[Flac-dev] Re: 0.9 problems

[Matt Zimmerman]

On Sat, May 19, 2001 at 08:03:54PM +0000, Christian Weisgerber wrote:

> Matt Zimmerman <mdz at debian.org> wrote:
> 
> > 0.9.  As I said, I was using an 8-bit sample,
> 
> Ah, that didn't quite register with me.  I'm using a CD-style
> 44.1kHz/stereo/16-bit test file.

I repeated my test with a 44.1kHz/stereo/16bit file.  I get a floating point
exception when encoding at fixed.c:84.  Decoding works fine for me.

The floating point exception occurs here:

84              residual_bits_per_sample[0] = (real)((data_len > 0) ? log(M_LN2 * (real)total_error_0  / (real) data_len) / M_LN2 : 0.0);

At the point of failure, the part inside the log() is zero, so the result of
the log is -inf.  On i386, dividing -inf / M_LN2 seems to evaluate to zero,
while on alpha it gives a floating point exception.  Here's a small test
program:

#include <math.h>

int main() {
  double foo = log((double)0);
  printf("%e\n", foo);
  printf("%e\n", foo / M_LN2);
  return 0;
}

I think this could be fixed by changing the (data_len > 0) test to be (data_len
> 0 && total_error_X > 0).

-- 
 - mdz