unofficial mirror of libc-alpha@sourceware.org
 help / color / mirror / Atom feed
From: "Gabriel F. T. Gomes" <gabriel@inconstante.eti.br>
To: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Cc: <libc-alpha@sourceware.org>
Subject: Re: [PATCH 06/28] powerpc: round/roundf refactor
Date: Fri, 3 May 2019 22:10:31 -0300	[thread overview]
Message-ID: <20190504011031.5pinthodq7fztpov@tereshkova> (raw)
In-Reply-To: <20190329133529.22523-7-adhemerval.zanella@linaro.org>

On Fri, Mar 29 2019, Adhemerval Zanella wrote:
> 
> 	* sysdeps/powerpc/fpu/round_to_integer.h (set_rounding_mode): Add
> 	ROUND handling.
> 	(round_to_integer_float): Likewise.

Similar to what I mentioned for the floor patch, just make sure you
mention the addition of the ROUND definition.

> diff --git a/sysdeps/powerpc/fpu/round_to_integer.h b/sysdeps/powerpc/fpu/round_to_integer.h
> index 77d9fc1f86..3fa2b77b6d 100644
> --- a/sysdeps/powerpc/fpu/round_to_integer.h
> +++ b/sysdeps/powerpc/fpu/round_to_integer.h
> @@ -38,6 +38,7 @@ set_fenv_mode (enum round_mode mode)
>    {
>    case CEIL:  rmode = FE_UPWARD; break;
>    case FLOOR: rmode = FE_DOWNWARD; break;
> +  case ROUND: rmode = FE_TOWARDZERO; break;
>    default:    rmode = FE_TONEAREST; break;
>    }

Likewise, check that the FLOOR definition is added to round_mode.

This patch looks good to me with these changes.

Reviewed-by: Gabriel F. T. Gomes <gabriel@inconstante.eti.br>


> @@ -60,12 +61,24 @@ round_to_integer_float (enum round_mode mode, float x)
>    set_fenv_mode (mode);
>    if (x > 0.0)
>      {
> +      /* IEEE 1003.1 round function.  IEEE specifies "round to the nearest
> +	 integer value, rounding halfway cases away from zero, regardless of
> +	 the current rounding mode."  However PowerPC Architecture defines
> +	 "Round to Nearest" as "Choose the best approximation. In case of a
> +	 tie, choose the one that is even (least significant bit o).".
> +	 So we can't use the PowerPC "Round to Nearest" mode. Instead we set
> +	 "Round toward Zero" mode and round by adding +-0.5 before rounding
> +	 to the integer value.  */
> +      if (mode == ROUND)
> +	r += 0.5f;
>        r += 0x1p+23;
>        r -= 0x1p+23;
>        r = fabs (r);
>      }
>    else if (x < 0.0)
>      {
> +      if (mode == ROUND)
> +	r -= 0.5f;
>        r -= 0x1p+23;
>        r += 0x1p+23;
>        r = -fabs (r);

OK.  Restored from the .S files.

> @@ -92,12 +105,16 @@ round_to_integer_double (enum round_mode mode, double x)
>    set_fenv_mode (mode);
>    if (x > 0.0)
>      {
> +      if (mode == ROUND)
> +	r += 0.5;
>        r += 0x1p+52;
>        r -= 0x1p+52;
>        r = fabs (r);
>      }
>    else if (x < 0.0)
>      {
> +      if (mode == ROUND)
> +	r -= 0.5;
>        r -= 0x1p+52;
>        r += 0x1p+52;
>        r = -fabs (r);

OK.  Likewise.

> +double
> +__round (double x)
> +{
> +#ifdef _ARCH_PWR5X
> +  return __builtin_round (x);
> +#else
> +  return round_to_integer_double (ROUND, x);
> +#endif
> +}

OK.  Arch check looks correct.

