-4

I have found a lot of answers that have to do with release version problems but none with the exact opposite.

I have a loop similar to the following:

while(index < 7 && FlagIsUp)
{
  // process
  Inner Loop
    Inner Inner Loop
    Array[index] = number;
  ++index;
}

Problem is that index changes radically from 6 (the last iteration) to 17209 for an int16_t and 1133165442 for size_t. NOTHING IN THE LOOP changes the index except the ++index. I replaced the while with a for and it still happens.

It only happens in debug mode, in release version it does finish without issues.

I also added volatile to the index and results were the same, it still overflowed.

Any ideas, pointers, would be appreciated. I can't provide a working copy of the bug so any theories are welcomed, I want to exhaust my options to find the problem.

EDIT: Yes I'm sorry. I gave to little information. First off I'm working with QNX Momentics Version: 4.6.0 and my debugger is part of the GNU Compiler Collection 4.3.3.

Now the inner loop is this:

cSignalNoIndex = 0;
  while ((cSignalNoIndex < (2 * NO_PHASES + 1)) && !ShutDownFlag)
  {
    wSF0 = 0;
    wExtSF = 0;
    dwSFAcc = 0;
    dwExtSFAcc = 0;
    std::string SignalNo= " Waveform number " + Tool::toString(cSignalNoIndex);
    Results[cSignalNoIndex].printWaveForm(SignalNo);
    // Prepare Input vectors for FFT compute
    cComponent = 0;
    while (cComponent < (HCYCLE_SAMPLES << 1))
    {
      awReal[cComponent] = static_cast<int>(Results[cSignalNoIndex].WaveForm[cComponent/64][cComponent % 64]);
      awImg[cComponent] = 0;
      pwSource++;
      cComponent++;
    }
    Results[cSignalNoIndex].printWaveForm(SignalNo);
    // Get FFT (forward)
    // Changed the wPwr from 7 to something else
    iFft(&awReal[0], &awImg[0], wPwr, 1);
    Results[cSignalNoIndex].printWaveForm(SignalNo);
    // Compute magnitudes
    //fMult = pInBlock3->fMult[cSignalNoIndex];         // Get Multiplier
    fMult = 1;
    for (cComponent = 0; cComponent < HCYCLE_SAMPLES && !ShutDownFlag; cComponent++)
    {
      int64_t dlOp = static_cast<int64_t>(awReal[cComponent]) * awReal[cComponent] + static_cast<int64_t>(awImg[cComponent]) * awImg[cComponent];
      dlOp <<= 1;       // Apply sqrt(2) term to result
      dlOp = static_cast<int>(fMult * isqrt64(dlOp));

      // Store into FFT object
      oFFTMag3.wFFT[cSignalNoIndex][cComponent] = static_cast<int16_t>( dlOp );

      // Set Base frequency magnitude and accumulate harmonics
      if (cComponent == 1)  // Base
      {
        wSF0 = dlOp;

        if(cSignalNoIndex == 6)
        {
          wRefMagnitude = static_cast<int16_t> ( 0.4 * wSF0 );
        }

        if(awReal[1] != 0)  // Also get phase for Base
        {
          dfPhase = std::atan((double)((float)awImg[1]/awReal[1])) * 180.0 / PI;
        }
        else
        {
          if(awImg[1] >= 0)
          {
            dfPhase = 90.0;
          }
          else
          {
            dfPhase = -90.0;
          }
        }

        if(awReal[1] < 0)   // convert to 2*PI range
        {
          dfPhase += 180.0;
        }
        else if(awImg[1] < 0)
        {
          dfPhase += 360.0;
        }
        //// THIS IS THE LINE
        fPhase[cSignalNoIndex] = dfPhase; ////////// WTF! cSignalNoIndex = 6 - cComponent = 2
        /// HERE cSignalNoIndex is overflown
      }
    }
4

1 回答 1

3

You haven't really posted enough code but my best bet is Array[index] = number overwrites index at some point. The fact that it only happens sometimes (in your case, when debugging) is a good example of "undefined behavior".

于 2013-03-04T17:31:47.560 回答