VadCore: Allow signed multiplication overflow that we don't know how to fix

webrtc/common_audio/vad/vad_core.c

 /*
  *  Copyright (c) 2012 The WebRTC 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.
  */
 
 #include "webrtc/common_audio/vad/vad_core.h"
 
+#include "webrtc/base/sanitizer.h"
 #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
 #include "webrtc/common_audio/vad/vad_filterbank.h"
 #include "webrtc/common_audio/vad/vad_gmm.h"
 #include "webrtc/common_audio/vad/vad_sp.h"
 #include "webrtc/typedefs.h"
 
 // Spectrum Weighting
 static const int16_t kSpectrumWeight[kNumChannels] = { 6, 8, 10, 12, 14, 16 };
 static const int16_t kNoiseUpdateConst = 655; // Q15
 static const int16_t kSpeechUpdateConst = 6554; // Q15
@@ skipping 80 matching lines @@
   int k;
   int32_t weighted_average = 0;
 
   for (k = 0; k < kNumGaussians; k++) {
     data[k * kNumChannels] += offset;
     weighted_average += data[k * kNumChannels] * weights[k * kNumChannels];
   }
   return weighted_average;
 }
 
+// An s16 x s32 -> s32 multiplication that's allowed to overflow. (It's still
+// undefined behavior, so not a good idea; this just makes UBSan ignore the
+// violation, so that our old code can continue to do what it's always been
+// doing.)
+static inline int32_t OverflowingMulS16ByS32ToS32(int16_t a, int32_t b)
+    RTC_NO_SANITIZE("signed-integer-overflow") {
+  return a * b;
+}
+
 // Calculates the probabilities for both speech and background noise using
 // Gaussian Mixture Models (GMM). A hypothesis-test is performed to decide which
 // type of signal is most probable.
 //
 // - self           [i/o] : Pointer to VAD instance
 // - features       [i]   : Feature vector of length |kNumChannels|
 //                          = log10(energy in frequency band)
 // - total_power    [i]   : Total power in audio frame.
 // - frame_length   [i]   : Number of input samples
 //
@@ skipping 248 matching lines @@
           // Update GMM variance vectors.
           // deltaN * (features[channel] - nmk) - 1
           // Q4 - (Q7 >> 3) = Q4.
           tmp_s16 = features[channel] - (nmk >> 3);
           // (Q11 * Q4 >> 3) = Q12.
           tmp1_s32 = (deltaN[gaussian] * tmp_s16) >> 3;
           tmp1_s32 -= 4096;
 
           // (Q14 >> 2) * Q12 = Q24.
           tmp_s16 = (ngprvec[gaussian] + 2) >> 2;
-          tmp2_s32 = tmp_s16 * tmp1_s32;
+          tmp2_s32 = OverflowingMulS16ByS32ToS32(tmp_s16, tmp1_s32);
           // Q20  * approx 0.001 (2^-10=0.0009766), hence,
           // (Q24 >> 14) = (Q24 >> 4) / 2^10 = Q20.
           tmp1_s32 = tmp2_s32 >> 14;
 
           // Q20 / Q7 = Q13.
           if (tmp1_s32 > 0) {
             tmp_s16 = (int16_t) WebRtcSpl_DivW32W16(tmp1_s32, nsk);
           } else {
             tmp_s16 = (int16_t) WebRtcSpl_DivW32W16(-tmp1_s32, nsk);
             tmp_s16 = -tmp_s16;
@@ skipping 275 matching lines @@
 
     // Get power in the bands
     total_power = WebRtcVad_CalculateFeatures(inst, speech_frame, frame_length,
                                               feature_vector);
 
     // Make a VAD
     inst->vad = GmmProbability(inst, feature_vector, total_power, frame_length);
 
     return inst->vad;
 }