diff --git a/vp10/encoder/encoder.c b/vp10/encoder/encoder.c
index 4f97c795a..55ec9c17e 100644
--- a/vp10/encoder/encoder.c
+++ b/vp10/encoder/encoder.c
@@ -50,7 +50,7 @@
 #include "./vp10_rtcd.h"
 #include "./vpx_dsp_rtcd.h"
 #include "./vpx_scale_rtcd.h"
-#include "vpx/internal/vpx_psnr.h"
+#include "vpx_dsp/psnr.h"
 #if CONFIG_INTERNAL_STATS
 #include "vpx_dsp/ssim.h"
 #endif
@@ -2033,261 +2033,6 @@ void vp10_remove_compressor(VP10_COMP *cpi) {
 #endif
 }
 
-/* TODO(yaowu): The block_variance calls the unoptimized versions of variance()
- * and highbd_8_variance(). It should not.
- */
-static void encoder_variance(const uint8_t *a, int  a_stride,
-                             const uint8_t *b, int  b_stride,
-                             int  w, int  h, unsigned int *sse, int *sum) {
-  int i, j;
-
-  *sum = 0;
-  *sse = 0;
-
-  for (i = 0; i < h; i++) {
-    for (j = 0; j < w; j++) {
-      const int diff = a[j] - b[j];
-      *sum += diff;
-      *sse += diff * diff;
-    }
-
-    a += a_stride;
-    b += b_stride;
-  }
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static void encoder_highbd_variance64(const uint8_t *a8, int  a_stride,
-                                      const uint8_t *b8, int  b_stride,
-                                      int w, int h, uint64_t *sse,
-                                      uint64_t *sum) {
-  int i, j;
-
-  uint16_t *a = CONVERT_TO_SHORTPTR(a8);
-  uint16_t *b = CONVERT_TO_SHORTPTR(b8);
-  *sum = 0;
-  *sse = 0;
-
-  for (i = 0; i < h; i++) {
-    for (j = 0; j < w; j++) {
-      const int diff = a[j] - b[j];
-      *sum += diff;
-      *sse += diff * diff;
-    }
-    a += a_stride;
-    b += b_stride;
-  }
-}
-
-static void encoder_highbd_8_variance(const uint8_t *a8, int  a_stride,
-                                      const uint8_t *b8, int  b_stride,
-                                      int w, int h,
-                                      unsigned int *sse, int *sum) {
-  uint64_t sse_long = 0;
-  uint64_t sum_long = 0;
-  encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h,
-                            &sse_long, &sum_long);
-  *sse = (unsigned int)sse_long;
-  *sum = (int)sum_long;
-}
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
-static int64_t get_sse(const uint8_t *a, int a_stride,
-                       const uint8_t *b, int b_stride,
-                       int width, int height) {
-  const int dw = width % 16;
-  const int dh = height % 16;
-  int64_t total_sse = 0;
-  unsigned int sse = 0;
-  int sum = 0;
-  int x, y;
-
-  if (dw > 0) {
-    encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
-                     dw, height, &sse, &sum);
-    total_sse += sse;
-  }
-
-  if (dh > 0) {
-    encoder_variance(&a[(height - dh) * a_stride], a_stride,
-                     &b[(height - dh) * b_stride], b_stride,
-                     width - dw, dh, &sse, &sum);
-    total_sse += sse;
-  }
-
-  for (y = 0; y < height / 16; ++y) {
-    const uint8_t *pa = a;
-    const uint8_t *pb = b;
-    for (x = 0; x < width / 16; ++x) {
-      vpx_mse16x16(pa, a_stride, pb, b_stride, &sse);
-      total_sse += sse;
-
-      pa += 16;
-      pb += 16;
-    }
-
-    a += 16 * a_stride;
-    b += 16 * b_stride;
-  }
-
-  return total_sse;
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
-                                    const uint8_t *b8, int b_stride,
-                                    int width, int height,
-                                    unsigned int input_shift) {
-  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
-  const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
-  int64_t total_sse = 0;
-  int x, y;
-  for (y = 0; y < height; ++y) {
-    for (x = 0; x < width; ++x) {
-      int64_t diff;
-      diff = (a[x] >> input_shift) - (b[x] >> input_shift);
-      total_sse += diff * diff;
-    }
-    a += a_stride;
-    b += b_stride;
-  }
-  return total_sse;
-}
-
-static int64_t highbd_get_sse(const uint8_t *a, int a_stride,
-                              const uint8_t *b, int b_stride,
-                              int width, int height) {
-  int64_t total_sse = 0;
-  int x, y;
-  const int dw = width % 16;
-  const int dh = height % 16;
-  unsigned int sse = 0;
-  int sum = 0;
-  if (dw > 0) {
-    encoder_highbd_8_variance(&a[width - dw], a_stride,
-                              &b[width - dw], b_stride,
-                              dw, height, &sse, &sum);
-    total_sse += sse;
-  }
-  if (dh > 0) {
-    encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
-                              &b[(height - dh) * b_stride], b_stride,
-                              width - dw, dh, &sse, &sum);
-    total_sse += sse;
-  }
-  for (y = 0; y < height / 16; ++y) {
-    const uint8_t *pa = a;
-    const uint8_t *pb = b;
-    for (x = 0; x < width / 16; ++x) {
-      vpx_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse);
-      total_sse += sse;
-      pa += 16;
-      pb += 16;
-    }
-    a += 16 * a_stride;
-    b += 16 * b_stride;
-  }
-  return total_sse;
-}
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
-typedef struct {
-  double psnr[4];       // total/y/u/v
-  uint64_t sse[4];      // total/y/u/v
-  uint32_t samples[4];  // total/y/u/v
-} PSNR_STATS;
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static void calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
-                             const YV12_BUFFER_CONFIG *b,
-                             PSNR_STATS *psnr,
-                             unsigned int bit_depth,
-                             unsigned int in_bit_depth) {
-  const int widths[3] =
-      {a->y_crop_width,  a->uv_crop_width,  a->uv_crop_width };
-  const int heights[3] =
-      {a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
-  const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer,  a->v_buffer };
-  const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
-  const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer,  b->v_buffer };
-  const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
-  int i;
-  uint64_t total_sse = 0;
-  uint32_t total_samples = 0;
-  const double peak = (double)((1 << in_bit_depth) - 1);
-  const unsigned int input_shift = bit_depth - in_bit_depth;
-
-  for (i = 0; i < 3; ++i) {
-    const int w = widths[i];
-    const int h = heights[i];
-    const uint32_t samples = w * h;
-    uint64_t sse;
-    if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
-      if (input_shift) {
-        sse = highbd_get_sse_shift(a_planes[i], a_strides[i],
-                                   b_planes[i], b_strides[i], w, h,
-                                   input_shift);
-      } else {
-        sse = highbd_get_sse(a_planes[i], a_strides[i],
-                             b_planes[i], b_strides[i], w, h);
-      }
-    } else {
-      sse = get_sse(a_planes[i], a_strides[i],
-                    b_planes[i], b_strides[i],
-                    w, h);
-    }
-    psnr->sse[1 + i] = sse;
-    psnr->samples[1 + i] = samples;
-    psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
-
-    total_sse += sse;
-    total_samples += samples;
-  }
-
-  psnr->sse[0] = total_sse;
-  psnr->samples[0] = total_samples;
-  psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
-                                  (double)total_sse);
-}
-
-#else  // !CONFIG_VP9_HIGHBITDEPTH
-
-static void calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b,
-                      PSNR_STATS *psnr) {
-  static const double peak = 255.0;
-  const int widths[3]        = {
-      a->y_crop_width, a->uv_crop_width, a->uv_crop_width};
-  const int heights[3]       = {
-      a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
-  const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer};
-  const int a_strides[3]     = {a->y_stride, a->uv_stride, a->uv_stride};
-  const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer};
-  const int b_strides[3]     = {b->y_stride, b->uv_stride, b->uv_stride};
-  int i;
-  uint64_t total_sse = 0;
-  uint32_t total_samples = 0;
-
-  for (i = 0; i < 3; ++i) {
-    const int w = widths[i];
-    const int h = heights[i];
-    const uint32_t samples = w * h;
-    const uint64_t sse = get_sse(a_planes[i], a_strides[i],
-                                 b_planes[i], b_strides[i],
-                                 w, h);
-    psnr->sse[1 + i] = sse;
-    psnr->samples[1 + i] = samples;
-    psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
-
-    total_sse += sse;
-    total_samples += samples;
-  }
-
-  psnr->sse[0] = total_sse;
-  psnr->samples[0] = total_samples;
-  psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
-                                  (double)total_sse);
-}
-#endif  // CONFIG_VP9_HIGHBITDEPTH
 
