -821,19 +843,21 @@ static INLINE void highbd_smooth_predictor(uint16_t *dst, ptrdiff_t stride,
 
 #if CONFIG_SMOOTH_HV
 static INLINE void highbd_smooth_v_predictor(uint16_t *dst, ptrdiff_t stride,
-                                             int bs, const uint16_t *above,
+                                             int bw, int bh,
+                                             const uint16_t *above,
                                              const uint16_t *left, int bd) {
-  const uint16_t below_pred = left[bs - 1];  // estimated by bottom-left pixel
-  const uint8_t *const sm_weights = sm_weight_arrays + bs;
+  const uint16_t below_pred = left[bh - 1];  // estimated by bottom-left pixel
+  const uint8_t *const sm_weights = sm_weight_arrays + bh;
   // scale = 2^sm_weight_log2_scale
   const int log2_scale = sm_weight_log2_scale;
   const uint16_t scale = (1 << sm_weight_log2_scale);
-  sm_weights_sanity_checks(sm_weights, scale, log2_scale + sizeof(*dst));
+  sm_weights_sanity_checks(sm_weights, sm_weights, scale,
+                           log2_scale + sizeof(*dst));
 
   int r;
-  for (r = 0; r < bs; r++) {
+  for (r = 0; r < bh; r++) {
     int c;
-    for (c = 0; c < bs; ++c) {
+    for (c = 0; c < bw; ++c) {
       const uint16_t pixels[] = { above[c], below_pred };
       const uint8_t weights[] = { sm_weights[r], scale - sm_weights[r] };
       uint32_t this_pred = 0;
@@ -849,19 +873,21 @@ static INLINE void highbd_smooth_v_predictor(uint16_t *dst, ptrdiff_t stride,
 }
 
 static INLINE void highbd_smooth_h_predictor(uint16_t *dst, ptrdiff_t stride,
-                                             int bs, const uint16_t *above,
+                                             int bw, int bh,
+                                             const uint16_t *above,
                                              const uint16_t *left, int bd) {
-  const uint16_t right_pred = above[bs - 1];  // estimated by top-right pixel
-  const uint8_t *const sm_weights = sm_weight_arrays + bs;
+  const uint16_t right_pred = above[bw - 1];  // estimated by top-right pixel
+  const uint8_t *const sm_weights = sm_weight_arrays + bw;
   // scale = 2^sm_weight_log2_scale
   const int log2_scale = sm_weight_log2_scale;
   const uint16_t scale = (1 << sm_weight_log2_scale);
-  sm_weights_sanity_checks(sm_weights, scale, log2_scale + sizeof(*dst));
+  sm_weights_sanity_checks(sm_weights, sm_weights, scale,
+                           log2_scale + sizeof(*dst));
 
   int r;
-  for (r = 0; r < bs; r++) {
+  for (r = 0; r < bh; r++) {
     int c;
-    for (c = 0; c < bs; ++c) {
+    for (c = 0; c < bw; ++c) {
       const uint16_t pixels[] = { left[r], right_pred };
       const uint8_t weights[] = { sm_weights[c], scale - sm_weights[c] };
       uint32_t this_pred = 0;
@@ -878,15 +904,15 @@ static INLINE void highbd_smooth_h_predictor(uint16_t *dst, ptrdiff_t stride,
 #endif
 
 #else
-static INLINE void highbd_tm_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
-                                       const uint16_t *above,
+static INLINE void highbd_tm_predictor(uint16_t *dst, ptrdiff_t stride, int bw,
+                                       int bh, const uint16_t *above,
                                        const uint16_t *left, int bd) {
   int r, c;
   int ytop_left = above[-1];
   (void)bd;
 
-  for (r = 0; r < bs; r++) {
-    for (c = 0; c < bs; c++)
+  for (r = 0; r < bh; r++) {
+    for (c = 0; c < bw; c++)
       dst[c] = clip_pixel_highbd(left[r] + above[c] - ytop_left, bd);
     dst += stride;
   }
@@ -894,66 +920,71 @@ static INLINE void highbd_tm_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
 #endif  // CONFIG_ALT_INTRA
 
 static INLINE void highbd_dc_128_predictor(uint16_t *dst, ptrdiff_t stride,
-                                           int bs, const uint16_t *above,
+                                           int bw, int bh,
+                                           const uint16_t *above,
                                            const uint16_t *left, int bd) {
   int r;
   (void)above;
   (void)left;
 
-  for (r = 0; r < bs; r++) {
-    aom_memset16(dst, 128 << (bd - 8), bs);
+  for (r = 0; r < bh; r++) {
+    aom_memset16(dst, 128 << (bd - 8), bw);
     dst += stride;
   }
 }
 
 static INLINE void highbd_dc_left_predictor(uint16_t *dst, ptrdiff_t stride,
-                                            int bs, const uint16_t *above,
+                                            int bw, int bh,
+                                            const uint16_t *above,
                                             const uint16_t *left, int bd) {
   int i, r, expected_dc, sum = 0;
   (void)above;
   (void)bd;
 
-  for (i = 0; i < bs; i++) sum += left[i];
-  expected_dc = (sum + (bs >> 1)) / bs;
+  for (i = 0; i < bh; i++) sum += left[i];
+  expected_dc = (sum + (bh >> 1)) / bh;
 
-  for (r = 0; r < bs; r++) {
-    aom_memset16(dst, expected_dc, bs);
+  for (r = 0; r < bh; r++) {
+    aom_memset16(dst, expected_dc, bw);
     dst += stride;
   }
 }
 
 static INLINE void highbd_dc_top_predictor(uint16_t *dst, ptrdiff_t stride,
-                                           int bs, const uint16_t *above,
+                                           int bw, int bh,
+                                           const uint16_t *above,
                                            const uint16_t *left, int bd) {
   int i, r, expected_dc, sum = 0;
   (void)left;
   (void)bd;
 
-  for (i = 0; i < bs; i++) sum += above[i];
-  expected_dc = (sum + (bs >> 1)) / bs;
+  for (i = 0; i < bw; i++) sum += above[i];
+  expected_dc = (sum + (bw >> 1)) / bw;
 
-  for (r = 0; r < bs; r++) {
-    aom_memset16(dst, expected_dc, bs);
+  for (r = 0; r < bh; r++) {
+    aom_memset16(dst, expected_dc, bw);
     dst += stride;
   }
 }
 
-static INLINE void highbd_dc_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
-                                       const uint16_t *above,
+static INLINE void highbd_dc_predictor(uint16_t *dst, ptrdiff_t stride, int bw,
+                                       int bh, const uint16_t *above,
                                        const uint16_t *left, int bd) {
   int i, r, expected_dc, sum = 0;
-  const int count = 2 * bs;
+  const int count = bw + bh;
   (void)bd;
 
-  for (i = 0; i < bs; i++) {
+  for (i = 0; i < bw; i++) {
     sum += above[i];
+  }
+  for (i = 0; i < bh; i++) {
     sum += left[i];
   }
 
   expected_dc = (sum + (count >> 1)) / count;
 
-  for (r = 0; r < bs; r++) {
-    aom_memset16(dst, expected_dc, bs);
+  for (r = 0; r < bh; r++) {
+    aom_memset16(dst, expected_dc, bw);
     dst += stride;
   }
 }
@@ -962,99 +993,121 @@ static INLINE void highbd_dc_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
 // This serves as a wrapper function, so that all the prediction functions
 // can be unified and accessed as a pointer array. Note that the boundary
 // above and left are not necessarily used all the time.
-#define intra_pred_sized(type, size)                        \
-  void aom_##type##_predictor_##size##x##size##_c(          \
-      uint8_t *dst, ptrdiff_t stride, const uint8_t *above, \
-      const uint8_t *left) {                                \
-    type##_predictor(dst, stride, size, above, left);       \
+#define intra_pred_sized(type, width, height)                  \
+  void aom_##type##_predictor_##width##x##height##_c(          \
+      uint8_t *dst, ptrdiff_t stride, const uint8_t *above,    \
+      const uint8_t *left) {                                   \
+    type##_predictor(dst, stride, width, height, above, left); \
   }
 
 #if CONFIG_HIGHBITDEPTH
-#define intra_pred_highbd_sized(type, size)                        \
-  void aom_highbd_##type##_predictor_##size##x##size##_c(          \
-      uint16_t *dst, ptrdiff_t stride, const uint16_t *above,      \
-      const uint16_t *left, int bd) {                              \
-    highbd_##type##_predictor(dst, stride, size, above, left, bd); \
+#define intra_pred_highbd_sized(type, width, height)                        \
+  void aom_highbd_##type##_predictor_##width##x##height##_c(                \
+      uint16_t *dst, ptrdiff_t stride, const uint16_t *above,               \
+      const uint16_t *left, int bd) {                                       \
+    highbd_##type##_predictor(dst, stride, width, height, above, left, bd); \
   }
 
 /* clang-format off */
 #if CONFIG_TX64X64
-#define intra_pred_allsizes(type) \
-  intra_pred_sized(type, 2) \
-  intra_pred_sized(type, 4) \
-  intra_pred_sized(type, 8) \
-  intra_pred_sized(type, 16) \
-  intra_pred_sized(type, 32) \
-  intra_pred_sized(type, 64) \
-  intra_pred_highbd_sized(type, 2) \
-  intra_pred_highbd_sized(type, 4) \
-  intra_pred_highbd_sized(type, 8) \
-  intra_pred_highbd_sized(type, 16) \
-  intra_pred_highbd_sized(type, 32) \
-  intra_pred_highbd_sized(type, 64)
-
+#define intra_pred_rectangular(type) \
+  intra_pred_sized(type, 4, 8) \
+  intra_pred_sized(type, 8, 4) \
+  intra_pred_sized(type, 8, 16) \
+  intra_pred_sized(type, 16, 8) \
+  intra_pred_sized(type, 16, 32) \
+  intra_pred_sized(type, 32, 16) \
+  intra_pred_highbd_sized(type, 4, 8) \
+  intra_pred_highbd_sized(type, 8, 4) \
+  intra_pred_highbd_sized(type, 8, 16) \
+  intra_pred_highbd_sized(type, 16, 8) \
+  intra_pred_highbd_sized(type, 16, 32) \
+  intra_pred_highbd_sized(type, 32, 16)
 #define intra_pred_above_4x4(type) \
-  intra_pred_sized(type, 8) \
-  intra_pred_sized(type, 16) \
-  intra_pred_sized(type, 32) \
-  intra_pred_sized(type, 64) \
-  intra_pred_highbd_sized(type, 4) \
-  intra_pred_highbd_sized(type, 8) \
-  intra_pred_highbd_sized(type, 16) \
-  intra_pred_highbd_sized(type, 32) \
-  intra_pred_highbd_sized(type, 64)
-#else  // CONFIG_TX64X64
+  intra_pred_sized(type, 8, 8) \
+  intra_pred_sized(type, 16, 16) \
+  intra_pred_sized(type, 32, 32) \
+  intra_pred_sized(type, 64, 64) \
+  intra_pred_highbd_sized(type, 4, 4) \
+  intra_pred_highbd_sized(type, 8, 8) \
+  intra_pred_highbd_sized(type, 16, 16) \
+  intra_pred_highbd_sized(type, 32, 32) \
+  intra_pred_highbd_sized(type, 64, 64) \
+  intra_pred_rectangular(type)
 #define intra_pred_allsizes(type) \
-  intra_pred_sized(type, 2) \
-  intra_pred_sized(type, 4) \
-  intra_pred_sized(type, 8) \
-  intra_pred_sized(type, 16) \
-  intra_pred_sized(type, 32) \
-  intra_pred_highbd_sized(type, 2) \
-  intra_pred_highbd_sized(type, 4) \
-  intra_pred_highbd_sized(type, 8) \
-  intra_pred_highbd_sized(type, 16) \
-  intra_pred_highbd_sized(type, 32)
-
+  intra_pred_sized(type, 2, 2) \
+  intra_pred_sized(type, 4, 4) \
+  intra_pred_highbd_sized(type, 2, 2) \
+  intra_pred_above_4x4(type)
+#else  // CONFIG_TX64X64
+#define intra_pred_rectangular(type) \
+  intra_pred_sized(type, 4, 8) \
+  intra_pred_sized(type, 8, 4) \
+  intra_pred_sized(type, 8, 16) \
+  intra_pred_sized(type, 16, 8) \
+  intra_pred_sized(type, 16, 32) \
+  intra_pred_sized(type, 32, 16) \
+  intra_pred_highbd_sized(type, 4, 8) \
+  intra_pred_highbd_sized(type, 8, 4) \
+  intra_pred_highbd_sized(type, 8, 16) \
+  intra_pred_highbd_sized(type, 16, 8) \
+  intra_pred_highbd_sized(type, 16, 32) \
+  intra_pred_highbd_sized(type, 32, 16)
 #define intra_pred_above_4x4(type) \
-  intra_pred_sized(type, 8) \
-  intra_pred_sized(type, 16) \
-  intra_pred_sized(type, 32) \
-  intra_pred_highbd_sized(type, 4) \
-  intra_pred_highbd_sized(type, 8) \
-  intra_pred_highbd_sized(type, 16) \
-  intra_pred_highbd_sized(type, 32)
+  intra_pred_sized(type, 8, 8) \
+  intra_pred_sized(type, 16, 16) \
+  intra_pred_sized(type, 32, 32) \
+  intra_pred_highbd_sized(type, 4, 4) \
+  intra_pred_highbd_sized(type, 8, 8) \
+  intra_pred_highbd_sized(type, 16, 16) \
+  intra_pred_highbd_sized(type, 32, 32) \
+  intra_pred_rectangular(type)
+#define intra_pred_allsizes(type) \
+  intra_pred_sized(type, 2, 2) \
+  intra_pred_sized(type, 4, 4) \
+  intra_pred_highbd_sized(type, 2, 2) \
+  intra_pred_above_4x4(type)
 #endif  // CONFIG_TX64X64
 
 #else
 
 #if CONFIG_TX64X64
-#define intra_pred_allsizes(type) \
-  intra_pred_sized(type, 2) \
-  intra_pred_sized(type, 4) \
-  intra_pred_sized(type, 8) \
-  intra_pred_sized(type, 16) \
-  intra_pred_sized(type, 32) \
-  intra_pred_sized(type, 64)
-
+#define intra_pred_rectangular(type) \
+  intra_pred_sized(type, 4, 8) \
+  intra_pred_sized(type, 8, 4) \
+  intra_pred_sized(type, 8, 16) \
+  intra_pred_sized(type, 16, 8) \
+  intra_pred_sized(type, 16, 32) \
+  intra_pred_sized(type, 32, 16)
 #define intra_pred_above_4x4(type) \
-  intra_pred_sized(type, 8) \
-  intra_pred_sized(type, 16) \
-  intra_pred_sized(type, 32) \
-  intra_pred_sized(type, 64)
-#else  // CONFIG_TX64X64
+  intra_pred_sized(type, 8, 8) \
+  intra_pred_sized(type, 16, 16) \
+  intra_pred_sized(type, 32, 32) \
+  intra_pred_sized(type, 64, 64) \
+  intra_pred_rectangular(type)
 #define intra_pred_allsizes(type) \
-  intra_pred_sized(type, 2) \
-  intra_pred_sized(type, 4) \
-  intra_pred_sized(type, 8) \
-  intra_pred_sized(type, 16) \
-  intra_pred_sized(type, 32)
-
+  intra_pred_sized(type, 2, 2) \
+  intra_pred_sized(type, 4, 4) \
+  intra_pred_above_4x4(type)
+#else  // CONFIG_TX64X64
+#define intra_pred_rectangular(type) \
+  intra_pred_sized(type, 4, 8) \
+  intra_pred_sized(type, 8, 4) \
+  intra_pred_sized(type, 8, 16) \
+  intra_pred_sized(type, 16, 8) \
+  intra_pred_sized(type, 16, 32) \
+  intra_pred_sized(type, 32, 16)
 #define intra_pred_above_4x4(type) \
-  intra_pred_sized(type, 8) \
-  intra_pred_sized(type, 16) \
-  intra_pred_sized(type, 32)
+  intra_pred_sized(type, 8, 8) \
+  intra_pred_sized(type, 16, 16) \
+  intra_pred_sized(type, 32, 32) \
+  intra_pred_rectangular(type)
+#define intra_pred_allsizes(type) \
+  intra_pred_sized(type, 2, 2) \
+  intra_pred_sized(type, 4, 4) \
+  intra_pred_above_4x4(type)
 #endif  // CONFIG_TX64X64
+
 #endif  // CONFIG_HIGHBITDEPTH
 
 intra_pred_allsizes(d207e)
diff --git a/third_party/aom/aom_dsp/inv_txfm.c b/third_party/aom/aom_dsp/inv_txfm.c
index 6e7d8c928..398eb0a12 100644
--- a/third_party/aom/aom_dsp/inv_txfm.c
+++ b/third_party/aom/aom_dsp/inv_txfm.c
@@ -14,6 +14,9 @@
 
 #include "./aom_dsp_rtcd.h"
 #include "aom_dsp/inv_txfm.h"
+#if CONFIG_DAALA_DCT4 || CONFIG_DAALA_DCT8
+#include "av1/common/daala_tx.h"
+#endif
 
 void aom_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
@@ -93,6 +96,18 @@ void aom_iwht4x4_1_add_c(const tran_low_t *in, uint8_t *dest, int dest_stride) {
   }
 }
 
+#if CONFIG_DAALA_DCT4
+void aom_idct4_c(const tran_low_t *input, tran_low_t *output) {
+  int i;
+  od_coeff x[4];
+  od_coeff y[4];
+  for (i = 0; i < 4; i++) y[i] = input[i];
+  od_bin_idct4(x, 1, y);
+  for (i = 0; i < 4; i++) output[i] = (tran_low_t)x[i];
+}
+
+#else
+
 void aom_idct4_c(const tran_low_t *input, tran_low_t *output) {
   tran_low_t step[4];
   tran_high_t temp1, temp2;
@@ -112,6 +127,7 @@ void aom_idct4_c(const tran_low_t *input, tran_low_t *output) {
   output[2] = WRAPLOW(step[1] - step[2]);
   output[3] = WRAPLOW(step[0] - step[3]);
 }
+#endif
 
 void aom_idct4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   tran_low_t out[4 * 4];
@@ -156,6 +172,18 @@ void aom_idct4x4_1_add_c(const tran_low_t *input, uint8_t *dest,
   }
 }
 
+#if CONFIG_DAALA_DCT8
+void aom_idct8_c(const tran_low_t *input, tran_low_t *output) {
+  int i;
+  od_coeff x[8];
+  od_coeff y[8];
+  for (i = 0; i < 8; i++) y[i] = (od_coeff)input[i];
+  od_bin_idct8(x, 1, y);
+  for (i = 0; i < 8; i++) output[i] = (tran_low_t)x[i];
+}
+
+#else
+
 void aom_idct8_c(const tran_low_t *input, tran_low_t *output) {
   tran_low_t step1[8], step2[8];
   tran_high_t temp1, temp2;
@@ -209,6 +237,7 @@ void aom_idct8_c(const tran_low_t *input, tran_low_t *output) {
   output[6] = WRAPLOW(step1[1] - step1[6]);
   output[7] = WRAPLOW(step1[0] - step1[7]);
 }
+#endif
 
 void aom_idct8x8_64_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   tran_low_t out[8 * 8];
@@ -284,6 +313,18 @@ void aom_iadst4_c(const tran_low_t *input, tran_low_t *output) {
   output[3] = WRAPLOW(dct_const_round_shift(s0 + s1 - s3));
 }
 
+#if CONFIG_DAALA_DCT8
+void aom_iadst8_c(const tran_low_t *input, tran_low_t *output) {
+  int i;
+  od_coeff x[8];
+  od_coeff y[8];
+  for (i = 0; i < 8; i++) y[i] = (od_coeff)input[i];
+  od_bin_idst8(x, 1, y);
+  for (i = 0; i < 8; i++) output[i] = (tran_low_t)x[i];
+}
+
+#else
+
 void aom_iadst8_c(const tran_low_t *input, tran_low_t *output) {
   int s0, s1, s2, s3, s4, s5, s6, s7;
 
@@ -361,6 +402,8 @@ void aom_iadst8_c(const tran_low_t *input, tran_low_t *output) {
   output[7] = WRAPLOW(-x1);
 }
 
+#endif
+
 void aom_idct8x8_12_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   tran_low_t out[8 * 8] = { 0 };
   tran_low_t *outptr = out;
@@ -1179,6 +1222,109 @@ void aom_idct32_c(const tran_low_t *input, tran_low_t *output) {
   output[31] = WRAPLOW(step1[0] - step1[31]);
 }
 
+#if CONFIG_MRC_TX
+void aom_imrc32x32_1024_add_c(const tran_low_t *input, uint8_t *dest,
+                              int stride, int *mask) {
+  tran_low_t out[32 * 32];
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[32], temp_out[32];
+
+  // Rows
+  for (i = 0; i < 32; ++i) {
+    int16_t zero_coeff[16];
+    for (j = 0; j < 16; ++j) zero_coeff[j] = input[2 * j] | input[2 * j + 1];
+    for (j = 0; j < 8; ++j)
+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+    for (j = 0; j < 4; ++j)
+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+    for (j = 0; j < 2; ++j)
+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+
+    if (zero_coeff[0] | zero_coeff[1])
+      aom_idct32_c(input, outptr);
+    else
+      memset(outptr, 0, sizeof(tran_low_t) * 32);
+    input += 32;
+    outptr += 32;
+  }
+
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    for (j = 0; j < 32; ++j) temp_in[j] = out[j * 32 + i];
+    aom_idct32_c(temp_in, temp_out);
+    for (j = 0; j < 32; ++j) {
+      // Only add the coefficient if the mask value is 1
+      int mask_val = mask[j * 32 + i];
+      dest[j * stride + i] =
+          mask_val ? clip_pixel_add(dest[j * stride + i],
+                                    ROUND_POWER_OF_TWO(temp_out[j], 6))
+                   : dest[j * stride + i];
+    }
+  }
+}
+
+void aom_imrc32x32_135_add_c(const tran_low_t *input, uint8_t *dest, int stride,
+                             int *mask) {
+  tran_low_t out[32 * 32] = { 0 };
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[32], temp_out[32];
+
+  // Rows
+  // only upper-left 16x16 has non-zero coeff
+  for (i = 0; i < 16; ++i) {
+    aom_idct32_c(input, outptr);
+    input += 32;
+    outptr += 32;
+  }
+
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    for (j = 0; j < 32; ++j) temp_in[j] = out[j * 32 + i];
+    aom_idct32_c(temp_in, temp_out);
+    for (j = 0; j < 32; ++j) {
+      // Only add the coefficient if the mask value is 1
+      int mask_val = mask[j * 32 + i];
+      dest[j * stride + i] =
+          mask_val ? clip_pixel_add(dest[j * stride + i],
+                                    ROUND_POWER_OF_TWO(temp_out[j], 6))
+                   : dest[j * stride + i];
+    }
+  }
+}
+
+void aom_imrc32x32_34_add_c(const tran_low_t *input, uint8_t *dest, int stride,
+                            int *mask) {
+  tran_low_t out[32 * 32] = { 0 };
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[32], temp_out[32];
+
+  // Rows
+  // only upper-left 8x8 has non-zero coeff
+  for (i = 0; i < 8; ++i) {
+    aom_idct32_c(input, outptr);
+    input += 32;
+    outptr += 32;
+  }
+
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    for (j = 0; j < 32; ++j) temp_in[j] = out[j * 32 + i];
+    aom_idct32_c(temp_in, temp_out);
+    for (j = 0; j < 32; ++j) {
+      // Only add the coefficient if the mask value is 1
+      int mask_val = mask[j * 32 + i];
+      dest[j * stride + i] =
+          mask_val ? clip_pixel_add(dest[j * stride + i],
+                                    ROUND_POWER_OF_TWO(temp_out[j], 6))
+                   : dest[j * stride + i];
+    }
+  }
+}
+#endif  // CONFIG_MRC_TX
+
 void aom_idct32x32_1024_add_c(const tran_low_t *input, uint8_t *dest,
                               int stride) {
   tran_low_t out[32 * 32];
@@ -1283,7 +1429,6 @@ void aom_idct32x32_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   }
 }
 
