diff --git a/av1/common/arm/resize_neon.c b/av1/common/arm/resize_neon.c
index 7169cd4ec..b570e15 100644
--- a/av1/common/arm/resize_neon.c
+++ b/av1/common/arm/resize_neon.c
@@ -428,6 +428,312 @@
   } while (x);
 }
 
+static INLINE uint8x8_t scale_filter_bilinear(const uint8x8_t *const s,
+                                              const uint8x8_t *const coef) {
+  const uint16x8_t h0 = vmull_u8(s[0], coef[0]);
+  const uint16x8_t h1 = vmlal_u8(h0, s[1], coef[1]);
+
+  return vrshrn_n_u16(h1, 7);
+}
+
+// Notes for 4 to 3 scaling:
+//
+// 1. 6 rows are calculated in each horizontal inner loop, so width_hor must be
+// multiple of 6, and no less than w.
+//
+// 2. 8 rows are calculated in each vertical inner loop, so width_ver must be
+// multiple of 8, and no less than w.
+//
+// 3. 8 columns are calculated in each horizontal inner loop for further
+// vertical scaling, so height_hor must be multiple of 8, and no less than
+// 4 * h / 3.
+//
+// 4. 6 columns are calculated in each vertical inner loop, so height_ver must
+// be multiple of 6, and no less than h.
+//
+// 5. The physical location of the last row of the 4 to 3 scaled frame is
+// decided by phase_scaler, and are always less than 1 pixel below the last row
+// of the original image.
+static void scale_plane_4_to_3_bilinear(const uint8_t *src,
+                                        const int src_stride, uint8_t *dst,
+                                        const int dst_stride, const int w,
+                                        const int h, const int phase_scaler,
+                                        uint8_t *const temp_buffer) {
+  static const int step_q4 = 16 * 4 / 3;
+  const int width_hor = (w + 5) - ((w + 5) % 6);
+  const int stride_hor = width_hor + 2;  // store 2 extra pixels
+  const int width_ver = (w + 7) & ~7;
+  // We only need 1 extra row below because there are only 2 bilinear
+  // coefficients.
+  const int height_hor = (4 * h / 3 + 1 + 7) & ~7;
+  const int height_ver = (h + 5) - ((h + 5) % 6);
+  int x, y = height_hor;
+  uint8_t *t = temp_buffer;
+  uint8x8_t s[9], d[8], c[6];
+  const InterpKernel *interp_kernel =
+      (const InterpKernel *)av1_interp_filter_params_list[BILINEAR].filter_ptr;
+  assert(w && h);
+
+  c[0] = vdup_n_u8((uint8_t)interp_kernel[phase_scaler][3]);
+  c[1] = vdup_n_u8((uint8_t)interp_kernel[phase_scaler][4]);
+  c[2] = vdup_n_u8(
+      (uint8_t)interp_kernel[(phase_scaler + 1 * step_q4) & SUBPEL_MASK][3]);
+  c[3] = vdup_n_u8(
+      (uint8_t)interp_kernel[(phase_scaler + 1 * step_q4) & SUBPEL_MASK][4]);
+  c[4] = vdup_n_u8(
+      (uint8_t)interp_kernel[(phase_scaler + 2 * step_q4) & SUBPEL_MASK][3]);
+  c[5] = vdup_n_u8(
+      (uint8_t)interp_kernel[(phase_scaler + 2 * step_q4) & SUBPEL_MASK][4]);
+
+  d[6] = vdup_n_u8(0);
+  d[7] = vdup_n_u8(0);
+
+  // horizontal 6x8
+  do {
+    load_u8_8x8(src, src_stride, &s[0], &s[1], &s[2], &s[3], &s[4], &s[5],
+                &s[6], &s[7]);
+    src += 1;
+    transpose_u8_8x8(&s[0], &s[1], &s[2], &s[3], &s[4], &s[5], &s[6], &s[7]);
+    x = width_hor;
+
+    do {
+      load_u8_8x8(src, src_stride, &s[1], &s[2], &s[3], &s[4], &s[5], &s[6],
+                  &s[7], &s[8]);
+      src += 8;
+      transpose_u8_8x8(&s[1], &s[2], &s[3], &s[4], &s[5], &s[6], &s[7], &s[8]);
+
+      // 00 10 20 30 40 50 60 70
+      // 01 11 21 31 41 51 61 71
+      // 02 12 22 32 42 52 62 72
+      // 03 13 23 33 43 53 63 73
+      // 04 14 24 34 44 54 64 74
+      // 05 15 25 35 45 55 65 75
+      d[0] = scale_filter_bilinear(&s[0], &c[0]);
+      d[1] =
+          scale_filter_bilinear(&s[(phase_scaler + 1 * step_q4) >> 4], &c[2]);
+      d[2] =
