Clean CONVERT_TO_BYTEPTR/SHORTPTR in convolve

Replace by CAST_TO_BYTEPTR/SHORTPTR.
The rule is: if a short ptr is casted to a byte ptr, any offset
operation on the byte ptr must be doubled. We do this by casting to
short ptr first, adding offset, then casting back to byte ptr.

BUG=webm:1388

Change-Id: I9e18a73ba45ddae58fc9dae470c0ff34951fe248

4 files changed, 22 insertions, 15 deletions

diff --git a/vp9/common/vp9_reconinter.h b/vp9/common/vp9_reconinter.h
index 4fed4f7f6..cb7d1c63a 100644
--- a/vp9/common/vp9_reconinter.h
+++ b/vp9/common/vp9_reconinter.h
@@ -37,8 +37,9 @@ static INLINE void highbd_inter_predictor(
     const int subpel_x, const int subpel_y, const struct scale_factors *sf,
     int w, int h, int ref, const InterpKernel *kernel, int xs, int ys, int bd) {
   sf->highbd_predict[subpel_x != 0][subpel_y != 0][ref](
-      src, src_stride, dst, dst_stride, kernel[subpel_x], xs, kernel[subpel_y],
-      ys, w, h, bd);
+      CAST_TO_BYTEPTR(CONVERT_TO_SHORTPTR(src)), src_stride,
+      CAST_TO_BYTEPTR(CONVERT_TO_SHORTPTR(dst)), dst_stride, kernel[subpel_x],
+      xs, kernel[subpel_y], ys, w, h, bd);
 }
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 591a85ee0..ebe6758c8 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -2417,10 +2417,11 @@ static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
         uint8_t *dst_ptr = dsts[i] + (y / factor) * dst_stride + (x / factor);
 
         if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
-          vpx_highbd_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
-                               kernel[x_q4 & 0xf], 16 * src_w / dst_w,
-                               kernel[y_q4 & 0xf], 16 * src_h / dst_h,
-                               16 / factor, 16 / factor, bd);
+          vpx_highbd_convolve8(
+              CAST_TO_BYTEPTR(CONVERT_TO_SHORTPTR(src_ptr)), src_stride,
+              CAST_TO_BYTEPTR(CONVERT_TO_SHORTPTR(dst_ptr)), dst_stride,
+              kernel[x_q4 & 0xf], 16 * src_w / dst_w, kernel[y_q4 & 0xf],
+              16 * src_h / dst_h, 16 / factor, 16 / factor, bd);
         } else {
           vpx_scaled_2d(src_ptr, src_stride, dst_ptr, dst_stride,
                         kernel[x_q4 & 0xf], 16 * src_w / dst_w,
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index db2bbe7c2..f177814d6 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -2053,9 +2053,11 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data,
         this_mode_pred = &tmp[get_pred_buffer(tmp, 3)];
 #if CONFIG_VP9_HIGHBITDEPTH
         if (cm->use_highbitdepth)
-          vpx_highbd_convolve_copy(best_pred->data, best_pred->stride,
-                                   this_mode_pred->data, this_mode_pred->stride,
-                                   NULL, 0, NULL, 0, bw, bh, xd->bd);
+          vpx_highbd_convolve_copy(
+              CAST_TO_BYTEPTR(CONVERT_TO_SHORTPTR(best_pred->data)),
+              best_pred->stride,
+              CAST_TO_BYTEPTR(CONVERT_TO_SHORTPTR(this_mode_pred->data)),
+              this_mode_pred->stride, NULL, 0, NULL, 0, bw, bh, xd->bd);
         else
           vpx_convolve_copy(best_pred->data, best_pred->stride,
                             this_mode_pred->data, this_mode_pred->stride, NULL,
@@ -2162,9 +2164,11 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data,
     if (best_pred->data != orig_dst.buf && is_inter_mode(mi->mode)) {
 #if CONFIG_VP9_HIGHBITDEPTH
       if (cm->use_highbitdepth)
-        vpx_highbd_convolve_copy(best_pred->data, best_pred->stride,
-                                 pd->dst.buf, pd->dst.stride, NULL, 0, NULL, 0,
-                                 bw, bh, xd->bd);
+        vpx_highbd_convolve_copy(
+            CAST_TO_BYTEPTR(CONVERT_TO_SHORTPTR(best_pred->data)),
+            best_pred->stride,
+            CAST_TO_BYTEPTR(CONVERT_TO_SHORTPTR(pd->dst.buf)), pd->dst.stride,
+            NULL, 0, NULL, 0, bw, bh, xd->bd);
       else
         vpx_convolve_copy(best_pred->data, best_pred->stride, pd->dst.buf,
                           pd->dst.stride, NULL, 0, NULL, 0, bw, bh);
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index d23d32446..2278ddc0f 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -599,9 +599,10 @@ static void dist_block(const VP9_COMP *cpi, MACROBLOCK *x, int plane,
 
 #if CONFIG_VP9_HIGHBITDEPTH
       if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-        recon = CONVERT_TO_BYTEPTR(recon);
-        vpx_highbd_convolve_copy(dst, dst_stride, recon, 32, NULL, 0, NULL, 0,
-                                 bs, bs, xd->bd);
+        vpx_highbd_convolve_copy(CAST_TO_BYTEPTR(CONVERT_TO_SHORTPTR(dst)),
+                                 dst_stride, recon, 32, NULL, 0, NULL, 0, bs,
+                                 bs, xd->bd);
+        recon = CONVERT_TO_BYTEPTR(recon16);
         if (xd->lossless) {
           vp9_highbd_iwht4x4_add(dqcoeff, recon, 32, *eob, xd->bd);
         } else {