libvpx: Pull from upstream

source/libvpx/vp9/common/vp9_convolve.c

 /*
  *  Copyright (c) 2013 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.
  */
 
@@ skipping 99 matching lines @@
   }
 }
 
 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,
                      const InterpKernel *const y_filters,
                      int y0_q4, int y_step_q4,
                      int w, int h) {
-  // Fixed size intermediate buffer places limits on parameters.
-  // Maximum intermediate_height is 324, for y_step_q4 == 80,
-  // h == 64, taps == 8.
-  // y_step_q4 of 80 allows for 1/10 scale for 5 layer svc
-  uint8_t temp[64 * 324];
+  // 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_q4 = 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[135 * 64];
   int intermediate_height = (((h - 1) * y_step_q4 + 15) >> 4) + SUBPEL_TAPS;
 
   assert(w <= 64);
   assert(h <= 64);
-  assert(y_step_q4 <= 80);
-  assert(x_step_q4 <= 80);
+  assert(y_step_q4 <= 32);
+  assert(x_step_q4 <= 32);
 
   if (intermediate_height < h)
     intermediate_height = h;
 
   convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride, temp, 64,
                  x_filters, x0_q4, x_step_q4, w, intermediate_height);
   convolve_vert(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, dst_stride,
                 y_filters, y0_q4, y_step_q4, w, h);
 }
 
@@ skipping 126 matching lines @@
   (void)filter_y;  (void)filter_y_stride;
 
   for (y = 0; y < h; ++y) {
     for (x = 0; x < w; ++x)
       dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
 
     src += src_stride;
     dst += dst_stride;
   }
 }