-#if CONFIG_HIGHBITDEPTH
 void aom_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest8,
                                  int stride, int bd) {
   /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
@@ -1374,936 +1519,3 @@ void aom_highbd_iwht4x4_1_add_c(const tran_low_t *in, uint8_t *dest8,
     dest++;
   }
 }
-
-void aom_highbd_idct4_c(const tran_low_t *input, tran_low_t *output, int bd) {
-  tran_low_t step[4];
-  tran_high_t temp1, temp2;
-  (void)bd;
-  // stage 1
-  temp1 = (input[0] + input[2]) * cospi_16_64;
-  temp2 = (input[0] - input[2]) * cospi_16_64;
-  step[0] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step[1] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
-  temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
-  step[2] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step[3] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  // stage 2
-  output[0] = HIGHBD_WRAPLOW(step[0] + step[3], bd);
-  output[1] = HIGHBD_WRAPLOW(step[1] + step[2], bd);
-  output[2] = HIGHBD_WRAPLOW(step[1] - step[2], bd);
-  output[3] = HIGHBD_WRAPLOW(step[0] - step[3], bd);
-}
-
-void aom_highbd_idct4x4_16_add_c(const tran_low_t *input, uint8_t *dest8,
-                                 int stride, int bd) {
-  tran_low_t out[4 * 4];
-  tran_low_t *outptr = out;
-  int i, j;
-  tran_low_t temp_in[4], temp_out[4];
-  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
-
-  // Rows
-  for (i = 0; i < 4; ++i) {
-    aom_highbd_idct4_c(input, outptr, bd);
-    input += 4;
-    outptr += 4;
-  }
-
-  // Columns
-  for (i = 0; i < 4; ++i) {
-    for (j = 0; j < 4; ++j) temp_in[j] = out[j * 4 + i];
-    aom_highbd_idct4_c(temp_in, temp_out, bd);
-    for (j = 0; j < 4; ++j) {
-      dest[j * stride + i] = highbd_clip_pixel_add(
-          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd);
-    }
-  }
-}
-
-void aom_highbd_idct4x4_1_add_c(const tran_low_t *input, uint8_t *dest8,
-                                int dest_stride, int bd) {
-  int i;
-  tran_high_t a1;
-  tran_low_t out =
-      HIGHBD_WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64), bd);
-  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
-
-  out = HIGHBD_WRAPLOW(dct_const_round_shift(out * cospi_16_64), bd);
-  a1 = ROUND_POWER_OF_TWO(out, 4);
-
-  for (i = 0; i < 4; i++) {
-    dest[0] = highbd_clip_pixel_add(dest[0], a1, bd);
-    dest[1] = highbd_clip_pixel_add(dest[1], a1, bd);
-    dest[2] = highbd_clip_pixel_add(dest[2], a1, bd);
-    dest[3] = highbd_clip_pixel_add(dest[3], a1, bd);
-    dest += dest_stride;
-  }
-}
-
-void aom_highbd_idct8_c(const tran_low_t *input, tran_low_t *output, int bd) {
-  tran_low_t step1[8], step2[8];
-  tran_high_t temp1, temp2;
-  // stage 1
-  step1[0] = input[0];
-  step1[2] = input[4];
-  step1[1] = input[2];
-  step1[3] = input[6];
-  temp1 = input[1] * cospi_28_64 - input[7] * cospi_4_64;
-  temp2 = input[1] * cospi_4_64 + input[7] * cospi_28_64;
-  step1[4] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[7] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = input[5] * cospi_12_64 - input[3] * cospi_20_64;
-  temp2 = input[5] * cospi_20_64 + input[3] * cospi_12_64;
-  step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  // stage 2 & stage 3 - even half
-  aom_highbd_idct4_c(step1, step1, bd);
-
-  // stage 2 - odd half
-  step2[4] = HIGHBD_WRAPLOW(step1[4] + step1[5], bd);
-  step2[5] = HIGHBD_WRAPLOW(step1[4] - step1[5], bd);
-  step2[6] = HIGHBD_WRAPLOW(-step1[6] + step1[7], bd);
-  step2[7] = HIGHBD_WRAPLOW(step1[6] + step1[7], bd);
-
-  // stage 3 - odd half
-  step1[4] = step2[4];
-  temp1 = (step2[6] - step2[5]) * cospi_16_64;
-  temp2 = (step2[5] + step2[6]) * cospi_16_64;
-  step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step1[7] = step2[7];
-
-  // stage 4
-  output[0] = HIGHBD_WRAPLOW(step1[0] + step1[7], bd);
-  output[1] = HIGHBD_WRAPLOW(step1[1] + step1[6], bd);
-  output[2] = HIGHBD_WRAPLOW(step1[2] + step1[5], bd);
-  output[3] = HIGHBD_WRAPLOW(step1[3] + step1[4], bd);
-  output[4] = HIGHBD_WRAPLOW(step1[3] - step1[4], bd);
-  output[5] = HIGHBD_WRAPLOW(step1[2] - step1[5], bd);
-  output[6] = HIGHBD_WRAPLOW(step1[1] - step1[6], bd);
-  output[7] = HIGHBD_WRAPLOW(step1[0] - step1[7], bd);
-}
-
-void aom_highbd_iadst4_c(const tran_low_t *input, tran_low_t *output, int bd) {
-  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
-
-  tran_low_t x0 = input[0];
-  tran_low_t x1 = input[1];
-  tran_low_t x2 = input[2];
-  tran_low_t x3 = input[3];
-  (void)bd;
-
-  if (!(x0 | x1 | x2 | x3)) {
-    memset(output, 0, 4 * sizeof(*output));
-    return;
-  }
-
-  s0 = sinpi_1_9 * x0;
-  s1 = sinpi_2_9 * x0;
-  s2 = sinpi_3_9 * x1;
-  s3 = sinpi_4_9 * x2;
-  s4 = sinpi_1_9 * x2;
-  s5 = sinpi_2_9 * x3;
-  s6 = sinpi_4_9 * x3;
-  s7 = (tran_high_t)HIGHBD_WRAPLOW(x0 - x2 + x3, bd);
-
-  s0 = s0 + s3 + s5;
-  s1 = s1 - s4 - s6;
-  s3 = s2;
-  s2 = sinpi_3_9 * s7;
-
-  // 1-D transform scaling factor is sqrt(2).
-  // The overall dynamic range is 14b (input) + 14b (multiplication scaling)
-  // + 1b (addition) = 29b.
-  // Hence the output bit depth is 15b.
-  output[0] = HIGHBD_WRAPLOW(dct_const_round_shift(s0 + s3), bd);
-  output[1] = HIGHBD_WRAPLOW(dct_const_round_shift(s1 + s3), bd);
-  output[2] = HIGHBD_WRAPLOW(dct_const_round_shift(s2), bd);
-  output[3] = HIGHBD_WRAPLOW(dct_const_round_shift(s0 + s1 - s3), bd);
-}
-
-void aom_highbd_iadst8_c(const tran_low_t *input, tran_low_t *output, int bd) {
-  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
-
-  tran_low_t x0 = input[7];
-  tran_low_t x1 = input[0];
-  tran_low_t x2 = input[5];
-  tran_low_t x3 = input[2];
-  tran_low_t x4 = input[3];
-  tran_low_t x5 = input[4];
-  tran_low_t x6 = input[1];
-  tran_low_t x7 = input[6];
-  (void)bd;
-
-  if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7)) {
-    memset(output, 0, 8 * sizeof(*output));
-    return;
-  }
-
-  // stage 1
-  s0 = cospi_2_64 * x0 + cospi_30_64 * x1;
-  s1 = cospi_30_64 * x0 - cospi_2_64 * x1;
-  s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
-  s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
-  s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
-  s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
-  s6 = cospi_26_64 * x6 + cospi_6_64 * x7;
-  s7 = cospi_6_64 * x6 - cospi_26_64 * x7;
-
-  x0 = HIGHBD_WRAPLOW(dct_const_round_shift(s0 + s4), bd);
-  x1 = HIGHBD_WRAPLOW(dct_const_round_shift(s1 + s5), bd);
-  x2 = HIGHBD_WRAPLOW(dct_const_round_shift(s2 + s6), bd);
-  x3 = HIGHBD_WRAPLOW(dct_const_round_shift(s3 + s7), bd);
-  x4 = HIGHBD_WRAPLOW(dct_const_round_shift(s0 - s4), bd);
-  x5 = HIGHBD_WRAPLOW(dct_const_round_shift(s1 - s5), bd);
-  x6 = HIGHBD_WRAPLOW(dct_const_round_shift(s2 - s6), bd);
-  x7 = HIGHBD_WRAPLOW(dct_const_round_shift(s3 - s7), bd);
-
-  // stage 2
-  s0 = x0;
-  s1 = x1;
-  s2 = x2;
-  s3 = x3;
-  s4 = cospi_8_64 * x4 + cospi_24_64 * x5;
-  s5 = cospi_24_64 * x4 - cospi_8_64 * x5;
-  s6 = -cospi_24_64 * x6 + cospi_8_64 * x7;
-  s7 = cospi_8_64 * x6 + cospi_24_64 * x7;
-
-  x0 = HIGHBD_WRAPLOW(s0 + s2, bd);
-  x1 = HIGHBD_WRAPLOW(s1 + s3, bd);
-  x2 = HIGHBD_WRAPLOW(s0 - s2, bd);
-  x3 = HIGHBD_WRAPLOW(s1 - s3, bd);
-  x4 = HIGHBD_WRAPLOW(dct_const_round_shift(s4 + s6), bd);
-  x5 = HIGHBD_WRAPLOW(dct_const_round_shift(s5 + s7), bd);
-  x6 = HIGHBD_WRAPLOW(dct_const_round_shift(s4 - s6), bd);
-  x7 = HIGHBD_WRAPLOW(dct_const_round_shift(s5 - s7), bd);
-
-  // stage 3
-  s2 = cospi_16_64 * (x2 + x3);
-  s3 = cospi_16_64 * (x2 - x3);
-  s6 = cospi_16_64 * (x6 + x7);
-  s7 = cospi_16_64 * (x6 - x7);
-
-  x2 = HIGHBD_WRAPLOW(dct_const_round_shift(s2), bd);
-  x3 = HIGHBD_WRAPLOW(dct_const_round_shift(s3), bd);
-  x6 = HIGHBD_WRAPLOW(dct_const_round_shift(s6), bd);
-  x7 = HIGHBD_WRAPLOW(dct_const_round_shift(s7), bd);
-
-  output[0] = HIGHBD_WRAPLOW(x0, bd);
-  output[1] = HIGHBD_WRAPLOW(-x4, bd);
-  output[2] = HIGHBD_WRAPLOW(x6, bd);
-  output[3] = HIGHBD_WRAPLOW(-x2, bd);
-  output[4] = HIGHBD_WRAPLOW(x3, bd);
-  output[5] = HIGHBD_WRAPLOW(-x7, bd);
-  output[6] = HIGHBD_WRAPLOW(x5, bd);
-  output[7] = HIGHBD_WRAPLOW(-x1, bd);
-}
-
-void aom_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd) {
-  tran_low_t step1[16], step2[16];
-  tran_high_t temp1, temp2;
-  (void)bd;
-
-  // stage 1
-  step1[0] = input[0 / 2];
-  step1[1] = input[16 / 2];
-  step1[2] = input[8 / 2];
-  step1[3] = input[24 / 2];
-  step1[4] = input[4 / 2];
-  step1[5] = input[20 / 2];
-  step1[6] = input[12 / 2];
-  step1[7] = input[28 / 2];
-  step1[8] = input[2 / 2];
-  step1[9] = input[18 / 2];
-  step1[10] = input[10 / 2];
-  step1[11] = input[26 / 2];
-  step1[12] = input[6 / 2];
-  step1[13] = input[22 / 2];
-  step1[14] = input[14 / 2];
-  step1[15] = input[30 / 2];
-
-  // stage 2
-  step2[0] = step1[0];
-  step2[1] = step1[1];
-  step2[2] = step1[2];
-  step2[3] = step1[3];
-  step2[4] = step1[4];
-  step2[5] = step1[5];
-  step2[6] = step1[6];
-  step2[7] = step1[7];
-
-  temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64;
-  temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64;
-  step2[8] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[15] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64;
-  temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64;
-  step2[9] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[14] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64;
-  temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64;
-  step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64;
-  temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64;
-  step2[11] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[12] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  // stage 3
-  step1[0] = step2[0];
-  step1[1] = step2[1];
-  step1[2] = step2[2];
-  step1[3] = step2[3];
-
-  temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64;
-  temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64;
-  step1[4] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[7] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64;
-  temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64;
-  step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  step1[8] = HIGHBD_WRAPLOW(step2[8] + step2[9], bd);
-  step1[9] = HIGHBD_WRAPLOW(step2[8] - step2[9], bd);
-  step1[10] = HIGHBD_WRAPLOW(-step2[10] + step2[11], bd);
-  step1[11] = HIGHBD_WRAPLOW(step2[10] + step2[11], bd);
-  step1[12] = HIGHBD_WRAPLOW(step2[12] + step2[13], bd);
-  step1[13] = HIGHBD_WRAPLOW(step2[12] - step2[13], bd);
-  step1[14] = HIGHBD_WRAPLOW(-step2[14] + step2[15], bd);
-  step1[15] = HIGHBD_WRAPLOW(step2[14] + step2[15], bd);
-
-  // stage 4
-  temp1 = (step1[0] + step1[1]) * cospi_16_64;
-  temp2 = (step1[0] - step1[1]) * cospi_16_64;
-  step2[0] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[1] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
-  temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64;
-  step2[2] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[3] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step2[4] = HIGHBD_WRAPLOW(step1[4] + step1[5], bd);
-  step2[5] = HIGHBD_WRAPLOW(step1[4] - step1[5], bd);
-  step2[6] = HIGHBD_WRAPLOW(-step1[6] + step1[7], bd);
-  step2[7] = HIGHBD_WRAPLOW(step1[6] + step1[7], bd);
-
-  step2[8] = step1[8];
-  step2[15] = step1[15];
-  temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64;
-  temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64;
-  step2[9] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[14] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64;
-  temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64;
-  step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step2[11] = step1[11];
-  step2[12] = step1[12];
-
-  // stage 5
-  step1[0] = HIGHBD_WRAPLOW(step2[0] + step2[3], bd);
-  step1[1] = HIGHBD_WRAPLOW(step2[1] + step2[2], bd);
-  step1[2] = HIGHBD_WRAPLOW(step2[1] - step2[2], bd);
-  step1[3] = HIGHBD_WRAPLOW(step2[0] - step2[3], bd);
-  step1[4] = step2[4];
-  temp1 = (step2[6] - step2[5]) * cospi_16_64;
-  temp2 = (step2[5] + step2[6]) * cospi_16_64;
-  step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step1[7] = step2[7];
-
-  step1[8] = HIGHBD_WRAPLOW(step2[8] + step2[11], bd);
-  step1[9] = HIGHBD_WRAPLOW(step2[9] + step2[10], bd);
-  step1[10] = HIGHBD_WRAPLOW(step2[9] - step2[10], bd);
-  step1[11] = HIGHBD_WRAPLOW(step2[8] - step2[11], bd);
-  step1[12] = HIGHBD_WRAPLOW(-step2[12] + step2[15], bd);
-  step1[13] = HIGHBD_WRAPLOW(-step2[13] + step2[14], bd);
-  step1[14] = HIGHBD_WRAPLOW(step2[13] + step2[14], bd);
-  step1[15] = HIGHBD_WRAPLOW(step2[12] + step2[15], bd);
-
-  // stage 6
-  step2[0] = HIGHBD_WRAPLOW(step1[0] + step1[7], bd);
-  step2[1] = HIGHBD_WRAPLOW(step1[1] + step1[6], bd);
-  step2[2] = HIGHBD_WRAPLOW(step1[2] + step1[5], bd);
-  step2[3] = HIGHBD_WRAPLOW(step1[3] + step1[4], bd);
-  step2[4] = HIGHBD_WRAPLOW(step1[3] - step1[4], bd);
-  step2[5] = HIGHBD_WRAPLOW(step1[2] - step1[5], bd);
-  step2[6] = HIGHBD_WRAPLOW(step1[1] - step1[6], bd);
-  step2[7] = HIGHBD_WRAPLOW(step1[0] - step1[7], bd);
-  step2[8] = step1[8];
-  step2[9] = step1[9];
-  temp1 = (-step1[10] + step1[13]) * cospi_16_64;
-  temp2 = (step1[10] + step1[13]) * cospi_16_64;
-  step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = (-step1[11] + step1[12]) * cospi_16_64;
-  temp2 = (step1[11] + step1[12]) * cospi_16_64;
-  step2[11] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[12] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step2[14] = step1[14];
-  step2[15] = step1[15];
-
-  // stage 7
-  output[0] = HIGHBD_WRAPLOW(step2[0] + step2[15], bd);
-  output[1] = HIGHBD_WRAPLOW(step2[1] + step2[14], bd);
-  output[2] = HIGHBD_WRAPLOW(step2[2] + step2[13], bd);
-  output[3] = HIGHBD_WRAPLOW(step2[3] + step2[12], bd);
-  output[4] = HIGHBD_WRAPLOW(step2[4] + step2[11], bd);
-  output[5] = HIGHBD_WRAPLOW(step2[5] + step2[10], bd);
-  output[6] = HIGHBD_WRAPLOW(step2[6] + step2[9], bd);
-  output[7] = HIGHBD_WRAPLOW(step2[7] + step2[8], bd);
-  output[8] = HIGHBD_WRAPLOW(step2[7] - step2[8], bd);
-  output[9] = HIGHBD_WRAPLOW(step2[6] - step2[9], bd);
-  output[10] = HIGHBD_WRAPLOW(step2[5] - step2[10], bd);
-  output[11] = HIGHBD_WRAPLOW(step2[4] - step2[11], bd);
-  output[12] = HIGHBD_WRAPLOW(step2[3] - step2[12], bd);
-  output[13] = HIGHBD_WRAPLOW(step2[2] - step2[13], bd);
-  output[14] = HIGHBD_WRAPLOW(step2[1] - step2[14], bd);
-  output[15] = HIGHBD_WRAPLOW(step2[0] - step2[15], bd);
-}
-
-void aom_highbd_iadst16_c(const tran_low_t *input, tran_low_t *output, int bd) {
-  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7, s8;
-  tran_high_t s9, s10, s11, s12, s13, s14, s15;
-
-  tran_low_t x0 = input[15];
-  tran_low_t x1 = input[0];
-  tran_low_t x2 = input[13];
-  tran_low_t x3 = input[2];
-  tran_low_t x4 = input[11];
-  tran_low_t x5 = input[4];
-  tran_low_t x6 = input[9];
-  tran_low_t x7 = input[6];
-  tran_low_t x8 = input[7];
-  tran_low_t x9 = input[8];
-  tran_low_t x10 = input[5];
-  tran_low_t x11 = input[10];
-  tran_low_t x12 = input[3];
-  tran_low_t x13 = input[12];
-  tran_low_t x14 = input[1];
-  tran_low_t x15 = input[14];
-  (void)bd;
-
-  if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7 | x8 | x9 | x10 | x11 | x12 |
-        x13 | x14 | x15)) {
-    memset(output, 0, 16 * sizeof(*output));
-    return;
-  }
-
-  // stage 1
-  s0 = x0 * cospi_1_64 + x1 * cospi_31_64;
-  s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
-  s2 = x2 * cospi_5_64 + x3 * cospi_27_64;
-  s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
-  s4 = x4 * cospi_9_64 + x5 * cospi_23_64;
-  s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
-  s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
-  s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
-  s8 = x8 * cospi_17_64 + x9 * cospi_15_64;
-  s9 = x8 * cospi_15_64 - x9 * cospi_17_64;
-  s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
-  s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
-  s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
-  s13 = x12 * cospi_7_64 - x13 * cospi_25_64;
-  s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
-  s15 = x14 * cospi_3_64 - x15 * cospi_29_64;
-
-  x0 = HIGHBD_WRAPLOW(dct_const_round_shift(s0 + s8), bd);
-  x1 = HIGHBD_WRAPLOW(dct_const_round_shift(s1 + s9), bd);
-  x2 = HIGHBD_WRAPLOW(dct_const_round_shift(s2 + s10), bd);
-  x3 = HIGHBD_WRAPLOW(dct_const_round_shift(s3 + s11), bd);
-  x4 = HIGHBD_WRAPLOW(dct_const_round_shift(s4 + s12), bd);
-  x5 = HIGHBD_WRAPLOW(dct_const_round_shift(s5 + s13), bd);
-  x6 = HIGHBD_WRAPLOW(dct_const_round_shift(s6 + s14), bd);
-  x7 = HIGHBD_WRAPLOW(dct_const_round_shift(s7 + s15), bd);
-  x8 = HIGHBD_WRAPLOW(dct_const_round_shift(s0 - s8), bd);
-  x9 = HIGHBD_WRAPLOW(dct_const_round_shift(s1 - s9), bd);
-  x10 = HIGHBD_WRAPLOW(dct_const_round_shift(s2 - s10), bd);
-  x11 = HIGHBD_WRAPLOW(dct_const_round_shift(s3 - s11), bd);
-  x12 = HIGHBD_WRAPLOW(dct_const_round_shift(s4 - s12), bd);
-  x13 = HIGHBD_WRAPLOW(dct_const_round_shift(s5 - s13), bd);
-  x14 = HIGHBD_WRAPLOW(dct_const_round_shift(s6 - s14), bd);
-  x15 = HIGHBD_WRAPLOW(dct_const_round_shift(s7 - s15), bd);
-
-  // stage 2
-  s0 = x0;
-  s1 = x1;
-  s2 = x2;
-  s3 = x3;
-  s4 = x4;
-  s5 = x5;
-  s6 = x6;
-  s7 = x7;
-  s8 = x8 * cospi_4_64 + x9 * cospi_28_64;
-  s9 = x8 * cospi_28_64 - x9 * cospi_4_64;
-  s10 = x10 * cospi_20_64 + x11 * cospi_12_64;
-  s11 = x10 * cospi_12_64 - x11 * cospi_20_64;
-  s12 = -x12 * cospi_28_64 + x13 * cospi_4_64;
-  s13 = x12 * cospi_4_64 + x13 * cospi_28_64;
-  s14 = -x14 * cospi_12_64 + x15 * cospi_20_64;
-  s15 = x14 * cospi_20_64 + x15 * cospi_12_64;
-
-  x0 = HIGHBD_WRAPLOW(s0 + s4, bd);
-  x1 = HIGHBD_WRAPLOW(s1 + s5, bd);
-  x2 = HIGHBD_WRAPLOW(s2 + s6, bd);
-  x3 = HIGHBD_WRAPLOW(s3 + s7, bd);
-  x4 = HIGHBD_WRAPLOW(s0 - s4, bd);
-  x5 = HIGHBD_WRAPLOW(s1 - s5, bd);
-  x6 = HIGHBD_WRAPLOW(s2 - s6, bd);
-  x7 = HIGHBD_WRAPLOW(s3 - s7, bd);
-  x8 = HIGHBD_WRAPLOW(dct_const_round_shift(s8 + s12), bd);
-  x9 = HIGHBD_WRAPLOW(dct_const_round_shift(s9 + s13), bd);
-  x10 = HIGHBD_WRAPLOW(dct_const_round_shift(s10 + s14), bd);
-  x11 = HIGHBD_WRAPLOW(dct_const_round_shift(s11 + s15), bd);
-  x12 = HIGHBD_WRAPLOW(dct_const_round_shift(s8 - s12), bd);
-  x13 = HIGHBD_WRAPLOW(dct_const_round_shift(s9 - s13), bd);
-  x14 = HIGHBD_WRAPLOW(dct_const_round_shift(s10 - s14), bd);
-  x15 = HIGHBD_WRAPLOW(dct_const_round_shift(s11 - s15), bd);
-
-  // stage 3
-  s0 = x0;
-  s1 = x1;
-  s2 = x2;
-  s3 = x3;
-  s4 = x4 * cospi_8_64 + x5 * cospi_24_64;
-  s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
-  s6 = -x6 * cospi_24_64 + x7 * cospi_8_64;
-  s7 = x6 * cospi_8_64 + x7 * cospi_24_64;
-  s8 = x8;
-  s9 = x9;
-  s10 = x10;
-  s11 = x11;
-  s12 = x12 * cospi_8_64 + x13 * cospi_24_64;
-  s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
-  s14 = -x14 * cospi_24_64 + x15 * cospi_8_64;
-  s15 = x14 * cospi_8_64 + x15 * cospi_24_64;
-
-  x0 = HIGHBD_WRAPLOW(s0 + s2, bd);
-  x1 = HIGHBD_WRAPLOW(s1 + s3, bd);
-  x2 = HIGHBD_WRAPLOW(s0 - s2, bd);
-  x3 = HIGHBD_WRAPLOW(s1 - s3, bd);
-  x4 = HIGHBD_WRAPLOW(dct_const_round_shift(s4 + s6), bd);
-  x5 = HIGHBD_WRAPLOW(dct_const_round_shift(s5 + s7), bd);
-  x6 = HIGHBD_WRAPLOW(dct_const_round_shift(s4 - s6), bd);
-  x7 = HIGHBD_WRAPLOW(dct_const_round_shift(s5 - s7), bd);
-  x8 = HIGHBD_WRAPLOW(s8 + s10, bd);
-  x9 = HIGHBD_WRAPLOW(s9 + s11, bd);
-  x10 = HIGHBD_WRAPLOW(s8 - s10, bd);
-  x11 = HIGHBD_WRAPLOW(s9 - s11, bd);
-  x12 = HIGHBD_WRAPLOW(dct_const_round_shift(s12 + s14), bd);
-  x13 = HIGHBD_WRAPLOW(dct_const_round_shift(s13 + s15), bd);
-  x14 = HIGHBD_WRAPLOW(dct_const_round_shift(s12 - s14), bd);
-  x15 = HIGHBD_WRAPLOW(dct_const_round_shift(s13 - s15), bd);
-
-  // stage 4
-  s2 = (-cospi_16_64) * (x2 + x3);
-  s3 = cospi_16_64 * (x2 - x3);
-  s6 = cospi_16_64 * (x6 + x7);
-  s7 = cospi_16_64 * (-x6 + x7);
-  s10 = cospi_16_64 * (x10 + x11);
-  s11 = cospi_16_64 * (-x10 + x11);
-  s14 = (-cospi_16_64) * (x14 + x15);
-  s15 = cospi_16_64 * (x14 - x15);
-
-  x2 = HIGHBD_WRAPLOW(dct_const_round_shift(s2), bd);
-  x3 = HIGHBD_WRAPLOW(dct_const_round_shift(s3), bd);
-  x6 = HIGHBD_WRAPLOW(dct_const_round_shift(s6), bd);
-  x7 = HIGHBD_WRAPLOW(dct_const_round_shift(s7), bd);
-  x10 = HIGHBD_WRAPLOW(dct_const_round_shift(s10), bd);
-  x11 = HIGHBD_WRAPLOW(dct_const_round_shift(s11), bd);
-  x14 = HIGHBD_WRAPLOW(dct_const_round_shift(s14), bd);
-  x15 = HIGHBD_WRAPLOW(dct_const_round_shift(s15), bd);
-
-  output[0] = HIGHBD_WRAPLOW(x0, bd);
-  output[1] = HIGHBD_WRAPLOW(-x8, bd);
-  output[2] = HIGHBD_WRAPLOW(x12, bd);
-  output[3] = HIGHBD_WRAPLOW(-x4, bd);
-  output[4] = HIGHBD_WRAPLOW(x6, bd);
-  output[5] = HIGHBD_WRAPLOW(x14, bd);
-  output[6] = HIGHBD_WRAPLOW(x10, bd);
-  output[7] = HIGHBD_WRAPLOW(x2, bd);
-  output[8] = HIGHBD_WRAPLOW(x3, bd);
-  output[9] = HIGHBD_WRAPLOW(x11, bd);
-  output[10] = HIGHBD_WRAPLOW(x15, bd);
-  output[11] = HIGHBD_WRAPLOW(x7, bd);
-  output[12] = HIGHBD_WRAPLOW(x5, bd);
-  output[13] = HIGHBD_WRAPLOW(-x13, bd);
-  output[14] = HIGHBD_WRAPLOW(x9, bd);
-  output[15] = HIGHBD_WRAPLOW(-x1, bd);
-}
-
-void aom_highbd_idct32_c(const tran_low_t *input, tran_low_t *output, int bd) {
-  tran_low_t step1[32], step2[32];
-  tran_high_t temp1, temp2;
-  (void)bd;
-
-  // stage 1
-  step1[0] = input[0];
-  step1[1] = input[16];
-  step1[2] = input[8];
-  step1[3] = input[24];
-  step1[4] = input[4];
-  step1[5] = input[20];
-  step1[6] = input[12];
-  step1[7] = input[28];
-  step1[8] = input[2];
-  step1[9] = input[18];
-  step1[10] = input[10];
-  step1[11] = input[26];
-  step1[12] = input[6];
-  step1[13] = input[22];
-  step1[14] = input[14];
-  step1[15] = input[30];
-
-  temp1 = input[1] * cospi_31_64 - input[31] * cospi_1_64;
-  temp2 = input[1] * cospi_1_64 + input[31] * cospi_31_64;
-  step1[16] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[31] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = input[17] * cospi_15_64 - input[15] * cospi_17_64;
-  temp2 = input[17] * cospi_17_64 + input[15] * cospi_15_64;
-  step1[17] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[30] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = input[9] * cospi_23_64 - input[23] * cospi_9_64;
-  temp2 = input[9] * cospi_9_64 + input[23] * cospi_23_64;
-  step1[18] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[29] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = input[25] * cospi_7_64 - input[7] * cospi_25_64;
-  temp2 = input[25] * cospi_25_64 + input[7] * cospi_7_64;
-  step1[19] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[28] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = input[5] * cospi_27_64 - input[27] * cospi_5_64;
-  temp2 = input[5] * cospi_5_64 + input[27] * cospi_27_64;
-  step1[20] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[27] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = input[21] * cospi_11_64 - input[11] * cospi_21_64;
-  temp2 = input[21] * cospi_21_64 + input[11] * cospi_11_64;
-  step1[21] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[26] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = input[13] * cospi_19_64 - input[19] * cospi_13_64;
-  temp2 = input[13] * cospi_13_64 + input[19] * cospi_19_64;
-  step1[22] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[25] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = input[29] * cospi_3_64 - input[3] * cospi_29_64;
-  temp2 = input[29] * cospi_29_64 + input[3] * cospi_3_64;
-  step1[23] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[24] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  // stage 2
-  step2[0] = step1[0];
-  step2[1] = step1[1];
-  step2[2] = step1[2];
-  step2[3] = step1[3];
-  step2[4] = step1[4];
-  step2[5] = step1[5];
-  step2[6] = step1[6];
-  step2[7] = step1[7];
-
-  temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64;
-  temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64;
-  step2[8] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[15] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64;
-  temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64;
-  step2[9] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[14] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64;
-  temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64;
-  step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64;
-  temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64;
-  step2[11] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[12] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  step2[16] = HIGHBD_WRAPLOW(step1[16] + step1[17], bd);
-  step2[17] = HIGHBD_WRAPLOW(step1[16] - step1[17], bd);
-  step2[18] = HIGHBD_WRAPLOW(-step1[18] + step1[19], bd);
-  step2[19] = HIGHBD_WRAPLOW(step1[18] + step1[19], bd);
-  step2[20] = HIGHBD_WRAPLOW(step1[20] + step1[21], bd);
-  step2[21] = HIGHBD_WRAPLOW(step1[20] - step1[21], bd);
-  step2[22] = HIGHBD_WRAPLOW(-step1[22] + step1[23], bd);
-  step2[23] = HIGHBD_WRAPLOW(step1[22] + step1[23], bd);
-  step2[24] = HIGHBD_WRAPLOW(step1[24] + step1[25], bd);
-  step2[25] = HIGHBD_WRAPLOW(step1[24] - step1[25], bd);
-  step2[26] = HIGHBD_WRAPLOW(-step1[26] + step1[27], bd);
-  step2[27] = HIGHBD_WRAPLOW(step1[26] + step1[27], bd);
-  step2[28] = HIGHBD_WRAPLOW(step1[28] + step1[29], bd);
-  step2[29] = HIGHBD_WRAPLOW(step1[28] - step1[29], bd);
-  step2[30] = HIGHBD_WRAPLOW(-step1[30] + step1[31], bd);
-  step2[31] = HIGHBD_WRAPLOW(step1[30] + step1[31], bd);
-
-  // stage 3
-  step1[0] = step2[0];
-  step1[1] = step2[1];
-  step1[2] = step2[2];
-  step1[3] = step2[3];
-
-  temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64;
-  temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64;
-  step1[4] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[7] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64;
-  temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64;
-  step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-
-  step1[8] = HIGHBD_WRAPLOW(step2[8] + step2[9], bd);
-  step1[9] = HIGHBD_WRAPLOW(step2[8] - step2[9], bd);
-  step1[10] = HIGHBD_WRAPLOW(-step2[10] + step2[11], bd);
-  step1[11] = HIGHBD_WRAPLOW(step2[10] + step2[11], bd);
-  step1[12] = HIGHBD_WRAPLOW(step2[12] + step2[13], bd);
-  step1[13] = HIGHBD_WRAPLOW(step2[12] - step2[13], bd);
-  step1[14] = HIGHBD_WRAPLOW(-step2[14] + step2[15], bd);
-  step1[15] = HIGHBD_WRAPLOW(step2[14] + step2[15], bd);
-
-  step1[16] = step2[16];
-  step1[31] = step2[31];
-  temp1 = -step2[17] * cospi_4_64 + step2[30] * cospi_28_64;
-  temp2 = step2[17] * cospi_28_64 + step2[30] * cospi_4_64;
-  step1[17] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[30] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = -step2[18] * cospi_28_64 - step2[29] * cospi_4_64;
-  temp2 = -step2[18] * cospi_4_64 + step2[29] * cospi_28_64;
-  step1[18] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[29] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step1[19] = step2[19];
-  step1[20] = step2[20];
-  temp1 = -step2[21] * cospi_20_64 + step2[26] * cospi_12_64;
-  temp2 = step2[21] * cospi_12_64 + step2[26] * cospi_20_64;
-  step1[21] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[26] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = -step2[22] * cospi_12_64 - step2[25] * cospi_20_64;
-  temp2 = -step2[22] * cospi_20_64 + step2[25] * cospi_12_64;
-  step1[22] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[25] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step1[23] = step2[23];
-  step1[24] = step2[24];
-  step1[27] = step2[27];
-  step1[28] = step2[28];
-
-  // stage 4
-  temp1 = (step1[0] + step1[1]) * cospi_16_64;
-  temp2 = (step1[0] - step1[1]) * cospi_16_64;
-  step2[0] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[1] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
-  temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64;
-  step2[2] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[3] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step2[4] = HIGHBD_WRAPLOW(step1[4] + step1[5], bd);
-  step2[5] = HIGHBD_WRAPLOW(step1[4] - step1[5], bd);
-  step2[6] = HIGHBD_WRAPLOW(-step1[6] + step1[7], bd);
-  step2[7] = HIGHBD_WRAPLOW(step1[6] + step1[7], bd);
-
-  step2[8] = step1[8];
-  step2[15] = step1[15];
-  temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64;
-  temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64;
-  step2[9] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[14] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64;
-  temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64;
-  step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step2[11] = step1[11];
-  step2[12] = step1[12];
-
-  step2[16] = HIGHBD_WRAPLOW(step1[16] + step1[19], bd);
-  step2[17] = HIGHBD_WRAPLOW(step1[17] + step1[18], bd);
-  step2[18] = HIGHBD_WRAPLOW(step1[17] - step1[18], bd);
-  step2[19] = HIGHBD_WRAPLOW(step1[16] - step1[19], bd);
-  step2[20] = HIGHBD_WRAPLOW(-step1[20] + step1[23], bd);
-  step2[21] = HIGHBD_WRAPLOW(-step1[21] + step1[22], bd);
-  step2[22] = HIGHBD_WRAPLOW(step1[21] + step1[22], bd);
-  step2[23] = HIGHBD_WRAPLOW(step1[20] + step1[23], bd);
-
-  step2[24] = HIGHBD_WRAPLOW(step1[24] + step1[27], bd);
-  step2[25] = HIGHBD_WRAPLOW(step1[25] + step1[26], bd);
-  step2[26] = HIGHBD_WRAPLOW(step1[25] - step1[26], bd);
-  step2[27] = HIGHBD_WRAPLOW(step1[24] - step1[27], bd);
-  step2[28] = HIGHBD_WRAPLOW(-step1[28] + step1[31], bd);
-  step2[29] = HIGHBD_WRAPLOW(-step1[29] + step1[30], bd);
-  step2[30] = HIGHBD_WRAPLOW(step1[29] + step1[30], bd);
-  step2[31] = HIGHBD_WRAPLOW(step1[28] + step1[31], bd);
-
-  // stage 5
-  step1[0] = HIGHBD_WRAPLOW(step2[0] + step2[3], bd);
-  step1[1] = HIGHBD_WRAPLOW(step2[1] + step2[2], bd);
-  step1[2] = HIGHBD_WRAPLOW(step2[1] - step2[2], bd);
-  step1[3] = HIGHBD_WRAPLOW(step2[0] - step2[3], bd);
-  step1[4] = step2[4];
-  temp1 = (step2[6] - step2[5]) * cospi_16_64;
-  temp2 = (step2[5] + step2[6]) * cospi_16_64;
-  step1[5] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[6] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step1[7] = step2[7];
-
-  step1[8] = HIGHBD_WRAPLOW(step2[8] + step2[11], bd);
-  step1[9] = HIGHBD_WRAPLOW(step2[9] + step2[10], bd);
-  step1[10] = HIGHBD_WRAPLOW(step2[9] - step2[10], bd);
-  step1[11] = HIGHBD_WRAPLOW(step2[8] - step2[11], bd);
-  step1[12] = HIGHBD_WRAPLOW(-step2[12] + step2[15], bd);
-  step1[13] = HIGHBD_WRAPLOW(-step2[13] + step2[14], bd);
-  step1[14] = HIGHBD_WRAPLOW(step2[13] + step2[14], bd);
-  step1[15] = HIGHBD_WRAPLOW(step2[12] + step2[15], bd);
-
-  step1[16] = step2[16];
-  step1[17] = step2[17];
-  temp1 = -step2[18] * cospi_8_64 + step2[29] * cospi_24_64;
-  temp2 = step2[18] * cospi_24_64 + step2[29] * cospi_8_64;
-  step1[18] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[29] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = -step2[19] * cospi_8_64 + step2[28] * cospi_24_64;
-  temp2 = step2[19] * cospi_24_64 + step2[28] * cospi_8_64;
-  step1[19] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[28] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = -step2[20] * cospi_24_64 - step2[27] * cospi_8_64;
-  temp2 = -step2[20] * cospi_8_64 + step2[27] * cospi_24_64;
-  step1[20] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[27] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = -step2[21] * cospi_24_64 - step2[26] * cospi_8_64;
-  temp2 = -step2[21] * cospi_8_64 + step2[26] * cospi_24_64;
-  step1[21] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[26] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step1[22] = step2[22];
-  step1[23] = step2[23];
-  step1[24] = step2[24];
-  step1[25] = step2[25];
-  step1[30] = step2[30];
-  step1[31] = step2[31];
-
-  // stage 6
-  step2[0] = HIGHBD_WRAPLOW(step1[0] + step1[7], bd);
-  step2[1] = HIGHBD_WRAPLOW(step1[1] + step1[6], bd);
-  step2[2] = HIGHBD_WRAPLOW(step1[2] + step1[5], bd);
-  step2[3] = HIGHBD_WRAPLOW(step1[3] + step1[4], bd);
-  step2[4] = HIGHBD_WRAPLOW(step1[3] - step1[4], bd);
-  step2[5] = HIGHBD_WRAPLOW(step1[2] - step1[5], bd);
-  step2[6] = HIGHBD_WRAPLOW(step1[1] - step1[6], bd);
-  step2[7] = HIGHBD_WRAPLOW(step1[0] - step1[7], bd);
-  step2[8] = step1[8];
-  step2[9] = step1[9];
-  temp1 = (-step1[10] + step1[13]) * cospi_16_64;
-  temp2 = (step1[10] + step1[13]) * cospi_16_64;
-  step2[10] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[13] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = (-step1[11] + step1[12]) * cospi_16_64;
-  temp2 = (step1[11] + step1[12]) * cospi_16_64;
-  step2[11] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step2[12] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step2[14] = step1[14];
-  step2[15] = step1[15];
-
-  step2[16] = HIGHBD_WRAPLOW(step1[16] + step1[23], bd);
-  step2[17] = HIGHBD_WRAPLOW(step1[17] + step1[22], bd);
-  step2[18] = HIGHBD_WRAPLOW(step1[18] + step1[21], bd);
-  step2[19] = HIGHBD_WRAPLOW(step1[19] + step1[20], bd);
-  step2[20] = HIGHBD_WRAPLOW(step1[19] - step1[20], bd);
-  step2[21] = HIGHBD_WRAPLOW(step1[18] - step1[21], bd);
-  step2[22] = HIGHBD_WRAPLOW(step1[17] - step1[22], bd);
-  step2[23] = HIGHBD_WRAPLOW(step1[16] - step1[23], bd);
-
-  step2[24] = HIGHBD_WRAPLOW(-step1[24] + step1[31], bd);
-  step2[25] = HIGHBD_WRAPLOW(-step1[25] + step1[30], bd);
-  step2[26] = HIGHBD_WRAPLOW(-step1[26] + step1[29], bd);
-  step2[27] = HIGHBD_WRAPLOW(-step1[27] + step1[28], bd);
-  step2[28] = HIGHBD_WRAPLOW(step1[27] + step1[28], bd);
-  step2[29] = HIGHBD_WRAPLOW(step1[26] + step1[29], bd);
-  step2[30] = HIGHBD_WRAPLOW(step1[25] + step1[30], bd);
-  step2[31] = HIGHBD_WRAPLOW(step1[24] + step1[31], bd);
-
-  // stage 7
-  step1[0] = HIGHBD_WRAPLOW(step2[0] + step2[15], bd);
-  step1[1] = HIGHBD_WRAPLOW(step2[1] + step2[14], bd);
-  step1[2] = HIGHBD_WRAPLOW(step2[2] + step2[13], bd);
-  step1[3] = HIGHBD_WRAPLOW(step2[3] + step2[12], bd);
-  step1[4] = HIGHBD_WRAPLOW(step2[4] + step2[11], bd);
-  step1[5] = HIGHBD_WRAPLOW(step2[5] + step2[10], bd);
-  step1[6] = HIGHBD_WRAPLOW(step2[6] + step2[9], bd);
-  step1[7] = HIGHBD_WRAPLOW(step2[7] + step2[8], bd);
-  step1[8] = HIGHBD_WRAPLOW(step2[7] - step2[8], bd);
-  step1[9] = HIGHBD_WRAPLOW(step2[6] - step2[9], bd);
-  step1[10] = HIGHBD_WRAPLOW(step2[5] - step2[10], bd);
-  step1[11] = HIGHBD_WRAPLOW(step2[4] - step2[11], bd);
-  step1[12] = HIGHBD_WRAPLOW(step2[3] - step2[12], bd);
-  step1[13] = HIGHBD_WRAPLOW(step2[2] - step2[13], bd);
-  step1[14] = HIGHBD_WRAPLOW(step2[1] - step2[14], bd);
-  step1[15] = HIGHBD_WRAPLOW(step2[0] - step2[15], bd);
-
-  step1[16] = step2[16];
-  step1[17] = step2[17];
-  step1[18] = step2[18];
-  step1[19] = step2[19];
-  temp1 = (-step2[20] + step2[27]) * cospi_16_64;
-  temp2 = (step2[20] + step2[27]) * cospi_16_64;
-  step1[20] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[27] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = (-step2[21] + step2[26]) * cospi_16_64;
-  temp2 = (step2[21] + step2[26]) * cospi_16_64;
-  step1[21] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[26] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = (-step2[22] + step2[25]) * cospi_16_64;
-  temp2 = (step2[22] + step2[25]) * cospi_16_64;
-  step1[22] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[25] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  temp1 = (-step2[23] + step2[24]) * cospi_16_64;
-  temp2 = (step2[23] + step2[24]) * cospi_16_64;
-  step1[23] = HIGHBD_WRAPLOW(dct_const_round_shift(temp1), bd);
-  step1[24] = HIGHBD_WRAPLOW(dct_const_round_shift(temp2), bd);
-  step1[28] = step2[28];
-  step1[29] = step2[29];
-  step1[30] = step2[30];
-  step1[31] = step2[31];
-
-  // final stage
-  output[0] = HIGHBD_WRAPLOW(step1[0] + step1[31], bd);
-  output[1] = HIGHBD_WRAPLOW(step1[1] + step1[30], bd);
-  output[2] = HIGHBD_WRAPLOW(step1[2] + step1[29], bd);
-  output[3] = HIGHBD_WRAPLOW(step1[3] + step1[28], bd);
-  output[4] = HIGHBD_WRAPLOW(step1[4] + step1[27], bd);
-  output[5] = HIGHBD_WRAPLOW(step1[5] + step1[26], bd);
-  output[6] = HIGHBD_WRAPLOW(step1[6] + step1[25], bd);
-  output[7] = HIGHBD_WRAPLOW(step1[7] + step1[24], bd);
-  output[8] = HIGHBD_WRAPLOW(step1[8] + step1[23], bd);
-  output[9] = HIGHBD_WRAPLOW(step1[9] + step1[22], bd);
-  output[10] = HIGHBD_WRAPLOW(step1[10] + step1[21], bd);
-  output[11] = HIGHBD_WRAPLOW(step1[11] + step1[20], bd);
-  output[12] = HIGHBD_WRAPLOW(step1[12] + step1[19], bd);
-  output[13] = HIGHBD_WRAPLOW(step1[13] + step1[18], bd);
-  output[14] = HIGHBD_WRAPLOW(step1[14] + step1[17], bd);
-  output[15] = HIGHBD_WRAPLOW(step1[15] + step1[16], bd);
-  output[16] = HIGHBD_WRAPLOW(step1[15] - step1[16], bd);
-  output[17] = HIGHBD_WRAPLOW(step1[14] - step1[17], bd);
-  output[18] = HIGHBD_WRAPLOW(step1[13] - step1[18], bd);
-  output[19] = HIGHBD_WRAPLOW(step1[12] - step1[19], bd);
-  output[20] = HIGHBD_WRAPLOW(step1[11] - step1[20], bd);
-  output[21] = HIGHBD_WRAPLOW(step1[10] - step1[21], bd);
-  output[22] = HIGHBD_WRAPLOW(step1[9] - step1[22], bd);
-  output[23] = HIGHBD_WRAPLOW(step1[8] - step1[23], bd);
-  output[24] = HIGHBD_WRAPLOW(step1[7] - step1[24], bd);
-  output[25] = HIGHBD_WRAPLOW(step1[6] - step1[25], bd);
-  output[26] = HIGHBD_WRAPLOW(step1[5] - step1[26], bd);
-  output[27] = HIGHBD_WRAPLOW(step1[4] - step1[27], bd);
-  output[28] = HIGHBD_WRAPLOW(step1[3] - step1[28], bd);
-  output[29] = HIGHBD_WRAPLOW(step1[2] - step1[29], bd);
-  output[30] = HIGHBD_WRAPLOW(step1[1] - step1[30], bd);
-  output[31] = HIGHBD_WRAPLOW(step1[0] - step1[31], bd);
-}
-
-#endif  // CONFIG_HIGHBITDEPTH
diff --git a/third_party/aom/aom_dsp/inv_txfm.h b/third_party/aom/aom_dsp/inv_txfm.h
index e64d463ea..a9c485e74 100644
--- a/third_party/aom/aom_dsp/inv_txfm.h
+++ b/third_party/aom/aom_dsp/inv_txfm.h
@@ -23,8 +23,7 @@ extern "C" {
 #endif
 
