Fix for integer overflow in NetEq.

webrtc/modules/audio_coding/neteq/expand.cc

Index: webrtc/modules/audio_coding/neteq/expand.cc
diff --git a/webrtc/modules/audio_coding/neteq/expand.cc b/webrtc/modules/audio_coding/neteq/expand.cc
index ffe4370e4e61e8766c50ce8efcf22c96bd9723bf..02d6290f2f589cfa26786920afb3f3808ef61665 100644
--- a/webrtc/modules/audio_coding/neteq/expand.cc
+++ b/webrtc/modules/audio_coding/neteq/expand.cc
@@ -681,10 +681,8 @@ void Expand::AnalyzeSignal(int16_t* random_vector) {
     } else {
       unvoiced_prescale = 0;
     }
-    int32_t unvoiced_energy = WebRtcSpl_DotProductWithScale(unvoiced_vector,
-                                                            unvoiced_vector,
-                                                            128,
-                                                            unvoiced_prescale);
+    int32_t unvoiced_energy = WebRtcSpl_DotProductWithScale_64b(

[hlundin-webrtc, 2016/12/12 21:40:41]

Will you not end up with the same overflow problem when casting the results to
int32_t?

[ivoc, 2016/12/13 08:43:09]

Oops, good catch, I'll change this to int64_t.

[kwiberg-webrtc, 2016/12/13 08:45:36]

I would imagine so. Since all the partial sums are positive, you can't have
overflow in an intermediate value without also having overflow in the final sum.

+        unvoiced_vector, unvoiced_vector, 128, unvoiced_prescale);
 
     // Normalize |unvoiced_energy| to 28 or 29 bits to preserve sqrt() accuracy.
     int16_t unvoiced_scale = WebRtcSpl_NormW32(unvoiced_energy) - 3;