+          scale_filter_bilinear(&s[(phase_scaler + 2 * step_q4) >> 4], &c[4]);
+      d[3] = scale_filter_bilinear(&s[4], &c[0]);
+      d[4] = scale_filter_bilinear(&s[4 + ((phase_scaler + 1 * step_q4) >> 4)],
+                                   &c[2]);
+      d[5] = scale_filter_bilinear(&s[4 + ((phase_scaler + 2 * step_q4) >> 4)],
+                                   &c[4]);
+
+      // 00 01 02 03 04 05 xx xx
+      // 10 11 12 13 14 15 xx xx
+      // 20 21 22 23 24 25 xx xx
+      // 30 31 32 33 34 35 xx xx
+      // 40 41 42 43 44 45 xx xx
+      // 50 51 52 53 54 55 xx xx
+      // 60 61 62 63 64 65 xx xx
+      // 70 71 72 73 74 75 xx xx
+      transpose_u8_8x8(&d[0], &d[1], &d[2], &d[3], &d[4], &d[5], &d[6], &d[7]);
+      // store 2 extra pixels
+      vst1_u8(t + 0 * stride_hor, d[0]);
+      vst1_u8(t + 1 * stride_hor, d[1]);
+      vst1_u8(t + 2 * stride_hor, d[2]);
+      vst1_u8(t + 3 * stride_hor, d[3]);
+      vst1_u8(t + 4 * stride_hor, d[4]);
+      vst1_u8(t + 5 * stride_hor, d[5]);
+      vst1_u8(t + 6 * stride_hor, d[6]);
+      vst1_u8(t + 7 * stride_hor, d[7]);
+
+      s[0] = s[8];
+
+      t += 6;
+      x -= 6;
+    } while (x);
+    src += 8 * src_stride - 4 * width_hor / 3 - 1;
+    t += 7 * stride_hor + 2;
+    y -= 8;
+  } while (y);
+
+  // vertical 8x6
+  x = width_ver;
+  t = temp_buffer;
+  do {
+    load_u8_8x8(t, stride_hor, &s[0], &s[1], &s[2], &s[3], &s[4], &s[5], &s[6],
+                &s[7]);
+    t += stride_hor;
+    y = height_ver;
+
+    do {
+      load_u8_8x8(t, stride_hor, &s[1], &s[2], &s[3], &s[4], &s[5], &s[6],
+                  &s[7], &s[8]);
+      t += 8 * stride_hor;
+
+      d[0] = scale_filter_bilinear(&s[0], &c[0]);
+      d[1] =
+          scale_filter_bilinear(&s[(phase_scaler + 1 * step_q4) >> 4], &c[2]);
+      d[2] =
+          scale_filter_bilinear(&s[(phase_scaler + 2 * step_q4) >> 4], &c[4]);
+      d[3] = scale_filter_bilinear(&s[4], &c[0]);
+      d[4] = scale_filter_bilinear(&s[4 + ((phase_scaler + 1 * step_q4) >> 4)],
+                                   &c[2]);
+      d[5] = scale_filter_bilinear(&s[4 + ((phase_scaler + 2 * step_q4) >> 4)],
+                                   &c[4]);
+      vst1_u8(dst + 0 * dst_stride, d[0]);
+      vst1_u8(dst + 1 * dst_stride, d[1]);
+      vst1_u8(dst + 2 * dst_stride, d[2]);
+      vst1_u8(dst + 3 * dst_stride, d[3]);
+      vst1_u8(dst + 4 * dst_stride, d[4]);
+      vst1_u8(dst + 5 * dst_stride, d[5]);
+
+      s[0] = s[8];
+
+      dst += 6 * dst_stride;
+      y -= 6;
+    } while (y);
+    t -= stride_hor * (4 * height_ver / 3 + 1);
+    t += 8;
+    dst -= height_ver * dst_stride;
+    dst += 8;
+    x -= 8;
+  } while (x);
+}
+
+static void scale_plane_4_to_3_general(const uint8_t *src, const int src_stride,
+                                       uint8_t *dst, const int dst_stride,
+                                       const int w, const int h,
+                                       const int16_t *const coef,
+                                       const int phase_scaler,
+                                       uint8_t *const temp_buffer) {
+  static const int step_q4 = 16 * 4 / 3;
+  const int width_hor = (w + 5) - ((w + 5) % 6);
+  const int stride_hor = width_hor + 2;  // store 2 extra pixels
+  const int width_ver = (w + 7) & ~7;
+  // We need (SUBPEL_TAPS - 1) extra rows: (SUBPEL_TAPS / 2 - 1) extra rows
+  // above and (SUBPEL_TAPS / 2) extra rows below.
+  const int height_hor = (4 * h / 3 + SUBPEL_TAPS - 1 + 7) & ~7;
+  const int height_ver = (h + 5) - ((h + 5) % 6);
+  const int16x8_t filters0 =
+      vld1q_s16(&coef[(phase_scaler + 0 * step_q4) & SUBPEL_MASK]);
+  const int16x8_t filters1 =