 static INLINE tran_high_t dct_const_round_shift(tran_high_t input) {
-  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
-  return rv;
+  return ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
 }
 
 static INLINE tran_high_t check_range(tran_high_t input, int bd) {
@@ -51,9 +50,19 @@ static INLINE tran_high_t check_range(tran_high_t input, int bd) {
 }
 
 #define WRAPLOW(x) ((int32_t)check_range(x, 8))
-#if CONFIG_HIGHBITDEPTH
 #define HIGHBD_WRAPLOW(x, bd) ((int32_t)check_range((x), bd))
-#endif  // CONFIG_HIGHBITDEPTH
+
+#if CONFIG_MRC_TX
+// These each perform dct but add coefficients based on a mask
+void aom_imrc32x32_1024_add_c(const tran_low_t *input, uint8_t *dest,
+                              int stride, int *mask);
+
+void aom_imrc32x32_135_add_c(const tran_low_t *input, uint8_t *dest, int stride,
+                             int *mask);
+
+void aom_imrc32x32_34_add_c(const tran_low_t *input, uint8_t *dest, int stride,
+                            int *mask);
+#endif  // CONFIG_MRC_TX
 
 void aom_idct4_c(const tran_low_t *input, tran_low_t *output);
 void aom_idct8_c(const tran_low_t *input, tran_low_t *output);
@@ -63,7 +72,6 @@ void aom_iadst4_c(const tran_low_t *input, tran_low_t *output);
 void aom_iadst8_c(const tran_low_t *input, tran_low_t *output);
 void aom_iadst16_c(const tran_low_t *input, tran_low_t *output);
 
