Re: [PATCH 2/2] MIPS: Hard-float rounding instructions support

[Joseph Myers <joseph@codesourcery.com>]

On Mon, 25 Dec 2023, Junxian Zhu wrote:

> diff --git a/sysdeps/mips/mips32/Implies b/sysdeps/mips/mips32/Implies
> index 6473f2517c..71b3678c6c 100644
> --- a/sysdeps/mips/mips32/Implies
> +++ b/sysdeps/mips/mips32/Implies
> @@ -1,3 +1,4 @@
> +mips/fpu
>  mips/ieee754
>  mips
>  wordsize-32
> diff --git a/sysdeps/mips/mips64/Implies b/sysdeps/mips/mips64/Implies
> index 826ff1541f..8885ebd564 100644
> --- a/sysdeps/mips/mips64/Implies
> +++ b/sysdeps/mips/mips64/Implies
> @@ -1,4 +1,5 @@
>  # MIPS uses IEEE 754 floating point.
> +mips/fpu
>  mips/ieee754
>  ieee754/flt-32
>  ieee754/dbl-64

These changes are clearly incorrect, as they would cause mips/fpu/ to be 
used for soft-float.  The fpu/ directories are selected automatically for 
hard-float, and must not be used for soft-float.

I'm guessing you're motivated by some ordering issue that results in the 
ieee754/ directories ending up being used before mips/fpu/.  I'm not sure 
of the best way to avoid such an ordering issue (noting that the ieee754/ 
directories go in Implies separately for mips32 and mips64 because only 
mips64 uses ldbl-128), but e.g. #includes like in 
sysdeps/mips/mips64/n32/fpu/s_fma.c should work if you can't get the 
preferred sysdeps directory ordering.

-- 
Joseph S. Myers
joseph@codesourcery.com