+      vld1q_s16(&coef[(phase_scaler + 1 * step_q4) & SUBPEL_MASK]);
+  const int16x8_t filters2 =
+      vld1q_s16(&coef[(phase_scaler + 2 * step_q4) & SUBPEL_MASK]);
+  int x, y = height_hor;
+  uint8_t *t = temp_buffer;
+  uint8x8_t s[15], d[8];
+
+  assert(w && h);
+
+  src -= (SUBPEL_TAPS / 2 - 1) * src_stride + SUBPEL_TAPS / 2;
+  d[6] = vdup_n_u8(0);
+  d[7] = vdup_n_u8(0);
+
+  // horizontal 6x8
+  do {
+    load_u8_8x8(src + 1, src_stride, &s[0], &s[1], &s[2], &s[3], &s[4], &s[5],
+                &s[6], &s[7]);
+    transpose_u8_8x8(&s[0], &s[1], &s[2], &s[3], &s[4], &s[5], &s[6], &s[7]);
+    x = width_hor;
+
+    do {
+      src += 8;
+      load_u8_8x8(src, src_stride, &s[7], &s[8], &s[9], &s[10], &s[11], &s[12],
+                  &s[13], &s[14]);
+      transpose_u8_8x8(&s[7], &s[8], &s[9], &s[10], &s[11], &s[12], &s[13],
+                       &s[14]);
+
+      // 00 10 20 30 40 50 60 70
+      // 01 11 21 31 41 51 61 71
+      // 02 12 22 32 42 52 62 72
+      // 03 13 23 33 43 53 63 73
+      // 04 14 24 34 44 54 64 74
+      // 05 15 25 35 45 55 65 75
+      d[0] = scale_filter_8(&s[0], filters0);
+      d[1] = scale_filter_8(&s[(phase_scaler + 1 * step_q4) >> 4], filters1);
+      d[2] = scale_filter_8(&s[(phase_scaler + 2 * step_q4) >> 4], filters2);
+      d[3] = scale_filter_8(&s[4], filters0);
+      d[4] =
+          scale_filter_8(&s[4 + ((phase_scaler + 1 * step_q4) >> 4)], filters1);
+      d[5] =
+          scale_filter_8(&s[4 + ((phase_scaler + 2 * step_q4) >> 4)], filters2);
+
+      // 00 01 02 03 04 05 xx xx
+      // 10 11 12 13 14 15 xx xx
+      // 20 21 22 23 24 25 xx xx
+      // 30 31 32 33 34 35 xx xx
+      // 40 41 42 43 44 45 xx xx
+      // 50 51 52 53 54 55 xx xx
+      // 60 61 62 63 64 65 xx xx
+      // 70 71 72 73 74 75 xx xx
+      transpose_u8_8x8(&d[0], &d[1], &d[2], &d[3], &d[4], &d[5], &d[6], &d[7]);
+      // store 2 extra pixels
+      vst1_u8(t + 0 * stride_hor, d[0]);
+      vst1_u8(t + 1 * stride_hor, d[1]);
+      vst1_u8(t + 2 * stride_hor, d[2]);
+      vst1_u8(t + 3 * stride_hor, d[3]);
+      vst1_u8(t + 4 * stride_hor, d[4]);
+      vst1_u8(t + 5 * stride_hor, d[5]);
+      vst1_u8(t + 6 * stride_hor, d[6]);
+      vst1_u8(t + 7 * stride_hor, d[7]);
+
+      s[0] = s[8];
+      s[1] = s[9];
+      s[2] = s[10];
+      s[3] = s[11];
+      s[4] = s[12];
+      s[5] = s[13];
+      s[6] = s[14];
+
+      t += 6;
+      x -= 6;
+    } while (x);
+    src += 8 * src_stride - 4 * width_hor / 3;
+    t += 7 * stride_hor + 2;
+    y -= 8;
+  } while (y);
+
+  // vertical 8x6
+  x = width_ver;
+  t = temp_buffer;
+  do {
+    load_u8_8x8(t, stride_hor, &s[0], &s[1], &s[2], &s[3], &s[4], &s[5], &s[6],
+                &s[7]);
+    t += 7 * stride_hor;
+    y = height_ver;
+
+    do {
+      load_u8_8x8(t, stride_hor, &s[7], &s[8], &s[9], &s[10], &s[11], &s[12],
+                  &s[13], &s[14]);
+      t += 8 * stride_hor;
+
+      d[0] = scale_filter_8(&s[0], filters0);
+      d[1] = scale_filter_8(&s[(phase_scaler + 1 * step_q4) >> 4], filters1);
+      d[2] = scale_filter_8(&s[(phase_scaler + 2 * step_q4) >> 4], filters2);
+      d[3] = scale_filter_8(&s[4], filters0);
+      d[4] =
+          scale_filter_8(&s[4 + ((phase_scaler + 1 * step_q4) >> 4)], filters1);
+      d[5] =
+          scale_filter_8(&s[4 + ((phase_scaler + 2 * step_q4) >> 4)], filters2);
+      vst1_u8(dst + 0 * dst_stride, d[0]);
+      vst1_u8(dst + 1 * dst_stride, d[1]);
+      vst1_u8(dst + 2 * dst_stride, d[2]);
+      vst1_u8(dst + 3 * dst_stride, d[3]);
+      vst1_u8(dst + 4 * dst_stride, d[4]);
+      vst1_u8(dst + 5 * dst_stride, d[5]);
+
+      s[0] = s[8];