-#if CONFIG_HIGHBITDEPTH
 void aom_highbd_idct4_c(const tran_low_t *input, tran_low_t *output, int bd);
 void aom_highbd_idct8_c(const tran_low_t *input, tran_low_t *output, int bd);
 void aom_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd);
@@ -78,7 +86,6 @@ static INLINE uint16_t highbd_clip_pixel_add(uint16_t dest, tran_high_t trans,
   trans = HIGHBD_WRAPLOW(trans, bd);
   return clip_pixel_highbd(dest + (int)trans, bd);
 }
-#endif
 
 static INLINE uint8_t clip_pixel_add(uint8_t dest, tran_high_t trans) {
   trans = WRAPLOW(trans);
diff --git a/third_party/aom/aom_dsp/mips/fwd_dct32x32_msa.c b/third_party/aom/aom_dsp/mips/fwd_dct32x32_msa.c
index dc9c63226..43dce8ba6 100644
--- a/third_party/aom/aom_dsp/mips/fwd_dct32x32_msa.c
+++ b/third_party/aom/aom_dsp/mips/fwd_dct32x32_msa.c
@@ -926,23 +926,3 @@ void aom_fdct32x32_rd_msa(const int16_t *input, int16_t *out,
                        out + (8 * i * 32));
   }
 }
-
-void aom_fdct32x32_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
-  int sum = LD_HADD(input, stride);
-  sum += LD_HADD(input + 8, stride);
-  sum += LD_HADD(input + 16, stride);
-  sum += LD_HADD(input + 24, stride);
-  sum += LD_HADD(input + 32 * 8, stride);
-  sum += LD_HADD(input + 32 * 8 + 8, stride);
-  sum += LD_HADD(input + 32 * 8 + 16, stride);
-  sum += LD_HADD(input + 32 * 8 + 24, stride);
-  sum += LD_HADD(input + 32 * 16, stride);
-  sum += LD_HADD(input + 32 * 16 + 8, stride);
-  sum += LD_HADD(input + 32 * 16 + 16, stride);
-  sum += LD_HADD(input + 32 * 16 + 24, stride);
-  sum += LD_HADD(input + 32 * 24, stride);
-  sum += LD_HADD(input + 32 * 24 + 8, stride);
-  sum += LD_HADD(input + 32 * 24 + 16, stride);
-  sum += LD_HADD(input + 32 * 24 + 24, stride);
-  out[0] = (int16_t)(sum >> 3);
-}
diff --git a/third_party/aom/aom_dsp/mips/fwd_txfm_msa.c b/third_party/aom/aom_dsp/mips/fwd_txfm_msa.c
index f16d290c8..7a285b7b8 100644
--- a/third_party/aom/aom_dsp/mips/fwd_txfm_msa.c
+++ b/third_party/aom/aom_dsp/mips/fwd_txfm_msa.c
@@ -236,11 +236,3 @@ void aom_fdct16x16_msa(const int16_t *input, int16_t *output,
     fdct16x8_1d_row((&tmp_buf[0] + (128 * i)), (output + (128 * i)));
   }
 }
-
-void aom_fdct16x16_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
-  int sum = LD_HADD(input, stride);
-  sum += LD_HADD(input + 8, stride);
-  sum += LD_HADD(input + 16 * 8, stride);
-  sum += LD_HADD(input + 16 * 8 + 8, stride);
-  out[0] = (int16_t)(sum >> 1);
-}
diff --git a/third_party/aom/aom_dsp/prob.c b/third_party/aom/aom_dsp/prob.c
index eefe7521f..a42fb806b 100644
--- a/third_party/aom/aom_dsp/prob.c
+++ b/third_party/aom/aom_dsp/prob.c
@@ -186,10 +186,8 @@ int tree_to_cdf(const aom_tree_index *tree, const aom_prob *probs,
   for (i = 1; i < nsymbs; i++) {
     cdf[i] = AOM_ICDF(AOM_ICDF(cdf[i - 1]) + cdf[i]);
   }
-// Store symbol count at the end of the CDF
-#if CONFIG_EC_ADAPT
+  // Store symbol count at the end of the CDF
   cdf[nsymbs] = 0;
-#endif
   return nsymbs;
 }
 
diff --git a/third_party/aom/aom_dsp/prob.h b/third_party/aom/aom_dsp/prob.h
index ec6654ab7..35db134e5 100644
--- a/third_party/aom/aom_dsp/prob.h
+++ b/third_party/aom/aom_dsp/prob.h
@@ -148,7 +148,6 @@ static INLINE void av1_tree_to_cdf(const aom_tree_index *tree,
 
 void av1_indices_from_tree(int *ind, int *inv, const aom_tree_index *tree);
 
-#if CONFIG_EC_ADAPT
 static INLINE void update_cdf(aom_cdf_prob *cdf, int val, int nsymbs) {
   const int rate = 4 + (cdf[nsymbs] > 31) + get_msb(nsymbs);
   const int rate2 = 5;
@@ -183,7 +182,6 @@ static INLINE void update_cdf(aom_cdf_prob *cdf, int val, int nsymbs) {
 #endif
   cdf[nsymbs] += (cdf[nsymbs] < 32);
 }
-#endif
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/third_party/aom/aom_dsp/quantize.c b/third_party/aom/aom_dsp/quantize.c
index 0759c22e3..fe98b6028 100644
--- a/third_party/aom/aom_dsp/quantize.c
+++ b/third_party/aom/aom_dsp/quantize.c
@@ -256,7 +256,6 @@ void aom_quantize_dc_64x64(const tran_low_t *coeff_ptr, int skip_block,
 }
 #endif  // CONFIG_TX64X64
 
-#if CONFIG_HIGHBITDEPTH
 void aom_highbd_quantize_dc(const tran_low_t *coeff_ptr, int n_coeffs,
                             int skip_block, const int16_t *round_ptr,
                             const int16_t quant, tran_low_t *qcoeff_ptr,
@@ -523,7 +522,6 @@ void aom_highbd_quantize_b_64x64_c(
   *eob_ptr = eob + 1;
 }
 #endif  // CONFIG_TX64X64
-#endif  // CONFIG_HIGHBITDEPTH
 
 #else  // CONFIG_AOM_QM
 
@@ -602,7 +600,6 @@ void aom_quantize_dc_64x64(const tran_low_t *coeff_ptr, int skip_block,
 }
 #endif  // CONFIG_TX64X64
 
-#if CONFIG_HIGHBITDEPTH
 void aom_highbd_quantize_dc(const tran_low_t *coeff_ptr, int n_coeffs,
                             int skip_block, const int16_t *round_ptr,
                             const int16_t quant, tran_low_t *qcoeff_ptr,
@@ -716,8 +713,7 @@ void aom_highbd_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
       if (abs_coeff >= zbins[rc != 0]) {
         const int64_t tmp1 = abs_coeff + round_ptr[rc != 0];
         const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
-        const uint32_t abs_qcoeff =
-            (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> 16);
+        const int abs_qcoeff = (int)((tmp2 * quant_shift_ptr[rc != 0]) >> 16);
         qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
         dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
         if (abs_qcoeff) eob = i;
@@ -767,8 +763,7 @@ void aom_highbd_quantize_b_32x32_c(
       const int64_t tmp1 =
           abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
       const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
-      const uint32_t abs_qcoeff =
-          (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> 15);
+      const int abs_qcoeff = (int)((tmp2 * quant_shift_ptr[rc != 0]) >> 15);
       qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
       dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
       if (abs_qcoeff) eob = idx_arr[i];
@@ -818,8 +813,7 @@ void aom_highbd_quantize_b_64x64_c(
       const int64_t tmp1 =
           abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 2);
       const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
-      const uint32_t abs_qcoeff =
-          (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> 14);
+      const int abs_qcoeff = (int)((tmp2 * quant_shift_ptr[rc != 0]) >> 14);
       qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
       dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 4;
       if (abs_qcoeff) eob = idx_arr[i];
@@ -828,5 +822,4 @@ void aom_highbd_quantize_b_64x64_c(
   *eob_ptr = eob + 1;
 }
 #endif  // CONFIG_TX64X64
-#endif  // CONFIG_HIGHBITDEPTH
 #endif  // CONFIG_AOM_QM
diff --git a/third_party/aom/aom_dsp/sad.c b/third_party/aom/aom_dsp/sad.c
index 2cc172ba5..b9c789ce5 100644
--- a/third_party/aom/aom_dsp/sad.c
+++ b/third_party/aom/aom_dsp/sad.c
@@ -153,10 +153,21 @@ sadMxN(4, 4)
 sadMxNxK(4, 4, 3)
 sadMxNxK(4, 4, 8)
 sadMxNx4D(4, 4)
+
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+sadMxN(4, 16)
+sadMxNx4D(4, 16)
+sadMxN(16, 4)
+sadMxNx4D(16, 4)
+sadMxN(8, 32)
+sadMxNx4D(8, 32)
+sadMxN(32, 8)
+sadMxNx4D(32, 8)
+#endif
 /* clang-format on */
 
 #if CONFIG_HIGHBITDEPTH
-        static INLINE
+                static INLINE
     unsigned int highbd_sad(const uint8_t *a8, int a_stride, const uint8_t *b8,
                             int b_stride, int width, int height) {
   int y, x;
@@ -307,11 +318,22 @@ highbd_sadMxN(4, 4)
 highbd_sadMxNxK(4, 4, 3)
 highbd_sadMxNxK(4, 4, 8)
 highbd_sadMxNx4D(4, 4)
+
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+highbd_sadMxN(4, 16)
+highbd_sadMxNx4D(4, 16)
+highbd_sadMxN(16, 4)
+highbd_sadMxNx4D(16, 4)
+highbd_sadMxN(8, 32)
+highbd_sadMxNx4D(8, 32)
+highbd_sadMxN(32, 8)
+highbd_sadMxNx4D(32, 8)
+#endif
 /* clang-format on */
 #endif  // CONFIG_HIGHBITDEPTH
 
 #if CONFIG_AV1 && CONFIG_EXT_INTER
-            static INLINE
+                            static INLINE
     unsigned int masked_sad(const uint8_t *src, int src_stride,
                             const uint8_t *a, int a_stride, const uint8_t *b,
                             int b_stride, const uint8_t *m, int m_stride,
@@ -367,10 +389,17 @@ MASKSADMxN(8, 8)
 MASKSADMxN(8, 4)
 MASKSADMxN(4, 8)
 MASKSADMxN(4, 4)
+
+#if CONFIG_EXT_PARTITION_TYPES
+MASKSADMxN(4, 16)
+MASKSADMxN(16, 4)
+MASKSADMxN(8, 32)
+MASKSADMxN(32, 8)
+#endif
 /* clang-format on */
 
 #if CONFIG_HIGHBITDEPTH
-                    static INLINE
+                            static INLINE
     unsigned int highbd_masked_sad(const uint8_t *src8, int src_stride,
                                    const uint8_t *a8, int a_stride,
                                    const uint8_t *b8, int b_stride,
@@ -429,6 +458,13 @@ HIGHBD_MASKSADMXN(8, 8)
 HIGHBD_MASKSADMXN(8, 4)
 HIGHBD_MASKSADMXN(4, 8)
 HIGHBD_MASKSADMXN(4, 4)
+
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+HIGHBD_MASKSADMXN(4, 16)
+HIGHBD_MASKSADMXN(16, 4)
+HIGHBD_MASKSADMXN(8, 32)
+HIGHBD_MASKSADMXN(32, 8)
+#endif
 #endif  // CONFIG_HIGHBITDEPTH
 #endif  // CONFIG_AV1 && CONFIG_EXT_INTER
 
@@ -480,10 +516,17 @@ OBMCSADMxN(8, 8)
 OBMCSADMxN(8, 4)
 OBMCSADMxN(4, 8)
 OBMCSADMxN(4, 4)
+
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+OBMCSADMxN(4, 16)
+OBMCSADMxN(16, 4)
+OBMCSADMxN(8, 32)
+OBMCSADMxN(32, 8)
+#endif
 /* clang-format on */
 
 #if CONFIG_HIGHBITDEPTH
-                    static INLINE
+                            static INLINE
     unsigned int highbd_obmc_sad(const uint8_t *pre8, int pre_stride,
                                  const int32_t *wsrc, const int32_t *mask,
                                  int width, int height) {
@@ -529,6 +572,13 @@ HIGHBD_OBMCSADMXN(8, 8)
 HIGHBD_OBMCSADMXN(8, 4)
 HIGHBD_OBMCSADMXN(4, 8)
 HIGHBD_OBMCSADMXN(4, 4)
+
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+HIGHBD_OBMCSADMXN(4, 16)
+HIGHBD_OBMCSADMXN(16, 4)
+HIGHBD_OBMCSADMXN(8, 32)
+HIGHBD_OBMCSADMXN(32, 8)
+#endif
 /* clang-format on */
 #endif  // CONFIG_HIGHBITDEPTH
 #endif  // CONFIG_AV1 && CONFIG_MOTION_VAR
diff --git a/third_party/aom/aom_dsp/simd/v256_intrinsics_v128.h b/third_party/aom/aom_dsp/simd/v256_intrinsics_v128.h
index a4b334ea6..cbea55ca1 100644
--- a/third_party/aom/aom_dsp/simd/v256_intrinsics_v128.h
+++ b/third_party/aom/aom_dsp/simd/v256_intrinsics_v128.h
@@ -508,17 +508,19 @@ SIMD_INLINE v256 v256_shr_s32(v256 a, unsigned int c) {
    to enforce that. */
 #define v256_shl_n_byte(a, n)                                                 \
   ((n) < 16 ? v256_from_v128(v128_or(v128_shl_n_byte(a.hi, n),                \
-                                     v128_shr_n_byte(a.lo, 16 - (n))),        \
+                                     v128_shr_n_byte(a.lo, (16 - (n)) & 31)), \
                              v128_shl_n_byte(a.lo, (n)))                      \
-            : v256_from_v128((n) > 16 ? v128_shl_n_byte(a.lo, (n)-16) : a.lo, \
-                             v128_zero()))
-
-#define v256_shr_n_byte(a, n)                                          \
-  ((n) < 16 ? v256_from_v128(v128_shr_n_byte(a.hi, n),                 \
-                             v128_or(v128_shr_n_byte(a.lo, n),         \
-                                     v128_shl_n_byte(a.hi, 16 - (n)))) \
-            : v256_from_v128(v128_zero(),                              \
-                             (n) > 16 ? v128_shr_n_byte(a.hi, (n)-16) : a.hi))
+            : v256_from_v128(                                                 \
+                  (n) > 16 ? v128_shl_n_byte(a.lo, ((n)-16) & 31) : a.lo,     \
+                  v128_zero()))
+
+#define v256_shr_n_byte(a, n)                                                 \
+  ((n) < 16 ? v256_from_v128(v128_shr_n_byte(a.hi, n),                        \
+                             v128_or(v128_shr_n_byte(a.lo, n),                \
+                                     v128_shl_n_byte(a.hi, (16 - (n)) & 31))) \
+            : v256_from_v128(                                                 \
+                  v128_zero(),                                                \
+                  (n) > 16 ? v128_shr_n_byte(a.hi, ((n)-16) & 31) : a.hi))
 
 #define v256_align(a, b, c) \
   ((c) ? v256_or(v256_shr_n_byte(b, c), v256_shl_n_byte(a, 32 - (c))) : b)
diff --git a/third_party/aom/aom_dsp/txfm_common.h b/third_party/aom/aom_dsp/txfm_common.h
index a5e964aad..01732ae64 100644
--- a/third_party/aom/aom_dsp/txfm_common.h
+++ b/third_party/aom/aom_dsp/txfm_common.h
@@ -21,6 +21,27 @@
 #define UNIT_QUANT_SHIFT 2
 #define UNIT_QUANT_FACTOR (1 << UNIT_QUANT_SHIFT)
 
+typedef struct txfm_param {
+  // for both forward and inverse transforms
+  int tx_type;
+  int tx_size;
+  int lossless;
+  int bd;
+#if CONFIG_MRC_TX || CONFIG_LGT
+  int stride;
+  uint8_t *dst;
+#endif  // CONFIG_MRC_TX || CONFIG_LGT
+#if CONFIG_LGT
+  int is_inter;
+  int mode;
+#endif
+// for inverse transforms only
+#if CONFIG_ADAPT_SCAN
+  const int16_t *eob_threshold;
+#endif
+  int eob;
+} TxfmParam;
+
 // Constants:
 //  for (int i = 1; i< 32; ++i)
 //    printf("static const int cospi_%d_64 = %.0f;\n", i,
@@ -67,4 +88,62 @@ static const tran_high_t sinpi_4_9 = 15212;
 // 16384 * sqrt(2)
 static const tran_high_t Sqrt2 = 23170;
 
+static INLINE tran_high_t fdct_round_shift(tran_high_t input) {
+  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
+  return rv;
+}
+
+#if CONFIG_LGT
+// The Line Graph Transforms (LGTs) matrices are written as follows.
+// Each 2D array is 16384 times an LGT matrix, which is the matrix of
+// eigenvectors of the graph Laplacian matrices for the line graph.
+
+// LGT4 name: lgt4_140
+// Self loops: 1.400, 0.000, 0.000, 0.000
+// Edges: 1.000, 1.000, 1.000
+static const tran_high_t lgt4_140[4][4] = {
+  { 4206, 9518, 13524, 15674 },
+  { 11552, 14833, 1560, -13453 },
+  { 15391, -1906, -14393, 9445 },
+  { 12201, -14921, 12016, -4581 },
+};
+
+// LGT4 name: lgt4_170
+// Self loops: 1.700, 0.000, 0.000, 0.000
+// Edges: 1.000, 1.000, 1.000
+static const tran_high_t lgt4_170[4][4] = {
+  { 3636, 9287, 13584, 15902 },
+  { 10255, 15563, 2470, -13543 },
+  { 14786, 711, -15249, 9231 },
+  { 14138, -14420, 10663, -3920 },
+};
+
+// LGT8 name: lgt8_150
+// Self loops: 1.500, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000
+// Edges: 1.000, 1.000, 1.000, 1.000, 1.000, 1.000, 1.000
+static const tran_high_t lgt8_150[8][8] = {
+  { 2075, 5110, 7958, 10511, 12677, 14376, 15544, 16140 },
+  { 6114, 13307, 16196, 13845, 7015, -2084, -10509, -15534 },
+  { 9816, 16163, 8717, -6168, -15790, -11936, 2104, 14348 },
+  { 12928, 12326, -7340, -15653, 242, 15763, 6905, -12632 },
+  { 15124, 3038, -16033, 1758, 15507, -6397, -13593, 10463 },
+  { 15895, -7947, -7947, 15895, -7947, -7947, 15895, -7947 },
+  { 14325, -15057, 9030, 1050, -10659, 15483, -13358, 5236 },
+  { 9054, -12580, 14714, -15220, 14043, -11312, 7330, -2537 },
+};
+
+// LGT8 name: lgt8_170
+// Self loops: 1.700, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000
+// Edges: 1.000, 1.000, 1.000, 1.000, 1.000, 1.000, 1.000
+static const tran_high_t lgt8_170[8][8] = {
+  { 1858, 4947, 7850, 10458, 12672, 14411, 15607, 16217 },
+  { 5494, 13022, 16256, 14129, 7343, -1864, -10456, -15601 },
+  { 8887, 16266, 9500, -5529, -15749, -12273, 1876, 14394 },
+  { 11870, 13351, -6199, -15984, -590, 15733, 7273, -12644 },
+  { 14248, 5137, -15991, 291, 15893, -5685, -13963, 10425 },
+  { 15716, -5450, -10010, 15929, -6665, -8952, 16036, -7835 },
+  { 15533, -13869, 6559, 3421, -12009, 15707, -13011, 5018 },
+  { 11357, -13726, 14841, -14600, 13025, -10259, 6556, -2254 },
+};
+#endif  // CONFIG_LGT
 #endif  // AOM_DSP_TXFM_COMMON_H_
diff --git a/third_party/aom/aom_dsp/variance.c b/third_party/aom/aom_dsp/variance.c
index 79677c92f..a4c3616e7 100644
--- a/third_party/aom/aom_dsp/variance.c
+++ b/third_party/aom/aom_dsp/variance.c
@@ -9,6 +9,8 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 #include <stdlib.h>
+#include <string.h>
+#include <assert.h>
 
 #include "./aom_config.h"
 #include "./aom_dsp_rtcd.h"
@@ -20,6 +22,9 @@
 #include "aom_dsp/aom_filter.h"
 #include "aom_dsp/blend.h"
 
+#include "./av1_rtcd.h"
+#include "av1/common/filter.h"
+
 uint32_t aom_get4x4sse_cs_c(const uint8_t *a, int a_stride, const uint8_t *b,
                             int b_stride) {
   int distortion = 0;
@@ -246,6 +251,13 @@ VARIANCES(4, 2)
 VARIANCES(2, 4)
 VARIANCES(2, 2)
 
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+VARIANCES(4, 16)
+VARIANCES(16, 4)
+VARIANCES(8, 32)
+VARIANCES(32, 8)
+#endif
+
 GET_VAR(16, 16)
 GET_VAR(8, 8)
 
@@ -271,33 +283,66 @@ void aom_comp_avg_pred_c(uint8_t *comp_pred, const uint8_t *pred, int width,
 
 // Get pred block from up-sampled reference.
 void aom_upsampled_pred_c(uint8_t *comp_pred, int width, int height,
-                          const uint8_t *ref, int ref_stride) {
-  int i, j, k;
-  int stride = ref_stride << 3;
-
-  for (i = 0; i < height; i++) {
-    for (j = 0, k = 0; j < width; j++, k += 8) {
-      comp_pred[j] = ref[k];
+                          int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
+                          int ref_stride) {
+  if (!subpel_x_q3 && !subpel_y_q3) {
+    int i;
+    for (i = 0; i < height; i++) {
+      memcpy(comp_pred, ref, width * sizeof(*comp_pred));
+      comp_pred += width;
+      ref += ref_stride;
+    }
+  } else {
+    InterpFilterParams filter;
+    filter = av1_get_interp_filter_params(EIGHTTAP_REGULAR);
+    if (!subpel_y_q3) {
+      const int16_t *kernel;
+      kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+      /*Directly call C version to allow this to work for small (2x2) sizes.*/
+      aom_convolve8_horiz_c(ref, ref_stride, comp_pred, width, kernel, 16, NULL,
+                            -1, width, height);
+    } else if (!subpel_x_q3) {
+      const int16_t *kernel;
+      kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+      /*Directly call C version to allow this to work for small (2x2) sizes.*/
+      aom_convolve8_vert_c(ref, ref_stride, comp_pred, width, NULL, -1, kernel,
+                           16, width, height);
+    } else {
+      DECLARE_ALIGNED(16, uint8_t,
+                      temp[((MAX_SB_SIZE * 2 + 16) + 16) * MAX_SB_SIZE]);
+      const int16_t *kernel_x;
+      const int16_t *kernel_y;
+      int intermediate_height;
+      kernel_x = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+      kernel_y = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+      intermediate_height =
+          (((height - 1) * 8 + subpel_y_q3) >> 3) + filter.taps;
+      assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
+      /*Directly call C versions to allow this to work for small (2x2) sizes.*/
+      aom_convolve8_horiz_c(ref - ref_stride * ((filter.taps >> 1) - 1),
+                            ref_stride, temp, MAX_SB_SIZE, kernel_x, 16, NULL,
+                            -1, width, intermediate_height);
+      aom_convolve8_vert_c(temp + MAX_SB_SIZE * ((filter.taps >> 1) - 1),
+                           MAX_SB_SIZE, comp_pred, width, NULL, -1, kernel_y,
+                           16, width, height);
     }
-    comp_pred += width;
-    ref += stride;
   }
 }
 
 void aom_comp_avg_upsampled_pred_c(uint8_t *comp_pred, const uint8_t *pred,
-                                   int width, int height, const uint8_t *ref,
+                                   int width, int height, int subpel_x_q3,
+                                   int subpel_y_q3, const uint8_t *ref,
                                    int ref_stride) {
   int i, j;
-  int stride = ref_stride << 3;
 
+  aom_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3, ref,
+                     ref_stride);
   for (i = 0; i < height; i++) {
     for (j = 0; j < width; j++) {
-      const int tmp = ref[(j << 3)] + pred[j];
-      comp_pred[j] = ROUND_POWER_OF_TWO(tmp, 1);
+      comp_pred[j] = ROUND_POWER_OF_TWO(comp_pred[j] + pred[j], 1);
     }
     comp_pred += width;
     pred += width;
-    ref += stride;
   }
 }
 
@@ -611,6 +656,13 @@ HIGHBD_VARIANCES(4, 2)
 HIGHBD_VARIANCES(2, 4)
 HIGHBD_VARIANCES(2, 2)
 
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+HIGHBD_VARIANCES(4, 16)
+HIGHBD_VARIANCES(16, 4)
+HIGHBD_VARIANCES(8, 32)
+HIGHBD_VARIANCES(32, 8)
+#endif
+
 HIGHBD_GET_VAR(8)
 HIGHBD_GET_VAR(16)
 