 static void generate_psnr_packet(VP10_COMP *cpi) {
   struct vpx_codec_cx_pkt pkt;
@@ -2955,7 +2700,7 @@ static void output_frame_level_debug_stats(VP10_COMP *cpi) {
 
   vpx_clear_system_state();
 
-  recon_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+  recon_err = vpx_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
 
   if (cpi->twopass.total_left_stats.coded_error != 0.0)
     fprintf(f, "%10u %dx%d  %10d %10d %d %d %10d %10d %10d %10d"
@@ -3380,12 +3125,12 @@ static void encode_with_recode_loop(VP10_COMP *cpi,
 
 #if CONFIG_VP9_HIGHBITDEPTH
         if (cm->use_highbitdepth) {
-          kf_err = vp10_highbd_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+          kf_err = vpx_highbd_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
         } else {
-          kf_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+          kf_err = vpx_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
         }
 #else
-        kf_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+        kf_err = vpx_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
         // Prevent possible divide by zero error below for perfect KF
@@ -3804,13 +3549,13 @@ static void encode_frame_to_data_rate(VP10_COMP *cpi,
   if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) {
 #if CONFIG_VP9_HIGHBITDEPTH
     if (cm->use_highbitdepth) {
-      cpi->ambient_err = vp10_highbd_get_y_sse(cpi->Source,
+      cpi->ambient_err = vpx_highbd_get_y_sse(cpi->Source,
                                               get_frame_new_buffer(cm));
     } else {
-      cpi->ambient_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+      cpi->ambient_err = vpx_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
     }
 #else
-    cpi->ambient_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+    cpi->ambient_err = vpx_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
 #endif  // CONFIG_VP9_HIGHBITDEPTH
   }
 
@@ -4557,28 +4302,6 @@ int vp10_set_size_literal(VP10_COMP *cpi, unsigned int width,
   return 0;
 }
 