+      s[1] = s[9];
+      s[2] = s[10];
+      s[3] = s[11];
+      s[4] = s[12];
+      s[5] = s[13];
+      s[6] = s[14];
+
+      dst += 6 * dst_stride;
+      y -= 6;
+    } while (y);
+    t -= stride_hor * (4 * height_ver / 3 + 7);
+    t += 8;
+    dst -= height_ver * dst_stride;
+    dst += 8;
+    x -= 8;
+  } while (x);
+}
+
 void av1_resize_and_extend_frame_neon(const YV12_BUFFER_CONFIG *src,
                                       YV12_BUFFER_CONFIG *dst,
                                       const InterpFilter filter,
@@ -495,6 +801,27 @@
           free(temp_buffer);
         }
       }
+    } else if (4 * dst_w == 3 * src_w && 4 * dst_h == 3 * src_h) {
+      // 4 to 3
+      const int buffer_stride = (dst_w + 5) - ((dst_w + 5) % 6) + 2;
+      const int buffer_height = (4 * dst_h / 3 + SUBPEL_TAPS - 1 + 7) & ~7;
+      uint8_t *const temp_buffer =
+          (uint8_t *)malloc(buffer_stride * buffer_height);
+      if (temp_buffer) {
+        if (filter == BILINEAR) {
+          scale_plane_4_to_3_bilinear(src->buffers[i], src->strides[is_uv],
+                                      dst->buffers[i], dst->strides[is_uv],
+                                      dst_w, dst_h, phase, temp_buffer);
+        } else {
+          const InterpKernel *interp_kernel =
+              (const InterpKernel *)av1_interp_filter_params_list[filter]
+                  .filter_ptr;
+          scale_plane_4_to_3_general(src->buffers[i], src->strides[is_uv],
+                                     dst->buffers[i], dst->strides[is_uv],
+                                     dst_w, dst_h, interp_kernel[phase], phase,
+                                     temp_buffer);
+        }
+      }
     } else {
       av1_resize_plane(src->buffers[i], src_h, src_w, src->strides[is_uv],
                        dst->buffers[i], dst_h, dst_w, dst->strides[is_uv]);
diff --git a/av1/common/resize.c b/av1/common/resize.c
index f39ca31..74847ba 100644
--- a/av1/common/resize.c
+++ b/av1/common/resize.c
@@ -1341,22 +1341,20 @@
                                           YV12_BUFFER_CONFIG *unscaled,
                                           YV12_BUFFER_CONFIG *scaled,
                                           const InterpFilter filter,
-                                          const int phase) {
+                                          const int phase,
+                                          const int use_optimized_scaler) {
   const int num_planes = av1_num_planes(cm);
   if (cm->width != unscaled->y_crop_width ||
       cm->height != unscaled->y_crop_height) {
 #if CONFIG_AV1_HIGHBITDEPTH
-    if (cm->width <= (unscaled->y_crop_width >> 1) &&
-        cm->height <= (unscaled->y_crop_height >> 1) &&
-        cm->seq_params.bit_depth == AOM_BITS_8) {
+    if (use_optimized_scaler && cm->seq_params.bit_depth == AOM_BITS_8) {
       av1_resize_and_extend_frame(unscaled, scaled, filter, phase, num_planes);
     } else {
       av1_resize_and_extend_frame_nonnormative(
           unscaled, scaled, (int)cm->seq_params.bit_depth, num_planes);
     }
 #else
-    if (cm->width <= (unscaled->y_crop_width >> 1) &&
-        cm->height <= (unscaled->y_crop_height >> 1)) {
+    if (use_optimized_scaler) {
       av1_resize_and_extend_frame(unscaled, scaled, filter, phase, num_planes);
     } else {
       av1_resize_and_extend_frame_nonnormative(
diff --git a/av1/common/resize.h b/av1/common/resize.h
index f5c84b0..af6eeb7 100644
--- a/av1/common/resize.h
+++ b/av1/common/resize.h
@@ -71,11 +71,9 @@
                                             const YV12_BUFFER_CONFIG *src,
                                             YV12_BUFFER_CONFIG *dst);
 