@@ -637,37 +689,74 @@ void aom_highbd_comp_avg_pred_c(uint16_t *comp_pred, const uint8_t *pred8,
 }
 
 void aom_highbd_upsampled_pred_c(uint16_t *comp_pred, int width, int height,
-                                 const uint8_t *ref8, int ref_stride) {
-  int i, j;
-  int stride = ref_stride << 3;
-
-  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
-  for (i = 0; i < height; ++i) {
-    for (j = 0; j < width; ++j) {
-      comp_pred[j] = ref[(j << 3)];
+                                 int subpel_x_q3, int subpel_y_q3,
+                                 const uint8_t *ref8, int ref_stride, int bd) {
+  if (!subpel_x_q3 && !subpel_y_q3) {
+    const uint16_t *ref;
+    int i;
+    ref = CONVERT_TO_SHORTPTR(ref8);
+    for (i = 0; i < height; i++) {
+      memcpy(comp_pred, ref, width * sizeof(*comp_pred));
+      comp_pred += width;
+      ref += ref_stride;
+    }
+  } else {
+    InterpFilterParams filter;
+    filter = av1_get_interp_filter_params(EIGHTTAP_REGULAR);
+    if (!subpel_y_q3) {
+      const int16_t *kernel;
+      kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+      /*Directly call C version to allow this to work for small (2x2) sizes.*/
+      aom_highbd_convolve8_horiz_c(ref8, ref_stride,
+                                   CONVERT_TO_BYTEPTR(comp_pred), width, kernel,
+                                   16, NULL, -1, width, height, bd);
+    } else if (!subpel_x_q3) {
+      const int16_t *kernel;
+      kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+      /*Directly call C version to allow this to work for small (2x2) sizes.*/
+      aom_highbd_convolve8_vert_c(ref8, ref_stride,
+                                  CONVERT_TO_BYTEPTR(comp_pred), width, NULL,
+                                  -1, kernel, 16, width, height, bd);
+    } else {
+      DECLARE_ALIGNED(16, uint16_t,
+                      temp[((MAX_SB_SIZE + 16) + 16) * MAX_SB_SIZE]);
+      const int16_t *kernel_x;
+      const int16_t *kernel_y;
+      int intermediate_height;
+      kernel_x = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+      kernel_y = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+      intermediate_height =
+          (((height - 1) * 8 + subpel_y_q3) >> 3) + filter.taps;
+      assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
+      /*Directly call C versions to allow this to work for small (2x2) sizes.*/
+      aom_highbd_convolve8_horiz_c(ref8 - ref_stride * ((filter.taps >> 1) - 1),
+                                   ref_stride, CONVERT_TO_BYTEPTR(temp),
+                                   MAX_SB_SIZE, kernel_x, 16, NULL, -1, width,
+                                   intermediate_height, bd);
+      aom_highbd_convolve8_vert_c(
+          CONVERT_TO_BYTEPTR(temp + MAX_SB_SIZE * ((filter.taps >> 1) - 1)),
+          MAX_SB_SIZE, CONVERT_TO_BYTEPTR(comp_pred), width, NULL, -1, kernel_y,
+          16, width, height, bd);
     }
-    comp_pred += width;
-    ref += stride;
   }
 }
 
 void aom_highbd_comp_avg_upsampled_pred_c(uint16_t *comp_pred,
                                           const uint8_t *pred8, int width,
-                                          int height, const uint8_t *ref8,
-                                          int ref_stride) {
+                                          int height, int subpel_x_q3,
+                                          int subpel_y_q3, const uint8_t *ref8,
+                                          int ref_stride, int bd) {
   int i, j;
-  int stride = ref_stride << 3;
 
-  uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
-  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
+  const uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
+  aom_highbd_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3,
+                            ref8, ref_stride, bd);
   for (i = 0; i < height; ++i) {
     for (j = 0; j < width; ++j) {
-      const int tmp = pred[j] + ref[(j << 3)];
-      comp_pred[j] = ROUND_POWER_OF_TWO(tmp, 1);
+      comp_pred[j] = ROUND_POWER_OF_TWO(pred[j] + comp_pred[j], 1);
     }
     comp_pred += width;
     pred += width;
-    ref += stride;
   }
 }
 #endif  // CONFIG_HIGHBITDEPTH
@@ -694,22 +783,23 @@ void aom_comp_mask_pred_c(uint8_t *comp_pred, const uint8_t *pred, int width,
 }
 
 void aom_comp_mask_upsampled_pred_c(uint8_t *comp_pred, const uint8_t *pred,
-                                    int width, int height, const uint8_t *ref,
+                                    int width, int height, int subpel_x_q3,
+                                    int subpel_y_q3, const uint8_t *ref,
                                     int ref_stride, const uint8_t *mask,
                                     int mask_stride, int invert_mask) {
   int i, j;
-  int stride = ref_stride << 3;
 
+  aom_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3, ref,
+                     ref_stride);
   for (i = 0; i < height; i++) {
     for (j = 0; j < width; j++) {
       if (!invert_mask)
-        comp_pred[j] = AOM_BLEND_A64(mask[j], ref[(j << 3)], pred[j]);
+        comp_pred[j] = AOM_BLEND_A64(mask[j], comp_pred[j], pred[j]);
       else
-        comp_pred[j] = AOM_BLEND_A64(mask[j], pred[j], ref[(j << 3)]);
+        comp_pred[j] = AOM_BLEND_A64(mask[j], pred[j], comp_pred[j]);
     }
     comp_pred += width;
     pred += width;
-    ref += stride;
     mask += mask_stride;
   }
 }
@@ -753,6 +843,13 @@ MASK_SUBPIX_VAR(128, 64)
 MASK_SUBPIX_VAR(128, 128)
 #endif  // CONFIG_EXT_PARTITION
 
+#if CONFIG_EXT_PARTITION_TYPES
+MASK_SUBPIX_VAR(4, 16)
+MASK_SUBPIX_VAR(16, 4)
+MASK_SUBPIX_VAR(8, 32)
+MASK_SUBPIX_VAR(32, 8)
+#endif
+
 #if CONFIG_HIGHBITDEPTH
 void aom_highbd_comp_mask_pred_c(uint16_t *comp_pred, const uint8_t *pred8,
                                  int width, int height, const uint8_t *ref8,
@@ -775,26 +872,24 @@ void aom_highbd_comp_mask_pred_c(uint16_t *comp_pred, const uint8_t *pred8,
   }
 }
 
-void aom_highbd_comp_mask_upsampled_pred_c(uint16_t *comp_pred,
-                                           const uint8_t *pred8, int width,
-                                           int height, const uint8_t *ref8,
-                                           int ref_stride, const uint8_t *mask,
-                                           int mask_stride, int invert_mask) {
+void aom_highbd_comp_mask_upsampled_pred_c(
+    uint16_t *comp_pred, const uint8_t *pred8, int width, int height,
+    int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8, int ref_stride,
+    const uint8_t *mask, int mask_stride, int invert_mask, int bd) {
   int i, j;
-  int stride = ref_stride << 3;
 
   uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
-  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
+  aom_highbd_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3,
+                            ref8, ref_stride, bd);
   for (i = 0; i < height; ++i) {
     for (j = 0; j < width; ++j) {
       if (!invert_mask)
-        comp_pred[j] = AOM_BLEND_A64(mask[j], ref[j << 3], pred[j]);
+        comp_pred[j] = AOM_BLEND_A64(mask[j], comp_pred[j], pred[j]);
       else
-        comp_pred[j] = AOM_BLEND_A64(mask[j], pred[j], ref[j << 3]);
+        comp_pred[j] = AOM_BLEND_A64(mask[j], pred[j], comp_pred[j]);
     }
     comp_pred += width;
     pred += width;
-    ref += stride;
     mask += mask_stride;
   }
 }
@@ -884,6 +979,13 @@ HIGHBD_MASK_SUBPIX_VAR(64, 128)
 HIGHBD_MASK_SUBPIX_VAR(128, 64)
 HIGHBD_MASK_SUBPIX_VAR(128, 128)
 #endif  // CONFIG_EXT_PARTITION
+
+#if CONFIG_EXT_PARTITION_TYPES
+HIGHBD_MASK_SUBPIX_VAR(4, 16)
+HIGHBD_MASK_SUBPIX_VAR(16, 4)
+HIGHBD_MASK_SUBPIX_VAR(8, 32)
+HIGHBD_MASK_SUBPIX_VAR(32, 8)
+#endif
 #endif  // CONFIG_HIGHBITDEPTH
 #endif  // CONFIG_AV1 && CONFIG_EXT_INTER
 
@@ -983,6 +1085,17 @@ OBMC_VAR(128, 128)
 OBMC_SUBPIX_VAR(128, 128)
 #endif  // CONFIG_EXT_PARTITION
 
+#if CONFIG_EXT_PARTITION_TYPES
+OBMC_VAR(4, 16)
+OBMC_SUBPIX_VAR(4, 16)
+OBMC_VAR(16, 4)
+OBMC_SUBPIX_VAR(16, 4)
+OBMC_VAR(8, 32)
+OBMC_SUBPIX_VAR(8, 32)
+OBMC_VAR(32, 8)
+OBMC_SUBPIX_VAR(32, 8)
+#endif
+
 #if CONFIG_HIGHBITDEPTH
 static INLINE void highbd_obmc_variance64(const uint8_t *pre8, int pre_stride,
                                           const int32_t *wsrc,
@@ -1164,5 +1277,16 @@ HIGHBD_OBMC_SUBPIX_VAR(128, 64)
 HIGHBD_OBMC_VAR(128, 128)
 HIGHBD_OBMC_SUBPIX_VAR(128, 128)
 #endif  // CONFIG_EXT_PARTITION
+
+#if CONFIG_EXT_PARTITION_TYPES
+HIGHBD_OBMC_VAR(4, 16)
+HIGHBD_OBMC_SUBPIX_VAR(4, 16)
+HIGHBD_OBMC_VAR(16, 4)
+HIGHBD_OBMC_SUBPIX_VAR(16, 4)
+HIGHBD_OBMC_VAR(8, 32)
+HIGHBD_OBMC_SUBPIX_VAR(8, 32)
+HIGHBD_OBMC_VAR(32, 8)
+HIGHBD_OBMC_SUBPIX_VAR(32, 8)
+#endif
 #endif  // CONFIG_HIGHBITDEPTH
 #endif  // CONFIG_AV1 && CONFIG_MOTION_VAR
diff --git a/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.c b/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.c
index a337e618d..657dcfa22 100644
--- a/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.c
+++ b/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.c
@@ -85,147 +85,6 @@ void aom_fdct8x8_1_sse2(const int16_t *input, tran_low_t *output, int stride) {
   output[0] = (tran_low_t)_mm_cvtsi128_si32(in1);
 }
 
-void aom_fdct16x16_1_sse2(const int16_t *input, tran_low_t *output,
-                          int stride) {
-  __m128i in0, in1, in2, in3;
-  __m128i u0, u1;
-  __m128i sum = _mm_setzero_si128();
-  int i;
-
-  for (i = 0; i < 2; ++i) {
-    in0 = _mm_load_si128((const __m128i *)(input + 0 * stride + 0));
-    in1 = _mm_load_si128((const __m128i *)(input + 0 * stride + 8));
-    in2 = _mm_load_si128((const __m128i *)(input + 1 * stride + 0));
-    in3 = _mm_load_si128((const __m128i *)(input + 1 * stride + 8));
-
-    u0 = _mm_add_epi16(in0, in1);
-    u1 = _mm_add_epi16(in2, in3);
-    sum = _mm_add_epi16(sum, u0);
-
-    in0 = _mm_load_si128((const __m128i *)(input + 2 * stride + 0));
-    in1 = _mm_load_si128((const __m128i *)(input + 2 * stride + 8));
-    in2 = _mm_load_si128((const __m128i *)(input + 3 * stride + 0));
-    in3 = _mm_load_si128((const __m128i *)(input + 3 * stride + 8));
-
-    sum = _mm_add_epi16(sum, u1);
-    u0 = _mm_add_epi16(in0, in1);
-    u1 = _mm_add_epi16(in2, in3);
-    sum = _mm_add_epi16(sum, u0);
-
-    in0 = _mm_load_si128((const __m128i *)(input + 4 * stride + 0));
-    in1 = _mm_load_si128((const __m128i *)(input + 4 * stride + 8));
-    in2 = _mm_load_si128((const __m128i *)(input + 5 * stride + 0));
-    in3 = _mm_load_si128((const __m128i *)(input + 5 * stride + 8));
-
-    sum = _mm_add_epi16(sum, u1);
-    u0 = _mm_add_epi16(in0, in1);
-    u1 = _mm_add_epi16(in2, in3);
-    sum = _mm_add_epi16(sum, u0);
-
-    in0 = _mm_load_si128((const __m128i *)(input + 6 * stride + 0));
-    in1 = _mm_load_si128((const __m128i *)(input + 6 * stride + 8));
-    in2 = _mm_load_si128((const __m128i *)(input + 7 * stride + 0));
-    in3 = _mm_load_si128((const __m128i *)(input + 7 * stride + 8));
-
-    sum = _mm_add_epi16(sum, u1);
-    u0 = _mm_add_epi16(in0, in1);
-    u1 = _mm_add_epi16(in2, in3);
-    sum = _mm_add_epi16(sum, u0);
-
-    sum = _mm_add_epi16(sum, u1);
-    input += 8 * stride;
-  }
-
-  u0 = _mm_setzero_si128();
-  in0 = _mm_unpacklo_epi16(u0, sum);
-  in1 = _mm_unpackhi_epi16(u0, sum);
-  in0 = _mm_srai_epi32(in0, 16);
-  in1 = _mm_srai_epi32(in1, 16);
-
-  sum = _mm_add_epi32(in0, in1);
-  in0 = _mm_unpacklo_epi32(sum, u0);
-  in1 = _mm_unpackhi_epi32(sum, u0);
-
-  sum = _mm_add_epi32(in0, in1);
-  in0 = _mm_srli_si128(sum, 8);
-
-  in1 = _mm_add_epi32(sum, in0);
-  in1 = _mm_srai_epi32(in1, 1);
-  output[0] = (tran_low_t)_mm_cvtsi128_si32(in1);
-}
-
-void aom_fdct32x32_1_sse2(const int16_t *input, tran_low_t *output,
-                          int stride) {
-  __m128i in0, in1, in2, in3;
-  __m128i u0, u1;
-  __m128i sum = _mm_setzero_si128();
-  int i;
-
-  for (i = 0; i < 8; ++i) {
-    in0 = _mm_load_si128((const __m128i *)(input + 0));
-    in1 = _mm_load_si128((const __m128i *)(input + 8));
-    in2 = _mm_load_si128((const __m128i *)(input + 16));
-    in3 = _mm_load_si128((const __m128i *)(input + 24));
-
-    input += stride;
-    u0 = _mm_add_epi16(in0, in1);
-    u1 = _mm_add_epi16(in2, in3);
-    sum = _mm_add_epi16(sum, u0);
-
-    in0 = _mm_load_si128((const __m128i *)(input + 0));
-    in1 = _mm_load_si128((const __m128i *)(input + 8));
-    in2 = _mm_load_si128((const __m128i *)(input + 16));
-    in3 = _mm_load_si128((const __m128i *)(input + 24));
-
-    input += stride;
-    sum = _mm_add_epi16(sum, u1);
-    u0 = _mm_add_epi16(in0, in1);
-    u1 = _mm_add_epi16(in2, in3);
-    sum = _mm_add_epi16(sum, u0);
-
-    in0 = _mm_load_si128((const __m128i *)(input + 0));
-    in1 = _mm_load_si128((const __m128i *)(input + 8));
-    in2 = _mm_load_si128((const __m128i *)(input + 16));
-    in3 = _mm_load_si128((const __m128i *)(input + 24));
-
-    input += stride;
-    sum = _mm_add_epi16(sum, u1);
-    u0 = _mm_add_epi16(in0, in1);
-    u1 = _mm_add_epi16(in2, in3);
-    sum = _mm_add_epi16(sum, u0);
-
-    in0 = _mm_load_si128((const __m128i *)(input + 0));
-    in1 = _mm_load_si128((const __m128i *)(input + 8));
-    in2 = _mm_load_si128((const __m128i *)(input + 16));
-    in3 = _mm_load_si128((const __m128i *)(input + 24));
-
-    input += stride;
-    sum = _mm_add_epi16(sum, u1);
-    u0 = _mm_add_epi16(in0, in1);
-    u1 = _mm_add_epi16(in2, in3);
-    sum = _mm_add_epi16(sum, u0);
-
-    sum = _mm_add_epi16(sum, u1);
-  }
-
-  u0 = _mm_setzero_si128();
-  in0 = _mm_unpacklo_epi16(u0, sum);
-  in1 = _mm_unpackhi_epi16(u0, sum);
-  in0 = _mm_srai_epi32(in0, 16);
-  in1 = _mm_srai_epi32(in1, 16);
-
-  sum = _mm_add_epi32(in0, in1);
-  in0 = _mm_unpacklo_epi32(sum, u0);
-  in1 = _mm_unpackhi_epi32(sum, u0);
-
-  sum = _mm_add_epi32(in0, in1);
-  in0 = _mm_srli_si128(sum, 8);
-
-  in1 = _mm_add_epi32(sum, in0);
-  in1 = _mm_srai_epi32(in1, 3);
-  output[0] = (tran_low_t)_mm_cvtsi128_si32(in1);
-}
-
 #define DCT_HIGH_BIT_DEPTH 0
 #define FDCT4x4_2D aom_fdct4x4_sse2
 #define FDCT8x8_2D aom_fdct8x8_sse2
diff --git a/third_party/aom/aom_dsp/x86/highbd_quantize_intrin_avx2.c b/third_party/aom/aom_dsp/x86/highbd_quantize_intrin_avx2.c
new file mode 100644
index 000000000..2bbf15ef2
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/highbd_quantize_intrin_avx2.c
@@ -0,0 +1,171 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <immintrin.h>
+
+#include "./aom_dsp_rtcd.h"
+#include "aom/aom_integer.h"
+
+static INLINE void init_one_qp(const __m128i *p, __m256i *qp) {
+  const __m128i sign = _mm_srai_epi16(*p, 15);
+  const __m128i dc = _mm_unpacklo_epi16(*p, sign);
+  const __m128i ac = _mm_unpackhi_epi16(*p, sign);
+  *qp = _mm256_insertf128_si256(_mm256_castsi128_si256(dc), ac, 1);
+}
+
+static INLINE void update_qp(__m256i *qp) {
+  int i;
+  for (i = 0; i < 5; ++i) {
+    qp[i] = _mm256_permute2x128_si256(qp[i], qp[i], 0x11);
+  }
+}
+
+static INLINE void init_qp(const int16_t *zbin_ptr, const int16_t *round_ptr,
+                           const int16_t *quant_ptr, const int16_t *dequant_ptr,
+                           const int16_t *quant_shift_ptr, __m256i *qp) {
+  const __m128i zbin = _mm_loadu_si128((const __m128i *)zbin_ptr);
+  const __m128i round = _mm_loadu_si128((const __m128i *)round_ptr);
+  const __m128i quant = _mm_loadu_si128((const __m128i *)quant_ptr);
+  const __m128i dequant = _mm_loadu_si128((const __m128i *)dequant_ptr);
+  const __m128i quant_shift = _mm_loadu_si128((const __m128i *)quant_shift_ptr);
+  init_one_qp(&zbin, &qp[0]);
+  init_one_qp(&round, &qp[1]);
+  init_one_qp(&quant, &qp[2]);
+  init_one_qp(&dequant, &qp[3]);
+  init_one_qp(&quant_shift, &qp[4]);
+}
+
+// Note:
+// *x is vector multiplied by *y which is 16 int32_t parallel multiplication
+// and right shift 16.  The output, 16 int32_t is save in *p.
+static INLINE void mm256_mul_shift_epi32(const __m256i *x, const __m256i *y,
+                                         __m256i *p) {
+  __m256i prod_lo = _mm256_mul_epi32(*x, *y);
+  __m256i prod_hi = _mm256_srli_epi64(*x, 32);
+  const __m256i mult_hi = _mm256_srli_epi64(*y, 32);
+  prod_hi = _mm256_mul_epi32(prod_hi, mult_hi);
+
+  prod_lo = _mm256_srli_epi64(prod_lo, 16);
+  const __m256i mask = _mm256_set_epi32(0, -1, 0, -1, 0, -1, 0, -1);
+  prod_lo = _mm256_and_si256(prod_lo, mask);
+  prod_hi = _mm256_srli_epi64(prod_hi, 16);
+
+  prod_hi = _mm256_slli_epi64(prod_hi, 32);
+  *p = _mm256_or_si256(prod_lo, prod_hi);
+}
+
+static INLINE void quantize(const __m256i *qp, __m256i *c,
+                            const int16_t *iscan_ptr, tran_low_t *qcoeff,
+                            tran_low_t *dqcoeff, __m256i *eob) {
+  const __m256i abs = _mm256_abs_epi32(*c);
+  const __m256i flag1 = _mm256_cmpgt_epi32(abs, qp[0]);
+  __m256i flag2 = _mm256_cmpeq_epi32(abs, qp[0]);
+  flag2 = _mm256_or_si256(flag1, flag2);
+  const int32_t nzflag = _mm256_movemask_epi8(flag2);
+
+  if (LIKELY(nzflag)) {
+    __m256i q = _mm256_add_epi32(abs, qp[1]);
+    __m256i tmp;
+    mm256_mul_shift_epi32(&q, &qp[2], &tmp);
+    q = _mm256_add_epi32(tmp, q);
+
+    mm256_mul_shift_epi32(&q, &qp[4], &q);
+    __m256i dq = _mm256_mullo_epi32(q, qp[3]);
+
+    q = _mm256_sign_epi32(q, *c);
+    dq = _mm256_sign_epi32(dq, *c);
+    q = _mm256_and_si256(q, flag2);
+    dq = _mm256_and_si256(dq, flag2);
+
+    _mm256_storeu_si256((__m256i *)qcoeff, q);
+    _mm256_storeu_si256((__m256i *)dqcoeff, dq);
+
+    const __m128i isc = _mm_loadu_si128((const __m128i *)iscan_ptr);
+    const __m128i zr = _mm_setzero_si128();
+    const __m128i lo = _mm_unpacklo_epi16(isc, zr);
+    const __m128i hi = _mm_unpackhi_epi16(isc, zr);
+    const __m256i iscan =
+        _mm256_insertf128_si256(_mm256_castsi128_si256(lo), hi, 1);
+
+    const __m256i zero = _mm256_setzero_si256();
+    const __m256i zc = _mm256_cmpeq_epi32(dq, zero);
+    const __m256i nz = _mm256_cmpeq_epi32(zc, zero);
+    __m256i cur_eob = _mm256_sub_epi32(iscan, nz);
+    cur_eob = _mm256_and_si256(cur_eob, nz);
+    *eob = _mm256_max_epi32(cur_eob, *eob);
+  } else {
+    const __m256i zero = _mm256_setzero_si256();
+    _mm256_storeu_si256((__m256i *)qcoeff, zero);
+    _mm256_storeu_si256((__m256i *)dqcoeff, zero);
+  }
+}
+
+void aom_highbd_quantize_b_avx2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
+                                int skip_block, const int16_t *zbin_ptr,
+                                const int16_t *round_ptr,
+                                const int16_t *quant_ptr,
+                                const int16_t *quant_shift_ptr,
+                                tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
+                                const int16_t *dequant_ptr, uint16_t *eob_ptr,
+                                const int16_t *scan, const int16_t *iscan) {
+  (void)scan;
+  const unsigned int step = 8;
+
+  if (LIKELY(!skip_block)) {
+    __m256i qp[5], coeff;
+    init_qp(zbin_ptr, round_ptr, quant_ptr, dequant_ptr, quant_shift_ptr, qp);
+    coeff = _mm256_loadu_si256((const __m256i *)coeff_ptr);
+
+    __m256i eob = _mm256_setzero_si256();
+    quantize(qp, &coeff, iscan, qcoeff_ptr, dqcoeff_ptr, &eob);
+
+    coeff_ptr += step;
+    qcoeff_ptr += step;
+    dqcoeff_ptr += step;
+    iscan += step;
+    n_coeffs -= step;
+
+    update_qp(qp);
+
+    while (n_coeffs > 0) {
+      coeff = _mm256_loadu_si256((const __m256i *)coeff_ptr);
+      quantize(qp, &coeff, iscan, qcoeff_ptr, dqcoeff_ptr, &eob);
+
+      coeff_ptr += step;
+      qcoeff_ptr += step;
+      dqcoeff_ptr += step;
+      iscan += step;
+      n_coeffs -= step;
+    }
+    {
+      __m256i eob_s;
+      eob_s = _mm256_shuffle_epi32(eob, 0xe);
+      eob = _mm256_max_epi16(eob, eob_s);
+      eob_s = _mm256_shufflelo_epi16(eob, 0xe);
+      eob = _mm256_max_epi16(eob, eob_s);
+      eob_s = _mm256_shufflelo_epi16(eob, 1);
+      eob = _mm256_max_epi16(eob, eob_s);
+      const __m128i final_eob = _mm_max_epi16(_mm256_castsi256_si128(eob),
+                                              _mm256_extractf128_si256(eob, 1));
+      *eob_ptr = _mm_extract_epi16(final_eob, 0);
+    }
+  } else {
+    do {
+      const __m256i zero = _mm256_setzero_si256();
+      _mm256_storeu_si256((__m256i *)qcoeff_ptr, zero);
+      _mm256_storeu_si256((__m256i *)dqcoeff_ptr, zero);
+      qcoeff_ptr += step;
+      dqcoeff_ptr += step;
+      n_coeffs -= step;
+    } while (n_coeffs > 0);
+    *eob_ptr = 0;
+  }
+}
diff --git a/third_party/aom/aom_dsp/x86/highbd_quantize_intrin_sse2.c b/third_party/aom/aom_dsp/x86/highbd_quantize_intrin_sse2.c
index 3ee24ab16..5570ca5b7 100644
--- a/third_party/aom/aom_dsp/x86/highbd_quantize_intrin_sse2.c
+++ b/third_party/aom/aom_dsp/x86/highbd_quantize_intrin_sse2.c
@@ -15,7 +15,6 @@
 #include "aom_mem/aom_mem.h"
 #include "aom_ports/mem.h"
 