-int64_t vp10_get_y_sse(const YV12_BUFFER_CONFIG *a,
-                      const YV12_BUFFER_CONFIG *b) {
-  assert(a->y_crop_width == b->y_crop_width);
-  assert(a->y_crop_height == b->y_crop_height);
-
-  return get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
-                 a->y_crop_width, a->y_crop_height);
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-int64_t vp10_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
-                             const YV12_BUFFER_CONFIG *b) {
-  assert(a->y_crop_width == b->y_crop_width);
-  assert(a->y_crop_height == b->y_crop_height);
-  assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
-  assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
-
-  return highbd_get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
-                        a->y_crop_width, a->y_crop_height);
-}
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
 int vp10_get_quantizer(VP10_COMP *cpi) {
   return cpi->common.base_qindex;
 }
diff --git a/vp10/encoder/encoder.h b/vp10/encoder/encoder.h
index cc2076564..59c768259 100644
--- a/vp10/encoder/encoder.h
+++ b/vp10/encoder/encoder.h
@@ -633,12 +633,6 @@ static INLINE int allocated_tokens(TileInfo tile) {
   return get_token_alloc(tile_mb_rows, tile_mb_cols);
 }
 
-int64_t vp10_get_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b);
-#if CONFIG_VP9_HIGHBITDEPTH
-int64_t vp10_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
-                             const YV12_BUFFER_CONFIG *b);
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
 void vp10_alloc_compressor_data(VP10_COMP *cpi);
 
 void vp10_scale_references(VP10_COMP *cpi);
diff --git a/vp10/encoder/picklpf.c b/vp10/encoder/picklpf.c
index f116c0008..cb2c1c766 100644
--- a/vp10/encoder/picklpf.c
+++ b/vp10/encoder/picklpf.c
@@ -13,6 +13,7 @@
 
 #include "./vpx_scale_rtcd.h"
 
+#include "vpx_dsp/psnr.h"
 #include "vpx_dsp/vpx_dsp_common.h"
 #include "vpx_mem/vpx_mem.h"
 #include "vpx_ports/mem.h"
@@ -56,12 +57,12 @@ static int64_t try_filter_frame(const YV12_BUFFER_CONFIG *sd,
 
 #if CONFIG_VP9_HIGHBITDEPTH
   if (cm->use_highbitdepth) {
-    filt_err = vp10_highbd_get_y_sse(sd, cm->frame_to_show);
+    filt_err = vpx_highbd_get_y_sse(sd, cm->frame_to_show);
   } else {
-    filt_err = vp10_get_y_sse(sd, cm->frame_to_show);
+    filt_err = vpx_get_y_sse(sd, cm->frame_to_show);
   }
 #else
-  filt_err = vp10_get_y_sse(sd, cm->frame_to_show);
+  filt_err = vpx_get_y_sse(sd, cm->frame_to_show);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
   // Re-instate the unfiltered frame
diff --git a/vp10/encoder/pickrst.c b/vp10/encoder/pickrst.c
index 79cda43ab..998283674 100644
--- a/vp10/encoder/pickrst.c
+++ b/vp10/encoder/pickrst.c
@@ -14,6 +14,7 @@
 
 #include "./vpx_scale_rtcd.h"
 
+#include "vpx_dsp/psnr.h"
 #include "vpx_dsp/vpx_dsp_common.h"
 #include "vpx_mem/vpx_mem.h"
 #include "vpx_ports/mem.h"
@@ -36,12 +37,12 @@ static int try_restoration_frame(const YV12_BUFFER_CONFIG *sd,
                               rsi, 1, partial_frame);
 #if CONFIG_VP9_HIGHBITDEPTH
   if (cm->use_highbitdepth) {
-    filt_err = vp10_highbd_get_y_sse(sd, cm->frame_to_show);
+    filt_err = vpx_highbd_get_y_sse(sd, cm->frame_to_show);
   } else {
-    filt_err = vp10_get_y_sse(sd, cm->frame_to_show);
+    filt_err = vpx_get_y_sse(sd, cm->frame_to_show);
   }
 #else
-  filt_err = vp10_get_y_sse(sd, cm->frame_to_show);
+  filt_err = vpx_get_y_sse(sd, cm->frame_to_show);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
   // Re-instate the unfiltered frame
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index b1c2e11c9..edfd60c37 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -21,7 +21,7 @@
 #include "vp8/common/alloccommon.h"
 #include "mcomp.h"
 #include "firstpass.h"
-#include "vpx/internal/vpx_psnr.h"
+#include "vpx_dsp/psnr.h"
 #include "vpx_scale/vpx_scale.h"
 #include "vp8/common/extend.h"
 #include "ratectrl.h"
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 69d5fa74b..713b5f7e1 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -16,7 +16,7 @@
 #include "./vpx_config.h"
 #include "./vpx_dsp_rtcd.h"
 #include "./vpx_scale_rtcd.h"
-#include "vpx/internal/vpx_psnr.h"
+#include "vpx_dsp/psnr.h"
 #include "vpx_dsp/vpx_dsp_common.h"
 #include "vpx_dsp/vpx_filter.h"
 #if CONFIG_INTERNAL_STATS
@@ -2140,262 +2140,6 @@ void vp9_remove_compressor(VP9_COMP *cpi) {
 #endif
 }
 
