MIPS cross compiler soft float problem

[Michael Meissner]

On Mon, Jun 16, 2003 at 10:14:34AM +0200, Stefan Fagerström wrote:
> Hi,
> 
> I have a stange problem with my cross compiler for mips.
> It returns NaN when calling function with float arguments.
> 
> Example:
> #include <stdio.h>
> 
> float a(float a, float b, float c, float d)
> {
>   return a*b*c*d;
> }
> 
> int main()
> {
> 
>   float b = a(1.0, 0.3, 0.4, 0.5);
> 
>   printf("b = %f\n", b);
>  
> 
>   printf("mul 2.4*4.5 = %f\n", 2.4*4.5);
>   return 0;
> }
> 
> Comipled with:
> /usr/local/devenv/bin/mipsel-linux-gnu-gcc -o testfloatcall -O2 -mips2
> -mcpu=r4600 -msoft-float -msoft-float testfloatcall.c
> 
> 
> Results:
> b = nan
> mul 2.4*4.5 = 10.800000

My guess is your glibc was not compiled with -msoft-float.  Glibc's under Linux
typically don't have multilib support.  I believe the MIPS calling sequence
passes floating point values to variable argument functions in floating point
registers, rather than integer registers.

-- 
Michael Meissner
email: gnu@the-meissners.org
http://www.the-meissners.org