-#if CONFIG_HIGHBITDEPTH
 void aom_highbd_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t count,
                                 int skip_block, const int16_t *zbin_ptr,
                                 const int16_t *round_ptr,
@@ -152,4 +151,3 @@ void aom_highbd_quantize_b_32x32_sse2(
   }
   *eob_ptr = eob + 1;
 }
-#endif
diff --git a/third_party/aom/aom_dsp/x86/highbd_sad4d_sse2.asm b/third_party/aom/aom_dsp/x86/highbd_sad4d_sse2.asm
index 0c7cb3998..9c3bbdd69 100644
--- a/third_party/aom/aom_dsp/x86/highbd_sad4d_sse2.asm
+++ b/third_party/aom/aom_dsp/x86/highbd_sad4d_sse2.asm
@@ -288,3 +288,9 @@ HIGH_SADNXN4D  8,  8
 HIGH_SADNXN4D  8,  4
 HIGH_SADNXN4D  4,  8
 HIGH_SADNXN4D  4,  4
+%if CONFIG_EXT_PARTITION_TYPES
+HIGH_SADNXN4D  4, 16
+HIGH_SADNXN4D 16,  4
+HIGH_SADNXN4D  8, 32
+HIGH_SADNXN4D 32,  8
+%endif
diff --git a/third_party/aom/aom_dsp/x86/highbd_sad_sse2.asm b/third_party/aom/aom_dsp/x86/highbd_sad_sse2.asm
index 8427b891c..248b98ef5 100644
--- a/third_party/aom/aom_dsp/x86/highbd_sad_sse2.asm
+++ b/third_party/aom/aom_dsp/x86/highbd_sad_sse2.asm
@@ -227,6 +227,10 @@ HIGH_SAD32XN 16 ; highbd_sad32x16_sse2
 HIGH_SAD32XN 64, 1 ; highbd_sad32x64_avg_sse2
 HIGH_SAD32XN 32, 1 ; highbd_sad32x32_avg_sse2
 HIGH_SAD32XN 16, 1 ; highbd_sad32x16_avg_sse2
+%if CONFIG_EXT_PARTITION_TYPES
+HIGH_SAD32XN 8 ; highbd_sad_32x8_sse2
+HIGH_SAD32XN 8, 1 ; highbd_sad_32x8_avg_sse2
+%endif
 
 ; unsigned int aom_highbd_sad16x{8,16,32}_sse2(uint8_t *src, int src_stride,
 ;                                    uint8_t *ref, int ref_stride);
@@ -295,7 +299,10 @@ HIGH_SAD16XN  8 ; highbd_sad16x8_sse2
 HIGH_SAD16XN 32, 1 ; highbd_sad16x32_avg_sse2
 HIGH_SAD16XN 16, 1 ; highbd_sad16x16_avg_sse2
 HIGH_SAD16XN  8, 1 ; highbd_sad16x8_avg_sse2
-
+%if CONFIG_EXT_PARTITION_TYPES
+HIGH_SAD16XN 4 ; highbd_sad_16x4_sse2
+HIGH_SAD16XN 4, 1 ; highbd_sad_16x4_avg_sse2
+%endif
 
 ; unsigned int aom_highbd_sad8x{4,8,16}_sse2(uint8_t *src, int src_stride,
 ;                                    uint8_t *ref, int ref_stride);
@@ -364,3 +371,7 @@ HIGH_SAD8XN  4 ; highbd_sad8x4_sse2
 HIGH_SAD8XN 16, 1 ; highbd_sad8x16_avg_sse2
 HIGH_SAD8XN  8, 1 ; highbd_sad8x8_avg_sse2
 HIGH_SAD8XN  4, 1 ; highbd_sad8x4_avg_sse2
+%if CONFIG_EXT_PARTITION_TYPES
+HIGH_SAD8XN 32 ; highbd_sad_8x32_sse2
+HIGH_SAD8XN 32, 1 ; highbd_sad_8x32_avg_sse2
+%endif
diff --git a/third_party/aom/aom_dsp/x86/highbd_subpel_variance_impl_sse2.asm b/third_party/aom/aom_dsp/x86/highbd_subpel_variance_impl_sse2.asm
index 797e9c1d4..ee19796e3 100644
--- a/third_party/aom/aom_dsp/x86/highbd_subpel_variance_impl_sse2.asm
+++ b/third_party/aom/aom_dsp/x86/highbd_subpel_variance_impl_sse2.asm
@@ -75,7 +75,7 @@ SECTION .text
   paddd                m6, m4
   mov                  r1, ssem         ; r1 = unsigned int *sse
   movd               [r1], m7           ; store sse
-  movd                rax, m6           ; store sum as return value
+  movd                eax, m6           ; store sum as return value
 %endif
   RET
 %endmacro
diff --git a/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c b/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c
index 29f96ce24..93923ffb0 100644
--- a/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c
+++ b/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c
@@ -9,6 +9,7 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
+#include <assert.h>
 #include <emmintrin.h>  // SSE2
 
 #include "./aom_config.h"
@@ -16,6 +17,9 @@
 
 #include "aom_ports/mem.h"
 
+#include "./av1_rtcd.h"
+#include "av1/common/filter.h"
+
 typedef uint32_t (*high_variance_fn_t)(const uint16_t *src, int src_stride,
                                        const uint16_t *ref, int ref_stride,
                                        uint32_t *sse, int *sum);
@@ -181,6 +185,11 @@ VAR_FN(16, 16, 16, 8);
 VAR_FN(16, 8, 8, 7);
 VAR_FN(8, 16, 8, 7);
 VAR_FN(8, 8, 8, 6);
+#if CONFIG_EXT_PARTITION_TYPES
+VAR_FN(16, 4, 16, 6);
+VAR_FN(8, 32, 8, 8);
+VAR_FN(32, 8, 16, 8);
+#endif
 
 #undef VAR_FN
 
@@ -387,6 +396,7 @@ DECLS(sse2);
     return (var >= 0) ? (uint32_t)var : 0;                                     \
   }
 
+#if CONFIG_EXT_PARTITION_TYPES
 #define FNS(opt)                        \
   FN(64, 64, 16, 6, 6, opt, (int64_t)); \
   FN(64, 32, 16, 6, 5, opt, (int64_t)); \
@@ -398,7 +408,24 @@ DECLS(sse2);
   FN(16, 8, 16, 4, 3, opt, (int64_t));  \
   FN(8, 16, 8, 3, 4, opt, (int64_t));   \
   FN(8, 8, 8, 3, 3, opt, (int64_t));    \
-  FN(8, 4, 8, 3, 2, opt, (int64_t));
+  FN(8, 4, 8, 3, 2, opt, (int64_t));    \
+  FN(16, 4, 16, 4, 2, opt, (int64_t));  \
+  FN(8, 32, 8, 3, 5, opt, (int64_t));   \
+  FN(32, 8, 16, 5, 3, opt, (int64_t))
+#else
+#define FNS(opt)                        \
+  FN(64, 64, 16, 6, 6, opt, (int64_t)); \
+  FN(64, 32, 16, 6, 5, opt, (int64_t)); \
+  FN(32, 64, 16, 5, 6, opt, (int64_t)); \
+  FN(32, 32, 16, 5, 5, opt, (int64_t)); \
+  FN(32, 16, 16, 5, 4, opt, (int64_t)); \
+  FN(16, 32, 16, 4, 5, opt, (int64_t)); \
+  FN(16, 16, 16, 4, 4, opt, (int64_t)); \
+  FN(16, 8, 16, 4, 3, opt, (int64_t));  \
+  FN(8, 16, 8, 3, 4, opt, (int64_t));   \
+  FN(8, 8, 8, 3, 3, opt, (int64_t));    \
+  FN(8, 4, 8, 3, 2, opt, (int64_t))
+#endif
 
 FNS(sse2);
 
@@ -412,9 +439,9 @@ FNS(sse2);
       const uint16_t *dst, ptrdiff_t dst_stride, const uint16_t *sec,        \
       ptrdiff_t sec_stride, int height, unsigned int *sse, void *unused0,    \
       void *unused);
-#define DECLS(opt1) \
-  DECL(16, opt1)    \
-  DECL(8, opt1)
+#define DECLS(opt) \
+  DECL(16, opt)    \
+  DECL(8, opt)
 
 DECLS(sse2);
 #undef DECL
@@ -546,18 +573,36 @@ DECLS(sse2);
     return (var >= 0) ? (uint32_t)var : 0;                                     \
   }
 
-#define FNS(opt1)                        \
-  FN(64, 64, 16, 6, 6, opt1, (int64_t)); \
-  FN(64, 32, 16, 6, 5, opt1, (int64_t)); \
-  FN(32, 64, 16, 5, 6, opt1, (int64_t)); \
-  FN(32, 32, 16, 5, 5, opt1, (int64_t)); \
-  FN(32, 16, 16, 5, 4, opt1, (int64_t)); \
-  FN(16, 32, 16, 4, 5, opt1, (int64_t)); \
-  FN(16, 16, 16, 4, 4, opt1, (int64_t)); \
-  FN(16, 8, 16, 4, 3, opt1, (int64_t));  \
-  FN(8, 16, 8, 4, 3, opt1, (int64_t));   \
-  FN(8, 8, 8, 3, 3, opt1, (int64_t));    \
-  FN(8, 4, 8, 3, 2, opt1, (int64_t));
+#if CONFIG_EXT_PARTITION_TYPES
+#define FNS(opt)                        \
+  FN(64, 64, 16, 6, 6, opt, (int64_t)); \
+  FN(64, 32, 16, 6, 5, opt, (int64_t)); \
+  FN(32, 64, 16, 5, 6, opt, (int64_t)); \
+  FN(32, 32, 16, 5, 5, opt, (int64_t)); \
+  FN(32, 16, 16, 5, 4, opt, (int64_t)); \
+  FN(16, 32, 16, 4, 5, opt, (int64_t)); \
+  FN(16, 16, 16, 4, 4, opt, (int64_t)); \
+  FN(16, 8, 16, 4, 3, opt, (int64_t));  \
+  FN(8, 16, 8, 3, 4, opt, (int64_t));   \
+  FN(8, 8, 8, 3, 3, opt, (int64_t));    \
+  FN(8, 4, 8, 3, 2, opt, (int64_t));    \
+  FN(16, 4, 16, 4, 2, opt, (int64_t));  \
+  FN(8, 32, 8, 3, 5, opt, (int64_t));   \
+  FN(32, 8, 16, 5, 3, opt, (int64_t));
+#else
+#define FNS(opt)                        \
+  FN(64, 64, 16, 6, 6, opt, (int64_t)); \
+  FN(64, 32, 16, 6, 5, opt, (int64_t)); \
+  FN(32, 64, 16, 5, 6, opt, (int64_t)); \
+  FN(32, 32, 16, 5, 5, opt, (int64_t)); \
+  FN(32, 16, 16, 5, 4, opt, (int64_t)); \
+  FN(16, 32, 16, 4, 5, opt, (int64_t)); \
+  FN(16, 16, 16, 4, 4, opt, (int64_t)); \
+  FN(16, 8, 16, 4, 3, opt, (int64_t));  \
+  FN(8, 16, 8, 3, 4, opt, (int64_t));   \
+  FN(8, 8, 8, 3, 3, opt, (int64_t));    \
+  FN(8, 4, 8, 3, 2, opt, (int64_t));
+#endif
 
 FNS(sse2);
 
@@ -565,131 +610,94 @@ FNS(sse2);
 #undef FN
 
 void aom_highbd_upsampled_pred_sse2(uint16_t *comp_pred, int width, int height,
-                                    const uint8_t *ref8, int ref_stride) {
-  int i, j;
-  int stride = ref_stride << 3;
-  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
-
-  if (width >= 8) {
-    // read 8 points at one time
-    for (i = 0; i < height; i++) {
-      for (j = 0; j < width; j += 8) {
-        __m128i s0 = _mm_cvtsi32_si128(*(const uint32_t *)ref);
-        __m128i s1 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 8));
-        __m128i s2 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 16));
-        __m128i s3 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 24));
-        __m128i s4 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 32));
-        __m128i s5 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 40));
-        __m128i s6 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 48));
-        __m128i s7 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 56));
-        __m128i t0, t1, t2, t3;
-
-        t0 = _mm_unpacklo_epi16(s0, s1);
-        t1 = _mm_unpacklo_epi16(s2, s3);
-        t2 = _mm_unpacklo_epi16(s4, s5);
-        t3 = _mm_unpacklo_epi16(s6, s7);
-        t0 = _mm_unpacklo_epi32(t0, t1);
-        t2 = _mm_unpacklo_epi32(t2, t3);
-        t0 = _mm_unpacklo_epi64(t0, t2);
-
-        _mm_storeu_si128((__m128i *)(comp_pred), t0);
+                                    int subpel_x_q3, int subpel_y_q3,
+                                    const uint8_t *ref8, int ref_stride,
+                                    int bd) {
+  if (!subpel_x_q3 && !subpel_y_q3) {
+    uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
+    if (width >= 8) {
+      int i;
+      assert(!(width & 7));
+      /*Read 8 pixels one row at a time.*/
+      for (i = 0; i < height; i++) {
+        int j;
+        for (j = 0; j < width; j += 8) {
+          __m128i s0 = _mm_loadu_si128((const __m128i *)ref);
+          _mm_storeu_si128((__m128i *)comp_pred, s0);
+          comp_pred += 8;
+          ref += 8;
+        }
+        ref += ref_stride - width;
+      }
+    } else {
+      int i;
+      assert(!(width & 3));
+      /*Read 4 pixels two rows at a time.*/
+      for (i = 0; i < height; i += 2) {
+        __m128i s0 = _mm_loadl_epi64((const __m128i *)ref);
+        __m128i s1 = _mm_loadl_epi64((const __m128i *)(ref + ref_stride));
+        __m128i t0 = _mm_unpacklo_epi64(s0, s1);
+        _mm_storeu_si128((__m128i *)comp_pred, t0);
         comp_pred += 8;
-        ref += 64;  // 8 * 8;
+        ref += 2 * ref_stride;
       }
-      ref += stride - (width << 3);
     }
   } else {
-    // read 4 points at one time
-    for (i = 0; i < height; i++) {
-      for (j = 0; j < width; j += 4) {
-        __m128i s0 = _mm_cvtsi32_si128(*(const uint32_t *)ref);
-        __m128i s1 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 8));
-        __m128i s2 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 16));
-        __m128i s3 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 24));
-        __m128i t0, t1;
-
-        t0 = _mm_unpacklo_epi16(s0, s1);
-        t1 = _mm_unpacklo_epi16(s2, s3);
-        t0 = _mm_unpacklo_epi32(t0, t1);
-
-        _mm_storel_epi64((__m128i *)(comp_pred), t0);
-        comp_pred += 4;
-        ref += 4 * 8;
-      }
-      ref += stride - (width << 3);
+    InterpFilterParams filter;
+    filter = av1_get_interp_filter_params(EIGHTTAP_REGULAR);
+    if (!subpel_y_q3) {
+      const int16_t *kernel;
+      kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+      aom_highbd_convolve8_horiz(ref8, ref_stride,
+                                 CONVERT_TO_BYTEPTR(comp_pred), width, kernel,
+                                 16, NULL, -1, width, height, bd);
+    } else if (!subpel_x_q3) {
+      const int16_t *kernel;
+      kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+      aom_highbd_convolve8_vert(ref8, ref_stride, CONVERT_TO_BYTEPTR(comp_pred),
+                                width, NULL, -1, kernel, 16, width, height, bd);
+    } else {
+      DECLARE_ALIGNED(16, uint16_t,
+                      temp[((MAX_SB_SIZE + 16) + 16) * MAX_SB_SIZE]);
+      const int16_t *kernel_x;
+      const int16_t *kernel_y;
+      int intermediate_height;
+      kernel_x = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+      kernel_y = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+      intermediate_height =
+          (((height - 1) * 8 + subpel_y_q3) >> 3) + filter.taps;
+      assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
+      aom_highbd_convolve8_horiz(ref8 - ref_stride * ((filter.taps >> 1) - 1),
+                                 ref_stride, CONVERT_TO_BYTEPTR(temp),
+                                 MAX_SB_SIZE, kernel_x, 16, NULL, -1, width,
+                                 intermediate_height, bd);
+      aom_highbd_convolve8_vert(
+          CONVERT_TO_BYTEPTR(temp + MAX_SB_SIZE * ((filter.taps >> 1) - 1)),
+          MAX_SB_SIZE, CONVERT_TO_BYTEPTR(comp_pred), width, NULL, -1, kernel_y,
+          16, width, height, bd);
     }
   }
 }
 
 void aom_highbd_comp_avg_upsampled_pred_sse2(uint16_t *comp_pred,
                                              const uint8_t *pred8, int width,
-                                             int height, const uint8_t *ref8,
-                                             int ref_stride) {
-  const __m128i one = _mm_set1_epi16(1);
-  int i, j;
-  int stride = ref_stride << 3;
+                                             int height, int subpel_x_q3,
+                                             int subpel_y_q3,
+                                             const uint8_t *ref8,
+                                             int ref_stride, int bd) {
   uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
-  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
-
-  if (width >= 8) {
-    // read 8 points at one time
-    for (i = 0; i < height; i++) {
-      for (j = 0; j < width; j += 8) {
-        __m128i s0 = _mm_cvtsi32_si128(*(const uint32_t *)ref);
-        __m128i s1 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 8));
-        __m128i s2 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 16));
-        __m128i s3 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 24));
-        __m128i s4 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 32));
-        __m128i s5 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 40));
-        __m128i s6 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 48));
-        __m128i s7 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 56));
-        __m128i p0 = _mm_loadu_si128((const __m128i *)pred);
-        __m128i t0, t1, t2, t3;
-
-        t0 = _mm_unpacklo_epi16(s0, s1);
-        t1 = _mm_unpacklo_epi16(s2, s3);
-        t2 = _mm_unpacklo_epi16(s4, s5);
-        t3 = _mm_unpacklo_epi16(s6, s7);
-        t0 = _mm_unpacklo_epi32(t0, t1);
-        t2 = _mm_unpacklo_epi32(t2, t3);
-        t0 = _mm_unpacklo_epi64(t0, t2);
-
-        p0 = _mm_adds_epu16(t0, p0);
-        p0 = _mm_adds_epu16(p0, one);
-        p0 = _mm_srli_epi16(p0, 1);
-
-        _mm_storeu_si128((__m128i *)(comp_pred), p0);
-        comp_pred += 8;
-        pred += 8;
-        ref += 8 * 8;
-      }
-      ref += stride - (width << 3);
-    }
-  } else {
-    // read 4 points at one time
-    for (i = 0; i < height; i++) {
-      for (j = 0; j < width; j += 4) {
-        __m128i s0 = _mm_cvtsi32_si128(*(const uint32_t *)ref);
-        __m128i s1 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 8));
-        __m128i s2 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 16));
-        __m128i s3 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + 24));
-        __m128i p0 = _mm_loadl_epi64((const __m128i *)pred);
-        __m128i t0, t1;
-
-        t0 = _mm_unpacklo_epi16(s0, s1);
-        t1 = _mm_unpacklo_epi16(s2, s3);
-        t0 = _mm_unpacklo_epi32(t0, t1);
-
-        p0 = _mm_adds_epu16(t0, p0);
-        p0 = _mm_adds_epu16(p0, one);
-        p0 = _mm_srli_epi16(p0, 1);
-
-        _mm_storel_epi64((__m128i *)(comp_pred), p0);
-        comp_pred += 4;
-        pred += 4;
-        ref += 4 * 8;
-      }
-      ref += stride - (width << 3);
-    }
+  int n;
+  int i;
+  aom_highbd_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3,
+                            ref8, ref_stride, bd);
+  /*The total number of pixels must be a multiple of 8 (e.g., 4x4).*/
+  assert(!(width * height & 7));
+  n = width * height >> 3;
+  for (i = 0; i < n; i++) {
+    __m128i s0 = _mm_loadu_si128((const __m128i *)comp_pred);
+    __m128i p0 = _mm_loadu_si128((const __m128i *)pred);
+    _mm_storeu_si128((__m128i *)comp_pred, _mm_avg_epu16(s0, p0));
+    comp_pred += 8;
+    pred += 8;
   }
 }
diff --git a/third_party/aom/aom_dsp/x86/inv_txfm_sse2.c b/third_party/aom/aom_dsp/x86/inv_txfm_sse2.c
index be200df4c..86ce928b7 100644
--- a/third_party/aom/aom_dsp/x86/inv_txfm_sse2.c
+++ b/third_party/aom/aom_dsp/x86/inv_txfm_sse2.c
@@ -3498,237 +3498,3 @@ void idct32_8col(__m128i *in0, __m128i *in1) {
   in1[14] = _mm_sub_epi16(stp1_1, stp1_30);
   in1[15] = _mm_sub_epi16(stp1_0, stp1_31);
 }
-
-#if CONFIG_HIGHBITDEPTH
-static INLINE __m128i clamp_high_sse2(__m128i value, int bd) {
-  __m128i ubounded, retval;
-  const __m128i zero = _mm_set1_epi16(0);
-  const __m128i one = _mm_set1_epi16(1);
-  const __m128i max = _mm_subs_epi16(_mm_slli_epi16(one, bd), one);
-  ubounded = _mm_cmpgt_epi16(value, max);
-  retval = _mm_andnot_si128(ubounded, value);
-  ubounded = _mm_and_si128(ubounded, max);
-  retval = _mm_or_si128(retval, ubounded);
-  retval = _mm_and_si128(retval, _mm_cmpgt_epi16(retval, zero));
-  return retval;
-}
-
-void aom_highbd_idct4x4_16_add_sse2(const tran_low_t *input, uint8_t *dest8,
-                                    int stride, int bd) {
-  tran_low_t out[4 * 4];
-  tran_low_t *outptr = out;
-  int i, j;
-  __m128i inptr[4];
-  __m128i sign_bits[2];
-  __m128i temp_mm, min_input, max_input;
-  int test;
-  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
-  int optimised_cols = 0;
-  const __m128i zero = _mm_set1_epi16(0);
-  const __m128i eight = _mm_set1_epi16(8);
-  const __m128i max = _mm_set1_epi16(12043);
-  const __m128i min = _mm_set1_epi16(-12043);
-  // Load input into __m128i
-  inptr[0] = _mm_loadu_si128((const __m128i *)input);
-  inptr[1] = _mm_loadu_si128((const __m128i *)(input + 4));
-  inptr[2] = _mm_loadu_si128((const __m128i *)(input + 8));
-  inptr[3] = _mm_loadu_si128((const __m128i *)(input + 12));
-
-  // Pack to 16 bits
-  inptr[0] = _mm_packs_epi32(inptr[0], inptr[1]);
-  inptr[1] = _mm_packs_epi32(inptr[2], inptr[3]);
-
-  max_input = _mm_max_epi16(inptr[0], inptr[1]);
-  min_input = _mm_min_epi16(inptr[0], inptr[1]);
-  max_input = _mm_cmpgt_epi16(max_input, max);
-  min_input = _mm_cmplt_epi16(min_input, min);
-  temp_mm = _mm_or_si128(max_input, min_input);
-  test = _mm_movemask_epi8(temp_mm);
-
-  if (!test) {
-    // Do the row transform
-    aom_idct4_sse2(inptr);
-
-    // Check the min & max values
-    max_input = _mm_max_epi16(inptr[0], inptr[1]);
-    min_input = _mm_min_epi16(inptr[0], inptr[1]);
-    max_input = _mm_cmpgt_epi16(max_input, max);
-    min_input = _mm_cmplt_epi16(min_input, min);
-    temp_mm = _mm_or_si128(max_input, min_input);
-    test = _mm_movemask_epi8(temp_mm);
-
-    if (test) {
-      array_transpose_4x4(inptr);
-      sign_bits[0] = _mm_cmplt_epi16(inptr[0], zero);
-      sign_bits[1] = _mm_cmplt_epi16(inptr[1], zero);
-      inptr[3] = _mm_unpackhi_epi16(inptr[1], sign_bits[1]);
-      inptr[2] = _mm_unpacklo_epi16(inptr[1], sign_bits[1]);
-      inptr[1] = _mm_unpackhi_epi16(inptr[0], sign_bits[0]);
-      inptr[0] = _mm_unpacklo_epi16(inptr[0], sign_bits[0]);
-      _mm_storeu_si128((__m128i *)outptr, inptr[0]);
-      _mm_storeu_si128((__m128i *)(outptr + 4), inptr[1]);
-      _mm_storeu_si128((__m128i *)(outptr + 8), inptr[2]);
-      _mm_storeu_si128((__m128i *)(outptr + 12), inptr[3]);
-    } else {
-      // Set to use the optimised transform for the column
-      optimised_cols = 1;
-    }
-  } else {
-    // Run the un-optimised row transform
-    for (i = 0; i < 4; ++i) {
-      aom_highbd_idct4_c(input, outptr, bd);
-      input += 4;
-      outptr += 4;
-    }
-  }
-
-  if (optimised_cols) {
-    aom_idct4_sse2(inptr);
-
-    // Final round and shift
-    inptr[0] = _mm_add_epi16(inptr[0], eight);
-    inptr[1] = _mm_add_epi16(inptr[1], eight);
-
-    inptr[0] = _mm_srai_epi16(inptr[0], 4);
-    inptr[1] = _mm_srai_epi16(inptr[1], 4);
-
-    // Reconstruction and Store
-    {
-      __m128i d0 = _mm_loadl_epi64((const __m128i *)dest);
-      __m128i d2 = _mm_loadl_epi64((const __m128i *)(dest + stride * 2));
-      d0 = _mm_unpacklo_epi64(
-          d0, _mm_loadl_epi64((const __m128i *)(dest + stride)));
-      d2 = _mm_unpacklo_epi64(
-          d2, _mm_loadl_epi64((const __m128i *)(dest + stride * 3)));
-      d0 = clamp_high_sse2(_mm_adds_epi16(d0, inptr[0]), bd);
-      d2 = clamp_high_sse2(_mm_adds_epi16(d2, inptr[1]), bd);
-      // store input0
-      _mm_storel_epi64((__m128i *)dest, d0);
-      // store input1
-      d0 = _mm_srli_si128(d0, 8);
-      _mm_storel_epi64((__m128i *)(dest + stride), d0);
-      // store input2
-      _mm_storel_epi64((__m128i *)(dest + stride * 2), d2);
-      // store input3
-      d2 = _mm_srli_si128(d2, 8);
-      _mm_storel_epi64((__m128i *)(dest + stride * 3), d2);
-    }
-  } else {
-    // Run the un-optimised column transform
-    tran_low_t temp_in[4], temp_out[4];
-    // Columns
-    for (i = 0; i < 4; ++i) {
-      for (j = 0; j < 4; ++j) temp_in[j] = out[j * 4 + i];
-      aom_highbd_idct4_c(temp_in, temp_out, bd);
-      for (j = 0; j < 4; ++j) {
-        dest[j * stride + i] = highbd_clip_pixel_add(
-            dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd);
-      }
-    }
-  }
-}
-
-void aom_highbd_idct8x8_10_add_sse2(const tran_low_t *input, uint8_t *dest8,
-                                    int stride, int bd) {
-  tran_low_t out[8 * 8] = { 0 };
-  tran_low_t *outptr = out;
-  int i, j, test;
-  __m128i inptr[8];
-  __m128i min_input, max_input, temp1, temp2, sign_bits;
-  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
-  const __m128i zero = _mm_set1_epi16(0);
-  const __m128i sixteen = _mm_set1_epi16(16);
-  const __m128i max = _mm_set1_epi16(6201);
-  const __m128i min = _mm_set1_epi16(-6201);
-  int optimised_cols = 0;
-
-  // Load input into __m128i & pack to 16 bits
-  for (i = 0; i < 8; i++) {
-    temp1 = _mm_loadu_si128((const __m128i *)(input + 8 * i));
-    temp2 = _mm_loadu_si128((const __m128i *)(input + 8 * i + 4));
-    inptr[i] = _mm_packs_epi32(temp1, temp2);
-  }
-
-  // Find the min & max for the row transform
-  // only first 4 row has non-zero coefs
-  max_input = _mm_max_epi16(inptr[0], inptr[1]);
-  min_input = _mm_min_epi16(inptr[0], inptr[1]);
-  for (i = 2; i < 4; i++) {
-    max_input = _mm_max_epi16(max_input, inptr[i]);
-    min_input = _mm_min_epi16(min_input, inptr[i]);
-  }
-  max_input = _mm_cmpgt_epi16(max_input, max);
-  min_input = _mm_cmplt_epi16(min_input, min);
-  temp1 = _mm_or_si128(max_input, min_input);
-  test = _mm_movemask_epi8(temp1);
-
-  if (!test) {
-    // Do the row transform
-    aom_idct8_sse2(inptr);
-
-    // Find the min & max for the column transform
-    // N.B. Only first 4 cols contain non-zero coeffs
-    max_input = _mm_max_epi16(inptr[0], inptr[1]);
-    min_input = _mm_min_epi16(inptr[0], inptr[1]);
-    for (i = 2; i < 8; i++) {
-      max_input = _mm_max_epi16(max_input, inptr[i]);
-      min_input = _mm_min_epi16(min_input, inptr[i]);
-    }
-    max_input = _mm_cmpgt_epi16(max_input, max);
-    min_input = _mm_cmplt_epi16(min_input, min);
-    temp1 = _mm_or_si128(max_input, min_input);
-    test = _mm_movemask_epi8(temp1);
-
-    if (test) {
-      // Use fact only first 4 rows contain non-zero coeffs
-      array_transpose_4X8(inptr, inptr);
-      for (i = 0; i < 4; i++) {
-        sign_bits = _mm_cmplt_epi16(inptr[i], zero);
-        temp1 = _mm_unpackhi_epi16(inptr[i], sign_bits);
-        temp2 = _mm_unpacklo_epi16(inptr[i], sign_bits);
-        _mm_storeu_si128((__m128i *)(outptr + 4 * (2 * i + 1)), temp1);
-        _mm_storeu_si128((__m128i *)(outptr + 4 * (2 * i)), temp2);
-      }
-    } else {
-      // Set to use the optimised transform for the column
-      optimised_cols = 1;
-    }
-  } else {
-    // Run the un-optimised row transform
-    for (i = 0; i < 4; ++i) {
-      aom_highbd_idct8_c(input, outptr, bd);
-      input += 8;
-      outptr += 8;
-    }
-  }
-
-  if (optimised_cols) {
-    aom_idct8_sse2(inptr);
-
-    // Final round & shift and Reconstruction and Store
-    {
-      __m128i d[8];
-      for (i = 0; i < 8; i++) {
-        inptr[i] = _mm_add_epi16(inptr[i], sixteen);
-        d[i] = _mm_loadu_si128((const __m128i *)(dest + stride * i));
-        inptr[i] = _mm_srai_epi16(inptr[i], 5);
-        d[i] = clamp_high_sse2(_mm_adds_epi16(d[i], inptr[i]), bd);
-        // Store
-        _mm_storeu_si128((__m128i *)(dest + stride * i), d[i]);
-      }
-    }
-  } else {
-    // Run the un-optimised column transform
-    tran_low_t temp_in[8], temp_out[8];
-    for (i = 0; i < 8; ++i) {
-      for (j = 0; j < 8; ++j) temp_in[j] = out[j * 8 + i];
-      aom_highbd_idct8_c(temp_in, temp_out, bd);
-      for (j = 0; j < 8; ++j) {
-        dest[j * stride + i] = highbd_clip_pixel_add(
-            dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd);
-      }
-    }
-  }
-}
-
-#endif  // CONFIG_HIGHBITDEPTH
diff --git a/third_party/aom/aom_dsp/x86/masked_sad_intrin_ssse3.c b/third_party/aom/aom_dsp/x86/masked_sad_intrin_ssse3.c
index 9d16a3e84..6a73ac460 100644
--- a/third_party/aom/aom_dsp/x86/masked_sad_intrin_ssse3.c
+++ b/third_party/aom/aom_dsp/x86/masked_sad_intrin_ssse3.c
@@ -93,6 +93,12 @@ MASKSAD8XN_SSSE3(8)
 MASKSAD8XN_SSSE3(4)
 MASKSAD4XN_SSSE3(8)
 MASKSAD4XN_SSSE3(4)
+#if CONFIG_EXT_PARTITION_TYPES
+MASKSAD4XN_SSSE3(16)
+MASKSADMXN_SSSE3(16, 4)
+MASKSAD8XN_SSSE3(32)
+MASKSADMXN_SSSE3(32, 8)
+#endif
 