-/* TODO(yaowu): The block_variance calls the unoptimized versions of variance()
- * and highbd_8_variance(). It should not.
- */
-static void encoder_variance(const uint8_t *a, int  a_stride,
-                             const uint8_t *b, int  b_stride,
-                             int  w, int  h, unsigned int *sse, int *sum) {
-  int i, j;
-
-  *sum = 0;
-  *sse = 0;
-
-  for (i = 0; i < h; i++) {
-    for (j = 0; j < w; j++) {
-      const int diff = a[j] - b[j];
-      *sum += diff;
-      *sse += diff * diff;
-    }
-
-    a += a_stride;
-    b += b_stride;
-  }
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static void encoder_highbd_variance64(const uint8_t *a8, int  a_stride,
-                                      const uint8_t *b8, int  b_stride,
-                                      int w, int h, uint64_t *sse,
-                                      uint64_t *sum) {
-  int i, j;
-
-  uint16_t *a = CONVERT_TO_SHORTPTR(a8);
-  uint16_t *b = CONVERT_TO_SHORTPTR(b8);
-  *sum = 0;
-  *sse = 0;
-
-  for (i = 0; i < h; i++) {
-    for (j = 0; j < w; j++) {
-      const int diff = a[j] - b[j];
-      *sum += diff;
-      *sse += diff * diff;
-    }
-    a += a_stride;
-    b += b_stride;
-  }
-}
-
-static void encoder_highbd_8_variance(const uint8_t *a8, int  a_stride,
-                                      const uint8_t *b8, int  b_stride,
-                                      int w, int h,
-                                      unsigned int *sse, int *sum) {
-  uint64_t sse_long = 0;
-  uint64_t sum_long = 0;
-  encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h,
-                            &sse_long, &sum_long);
-  *sse = (unsigned int)sse_long;
-  *sum = (int)sum_long;
-}
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
-static int64_t get_sse(const uint8_t *a, int a_stride,
-                       const uint8_t *b, int b_stride,
-                       int width, int height) {
-  const int dw = width % 16;
-  const int dh = height % 16;
-  int64_t total_sse = 0;
-  unsigned int sse = 0;
-  int sum = 0;
-  int x, y;
-
-  if (dw > 0) {
-    encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
-                     dw, height, &sse, &sum);
-    total_sse += sse;
-  }
-
-  if (dh > 0) {
-    encoder_variance(&a[(height - dh) * a_stride], a_stride,
-                     &b[(height - dh) * b_stride], b_stride,
-                     width - dw, dh, &sse, &sum);
-    total_sse += sse;
-  }
-
-  for (y = 0; y < height / 16; ++y) {
-    const uint8_t *pa = a;
-    const uint8_t *pb = b;
-    for (x = 0; x < width / 16; ++x) {
-      vpx_mse16x16(pa, a_stride, pb, b_stride, &sse);
-      total_sse += sse;
-
-      pa += 16;
-      pb += 16;
-    }
-
-    a += 16 * a_stride;
-    b += 16 * b_stride;
-  }
-
-  return total_sse;
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
-                                    const uint8_t *b8, int b_stride,
-                                    int width, int height,
-                                    unsigned int input_shift) {
-  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
-  const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
-  int64_t total_sse = 0;
-  int x, y;
-  for (y = 0; y < height; ++y) {
-    for (x = 0; x < width; ++x) {
-      int64_t diff;
-      diff = (a[x] >> input_shift) - (b[x] >> input_shift);
-      total_sse += diff * diff;
-    }
-    a += a_stride;
-    b += b_stride;
-  }
-  return total_sse;
-}
-
-static int64_t highbd_get_sse(const uint8_t *a, int a_stride,
-                              const uint8_t *b, int b_stride,
-                              int width, int height) {
-  int64_t total_sse = 0;
-  int x, y;
-  const int dw = width % 16;
-  const int dh = height % 16;
-  unsigned int sse = 0;
-  int sum = 0;
-  if (dw > 0) {
-    encoder_highbd_8_variance(&a[width - dw], a_stride,
-                              &b[width - dw], b_stride,
-                              dw, height, &sse, &sum);
-    total_sse += sse;
-  }
-  if (dh > 0) {
-    encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
-                              &b[(height - dh) * b_stride], b_stride,
-                              width - dw, dh, &sse, &sum);
-    total_sse += sse;
-  }
-  for (y = 0; y < height / 16; ++y) {
-    const uint8_t *pa = a;
-    const uint8_t *pb = b;
-    for (x = 0; x < width / 16; ++x) {
-      vpx_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse);
-      total_sse += sse;
-      pa += 16;
-      pb += 16;
-    }
-    a += 16 * a_stride;
-    b += 16 * b_stride;
-  }
-  return total_sse;
-}
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
-typedef struct {
-  double psnr[4];       // total/y/u/v
-  uint64_t sse[4];      // total/y/u/v
-  uint32_t samples[4];  // total/y/u/v
-} PSNR_STATS;
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static void calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
-                             const YV12_BUFFER_CONFIG *b,
-                             PSNR_STATS *psnr,
-                             unsigned int bit_depth,
-                             unsigned int in_bit_depth) {
-  const int widths[3] =
-      {a->y_crop_width,  a->uv_crop_width,  a->uv_crop_width };
-  const int heights[3] =
-      {a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
-  const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer,  a->v_buffer };
-  const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
-  const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer,  b->v_buffer };
-  const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
-  int i;
-  uint64_t total_sse = 0;
-  uint32_t total_samples = 0;
-  const double peak = (double)((1 << in_bit_depth) - 1);
-  const unsigned int input_shift = bit_depth - in_bit_depth;
-
-  for (i = 0; i < 3; ++i) {
-    const int w = widths[i];
-    const int h = heights[i];
-    const uint32_t samples = w * h;
-    uint64_t sse;
-    if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
-      if (input_shift) {
-        sse = highbd_get_sse_shift(a_planes[i], a_strides[i],
-                                   b_planes[i], b_strides[i], w, h,
-                                   input_shift);
-      } else {
-        sse = highbd_get_sse(a_planes[i], a_strides[i],
-                             b_planes[i], b_strides[i], w, h);
-      }
-    } else {
-      sse = get_sse(a_planes[i], a_strides[i],
-                    b_planes[i], b_strides[i],
-                    w, h);
-    }
-    psnr->sse[1 + i] = sse;
-    psnr->samples[1 + i] = samples;
-    psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
-
-    total_sse += sse;
-    total_samples += samples;
-  }
-
-  psnr->sse[0] = total_sse;
-  psnr->samples[0] = total_samples;
-  psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
-                                  (double)total_sse);
-}
-
-#else  // !CONFIG_VP9_HIGHBITDEPTH
-
-static void calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b,
-                      PSNR_STATS *psnr) {
-  static const double peak = 255.0;
-  const int widths[3]        = {
-      a->y_crop_width, a->uv_crop_width, a->uv_crop_width};
-  const int heights[3]       = {
-      a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
-  const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer};
-  const int a_strides[3]     = {a->y_stride, a->uv_stride, a->uv_stride};
-  const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer};
-  const int b_strides[3]     = {b->y_stride, b->uv_stride, b->uv_stride};
-  int i;
-  uint64_t total_sse = 0;
-  uint32_t total_samples = 0;
-
-  for (i = 0; i < 3; ++i) {
-    const int w = widths[i];
-    const int h = heights[i];
-    const uint32_t samples = w * h;
-    const uint64_t sse = get_sse(a_planes[i], a_strides[i],
-                                 b_planes[i], b_strides[i],
-                                 w, h);
-    psnr->sse[1 + i] = sse;
-    psnr->samples[1 + i] = samples;
-    psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
-
-    total_sse += sse;
-    total_samples += samples;
-  }
-
-  psnr->sse[0] = total_sse;
-  psnr->samples[0] = total_samples;
-  psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
-                                  (double)total_sse);
-}
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
 static void generate_psnr_packet(VP9_COMP *cpi) {
   struct vpx_codec_cx_pkt pkt;
   int i;
@@ -3061,7 +2805,7 @@ static void output_frame_level_debug_stats(VP9_COMP *cpi) {
 
