This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH v2] powerpc: Refactor fenvinline.h
- From: Paul Clarke <pc at us dot ibm dot com>
- To: Rogerio Alves <rcardoso at linux dot ibm dot com>, libc-alpha at sourceware dot org
- Cc: adhemerval dot zanella at linaro dot org
- Date: Fri, 7 Feb 2020 08:50:47 -0600
- Subject: Re: [PATCH v2] powerpc: Refactor fenvinline.h
- References: <20200207143410.3948-1-rcardoso@linux.ibm.com>
On 2/7/20 8:34 AM, Rogerio Alves wrote:
> This patch refactor fenviline.h replaces some statements for
> builtins.
>
> ---
>
> * Changes in v2: Per Adhemerval review: don not redefine _buitins use a
> macro instead. Use of __builtin_popcount allows do eliminate the
> if (e == 0) test. __builtin_popcount(e) == 1 for e = 0 validate false
> while (e & (e - 1)) == 0 validate true if e = 0. Appended a comment
> about the weird asm constrain i#*X. The warning only appear at GCC
> versions 3 and below. If we even stops to support GCC 3 for installed
> headers we may replace that for plain `i`.
> sysdeps/powerpc/bits/fenvinline.h | 30 +++++++++++++++++-------------
> 1 file changed, 17 insertions(+), 13 deletions(-)
>
> diff --git a/sysdeps/powerpc/bits/fenvinline.h b/sysdeps/powerpc/bits/fenvinline.h
> index 70689664e2..7b2b75e5ef 100644
> --- a/sysdeps/powerpc/bits/fenvinline.h
> +++ b/sysdeps/powerpc/bits/fenvinline.h
> @@ -51,9 +51,19 @@
> # define fegetround() __fegetround ()
>
> # ifndef __NO_MATH_INLINES
> +
> +/* Builtins to mtfsb0 and mtfsb1 was introduced on GCC 9. */
> +# if !__GNUC_PREREQ(9, 0)
> /* The weird 'i#*X' constraints on the following suppress a gcc
> warning when __excepts is not a constant. Otherwise, they mean the
> - same as just plain 'i'. */
> + same as just plain 'i'. This warning only happens in old GCC
> + versions (gcc 3 or less). Otherwise plain 'i' works fine. */
> +# define __mtfsb0(__b) __asm__ __volatile__ ("mtfsb0 %0" : : "i#*X" (__b))
> +# define __mtfsb1(__b) __asm__ __volatile__ ("mtfsb1 %0" : : "i#*X" (__b))
> +# else
> +# define __mtfsb0(__b) __builtin_mtfsb0 (__b)
> +# define __mtfsb1(__b) __builtin_mtfsb1 (__b)
> +# endif
These names actually conflict with XL compiler builtins which serve the same purpose. https://www.ibm.com/support/knowledgecenter/SSXVZZ_16.1.1/com.ibm.xlcpp1611.lelinux.doc/compiler_ref/bifs_fpscr.html
Maybe all caps? __MTFSB0/__MTFSB1 ?
Of course, I don't think XL currently provides the __builtin versions, so the chances of an actual collision are vanishingly small. I would be nice to plan for better compatibility in the future, though. :-)
PC