-YV12_BUFFER_CONFIG *av1_scale_if_required(AV1_COMMON *cm,
-                                          YV12_BUFFER_CONFIG *unscaled,
-                                          YV12_BUFFER_CONFIG *scaled,
-                                          const InterpFilter filter,
-                                          const int phase);
+YV12_BUFFER_CONFIG *av1_scale_if_required(
+    AV1_COMMON *cm, YV12_BUFFER_CONFIG *unscaled, YV12_BUFFER_CONFIG *scaled,
+    const InterpFilter filter, const int phase, const int use_optimized_scaler);
 
 void av1_resize_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
                                               YV12_BUFFER_CONFIG *dst, int bd,
diff --git a/av1/encoder/encode_strategy.c b/av1/encoder/encode_strategy.c
index b9fbd23..63b55bf 100644
--- a/av1/encoder/encode_strategy.c
+++ b/av1/encoder/encode_strategy.c
@@ -948,7 +948,7 @@
   if (apply_filtering && is_psnr_calc_enabled(cpi)) {
     cpi->source =
         av1_scale_if_required(cm, source_kf_buffer, &cpi->scaled_source,
-                              cm->features.interp_filter, 0);
+                              cm->features.interp_filter, 0, 0);
     cpi->unscaled_source = source_kf_buffer;
   }
 