   vpx_clear_system_state();
 
-  recon_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+  recon_err = vpx_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
 
   if (cpi->twopass.total_left_stats.coded_error != 0.0)
     fprintf(f, "%10u %dx%d %10d %10d %d %d %10d %10d %10d %10d"
@@ -3571,12 +3315,12 @@ static void encode_with_recode_loop(VP9_COMP *cpi,
 
 #if CONFIG_VP9_HIGHBITDEPTH
         if (cm->use_highbitdepth) {
-          kf_err = vp9_highbd_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+          kf_err = vpx_highbd_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
         } else {
-          kf_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+          kf_err = vpx_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
         }
 #else
-        kf_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+        kf_err = vpx_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
         // Prevent possible divide by zero error below for perfect KF
@@ -3967,13 +3711,13 @@ static void encode_frame_to_data_rate(VP9_COMP *cpi,
   if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) {
 #if CONFIG_VP9_HIGHBITDEPTH
     if (cm->use_highbitdepth) {
-      cpi->ambient_err = vp9_highbd_get_y_sse(cpi->Source,
+      cpi->ambient_err = vpx_highbd_get_y_sse(cpi->Source,
                                               get_frame_new_buffer(cm));
     } else {
-      cpi->ambient_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+      cpi->ambient_err = vpx_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
     }
 #else
-    cpi->ambient_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+    cpi->ambient_err = vpx_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
 #endif  // CONFIG_VP9_HIGHBITDEPTH
   }
 
@@ -4844,28 +4588,6 @@ void vp9_set_svc(VP9_COMP *cpi, int use_svc) {
   return;
 }
 
-int64_t vp9_get_y_sse(const YV12_BUFFER_CONFIG *a,
-                      const YV12_BUFFER_CONFIG *b) {
-  assert(a->y_crop_width == b->y_crop_width);
-  assert(a->y_crop_height == b->y_crop_height);
-
-  return get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
-                 a->y_crop_width, a->y_crop_height);
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-int64_t vp9_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
-                             const YV12_BUFFER_CONFIG *b) {
-  assert(a->y_crop_width == b->y_crop_width);
-  assert(a->y_crop_height == b->y_crop_height);
-  assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
-  assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
-
-  return highbd_get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
-                        a->y_crop_width, a->y_crop_height);
-}
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
 int vp9_get_quantizer(VP9_COMP *cpi) {
   return cpi->common.base_qindex;
 }
diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index 8759cbe10..017fa61f4 100644
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -608,12 +608,6 @@ static INLINE int allocated_tokens(TileInfo tile) {
   return get_token_alloc(tile_mb_rows, tile_mb_cols);
 }
 
-int64_t vp9_get_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b);
-#if CONFIG_VP9_HIGHBITDEPTH
-int64_t vp9_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
-                             const YV12_BUFFER_CONFIG *b);
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
 void vp9_scale_references(VP9_COMP *cpi);
 
 void vp9_update_reference_frames(VP9_COMP *cpi);
diff --git a/vp9/encoder/vp9_picklpf.c b/vp9/encoder/vp9_picklpf.c
index f6b1dfcd5..80ab23898 100644
--- a/vp9/encoder/vp9_picklpf.c
+++ b/vp9/encoder/vp9_picklpf.c
@@ -12,7 +12,7 @@
 #include <limits.h>
 
 #include "./vpx_scale_rtcd.h"
-
+#include "vpx_dsp/psnr.h"
 #include "vpx_mem/vpx_mem.h"
 #include "vpx_ports/mem.h"
 
@@ -52,12 +52,12 @@ static int64_t try_filter_frame(const YV12_BUFFER_CONFIG *sd,
 
