question on FE_TOWARDZERO

[Francois Grieu]

Does ISO C insure this program's output is 0 ?

#include <fenv.h>
#include <math.h>
int main(void)
{
#ifdef FE_TOWARDZERO
unsigned long a;
double d;
int j;
fesetround(FE_TOWARDZERO);
a = 0xFFFFFFFF;
d = 0;
for(j=10;--j;)
d = (d+a)*((double)1/65536/65536);
do
if ((unsigned long)floor(d*a)==a)
return 1;
while (--a);
#endif
return 0;
}
As an aside, my <fenv.h> defines FE_TOWARDZERO as an enum,
isn't that a defect (assuming FE_TOWARDZERO is supported)?

François Grieu

[Kevin Bracey]

In message <fg**************************@news.fu-berlin.de>
Francois Grieu <fg****@micronet.fr> wrote:

Does ISO C insure this program's output is 0 ?

#include <fenv.h>
#include <math.h>
int main(void)
{
You need to put

#pragma STDC FENV_ACCESS ON

either here or before main, otherwise you're not allowed to change the
rounding mode.
#ifdef FE_TOWARDZERO
unsigned long a;
double d;
int j;
fesetround(FE_TOWARDZERO);
a = 0xFFFFFFFF;
d = 0;
for(j=10;--j;)
d = (d+a)*((double)1/65536/65536);
do
if ((unsigned long)floor(d*a)==a)
return 1;
while (--a);
#endif
return 0;
}
Looks like it's got to return 0 to me, yes.
As an aside, my <fenv.h> defines FE_TOWARDZERO as an enum,
isn't that a defect (assuming FE_TOWARDZERO is supported)?

FE_TOWARDZERO has to exist as a macro if the rounding mode is supported, so
that it can be tested for with #ifdef. The macro could expand to an enum
value though:

enum { FE_TONEAREST, FE_TOWARDZERO };
#define FE_TOWARDZERO FE_TOWARDZERO

--
Kevin Bracey, Principal Software Engineer
Tematic Ltd Tel: [phone]
182-190 Newmarket Road Fax: [phone]
Cambridge, CB5 8HE, United Kingdom WWW: http://www.tematic.com/

[Francois Grieu]

In article <f1****************@tematic.com>,
Kevin Bracey <ke**********@tematic.com> wrote:

You need to put

#pragma STDC FENV_ACCESS ON

either here or before main, otherwise you're not allowed to change the
rounding mode.
Many thanks; I did not realize this was a requirement, because
it is not one in my usual environements.

Looks like it's got to return 0 to me, yes.
My criptic question bounces with a terse answer.

Of course you guessed what I REALLY wan to know:
is it guaranteed that when FE_TOWARDZERO is in effect,
for any double d in range ]0,1[,
for any unsigned long a in range ]0,0xFFFFFFFF[
floor(d*a)<a holds.

FE_TOWARDZERO has to exist as a macro if the rounding mode is supported
The macro could expand to an enum value

The author of my <fenv.h> missed that requirement. I should report..
François Grieu