diff --git a/av1/encoder/encoder.c b/av1/encoder/encoder.c
index 7ae8079..328faeb 100644
--- a/av1/encoder/encoder.c
+++ b/av1/encoder/encoder.c
@@ -2058,7 +2058,7 @@
   aom_clear_system_state();
 
   cpi->source = av1_scale_if_required(cm, unscaled, &cpi->scaled_source,
-                                      filter_scaler, phase_scaler);
+                                      filter_scaler, phase_scaler, 1);
   if (frame_is_intra_only(cm) || resize_pending != 0) {
     memset(cpi->consec_zero_mv, 0,
            ((cm->mi_params.mi_rows * cm->mi_params.mi_cols) >> 2) *
@@ -2068,7 +2068,7 @@
   if (cpi->unscaled_last_source != NULL) {
     cpi->last_source = av1_scale_if_required(cm, cpi->unscaled_last_source,
                                              &cpi->scaled_last_source,
-                                             filter_scaler, phase_scaler);
+                                             filter_scaler, phase_scaler, 1);
   }
 
   if (cpi->sf.rt_sf.use_temporal_noise_estimate) {
@@ -2080,7 +2080,7 @@
   // use for newmv search, we can avoid scaling here.
   if (!frame_is_intra_only(cm) &&
       !(cpi->use_svc && cpi->svc.force_zero_mode_spatial_ref))
-    av1_scale_references(cpi, filter_scaler, phase_scaler);
+    av1_scale_references(cpi, filter_scaler, phase_scaler, 1);
 
   av1_set_quantizer(cm, q_cfg->qm_minlevel, q_cfg->qm_maxlevel, q,
                     q_cfg->enable_chroma_deltaq);
@@ -2239,19 +2239,19 @@
       }
     }
     cpi->source = av1_scale_if_required(
-        cm, cpi->unscaled_source, &cpi->scaled_source, EIGHTTAP_REGULAR, 0);
+        cm, cpi->unscaled_source, &cpi->scaled_source, EIGHTTAP_REGULAR, 0, 0);
 
     if (cpi->unscaled_last_source != NULL) {
-      cpi->last_source =
-          av1_scale_if_required(cm, cpi->unscaled_last_source,
-                                &cpi->scaled_last_source, EIGHTTAP_REGULAR, 0);
+      cpi->last_source = av1_scale_if_required(cm, cpi->unscaled_last_source,
+                                               &cpi->scaled_last_source,
+                                               EIGHTTAP_REGULAR, 0, 0);
     }
 
     if (!frame_is_intra_only(cm)) {
       if (loop_count > 0) {
         release_scaled_references(cpi);
       }
-      av1_scale_references(cpi, EIGHTTAP_REGULAR, 0);
+      av1_scale_references(cpi, EIGHTTAP_REGULAR, 0, 0);
     }
 #if CONFIG_TUNE_VMAF
     if (oxcf->tuning == AOM_TUNE_VMAF_WITH_PREPROCESSING ||
diff --git a/av1/encoder/encoder_utils.c b/av1/encoder/encoder_utils.c
index 72cc146..b55acb6 100644
--- a/av1/encoder/encoder_utils.c
+++ b/av1/encoder/encoder_utils.c
@@ -633,7 +633,7 @@
 }
 
 void av1_scale_references(AV1_COMP *cpi, const InterpFilter filter,
-                          const int phase) {
+                          const int phase, const int use_optimized_scaler) {
   AV1_COMMON *cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
   MV_REFERENCE_FRAME ref_frame;
@@ -692,17 +692,14 @@
                                "Failed to allocate frame buffer");
           }
 #if CONFIG_AV1_HIGHBITDEPTH
-          if (cm->width <= (ref->y_crop_width >> 1) &&
-              cm->height <= (ref->y_crop_height >> 1) &&
-              cm->seq_params.bit_depth == AOM_BITS_8)
+          if (use_optimized_scaler && cm->seq_params.bit_depth == AOM_BITS_8)
             av1_resize_and_extend_frame(ref, &new_fb->buf, filter, phase,
                                         num_planes);
           else
             av1_resize_and_extend_frame_nonnormative(
                 ref, &new_fb->buf, (int)cm->seq_params.bit_depth, num_planes);
 #else
-          if (cm->width <= (ref->y_crop_width >> 1) &&
-              cm->height <= (ref->y_crop_height >> 1))
+          if (use_optimized_scaler)
             av1_resize_and_extend_frame(ref, &new_fb->buf, filter, phase,
                                         num_planes);
           else
@@ -942,11 +939,11 @@
 
   cpi->source =
       av1_scale_if_required(cm, cpi->unscaled_source, &cpi->scaled_source,
-                            cm->features.interp_filter, 0);
+                            cm->features.interp_filter, 0, 0);
   if (cpi->unscaled_last_source != NULL) {
     cpi->last_source = av1_scale_if_required(cm, cpi->unscaled_last_source,
                                              &cpi->scaled_last_source,
-                                             cm->features.interp_filter, 0);
+                                             cm->features.interp_filter, 0, 0);
   }
 
   av1_setup_frame(cpi);
diff --git a/av1/encoder/encoder_utils.h b/av1/encoder/encoder_utils.h
index 55347ba..4cb480a3 100644
--- a/av1/encoder/encoder_utils.h
+++ b/av1/encoder/encoder_utils.h
@@ -787,7 +787,7 @@
                                       const AV1EncoderConfig *oxcf);
 
 void av1_scale_references(AV1_COMP *cpi, const InterpFilter filter,
-                          const int phase);
+                          const int phase, const int use_optimized_scaler);
 
 void av1_setup_frame(AV1_COMP *cpi);