 #if CONFIG_VP9_HIGHBITDEPTH
   if (cm->use_highbitdepth) {
-    filt_err = vp9_highbd_get_y_sse(sd, cm->frame_to_show);
+    filt_err = vpx_highbd_get_y_sse(sd, cm->frame_to_show);
   } else {
-    filt_err = vp9_get_y_sse(sd, cm->frame_to_show);
+    filt_err = vpx_get_y_sse(sd, cm->frame_to_show);
   }
 #else
-  filt_err = vp9_get_y_sse(sd, cm->frame_to_show);
+  filt_err = vpx_get_y_sse(sd, cm->frame_to_show);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
   // Re-instate the unfiltered frame
diff --git a/vpx/internal/vpx_psnr.h b/vpx/internal/vpx_psnr.h
deleted file mode 100644
index 0e900858e..000000000
--- a/vpx/internal/vpx_psnr.h
+++ /dev/null
@@ -1,36 +0,0 @@
-/*
- *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-#ifndef VPX_INTERNAL_VPX_PSNR_H_
-#define VPX_INTERNAL_VPX_PSNR_H_
-
-#define MAX_PSNR 100.0
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-// TODO(dkovalev) change vpx_sse_to_psnr signature: double -> int64_t
-
-/*!\brief Converts SSE to PSNR
- *
- * Converts sum of squared errros (SSE) to peak signal-to-noise ratio (PNSR).
- *
- * \param[in]    samples       Number of samples
- * \param[in]    peak          Max sample value
- * \param[in]    sse           Sum of squared errors
- */
-double vpx_sse_to_psnr(double samples, double peak, double sse);
-
-#ifdef __cplusplus
-}  // extern "C"
-#endif
-
-#endif  // VPX_INTERNAL_VPX_PSNR_H_
diff --git a/vpx/src/vpx_psnr.c b/vpx/src/vpx_psnr.c
deleted file mode 100644
index 27a6180ce..000000000
--- a/vpx/src/vpx_psnr.c
+++ /dev/null
@@ -1,23 +0,0 @@
-/*
- *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <math.h>
-
-#include "vpx/internal/vpx_psnr.h"
-
-
-double vpx_sse_to_psnr(double samples, double peak, double sse) {
-  if (sse > 0.0) {
-    const double psnr = 10.0 * log10(samples * peak * peak / sse);
-    return psnr > MAX_PSNR ? MAX_PSNR : psnr;
-  } else {
-    return MAX_PSNR;
-  }
-}
diff --git a/vpx/vpx_codec.mk b/vpx/vpx_codec.mk
index ccdef040c..b77f45817 100644
--- a/vpx/vpx_codec.mk
+++ b/vpx/vpx_codec.mk
@@ -36,10 +36,8 @@ API_SRCS-yes += vpx_decoder.h
 API_SRCS-yes += src/vpx_encoder.c
 API_SRCS-yes += vpx_encoder.h
 API_SRCS-yes += internal/vpx_codec_internal.h
-API_SRCS-yes += internal/vpx_psnr.h
 API_SRCS-yes += src/vpx_codec.c
 API_SRCS-yes += src/vpx_image.c
-API_SRCS-yes += src/vpx_psnr.c
 API_SRCS-yes += vpx_codec.h
 API_SRCS-yes += vpx_codec.mk
 API_SRCS-yes += vpx_frame_buffer.h
diff --git a/vpx_dsp/psnr.c b/vpx_dsp/psnr.c
new file mode 100644
index 000000000..ee803f6ac
--- /dev/null
+++ b/vpx_dsp/psnr.c
@@ -0,0 +1,298 @@
+/*
+*  Copyright (c) 2016 The WebM project authors. All Rights Reserved.
+*
+*  Use of this source code is governed by a BSD-style license
+*  that can be found in the LICENSE file in the root of the source
+*  tree. An additional intellectual property rights grant can be found
+*  in the file PATENTS.  All contributing project authors may
+*  be found in the AUTHORS file in the root of the source tree.
+*/
+
+#include <math.h>
+#include <assert.h>
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/psnr.h"
+#include "vpx_scale/yv12config.h"
+
+
+double vpx_sse_to_psnr(double samples, double peak, double sse) {
+  if (sse > 0.0) {
+    const double psnr = 10.0 * log10(samples * peak * peak / sse);
+    return psnr > MAX_PSNR ? MAX_PSNR : psnr;
+  } else {
+    return MAX_PSNR;
+  }
+}
+
+/* TODO(yaowu): The block_variance calls the unoptimized versions of variance()
+* and highbd_8_variance(). It should not.
+*/
+static void encoder_variance(const uint8_t *a, int  a_stride,
+  const uint8_t *b, int  b_stride,
+  int  w, int  h, unsigned int *sse, int *sum) {
+  int i, j;
+
+  *sum = 0;
+  *sse = 0;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      const int diff = a[j] - b[j];
+      *sum += diff;
+      *sse += diff * diff;
+    }
+
+    a += a_stride;
+    b += b_stride;
+  }
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static void encoder_highbd_variance64(const uint8_t *a8, int  a_stride,
+  const uint8_t *b8, int  b_stride,
+  int w, int h, uint64_t *sse,
+  uint64_t *sum) {
+  int i, j;
+
+  uint16_t *a = CONVERT_TO_SHORTPTR(a8);
+  uint16_t *b = CONVERT_TO_SHORTPTR(b8);
+  *sum = 0;
+  *sse = 0;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      const int diff = a[j] - b[j];
+      *sum += diff;
+      *sse += diff * diff;
+    }
+    a += a_stride;
+    b += b_stride;
+  }
+}
+
+static void encoder_highbd_8_variance(const uint8_t *a8, int  a_stride,
+  const uint8_t *b8, int  b_stride,
+  int w, int h,
+  unsigned int *sse, int *sum) {
+  uint64_t sse_long = 0;
+  uint64_t sum_long = 0;
+  encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h,
+    &sse_long, &sum_long);
+  *sse = (unsigned int)sse_long;
+  *sum = (int)sum_long;
