On Thu, 12 Mar 2020, Vineet Gupta via Libc-alpha wrote:

> diff --git a/sysdeps/arc/bits/fenv.h b/sysdeps/arc/bits/fenv.h

This is another example of one patch fixing up another.  You're using the 
same ABI for both hard and soft float, so the first patch adding a 
bits/fenv.h header should be using that ABI for fenv_t, rather than one 
patch adding it with one ABI then a subsequent patch changing the 
definition of that type.

> diff --git a/sysdeps/arc/fpu/libm-test-ulps b/sysdeps/arc/fpu/libm-test-ulps

This will need updating for the recent changes to remove separate inline 
function testing (so there should be no ifloat or idouble entries any 

> diff --git a/sysdeps/arc/tininess.h b/sysdeps/arc/tininess.h
> new file mode 100644
> index 000000000000..1db37790f881
> --- /dev/null
> +++ b/sysdeps/arc/tininess.h
> @@ -0,0 +1 @@

In the soft-float patch you define _FP_TININESS_AFTER_ROUNDING to 0.  
Formally it doesn't really matter since you aren't supporting exceptions 
for soft-float anyway.  But typically I'd expect the definition of 
_FP_TININESS_AFTER_ROUNDING, on architecture with support for both hard 
and soft float, to match the architecture's rule for tininess detection 
for hard-float.

Joseph S. Myers

linux-snps-arc mailing list

Reply via email to