shithub: libvpx

Download patch

ref: b6fa0f20565ce10b4724a120ff0c8d6faff0926b
parent: 2d0f3b23a908bb54dbeddf6d1dcc666600470e2e
author: Jorge E. Moreira <jemoreira@google.com>
date: Fri May 8 11:23:48 EDT 2020

Temporarily convert to 64 bits to avoid overflows

In the vp8_cost_branch function a couple of unsigned int are being
multiplied by integer coefficients and added to later be divided by
256. While the end result most likely fits an unsigned int, the
intermediary result of multiplying and adding sometimes doesn't (I was
able to reproduce it by leaving the encoder running at 60 fps for a
while). To avoid the multiplication overflow (which is undefined
behavior and causes a wrong result anyways) the calculation is
performed using unsigned long long instead and cast to unsigned int
for return.

Bug: b/154172422
Test: run cuttlefish with webrtc enabled for an hour
Change-Id: If7ebbda38b2450a59ed3c99ffbb59dc62431a324

--- a/vp8/encoder/treewriter.h
+++ b/vp8/encoder/treewriter.h
@@ -14,6 +14,8 @@
 /* Trees map alphabets into huffman-like codes suitable for an arithmetic
    bit coder.  Timothy S Murphy  11 October 2004 */
 
+#include <stdint.h>
+
 #include "./vpx_config.h"
 #include "vp8/common/treecoder.h"
 
@@ -48,7 +50,9 @@
                                            vp8_prob p) {
   /* Imitate existing calculation */
 
-  return ((ct[0] * vp8_cost_zero(p)) + (ct[1] * vp8_cost_one(p))) >> 8;
+  return (unsigned int)(((((uint64_t)ct[0]) * vp8_cost_zero(p)) +
+                         (((uint64_t)ct[1]) * vp8_cost_one(p))) >>
+                        8);
 }
 
 /* Small functions to write explicit values and tokens, as well as