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+static int64_t get_sse(const uint8_t *a, int a_stride,
+  const uint8_t *b, int b_stride,
+  int width, int height) {
+  const int dw = width % 16;
+  const int dh = height % 16;
+  int64_t total_sse = 0;
+  unsigned int sse = 0;
+  int sum = 0;
+  int x, y;
+
+  if (dw > 0) {
+    encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
+      dw, height, &sse, &sum);
+    total_sse += sse;
+  }
+
+  if (dh > 0) {
+    encoder_variance(&a[(height - dh) * a_stride], a_stride,
+      &b[(height - dh) * b_stride], b_stride,
+      width - dw, dh, &sse, &sum);
+    total_sse += sse;
+  }
+
+  for (y = 0; y < height / 16; ++y) {
+    const uint8_t *pa = a;
+    const uint8_t *pb = b;
+    for (x = 0; x < width / 16; ++x) {
+      vpx_mse16x16(pa, a_stride, pb, b_stride, &sse);
+      total_sse += sse;
+
+      pa += 16;
+      pb += 16;
+    }
+
+    a += 16 * a_stride;
+    b += 16 * b_stride;
+  }
+
+  return total_sse;
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
+  const uint8_t *b8, int b_stride,
+  int width, int height,
+  unsigned int input_shift) {
+  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
+  const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
+  int64_t total_sse = 0;
+  int x, y;
+  for (y = 0; y < height; ++y) {
+    for (x = 0; x < width; ++x) {
+      int64_t diff;
+      diff = (a[x] >> input_shift) - (b[x] >> input_shift);
+      total_sse += diff * diff;
+    }
+    a += a_stride;
+    b += b_stride;
+  }
+  return total_sse;
+}
+
+int64_t highbd_get_sse(const uint8_t *a, int a_stride,
+  const uint8_t *b, int b_stride,
+  int width, int height) {
+  int64_t total_sse = 0;
+  int x, y;
+  const int dw = width % 16;
+  const int dh = height % 16;
+  unsigned int sse = 0;
+  int sum = 0;
+  if (dw > 0) {
+    encoder_highbd_8_variance(&a[width - dw], a_stride,
+      &b[width - dw], b_stride,
+      dw, height, &sse, &sum);
+    total_sse += sse;
+  }
+  if (dh > 0) {
+    encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
+      &b[(height - dh) * b_stride], b_stride,
+      width - dw, dh, &sse, &sum);
+    total_sse += sse;
+  }
+  for (y = 0; y < height / 16; ++y) {
+    const uint8_t *pa = a;
+    const uint8_t *pb = b;
+    for (x = 0; x < width / 16; ++x) {
+      vpx_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse);
+      total_sse += sse;
+      pa += 16;
+      pb += 16;
+    }
+    a += 16 * a_stride;
+    b += 16 * b_stride;
+  }
+  return total_sse;
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+
+int64_t vpx_get_y_sse(const YV12_BUFFER_CONFIG *a,
+  const YV12_BUFFER_CONFIG *b) {
+  assert(a->y_crop_width == b->y_crop_width);
+  assert(a->y_crop_height == b->y_crop_height);
+
+  return get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
+    a->y_crop_width, a->y_crop_height);
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+int64_t vpx_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
+  const YV12_BUFFER_CONFIG *b) {
+  assert(a->y_crop_width == b->y_crop_width);
+  assert(a->y_crop_height == b->y_crop_height);
+  assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
+  assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
+
+  return highbd_get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
+    a->y_crop_width, a->y_crop_height);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
+  const YV12_BUFFER_CONFIG *b,
+  PSNR_STATS *psnr,
+  unsigned int bit_depth,
+  unsigned int in_bit_depth) {
+  const int widths[3] =
+  { a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
+  const int heights[3] =
+  { a->y_crop_height, a->uv_crop_height, a->uv_crop_height };
+  const uint8_t *a_planes[3] = { a->y_buffer, a->u_buffer, a->v_buffer };
+  const int a_strides[3] = { a->y_stride, a->uv_stride, a->uv_stride };
+  const uint8_t *b_planes[3] = { b->y_buffer, b->u_buffer, b->v_buffer };
+  const int b_strides[3] = { b->y_stride, b->uv_stride, b->uv_stride };
+  int i;
+  uint64_t total_sse = 0;
+  uint32_t total_samples = 0;
+  const double peak = (double)((1 << in_bit_depth) - 1);
+  const unsigned int input_shift = bit_depth - in_bit_depth;
+
+  for (i = 0; i < 3; ++i) {
+    const int w = widths[i];
+    const int h = heights[i];
+    const uint32_t samples = w * h;
+    uint64_t sse;
+    if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
+      if (input_shift) {
+        sse = highbd_get_sse_shift(a_planes[i], a_strides[i],
+          b_planes[i], b_strides[i], w, h,
+          input_shift);
+      } else {
+        sse = highbd_get_sse(a_planes[i], a_strides[i],
+          b_planes[i], b_strides[i], w, h);
+      }
+    } else {
+      sse = get_sse(a_planes[i], a_strides[i],
+        b_planes[i], b_strides[i],
+        w, h);
+    }
+    psnr->sse[1 + i] = sse;
+    psnr->samples[1 + i] = samples;