 static INLINE unsigned int masked_sad_ssse3(const uint8_t *src_ptr,
                                             int src_stride,
@@ -283,6 +289,12 @@ HIGHBD_MASKSADMXN_SSSE3(8, 8)
 HIGHBD_MASKSADMXN_SSSE3(8, 4)
 HIGHBD_MASKSAD4XN_SSSE3(8)
 HIGHBD_MASKSAD4XN_SSSE3(4)
+#if CONFIG_EXT_PARTITION_TYPES
+HIGHBD_MASKSAD4XN_SSSE3(16)
+HIGHBD_MASKSADMXN_SSSE3(16, 4)
+HIGHBD_MASKSADMXN_SSSE3(8, 32)
+HIGHBD_MASKSADMXN_SSSE3(32, 8)
+#endif
 
 static INLINE unsigned int highbd_masked_sad_ssse3(
     const uint8_t *src8, int src_stride, const uint8_t *a8, int a_stride,
diff --git a/third_party/aom/aom_dsp/x86/masked_variance_intrin_ssse3.c b/third_party/aom/aom_dsp/x86/masked_variance_intrin_ssse3.c
index be9d437d2..24e7ed1c6 100644
--- a/third_party/aom/aom_dsp/x86/masked_variance_intrin_ssse3.c
+++ b/third_party/aom/aom_dsp/x86/masked_variance_intrin_ssse3.c
@@ -126,6 +126,12 @@ MASK_SUBPIX_VAR8XH_SSSE3(8)
 MASK_SUBPIX_VAR8XH_SSSE3(4)
 MASK_SUBPIX_VAR4XH_SSSE3(8)
 MASK_SUBPIX_VAR4XH_SSSE3(4)
+#if CONFIG_EXT_PARTITION_TYPES
+MASK_SUBPIX_VAR4XH_SSSE3(16)
+MASK_SUBPIX_VAR_SSSE3(16, 4)
+MASK_SUBPIX_VAR8XH_SSSE3(32)
+MASK_SUBPIX_VAR_SSSE3(32, 8)
+#endif
 
 static INLINE __m128i filter_block(const __m128i a, const __m128i b,
                                    const __m128i filter) {
@@ -564,6 +570,7 @@ static void highbd_masked_variance4xh(const uint16_t *src_ptr, int src_stride,
       const uint8_t *msk, int msk_stride, int invert_mask, uint32_t *sse) { \
     uint64_t sse64;                                                         \
     int sum;                                                                \
+    int64_t var;                                                            \
     uint16_t temp[(H + 1) * W];                                             \
     const uint16_t *src = CONVERT_TO_SHORTPTR(src8);                        \
     const uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);                        \
@@ -579,7 +586,8 @@ static void highbd_masked_variance4xh(const uint16_t *src_ptr, int src_stride,
                              msk_stride, W, H, &sse64, &sum);               \
     *sse = (uint32_t)ROUND_POWER_OF_TWO(sse64, 4);                          \
     sum = ROUND_POWER_OF_TWO(sum, 2);                                       \
-    return *sse - (uint32_t)(((int64_t)sum * sum) / (W * H));               \
+    var = (int64_t)(*sse) - (((int64_t)sum * sum) / (W * H));               \
+    return (var >= 0) ? (uint32_t)var : 0;                                  \
   }                                                                         \
   unsigned int aom_highbd_12_masked_sub_pixel_variance##W##x##H##_ssse3(    \
       const uint8_t *src8, int src_stride, int xoffset, int yoffset,        \
@@ -587,6 +595,7 @@ static void highbd_masked_variance4xh(const uint16_t *src_ptr, int src_stride,
       const uint8_t *msk, int msk_stride, int invert_mask, uint32_t *sse) { \
     uint64_t sse64;                                                         \
     int sum;                                                                \
+    int64_t var;                                                            \
     uint16_t temp[(H + 1) * W];                                             \
     const uint16_t *src = CONVERT_TO_SHORTPTR(src8);                        \
     const uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);                        \
@@ -602,7 +611,8 @@ static void highbd_masked_variance4xh(const uint16_t *src_ptr, int src_stride,
                              msk_stride, W, H, &sse64, &sum);               \
     *sse = (uint32_t)ROUND_POWER_OF_TWO(sse64, 8);                          \
     sum = ROUND_POWER_OF_TWO(sum, 4);                                       \
-    return *sse - (uint32_t)(((int64_t)sum * sum) / (W * H));               \
+    var = (int64_t)(*sse) - (((int64_t)sum * sum) / (W * H));               \
+    return (var >= 0) ? (uint32_t)var : 0;                                  \
   }
 
 #define HIGHBD_MASK_SUBPIX_VAR4XH_SSSE3(H)                                  \
@@ -634,6 +644,7 @@ static void highbd_masked_variance4xh(const uint16_t *src_ptr, int src_stride,
       const uint8_t *msk, int msk_stride, int invert_mask, uint32_t *sse) { \
     int sse_;                                                               \
     int sum;                                                                \
+    int64_t var;                                                            \
     uint16_t temp[(H + 1) * 4];                                             \
     const uint16_t *src = CONVERT_TO_SHORTPTR(src8);                        \
     const uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);                        \
@@ -649,7 +660,8 @@ static void highbd_masked_variance4xh(const uint16_t *src_ptr, int src_stride,
                                 msk_stride, H, &sse_, &sum);                \
     *sse = (uint32_t)ROUND_POWER_OF_TWO(sse_, 4);                           \
     sum = ROUND_POWER_OF_TWO(sum, 2);                                       \
-    return *sse - (uint32_t)(((int64_t)sum * sum) / (4 * H));               \
+    var = (int64_t)(*sse) - (((int64_t)sum * sum) / (4 * H));               \
+    return (var >= 0) ? (uint32_t)var : 0;                                  \
   }                                                                         \
   unsigned int aom_highbd_12_masked_sub_pixel_variance4x##H##_ssse3(        \
       const uint8_t *src8, int src_stride, int xoffset, int yoffset,        \
@@ -657,6 +669,7 @@ static void highbd_masked_variance4xh(const uint16_t *src_ptr, int src_stride,
       const uint8_t *msk, int msk_stride, int invert_mask, uint32_t *sse) { \
     int sse_;                                                               \
     int sum;                                                                \
+    int64_t var;                                                            \
     uint16_t temp[(H + 1) * 4];                                             \
     const uint16_t *src = CONVERT_TO_SHORTPTR(src8);                        \
     const uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);                        \
@@ -672,7 +685,8 @@ static void highbd_masked_variance4xh(const uint16_t *src_ptr, int src_stride,
                                 msk_stride, H, &sse_, &sum);                \
     *sse = (uint32_t)ROUND_POWER_OF_TWO(sse_, 8);                           \
     sum = ROUND_POWER_OF_TWO(sum, 4);                                       \
-    return *sse - (uint32_t)(((int64_t)sum * sum) / (4 * H));               \
+    var = (int64_t)(*sse) - (((int64_t)sum * sum) / (4 * H));               \
+    return (var >= 0) ? (uint32_t)var : 0;                                  \
   }
 
 #if CONFIG_EXT_PARTITION
@@ -693,6 +707,12 @@ HIGHBD_MASK_SUBPIX_VAR_SSSE3(8, 8)
 HIGHBD_MASK_SUBPIX_VAR_SSSE3(8, 4)
 HIGHBD_MASK_SUBPIX_VAR4XH_SSSE3(8)
 HIGHBD_MASK_SUBPIX_VAR4XH_SSSE3(4)
+#if CONFIG_EXT_PARTITION_TYPES
+HIGHBD_MASK_SUBPIX_VAR4XH_SSSE3(16)
+HIGHBD_MASK_SUBPIX_VAR_SSSE3(16, 4)
+HIGHBD_MASK_SUBPIX_VAR_SSSE3(8, 32)
+HIGHBD_MASK_SUBPIX_VAR_SSSE3(32, 8)
+#endif
 
 static INLINE __m128i highbd_filter_block(const __m128i a, const __m128i b,
                                           const __m128i filter) {
diff --git a/third_party/aom/aom_dsp/x86/obmc_sad_sse4.c b/third_party/aom/aom_dsp/x86/obmc_sad_sse4.c
index 21632644f..3fd6f71e5 100644
--- a/third_party/aom/aom_dsp/x86/obmc_sad_sse4.c
+++ b/third_party/aom/aom_dsp/x86/obmc_sad_sse4.c
@@ -137,6 +137,12 @@ OBMCSADWXH(8, 8)
 OBMCSADWXH(8, 4)
 OBMCSADWXH(4, 8)
 OBMCSADWXH(4, 4)
+#if CONFIG_EXT_PARTITION_TYPES
+OBMCSADWXH(4, 16)
+OBMCSADWXH(16, 4)
+OBMCSADWXH(8, 32)
+OBMCSADWXH(32, 8)
+#endif
 
 ////////////////////////////////////////////////////////////////////////////////
 // High bit-depth
@@ -260,4 +266,10 @@ HBD_OBMCSADWXH(8, 8)
 HBD_OBMCSADWXH(8, 4)
 HBD_OBMCSADWXH(4, 8)
 HBD_OBMCSADWXH(4, 4)
+#if CONFIG_EXT_PARTITION_TYPES
+HBD_OBMCSADWXH(4, 16)
+HBD_OBMCSADWXH(16, 4)
+HBD_OBMCSADWXH(8, 32)
+HBD_OBMCSADWXH(32, 8)
+#endif
 #endif  // CONFIG_HIGHBITDEPTH
diff --git a/third_party/aom/aom_dsp/x86/obmc_variance_sse4.c b/third_party/aom/aom_dsp/x86/obmc_variance_sse4.c
index 1797ded80..44cfa8e28 100644
--- a/third_party/aom/aom_dsp/x86/obmc_variance_sse4.c
+++ b/third_party/aom/aom_dsp/x86/obmc_variance_sse4.c
@@ -146,6 +146,12 @@ OBMCVARWXH(8, 8)
 OBMCVARWXH(8, 4)
 OBMCVARWXH(4, 8)
 OBMCVARWXH(4, 4)
+#if CONFIG_EXT_PARTITION_TYPES
+OBMCVARWXH(4, 16)
+OBMCVARWXH(16, 4)
+OBMCVARWXH(8, 32)
+OBMCVARWXH(32, 8)
+#endif
 
 ////////////////////////////////////////////////////////////////////////////////
 // High bit-depth
@@ -353,4 +359,10 @@ HBD_OBMCVARWXH(8, 8)
 HBD_OBMCVARWXH(8, 4)
 HBD_OBMCVARWXH(4, 8)
 HBD_OBMCVARWXH(4, 4)
+#if CONFIG_EXT_PARTITION_TYPES
+HBD_OBMCVARWXH(4, 16)
+HBD_OBMCVARWXH(16, 4)
+HBD_OBMCVARWXH(8, 32)
+HBD_OBMCVARWXH(32, 8)
+#endif
 #endif  // CONFIG_HIGHBITDEPTH
diff --git a/third_party/aom/aom_dsp/x86/sad4d_sse2.asm b/third_party/aom/aom_dsp/x86/sad4d_sse2.asm
index 8f04ef2f3..4570e2ce6 100644
--- a/third_party/aom/aom_dsp/x86/sad4d_sse2.asm
+++ b/third_party/aom/aom_dsp/x86/sad4d_sse2.asm
@@ -251,3 +251,9 @@ SADNXN4D  8,  8
 SADNXN4D  8,  4
 SADNXN4D  4,  8
 SADNXN4D  4,  4
+%if CONFIG_EXT_PARTITION_TYPES
+SADNXN4D  4, 16
+SADNXN4D 16,  4
+SADNXN4D  8, 32
+SADNXN4D 32,  8
+%endif
diff --git a/third_party/aom/aom_dsp/x86/sad_highbd_avx2.c b/third_party/aom/aom_dsp/x86/sad_highbd_avx2.c
index 196394379..e8dd87a26 100644
--- a/third_party/aom/aom_dsp/x86/sad_highbd_avx2.c
+++ b/third_party/aom/aom_dsp/x86/sad_highbd_avx2.c
@@ -704,7 +704,12 @@ unsigned int aom_highbd_sad128x128_avg_avx2(const uint8_t *src, int src_stride,
 static INLINE void get_4d_sad_from_mm256_epi32(const __m256i *v,
                                                uint32_t *res) {
   __m256i u0, u1, u2, u3;
+#if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
+  const __m256i mask = _mm256_setr_epi32(UINT32_MAX, 0, UINT32_MAX, 0,
+                                         UINT32_MAX, 0, UINT32_MAX, 0);
+#else
   const __m256i mask = _mm256_set1_epi64x(UINT32_MAX);
+#endif
   __m128i sad;
 
   // 8 32-bit summation
diff --git a/third_party/aom/aom_dsp/x86/sad_sse2.asm b/third_party/aom/aom_dsp/x86/sad_sse2.asm
index e45457a57..88d427077 100644
--- a/third_party/aom/aom_dsp/x86/sad_sse2.asm
+++ b/third_party/aom/aom_dsp/x86/sad_sse2.asm
@@ -208,6 +208,10 @@ SAD32XN 16 ; sad32x16_sse2
 SAD32XN 64, 1 ; sad32x64_avg_sse2
 SAD32XN 32, 1 ; sad32x32_avg_sse2
 SAD32XN 16, 1 ; sad32x16_avg_sse2
+%if CONFIG_EXT_PARTITION_TYPES
+SAD32XN 8 ; sad_32x8_sse2
+SAD32XN 8, 1 ; sad_32x8_avg_sse2
+%endif
 
 ; unsigned int aom_sad16x{8,16}_sse2(uint8_t *src, int src_stride,
 ;                                    uint8_t *ref, int ref_stride);
@@ -254,6 +258,10 @@ SAD16XN  8 ; sad16x8_sse2
 SAD16XN 32, 1 ; sad16x32_avg_sse2
 SAD16XN 16, 1 ; sad16x16_avg_sse2
 SAD16XN  8, 1 ; sad16x8_avg_sse2
+%if CONFIG_EXT_PARTITION_TYPES
+SAD16XN 4 ; sad_16x4_sse2
+SAD16XN 4, 1 ; sad_16x4_avg_sse2
+%endif
 
 ; unsigned int aom_sad8x{8,16}_sse2(uint8_t *src, int src_stride,
 ;                                   uint8_t *ref, int ref_stride);
@@ -298,6 +306,10 @@ SAD8XN  4 ; sad8x4_sse2
 SAD8XN 16, 1 ; sad8x16_avg_sse2
 SAD8XN  8, 1 ; sad8x8_avg_sse2
 SAD8XN  4, 1 ; sad8x4_avg_sse2
+%if CONFIG_EXT_PARTITION_TYPES
+SAD8XN 32 ; sad_8x32_sse2
+SAD8XN 32, 1 ; sad_8x32_avg_sse2
+%endif
 
 ; unsigned int aom_sad4x{4, 8}_sse2(uint8_t *src, int src_stride,
 ;                                   uint8_t *ref, int ref_stride);
@@ -343,3 +355,7 @@ SAD4XN  8 ; sad4x8_sse
 SAD4XN  4 ; sad4x4_sse
 SAD4XN  8, 1 ; sad4x8_avg_sse
 SAD4XN  4, 1 ; sad4x4_avg_sse
+%if CONFIG_EXT_PARTITION_TYPES
+SAD4XN 16 ; sad_4x16_sse2
+SAD4XN 16, 1 ; sad_4x16_avg_sse2
+%endif
diff --git a/third_party/aom/aom_dsp/x86/variance_sse2.c b/third_party/aom/aom_dsp/x86/variance_sse2.c
index d9563aa7f..918844185 100644
--- a/third_party/aom/aom_dsp/x86/variance_sse2.c
+++ b/third_party/aom/aom_dsp/x86/variance_sse2.c
@@ -17,6 +17,9 @@
 
 #include "aom_ports/mem.h"
 
+#include "./av1_rtcd.h"
+#include "av1/common/filter.h"
+
 typedef void (*getNxMvar_fn_t)(const unsigned char *src, int src_stride,
                                const unsigned char *ref, int ref_stride,
                                unsigned int *sse, int *sum);
@@ -335,6 +338,52 @@ unsigned int aom_mse16x16_sse2(const uint8_t *src, int src_stride,
   return *sse;
 }
 
+#if CONFIG_EXT_PARTITION_TYPES
+unsigned int aom_variance4x16_sse2(const uint8_t *src, int src_stride,
+                                   const uint8_t *ref, int ref_stride,
+                                   unsigned int *sse) {
+  int sum;
+  variance_sse2(src, src_stride, ref, ref_stride, 4, 16, sse, &sum,
+                get4x4var_sse2, 4);
+  assert(sum <= 255 * 4 * 16);
+  assert(sum >= -255 * 4 * 16);
+  return *sse - (unsigned int)(((int64_t)sum * sum) >> 6);
+}
+
+unsigned int aom_variance16x4_sse2(const uint8_t *src, int src_stride,
+                                   const uint8_t *ref, int ref_stride,
+                                   unsigned int *sse) {
+  int sum;
+  variance_sse2(src, src_stride, ref, ref_stride, 16, 4, sse, &sum,
+                get4x4var_sse2, 4);
+  assert(sum <= 255 * 16 * 4);
+  assert(sum >= -255 * 16 * 4);
+  return *sse - (unsigned int)(((int64_t)sum * sum) >> 6);
+}
+
+unsigned int aom_variance8x32_sse2(const uint8_t *src, int src_stride,
+                                   const uint8_t *ref, int ref_stride,
+                                   unsigned int *sse) {
+  int sum;
+  variance_sse2(src, src_stride, ref, ref_stride, 8, 32, sse, &sum,
+                aom_get8x8var_sse2, 8);
+  assert(sum <= 255 * 8 * 32);
+  assert(sum >= -255 * 8 * 32);
+  return *sse - (unsigned int)(((int64_t)sum * sum) >> 8);
+}
+
+unsigned int aom_variance32x8_sse2(const uint8_t *src, int src_stride,
+                                   const uint8_t *ref, int ref_stride,
+                                   unsigned int *sse) {
+  int sum;
+  variance_sse2(src, src_stride, ref, ref_stride, 32, 8, sse, &sum,
+                aom_get8x8var_sse2, 8);
+  assert(sum <= 255 * 32 * 8);
+  assert(sum >= -255 * 32 * 8);
+  return *sse - (unsigned int)(((int64_t)sum * sum) >> 8);
+}
+#endif
+
 // The 2 unused parameters are place holders for PIC enabled build.
 // These definitions are for functions defined in subpel_variance.asm
 #define DECL(w, opt)                                                           \
@@ -342,13 +391,13 @@ unsigned int aom_mse16x16_sse2(const uint8_t *src, int src_stride,
       const uint8_t *src, ptrdiff_t src_stride, int x_offset, int y_offset,    \
       const uint8_t *dst, ptrdiff_t dst_stride, int height, unsigned int *sse, \
       void *unused0, void *unused)
-#define DECLS(opt1, opt2) \
-  DECL(4, opt1);          \
-  DECL(8, opt1);          \
-  DECL(16, opt1)
+#define DECLS(opt) \
+  DECL(4, opt);    \
+  DECL(8, opt);    \
+  DECL(16, opt)
 
