Duplicate reference variance code

Some places are using the unoptimized variance function. This was never
intended and does not fit into the optimization framework.

Change-Id: Id96238407aad03b0ffd4a46cd183555a026daedc

3 files changed, 132 insertions, 71 deletions

diff --git a/vp9/encoder/vp9_aq_variance.c b/vp9/encoder/vp9_aq_variance.c
index 9e5d9ee6a..f072717f1 100644
--- a/vp9/encoder/vp9_aq_variance.c
+++ b/vp9/encoder/vp9_aq_variance.c
@@ -82,6 +82,61 @@ void vp9_vaq_frame_setup(VP9_COMP *cpi) {
   }
 }
 
+/* TODO(agrange, paulwilkins): The block_variance calls the unoptimized versions
+ * of variance() and highbd_8_variance(). It should not.
+ */
+static void aq_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 aq_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 aq_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;
+  aq_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 unsigned int block_variance(VP9_COMP *cpi, MACROBLOCK *x,
                                    BLOCK_SIZE bs) {
@@ -98,18 +153,18 @@ static unsigned int block_variance(VP9_COMP *cpi, MACROBLOCK *x,
     int avg;
 #if CONFIG_VP9_HIGHBITDEPTH
     if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-      highbd_8_variance(x->plane[0].src.buf, x->plane[0].src.stride,
-                        CONVERT_TO_BYTEPTR(vp9_highbd_64_zeros), 0, bw, bh,
-                        &sse, &avg);
+      aq_highbd_8_variance(x->plane[0].src.buf, x->plane[0].src.stride,
+                           CONVERT_TO_BYTEPTR(vp9_highbd_64_zeros), 0, bw, bh,
+                           &sse, &avg);
       sse >>= 2 * (xd->bd - 8);
       avg >>= (xd->bd - 8);
     } else {
-      variance(x->plane[0].src.buf, x->plane[0].src.stride,
-               vp9_64_zeros, 0, bw, bh, &sse, &avg);
+      aq_variance(x->plane[0].src.buf, x->plane[0].src.stride,
+                  vp9_64_zeros, 0, bw, bh, &sse, &avg);
     }
 #else
-    variance(x->plane[0].src.buf, x->plane[0].src.stride,
-             vp9_64_zeros, 0, bw, bh, &sse, &avg);
+    aq_variance(x->plane[0].src.buf, x->plane[0].src.stride,
+                vp9_64_zeros, 0, bw, bh, &sse, &avg);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
     var = sse - (((int64_t)avg * avg) / (bw * bh));
     return (256 * var) / (bw * bh);
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 95f9bc488..b79bc00d2 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -2056,6 +2056,65 @@ 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) {
@@ -2067,15 +2126,15 @@ static int64_t get_sse(const uint8_t *a, int a_stride,
   int x, y;
 
   if (dw > 0) {
-    variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
-             dw, height, &sse, &sum);
+    encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
+                     dw, height, &sse, &sum);
     total_sse += sse;
   }
 
   if (dh > 0) {
-    variance(&a[(height - dh) * a_stride], a_stride,
-             &b[(height - dh) * b_stride], b_stride,
-             width - dw, dh, &sse, &sum);
+    encoder_variance(&a[(height - dh) * a_stride], a_stride,
+                     &b[(height - dh) * b_stride], b_stride,
+                     width - dw, dh, &sse, &sum);
     total_sse += sse;
   }
 
@@ -2128,14 +2187,15 @@ static int64_t highbd_get_sse(const uint8_t *a, int a_stride,
   unsigned int sse = 0;
   int sum = 0;
   if (dw > 0) {
-    highbd_8_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
-                      dw, height, &sse, &sum);
+    encoder_highbd_8_variance(&a[width - dw], a_stride,
+                              &b[width - dw], b_stride,
+                              dw, height, &sse, &sum);
     total_sse += sse;
   }
   if (dh > 0) {
-    highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
-                      &b[(height - dh) * b_stride], b_stride,
-                      width - dw, dh, &sse, &sum);
+    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) {
diff --git a/vp9/encoder/vp9_variance.h b/vp9/encoder/vp9_variance.h
index 8fc47a850..0a8739510 100644
--- a/vp9/encoder/vp9_variance.h
+++ b/vp9/encoder/vp9_variance.h
@@ -18,60 +18,6 @@
 extern "C" {
 #endif
 
-// TODO(johannkoenig): All functions which depend on
-// [highbd_][8|10|12_]variance should be refactored or moved to vpx_dsp.
-static void 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 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 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;
-  highbd_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
-  *sse = (unsigned int)sse_long;
-  *sum = (int)sum_long;
-}
-#endif
-
 typedef unsigned int(*vp9_sad_fn_t)(const uint8_t *src_ptr,
                                     int source_stride,
                                     const uint8_t *ref_ptr,