+    psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
+
+    total_sse += sse;
+    total_samples += samples;
+  }
+
+  psnr->sse[0] = total_sse;
+  psnr->samples[0] = total_samples;
+  psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
+    (double)total_sse);
+}
+
+#else  // !CONFIG_VP9_HIGHBITDEPTH
+
+void calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b,
+  PSNR_STATS *psnr) {
+  static const double peak = 255.0;
+  const int widths[3] = {
+    a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
+  const int heights[3] = {
+    a->y_crop_height, a->uv_crop_height, a->uv_crop_height };
+  const uint8_t *a_planes[3] = { a->y_buffer, a->u_buffer, a->v_buffer };
+  const int a_strides[3] = { a->y_stride, a->uv_stride, a->uv_stride };
+  const uint8_t *b_planes[3] = { b->y_buffer, b->u_buffer, b->v_buffer };
+  const int b_strides[3] = { b->y_stride, b->uv_stride, b->uv_stride };
+  int i;
+  uint64_t total_sse = 0;
+  uint32_t total_samples = 0;
+
+  for (i = 0; i < 3; ++i) {
+    const int w = widths[i];
+    const int h = heights[i];
+    const uint32_t samples = w * h;
+    const uint64_t sse = get_sse(a_planes[i], a_strides[i],
+      b_planes[i], b_strides[i],
+      w, h);
+    psnr->sse[1 + i] = sse;
+    psnr->samples[1 + i] = samples;
+    psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
+
+    total_sse += sse;
+    total_samples += samples;
+  }
+
+  psnr->sse[0] = total_sse;
+  psnr->samples[0] = total_samples;
+  psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
+    (double)total_sse);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
diff --git a/vpx_dsp/psnr.h b/vpx_dsp/psnr.h
new file mode 100644
index 000000000..e58eb3de0
--- /dev/null
+++ b/vpx_dsp/psnr.h
@@ -0,0 +1,65 @@
+/*
+*  Copyright (c) 2016 The WebM project authors. All Rights Reserved.
+*
+*  Use of this source code is governed by a BSD-style license
+*  that can be found in the LICENSE file in the root of the source
+*  tree. An additional intellectual property rights grant can be found
+*  in the file PATENTS.  All contributing project authors may
+*  be found in the AUTHORS file in the root of the source tree.
+*/
+
+#ifndef VPX_DSP_PSNR_H_
+#define VPX_DSP_PSNR_H_
+
+
+#include "vpx_scale/yv12config.h"
+
+#define MAX_PSNR 100.0
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct {
+  double psnr[4];       // total/y/u/v
+  uint64_t sse[4];      // total/y/u/v
+  uint32_t samples[4];  // total/y/u/v
+} PSNR_STATS;
+
+// TODO(dkovalev) change vpx_sse_to_psnr signature: double -> int64_t
+
+/*!\brief Converts SSE to PSNR
+*
+* Converts sum of squared errros (SSE) to peak signal-to-noise ratio (PNSR).
+*
+* \param[in]    samples       Number of samples
+* \param[in]    peak          Max sample value
+* \param[in]    sse           Sum of squared errors
+*/
+double vpx_sse_to_psnr(double samples, double peak, double sse);
+int64_t vpx_get_y_sse(const YV12_BUFFER_CONFIG *a,
+                      const YV12_BUFFER_CONFIG *b);
+#if CONFIG_VP9_HIGHBITDEPTH
+int64_t vpx_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
+                             const YV12_BUFFER_CONFIG *b);
+void calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
+                      const YV12_BUFFER_CONFIG *b,
+                      PSNR_STATS *psnr,
+                      unsigned int bit_depth,
+                      unsigned int in_bit_depth);
+int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
+                             const uint8_t *b8, int b_stride,
+                             int width, int height,
+                             unsigned int input_shift);
+#else
+void calc_psnr(const YV12_BUFFER_CONFIG *a,
+               const YV12_BUFFER_CONFIG *b,
+               PSNR_STATS *psnr);
+#endif
+int64_t highbd_get_sse(const uint8_t *a, int a_stride,
+                       const uint8_t *b, int b_stride,
+                       int width, int height);
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+#endif  // VPX_DSP_PSNR_H_
diff --git a/vpx_dsp/psnrhvs.c b/vpx_dsp/psnrhvs.c
index 4d3d6eea0..9b70c6a49 100644
--- a/vpx_dsp/psnrhvs.c
+++ b/vpx_dsp/psnrhvs.c
@@ -19,7 +19,7 @@
 #include "./vpx_dsp_rtcd.h"
 #include "vpx_dsp/ssim.h"
 #include "vpx_ports/system_state.h"
-#include "vpx/internal/vpx_psnr.h"
+#include "vpx_dsp/psnr.h"
 
 #if !defined(M_PI)
 # define M_PI (3.141592653589793238462643)
diff --git a/vpx_dsp/vpx_dsp.mk b/vpx_dsp/vpx_dsp.mk
index a44f948fa..dbb41aaea 100644
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -22,6 +22,8 @@ DSP_SRCS-yes += bitwriter.h
 DSP_SRCS-yes += bitwriter.c
 DSP_SRCS-yes += bitwriter_buffer.c
 DSP_SRCS-yes += bitwriter_buffer.h
+DSP_SRCS-yes += psnr.c
+DSP_SRCS-yes += psnr.h
 DSP_SRCS-$(CONFIG_INTERNAL_STATS) += ssim.c
 DSP_SRCS-$(CONFIG_INTERNAL_STATS) += ssim.h
 DSP_SRCS-$(CONFIG_INTERNAL_STATS) += psnrhvs.c