Re: MSVC fails to compile floor.c with -fp:strict

[Michael Goffioul]

On Thu, Feb 23, 2012 at 10:12 AM, Bruno Haible <address@hidden> wrote:
> Michael Goffioul wrote:
>> I encountered a compilation error with MSVC when compiling floor.c in
>> strict floating-point mode (-fp:strict). The error is:
>>
>> "error C2099: initializer is not a constant"
>>
>> The error is triggered by the initializer for the constant variable
>> TWO_MANT_DIG and is documented here:
>>
>> http://msdn.microsoft.com/en-us/library/t801az8a(v=vs.80).aspx
>
> Does adding the line
>  #pragma fenv_access (off)
> before the initialization help?
>
> *** lib/floor.c.orig    Thu Feb 23 11:11:48 2012
> --- lib/floor.c Thu Feb 23 11:11:43 2012
> ***************
> *** 42,47 ****
> --- 42,51 ----
>  # define L_(literal) literal##f
>  #endif
>
> + #ifdef _MSC_VER
> + # pragma fenv_access (off)
> + #endif
> +
>  /* 2^(MANT_DIG-1).  */
>  static const DOUBLE TWO_MANT_DIG =
>    /* Assume MANT_DIG <= 5 * 31.
>

To illustrate the issue, I've built a little test program:

#include <float.h>

# define DOUBLE double
# define MANT_DIG DBL_MANT_DIG

#ifdef _MSC_VER
# pragma fenv_access (off)
#endif

static const DOUBLE TWO_MANT_DIG =
  /* Assume MANT_DIG <= 5 * 31.
     Use the identity
       n = floor(n/5) + floor((n+1)/5) + ... + floor((n+4)/5).  */
  (DOUBLE) (1U << ((MANT_DIG - 1) / 5))
  * (DOUBLE) (1U << ((MANT_DIG - 1 + 1) / 5))
  * (DOUBLE) (1U << ((MANT_DIG - 1 + 2) / 5))
  * (DOUBLE) (1U << ((MANT_DIG - 1 + 3) / 5))
  * (DOUBLE) (1U << ((MANT_DIG - 1 + 4) / 5));

int main(int argc, char** argv)
{
  double x = TWO_MANT_DIG;
}

With the #pragma statement, the compilation error disappears, thanks.
I assume this will also solve the problem when compiling octave (which
is how I found out the issue initially). Note that I think the same
problem will appear in round.c and trunc.c (those are also used by
octave).

Michael.