-DECLS(sse2, sse2);
-DECLS(ssse3, ssse3);
+DECLS(sse2);
+DECLS(ssse3);
 #undef DECLS
 #undef DECL
 
@@ -384,23 +433,44 @@ DECLS(ssse3, ssse3);
     return sse - (unsigned int)(cast_prod(cast se * se) >> (wlog2 + hlog2));   \
   }
 
-#define FNS(opt1, opt2)                              \
-  FN(64, 64, 16, 6, 6, opt1, (int64_t), (int64_t));  \
-  FN(64, 32, 16, 6, 5, opt1, (int64_t), (int64_t));  \
-  FN(32, 64, 16, 5, 6, opt1, (int64_t), (int64_t));  \
-  FN(32, 32, 16, 5, 5, opt1, (int64_t), (int64_t));  \
-  FN(32, 16, 16, 5, 4, opt1, (int64_t), (int64_t));  \
-  FN(16, 32, 16, 4, 5, opt1, (int64_t), (int64_t));  \
-  FN(16, 16, 16, 4, 4, opt1, (uint32_t), (int64_t)); \
-  FN(16, 8, 16, 4, 3, opt1, (int32_t), (int32_t));   \
-  FN(8, 16, 8, 3, 4, opt1, (int32_t), (int32_t));    \
-  FN(8, 8, 8, 3, 3, opt1, (int32_t), (int32_t));     \
-  FN(8, 4, 8, 3, 2, opt1, (int32_t), (int32_t));     \
-  FN(4, 8, 4, 2, 3, opt1, (int32_t), (int32_t));     \
-  FN(4, 4, 4, 2, 2, opt1, (int32_t), (int32_t))
-
-FNS(sse2, sse2);
-FNS(ssse3, ssse3);
+#if CONFIG_EXT_PARTITION_TYPES
+#define FNS(opt)                                    \
+  FN(64, 64, 16, 6, 6, opt, (int64_t), (int64_t));  \
+  FN(64, 32, 16, 6, 5, opt, (int64_t), (int64_t));  \
+  FN(32, 64, 16, 5, 6, opt, (int64_t), (int64_t));  \
+  FN(32, 32, 16, 5, 5, opt, (int64_t), (int64_t));  \
+  FN(32, 16, 16, 5, 4, opt, (int64_t), (int64_t));  \
+  FN(16, 32, 16, 4, 5, opt, (int64_t), (int64_t));  \
+  FN(16, 16, 16, 4, 4, opt, (uint32_t), (int64_t)); \
+  FN(16, 8, 16, 4, 3, opt, (int32_t), (int32_t));   \
+  FN(8, 16, 8, 3, 4, opt, (int32_t), (int32_t));    \
+  FN(8, 8, 8, 3, 3, opt, (int32_t), (int32_t));     \
+  FN(8, 4, 8, 3, 2, opt, (int32_t), (int32_t));     \
+  FN(4, 8, 4, 2, 3, opt, (int32_t), (int32_t));     \
+  FN(4, 4, 4, 2, 2, opt, (int32_t), (int32_t));     \
+  FN(4, 16, 4, 2, 4, opt, (int32_t), (int32_t));    \
+  FN(16, 4, 16, 4, 2, opt, (int32_t), (int32_t));   \
+  FN(8, 32, 8, 3, 5, opt, (int32_t), (int32_t));    \
+  FN(32, 8, 16, 5, 3, opt, (int32_t), (int32_t))
+#else
+#define FNS(opt)                                    \
+  FN(64, 64, 16, 6, 6, opt, (int64_t), (int64_t));  \
+  FN(64, 32, 16, 6, 5, opt, (int64_t), (int64_t));  \
+  FN(32, 64, 16, 5, 6, opt, (int64_t), (int64_t));  \
+  FN(32, 32, 16, 5, 5, opt, (int64_t), (int64_t));  \
+  FN(32, 16, 16, 5, 4, opt, (int64_t), (int64_t));  \
+  FN(16, 32, 16, 4, 5, opt, (int64_t), (int64_t));  \
+  FN(16, 16, 16, 4, 4, opt, (uint32_t), (int64_t)); \
+  FN(16, 8, 16, 4, 3, opt, (int32_t), (int32_t));   \
+  FN(8, 16, 8, 3, 4, opt, (int32_t), (int32_t));    \
+  FN(8, 8, 8, 3, 3, opt, (int32_t), (int32_t));     \
+  FN(8, 4, 8, 3, 2, opt, (int32_t), (int32_t));     \
+  FN(4, 8, 4, 2, 3, opt, (int32_t), (int32_t));     \
+  FN(4, 4, 4, 2, 2, opt, (int32_t), (int32_t))
+#endif
+
+FNS(sse2);
+FNS(ssse3);
 
 #undef FNS
 #undef FN
@@ -412,13 +482,13 @@ FNS(ssse3, ssse3);
       const uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *sec,         \
       ptrdiff_t sec_stride, int height, unsigned int *sse, void *unused0,   \
       void *unused)
-#define DECLS(opt1, opt2) \
-  DECL(4, opt1);          \
-  DECL(8, opt1);          \
-  DECL(16, opt1)
+#define DECLS(opt) \
+  DECL(4, opt);    \
+  DECL(8, opt);    \
+  DECL(16, opt)
 
-DECLS(sse2, sse2);
-DECLS(ssse3, ssse3);
+DECLS(sse2);
+DECLS(ssse3);
 #undef DECL
 #undef DECLS
 
@@ -455,236 +525,149 @@ DECLS(ssse3, ssse3);
     return sse - (unsigned int)(cast_prod(cast se * se) >> (wlog2 + hlog2));   \
   }
 
-#define FNS(opt1, opt2)                              \
-  FN(64, 64, 16, 6, 6, opt1, (int64_t), (int64_t));  \
-  FN(64, 32, 16, 6, 5, opt1, (int64_t), (int64_t));  \
-  FN(32, 64, 16, 5, 6, opt1, (int64_t), (int64_t));  \
-  FN(32, 32, 16, 5, 5, opt1, (int64_t), (int64_t));  \
-  FN(32, 16, 16, 5, 4, opt1, (int64_t), (int64_t));  \
-  FN(16, 32, 16, 4, 5, opt1, (int64_t), (int64_t));  \
-  FN(16, 16, 16, 4, 4, opt1, (uint32_t), (int64_t)); \
-  FN(16, 8, 16, 4, 3, opt1, (uint32_t), (int32_t));  \
-  FN(8, 16, 8, 3, 4, opt1, (uint32_t), (int32_t));   \
-  FN(8, 8, 8, 3, 3, opt1, (uint32_t), (int32_t));    \
-  FN(8, 4, 8, 3, 2, opt1, (uint32_t), (int32_t));    \
-  FN(4, 8, 4, 2, 3, opt1, (uint32_t), (int32_t));    \
-  FN(4, 4, 4, 2, 2, opt1, (uint32_t), (int32_t))
-
-FNS(sse2, sse);
-FNS(ssse3, ssse3);
+#if CONFIG_EXT_PARTITION_TYPES
+#define FNS(opt)                                    \
+  FN(64, 64, 16, 6, 6, opt, (int64_t), (int64_t));  \
+  FN(64, 32, 16, 6, 5, opt, (int64_t), (int64_t));  \
+  FN(32, 64, 16, 5, 6, opt, (int64_t), (int64_t));  \
+  FN(32, 32, 16, 5, 5, opt, (int64_t), (int64_t));  \
+  FN(32, 16, 16, 5, 4, opt, (int64_t), (int64_t));  \
+  FN(16, 32, 16, 4, 5, opt, (int64_t), (int64_t));  \
+  FN(16, 16, 16, 4, 4, opt, (uint32_t), (int64_t)); \
+  FN(16, 8, 16, 4, 3, opt, (uint32_t), (int32_t));  \
+  FN(8, 16, 8, 3, 4, opt, (uint32_t), (int32_t));   \
+  FN(8, 8, 8, 3, 3, opt, (uint32_t), (int32_t));    \
+  FN(8, 4, 8, 3, 2, opt, (uint32_t), (int32_t));    \
+  FN(4, 8, 4, 2, 3, opt, (uint32_t), (int32_t));    \
+  FN(4, 4, 4, 2, 2, opt, (uint32_t), (int32_t));    \
+  FN(4, 16, 4, 2, 4, opt, (int32_t), (int32_t));    \
+  FN(16, 4, 16, 4, 2, opt, (int32_t), (int32_t));   \
+  FN(8, 32, 8, 3, 5, opt, (int32_t), (int32_t));    \
+  FN(32, 8, 16, 5, 3, opt, (int32_t), (int32_t))
+#else
+#define FNS(opt)                                    \
+  FN(64, 64, 16, 6, 6, opt, (int64_t), (int64_t));  \
+  FN(64, 32, 16, 6, 5, opt, (int64_t), (int64_t));  \
+  FN(32, 64, 16, 5, 6, opt, (int64_t), (int64_t));  \
+  FN(32, 32, 16, 5, 5, opt, (int64_t), (int64_t));  \
+  FN(32, 16, 16, 5, 4, opt, (int64_t), (int64_t));  \
+  FN(16, 32, 16, 4, 5, opt, (int64_t), (int64_t));  \
+  FN(16, 16, 16, 4, 4, opt, (uint32_t), (int64_t)); \
+  FN(16, 8, 16, 4, 3, opt, (uint32_t), (int32_t));  \
+  FN(8, 16, 8, 3, 4, opt, (uint32_t), (int32_t));   \
+  FN(8, 8, 8, 3, 3, opt, (uint32_t), (int32_t));    \
+  FN(8, 4, 8, 3, 2, opt, (uint32_t), (int32_t));    \
+  FN(4, 8, 4, 2, 3, opt, (uint32_t), (int32_t));    \
+  FN(4, 4, 4, 2, 2, opt, (uint32_t), (int32_t))
+#endif
+
+FNS(sse2);
+FNS(ssse3);
 
 #undef FNS
 #undef FN
 
 void aom_upsampled_pred_sse2(uint8_t *comp_pred, int width, int height,
+                             int subpel_x_q3, int subpel_y_q3,
                              const uint8_t *ref, int ref_stride) {
-  int i, j;
-  int stride = ref_stride << 3;
-
-  if (width >= 16) {
-    // read 16 points at one time
-    for (i = 0; i < height; i++) {
-      for (j = 0; j < width; j += 16) {
-        __m128i s0 = _mm_loadu_si128((const __m128i *)ref);
-        __m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
-        __m128i s2 = _mm_loadu_si128((const __m128i *)(ref + 32));
-        __m128i s3 = _mm_loadu_si128((const __m128i *)(ref + 48));
-        __m128i s4 = _mm_loadu_si128((const __m128i *)(ref + 64));
-        __m128i s5 = _mm_loadu_si128((const __m128i *)(ref + 80));
-        __m128i s6 = _mm_loadu_si128((const __m128i *)(ref + 96));
-        __m128i s7 = _mm_loadu_si128((const __m128i *)(ref + 112));
-        __m128i t0, t1, t2, t3;
-
-        t0 = _mm_unpacklo_epi8(s0, s1);
-        s1 = _mm_unpackhi_epi8(s0, s1);
-        t1 = _mm_unpacklo_epi8(s2, s3);
-        s3 = _mm_unpackhi_epi8(s2, s3);
-        t2 = _mm_unpacklo_epi8(s4, s5);
-        s5 = _mm_unpackhi_epi8(s4, s5);
-        t3 = _mm_unpacklo_epi8(s6, s7);
-        s7 = _mm_unpackhi_epi8(s6, s7);
-
-        s0 = _mm_unpacklo_epi8(t0, s1);
-        s2 = _mm_unpacklo_epi8(t1, s3);
-        s4 = _mm_unpacklo_epi8(t2, s5);
-        s6 = _mm_unpacklo_epi8(t3, s7);
-        s0 = _mm_unpacklo_epi32(s0, s2);
-        s4 = _mm_unpacklo_epi32(s4, s6);
-        s0 = _mm_unpacklo_epi64(s0, s4);
-
-        _mm_storeu_si128((__m128i *)(comp_pred), s0);
+  if (!subpel_x_q3 && !subpel_y_q3) {
+    if (width >= 16) {
+      int i;
+      assert(!(width & 15));
+      /*Read 16 pixels one row at a time.*/
+      for (i = 0; i < height; i++) {
+        int j;
+        for (j = 0; j < width; j += 16) {
+          __m128i s0 = _mm_loadu_si128((const __m128i *)ref);
+          _mm_storeu_si128((__m128i *)comp_pred, s0);
+          comp_pred += 16;
+          ref += 16;
+        }
+        ref += ref_stride - width;
+      }
+    } else if (width >= 8) {
+      int i;
+      assert(!(width & 7));
+      assert(!(height & 1));
+      /*Read 8 pixels two rows at a time.*/
+      for (i = 0; i < height; i += 2) {
+        __m128i s0 = _mm_loadl_epi64((const __m128i *)ref);
+        __m128i s1 = _mm_loadl_epi64((const __m128i *)(ref + ref_stride));
+        __m128i t0 = _mm_unpacklo_epi64(s0, s1);
+        _mm_storeu_si128((__m128i *)comp_pred, t0);
         comp_pred += 16;
-        ref += 16 * 8;
+        ref += 2 * ref_stride;
       }
-      ref += stride - (width << 3);
-    }
-  } else if (width >= 8) {
-    // read 8 points at one time
-    for (i = 0; i < height; i++) {
-      for (j = 0; j < width; j += 8) {
-        __m128i s0 = _mm_loadu_si128((const __m128i *)ref);
-        __m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
-        __m128i s2 = _mm_loadu_si128((const __m128i *)(ref + 32));
-        __m128i s3 = _mm_loadu_si128((const __m128i *)(ref + 48));
-        __m128i t0, t1;
-
-        t0 = _mm_unpacklo_epi8(s0, s1);
-        s1 = _mm_unpackhi_epi8(s0, s1);
-        t1 = _mm_unpacklo_epi8(s2, s3);
-        s3 = _mm_unpackhi_epi8(s2, s3);
-
-        s0 = _mm_unpacklo_epi8(t0, s1);
-        s2 = _mm_unpacklo_epi8(t1, s3);
-        s0 = _mm_unpacklo_epi32(s0, s2);
-
-        _mm_storel_epi64((__m128i *)(comp_pred), s0);
-        comp_pred += 8;
-        ref += 8 * 8;
+    } else {
+      int i;
+      assert(!(width & 3));
+      assert(!(height & 3));
+      /*Read 4 pixels four rows at a time.*/
+      for (i = 0; i < height; i++) {
+        __m128i s0 = _mm_cvtsi32_si128(*(const uint32_t *)ref);
+        __m128i s1 = _mm_cvtsi32_si128(*(const uint32_t *)(ref + ref_stride));
+        __m128i s2 =
+            _mm_cvtsi32_si128(*(const uint32_t *)(ref + 2 * ref_stride));
+        __m128i s3 =
+            _mm_cvtsi32_si128(*(const uint32_t *)(ref + 3 * ref_stride));
+        __m128i t0 = _mm_unpacklo_epi32(s0, s1);
+        __m128i t1 = _mm_unpacklo_epi32(s2, s3);
+        __m128i u0 = _mm_unpacklo_epi64(t0, t1);
+        _mm_storeu_si128((__m128i *)comp_pred, u0);
+        comp_pred += 16;
+        ref += 4 * ref_stride;
       }
-      ref += stride - (width << 3);
     }
   } else {
-    // read 4 points at one time
-    for (i = 0; i < height; i++) {
-      for (j = 0; j < width; j += 4) {
-        __m128i s0 = _mm_loadu_si128((const __m128i *)ref);
-        __m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
-        __m128i t0;
-
-        t0 = _mm_unpacklo_epi8(s0, s1);
-        s1 = _mm_unpackhi_epi8(s0, s1);
-        s0 = _mm_unpacklo_epi8(t0, s1);
-
-        *(int *)comp_pred = _mm_cvtsi128_si32(s0);
-        comp_pred += 4;
-        ref += 4 * 8;
-      }
-      ref += stride - (width << 3);
+    InterpFilterParams filter;
+    filter = av1_get_interp_filter_params(EIGHTTAP_REGULAR);
+    if (!subpel_y_q3) {
+      const int16_t *kernel;
+      kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+      aom_convolve8_horiz(ref, ref_stride, comp_pred, width, kernel, 16, NULL,
+                          -1, width, height);
+    } else if (!subpel_x_q3) {
+      const int16_t *kernel;
+      kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+      aom_convolve8_vert(ref, ref_stride, comp_pred, width, NULL, -1, kernel,
+                         16, width, height);
+    } else {
+      DECLARE_ALIGNED(16, uint8_t,
+                      temp[((MAX_SB_SIZE * 2 + 16) + 16) * MAX_SB_SIZE]);
+      const int16_t *kernel_x;
+      const int16_t *kernel_y;
+      int intermediate_height;
+      kernel_x = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+      kernel_y = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+      intermediate_height =
+          (((height - 1) * 8 + subpel_y_q3) >> 3) + filter.taps;
+      assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
+      aom_convolve8_horiz(ref - ref_stride * ((filter.taps >> 1) - 1),
+                          ref_stride, temp, MAX_SB_SIZE, kernel_x, 16, NULL, -1,
+                          width, intermediate_height);
+      aom_convolve8_vert(temp + MAX_SB_SIZE * ((filter.taps >> 1) - 1),
+                         MAX_SB_SIZE, comp_pred, width, NULL, -1, kernel_y, 16,
+                         width, height);
     }
   }
 }
 
 void aom_comp_avg_upsampled_pred_sse2(uint8_t *comp_pred, const uint8_t *pred,
-                                      int width, int height, const uint8_t *ref,
+                                      int width, int height, int subpel_x_q3,
+                                      int subpel_y_q3, const uint8_t *ref,
                                       int ref_stride) {
-  const __m128i zero = _mm_set1_epi16(0);
-  const __m128i one = _mm_set1_epi16(1);
-  int i, j;
-  int stride = ref_stride << 3;
-
-  if (width >= 16) {
-    // read 16 points at one time
-    for (i = 0; i < height; i++) {
-      for (j = 0; j < width; j += 16) {
-        __m128i s0 = _mm_loadu_si128((const __m128i *)ref);
-        __m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
-        __m128i s2 = _mm_loadu_si128((const __m128i *)(ref + 32));
-        __m128i s3 = _mm_loadu_si128((const __m128i *)(ref + 48));
-        __m128i s4 = _mm_loadu_si128((const __m128i *)(ref + 64));
-        __m128i s5 = _mm_loadu_si128((const __m128i *)(ref + 80));
-        __m128i s6 = _mm_loadu_si128((const __m128i *)(ref + 96));
-        __m128i s7 = _mm_loadu_si128((const __m128i *)(ref + 112));
-        __m128i p0 = _mm_loadu_si128((const __m128i *)pred);
-        __m128i p1;
-        __m128i t0, t1, t2, t3;
-
-        t0 = _mm_unpacklo_epi8(s0, s1);
-        s1 = _mm_unpackhi_epi8(s0, s1);
-        t1 = _mm_unpacklo_epi8(s2, s3);
-        s3 = _mm_unpackhi_epi8(s2, s3);
-        t2 = _mm_unpacklo_epi8(s4, s5);
-        s5 = _mm_unpackhi_epi8(s4, s5);
-        t3 = _mm_unpacklo_epi8(s6, s7);
-        s7 = _mm_unpackhi_epi8(s6, s7);
-
-        s0 = _mm_unpacklo_epi8(t0, s1);
-        s2 = _mm_unpacklo_epi8(t1, s3);
-        s4 = _mm_unpacklo_epi8(t2, s5);
-        s6 = _mm_unpacklo_epi8(t3, s7);
-
-        s0 = _mm_unpacklo_epi32(s0, s2);
-        s4 = _mm_unpacklo_epi32(s4, s6);
-        s0 = _mm_unpacklo_epi8(s0, zero);
-        s4 = _mm_unpacklo_epi8(s4, zero);
-
-        p1 = _mm_unpackhi_epi8(p0, zero);
-        p0 = _mm_unpacklo_epi8(p0, zero);
-        p0 = _mm_adds_epu16(s0, p0);
-        p1 = _mm_adds_epu16(s4, p1);
-        p0 = _mm_adds_epu16(p0, one);
-        p1 = _mm_adds_epu16(p1, one);
-
-        p0 = _mm_srli_epi16(p0, 1);
-        p1 = _mm_srli_epi16(p1, 1);
-        p0 = _mm_packus_epi16(p0, p1);
-
-        _mm_storeu_si128((__m128i *)(comp_pred), p0);
-        comp_pred += 16;
-        pred += 16;
-        ref += 16 * 8;
-      }
-      ref += stride - (width << 3);
-    }
-  } else if (width >= 8) {
-    // read 8 points at one time
-    for (i = 0; i < height; i++) {
-      for (j = 0; j < width; j += 8) {
-        __m128i s0 = _mm_loadu_si128((const __m128i *)ref);
-        __m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
-        __m128i s2 = _mm_loadu_si128((const __m128i *)(ref + 32));
-        __m128i s3 = _mm_loadu_si128((const __m128i *)(ref + 48));
-        __m128i p0 = _mm_loadl_epi64((const __m128i *)pred);
-        __m128i t0, t1;
-
-        t0 = _mm_unpacklo_epi8(s0, s1);
-        s1 = _mm_unpackhi_epi8(s0, s1);
-        t1 = _mm_unpacklo_epi8(s2, s3);
-        s3 = _mm_unpackhi_epi8(s2, s3);
-
-        s0 = _mm_unpacklo_epi8(t0, s1);
-        s2 = _mm_unpacklo_epi8(t1, s3);
-        s0 = _mm_unpacklo_epi32(s0, s2);
-        s0 = _mm_unpacklo_epi8(s0, zero);
-
-        p0 = _mm_unpacklo_epi8(p0, zero);
-        p0 = _mm_adds_epu16(s0, p0);
-        p0 = _mm_adds_epu16(p0, one);
-        p0 = _mm_srli_epi16(p0, 1);
-        p0 = _mm_packus_epi16(p0, zero);
-
-        _mm_storel_epi64((__m128i *)(comp_pred), p0);
-        comp_pred += 8;
-        pred += 8;
-        ref += 8 * 8;
-      }
-      ref += stride - (width << 3);
-    }
-  } else {
-    // read 4 points at one time
-    for (i = 0; i < height; i++) {
-      for (j = 0; j < width; j += 4) {
-        __m128i s0 = _mm_loadu_si128((const __m128i *)ref);
-        __m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
-        __m128i p0 = _mm_cvtsi32_si128(*(const uint32_t *)pred);
-        __m128i t0;
-
-        t0 = _mm_unpacklo_epi8(s0, s1);
-        s1 = _mm_unpackhi_epi8(s0, s1);
-        s0 = _mm_unpacklo_epi8(t0, s1);
-        s0 = _mm_unpacklo_epi8(s0, zero);
-
-        p0 = _mm_unpacklo_epi8(p0, zero);
-        p0 = _mm_adds_epu16(s0, p0);
-        p0 = _mm_adds_epu16(p0, one);
-        p0 = _mm_srli_epi16(p0, 1);
-        p0 = _mm_packus_epi16(p0, zero);
-
-        *(int *)comp_pred = _mm_cvtsi128_si32(p0);
-        comp_pred += 4;
-        pred += 4;
-        ref += 4 * 8;
-      }
-      ref += stride - (width << 3);
-    }
+  int n;
+  int i;
+  aom_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3, ref,
+                     ref_stride);
+  /*The total number of pixels must be a multiple of 16 (e.g., 4x4).*/
+  assert(!(width * height & 15));
+  n = width * height >> 4;
+  for (i = 0; i < n; i++) {
+    __m128i s0 = _mm_loadu_si128((const __m128i *)comp_pred);
+    __m128i p0 = _mm_loadu_si128((const __m128i *)pred);
+    _mm_storeu_si128((__m128i *)comp_pred, _mm_avg_epu8(s0, p0));
+    comp_pred += 16;
+    pred += 16;
   }
 }