public inbox for libc-alpha@sourceware.org
 help / color / mirror / Atom feed
From: Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>
To: Manjunath Matti <mmatti@linux.ibm.com>, libc-alpha@sourceware.org
Cc: rajis@linux.ibm.com, Carl Love <cel@us.ibm.com>
Subject: Re: [PATCH] [powerpc] fegetenv_and_set_rn now uses the builtins provided by GCC.
Date: Tue, 12 Sep 2023 10:27:51 -0300	[thread overview]
Message-ID: <526f2aac-a06c-d1cf-4c7b-1339846a1711@linaro.org> (raw)
In-Reply-To: <20230912082730.3951006-1-mmatti@linux.ibm.com>



On 12/09/23 05:27, Manjunath Matti wrote:
> On powerpc, SET_RESTORE_ROUND uses inline assembly to optimize the
> prologue get/save/set rounding mode operations for POWER9 and
> later by using 'mffscrn' where possible, this was introduced by
> commit f1c56cdff09f650ad721fae026eb6a3651631f3d.
> 
> GCC version 14 onwards supports builtins as __builtin_set_fpscr_rn
> which now returns the FPSCR fields in a double. This feature is
> available on Power9 when the __SET_FPSCR_RN_RETURNS_FPSCR__ macro
> is defined along with __builtin_set_fpscr_rn enabled.
> GCC commit ef3bbc69d15707e4db6e2f198c621effb636cc26 adds
> this feature.
> 
> Changes are done to use __builtin_set_fpscr_rn instead of mffscrn
> or mffscrni in __fe_mffscrn(rn).
> 
> Suggested-by: Carl Love <cel@us.ibm.com>
> ---
>  sysdeps/powerpc/fpu/fenv_libc.h | 23 ++++++++++++++++++++---
>  1 file changed, 20 insertions(+), 3 deletions(-)
> 
> diff --git a/sysdeps/powerpc/fpu/fenv_libc.h b/sysdeps/powerpc/fpu/fenv_libc.h
> index fa5e1c697e..55484eb229 100644
> --- a/sysdeps/powerpc/fpu/fenv_libc.h
> +++ b/sysdeps/powerpc/fpu/fenv_libc.h
> @@ -84,8 +84,15 @@ extern const fenv_t *__fe_mask_env (void) attribute_hidden;
>      __fr.fenv;								\
>    })
>  
> +/* GCC version 14 onwards supports builtins as __builtin_set_fpscr_rn and
> +   now returns the FPSCR fields in a double. This support is available
> +   on Power9 when the __SET_FPSCR_RN_RETURNS_FPSCR__ macro is defined.
> +   To retain backward compatibility with older GCC, we still retain the
> +   old inline assembly implementation.  */
> +#if defined _ARCH_PWR9 && defined __SET_FPSCR_RN_RETURNS_FPSCR__
> +#define fegetenv_and_set_rn(rn) __builtin_set_fpscr_rn (rn)
> +#elif defined _ARCH_PWR9
>  /* Like fegetenv_control, but also sets the rounding mode.  */
> -#ifdef _ARCH_PWR9
>  #define fegetenv_and_set_rn(rn) __fe_mffscrn (rn)
>  #else
>  /* 'mffscrn' will decode to 'mffs' on ARCH < 3_00, which is still necessary

I think the macro would be better defined as:

#ifdef __SET_FPSCR_RN_RETURNS_FPSCR__
# define __fe_mffscrn(rn)  __builtin_set_fpscr_rn (rn)
#else
# define __fe_mffscrn(rn) [...]
#endif

Then there is no need to redefine fegetenv_and_set_rn nor __fesetround_inline.

> @@ -148,7 +155,12 @@ typedef union
>  static inline int
>  __fesetround_inline (int round)
>  {
> -#ifdef _ARCH_PWR9
> +/* GCC version 14 onwards supports builtins as __builtin_set_fpscr_rn and
> +   now returns the FPSCR fields in a double. This support is available
> +   on Power9 when the __SET_FPSCR_RN_RETURNS_FPSCR__ macro is defined.  */
> +#if defined _ARCH_PWR9 && defined __SET_FPSCR_RN_RETURNS_FPSCR__
> +  __builtin_set_fpscr_rn (round);
> +#elif defined _ARCH_PWR9
>    __fe_mffscrn (round);
>  #else
>    if (__glibc_likely (GLRO(dl_hwcap2) & PPC_FEATURE2_ARCH_3_00))
> @@ -178,7 +190,12 @@ __fesetround_inline (int round)
>  static inline void
>  __fesetround_inline_nocheck (const int round)
>  {
> -#ifdef _ARCH_PWR9
> +/* GCC version 14 onwards supports builtins as __builtin_set_fpscr_rn and
> +   now returns the FPSCR fields in a double. This support is available
> +   on Power9 when the __SET_FPSCR_RN_RETURNS_FPSCR__ macro is defined.  */
> +#if defined _ARCH_PWR9 && defined __SET_FPSCR_RN_RETURNS_FPSCR__
> +  __builtin_set_fpscr_rn (round);
> +#elif defined _ARCH_PWR9
>    __fe_mffscrn (round);
>  #else
>    if (__glibc_likely (GLRO(dl_hwcap2) & PPC_FEATURE2_ARCH_3_00))

  reply	other threads:[~2023-09-12 13:27 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-09-12  8:27 Manjunath Matti
2023-09-12 13:27 ` Adhemerval Zanella Netto [this message]
2023-09-20 11:28   ` Manjunath S Matti
2023-09-20 13:33     ` Adhemerval Zanella Netto
2023-10-19  1:25     ` Peter Bergner
2023-09-23 15:49 Manjunath Matti
2023-09-27 16:56 ` Adhemerval Zanella Netto
2023-10-19  1:47 ` Peter Bergner

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=526f2aac-a06c-d1cf-4c7b-1339846a1711@linaro.org \
    --to=adhemerval.zanella@linaro.org \
    --cc=cel@us.ibm.com \
    --cc=libc-alpha@sourceware.org \
    --cc=mmatti@linux.ibm.com \
    --cc=rajis@linux.ibm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).