Avoid overflow in WebRtcSpl_DotProductWithScale

webrtc/common_audio/signal_processing/dot_product_with_scale.c

Index: webrtc/common_audio/signal_processing/dot_product_with_scale.c
diff --git a/webrtc/common_audio/signal_processing/dot_product_with_scale.c b/webrtc/common_audio/signal_processing/dot_product_with_scale.c
index 1302d62541266fdc45b07576de5ff5026d69032f..202d0ccd6c93890bd4cf8e5ef6583235c7b33a7d 100644
--- a/webrtc/common_audio/signal_processing/dot_product_with_scale.c
+++ b/webrtc/common_audio/signal_processing/dot_product_with_scale.c
@@ -14,7 +14,7 @@ int32_t WebRtcSpl_DotProductWithScale(const int16_t* vector1,
                                       const int16_t* vector2,
                                       size_t length,
                                       int scaling) {
-  int32_t sum = 0;
+  int64_t sum = 0;
   size_t i = 0;
 
   /* Unroll the loop to improve performance. */
@@ -28,5 +28,5 @@ int32_t WebRtcSpl_DotProductWithScale(const int16_t* vector1,
     sum += (vector1[i] * vector2[i]) >> scaling;
   }
 
-  return sum;
+  return (int32_t) sum;

[kwiberg-webrtc, 2017/02/28 10:07:47]

You'll get garbage if this cast truncates. And wasn't the reason for this change
that the int32_t sum was found to be overflowing?

Suggestion: This is a tiny file. Convert it to C++ so that you can use
saturated_cast.

[hlundin-webrtc, 2017/02/28 11:58:45]

Done.

 }