> -.LC1:	/* 0.5 */
> -	.long 0x3f000000
> -
> -/* double [fp1] round (double x [fp1])
> -   IEEE 1003.1 round function.  IEEE specifies "round to the nearest
> -   integer value, rounding halfway cases away from zero, regardless of
> -   the current rounding mode."  However PowerPC Architecture defines
> -   "Round to Nearest" as "Choose the best approximation. In case of a
> -   tie, choose the one that is even (least significant bit o).".
> -   So we can't use the PowerPC "Round to Nearest" mode. Instead we set
> -   "Round toward Zero" mode and round by adding +-0.5 before rounding
> -   to the integer value.  */
>
> [...]
>
> -#ifdef SHARED
> -	lfs	fp10,.LC1-.LC0(r9)
> -#else
> -	lis	r9,.LC1@ha
> -	lfs	fp10,.LC1@l(r9)
> -#endif
> -	ble-	cr6,.L4
> -	fadd	fp1,fp1,fp10	/* x+= 0.5;  */

OK.  Moved to round_to_integer_float and round_to_integer_double.

  reply	other threads:[~2019-05-04  1:10 UTC|newest]

Thread overview: 85+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-03-29 13:35 [PATCH 00/28] powerpc floating-point optimization refactor Adhemerval Zanella
2019-03-29 13:35 ` [PATCH 01/28] powerpc: Use generic fabs{f} implementations Adhemerval Zanella
2019-04-01 20:04   ` Joseph Myers
2019-04-03  1:04     ` Adhemerval Zanella
2019-04-15 20:23       ` Gabriel F. T. Gomes
2019-04-15 21:32         ` Tulio Magno Quites Machado Filho
2019-04-17 17:08           ` Adhemerval Zanella
2019-03-29 13:35 ` [PATCH 02/28] powerpc: fma using builtins Adhemerval Zanella
2019-04-01 20:05   ` Joseph Myers
2019-04-03  1:06     ` Adhemerval Zanella
2019-04-15 21:44       ` Gabriel F. T. Gomes
2019-04-17 21:10       ` Joseph Myers
2019-04-17 21:28         ` Adhemerval Zanella
2019-03-29 13:35 ` [PATCH 03/28] powerpc: Remove power4 mpa optimization Adhemerval Zanella
2019-04-24 21:51   ` Gabriel F. T. Gomes
2019-04-25 12:19     ` Adhemerval Zanella
2019-03-29 13:35 ` [PATCH 04/28] powerpc: ceil/ceilf refactor Adhemerval Zanella
2019-04-25  1:56   ` Gabriel F. T. Gomes
2019-04-25 21:58     ` Adhemerval Zanella
2019-05-02 18:41       ` Gabriel F. T. Gomes
2019-05-04 21:46         ` Gabriel F. T. Gomes
2019-05-06 12:22           ` Adhemerval Zanella
2019-05-09 19:56             ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 05/28] powerpc: floor/floorf refactor Adhemerval Zanella
2019-05-03 21:44   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 06/28] powerpc: round/roundf refactor Adhemerval Zanella
2019-05-04  1:10   ` Gabriel F. T. Gomes [this message]
2019-03-29 13:35 ` [PATCH 07/28] powerpc: trunc/truncf refactor Adhemerval Zanella
2019-05-09 20:06   ` Gabriel F. T. Gomes
2019-05-09 20:54     ` Adhemerval Zanella
2019-05-09 21:36       ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 08/28] powerpc: generic nearbyint/nearbyintf Adhemerval Zanella
2019-05-27 20:42   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 09/28] powerpc: consolidate rint Adhemerval Zanella
2019-05-30 14:36   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 10/28] powerpc: copysign cleanup Adhemerval Zanella
2019-05-31 14:14   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 11/28] benchtests: Add isnan/isinf/isfinite benchmark Adhemerval Zanella
2019-06-05 22:45   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 12/28] math: Use wordsize-64 version for isnan Adhemerval Zanella
2019-06-05 22:45   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 13/28] powerpc: Remove optimized isnan Adhemerval Zanella
2019-06-05 22:46   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 14/28] math: Use wordsize-64 version for isinf Adhemerval Zanella
2019-06-11 13:13   ` Gabriel F. T. Gomes
2019-06-13  8:57   ` Szabolcs Nagy
2019-06-13  9:11     ` Szabolcs Nagy
2019-03-29 13:35 ` [PATCH 15/28] powerpc: Remove optimized isinf Adhemerval Zanella
2019-06-11 13:45   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 16/28] math: Use wordsize-64 version for finite Adhemerval Zanella
2019-06-11 16:20   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 17/28] powerpc: Remove optimized finite Adhemerval Zanella
2019-06-11 18:08   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 18/28] powerpc: refactor powerpc64 lrint/lrintf/llrint/llrintf Adhemerval Zanella
2019-06-13 19:30   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 19/28] powerpc: Refactor powerpc32 lrint/lrintf/llrint/llrintf Adhemerval Zanella
2019-06-14 18:34   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 20/28] powerpc: Refactor powerpc64 lround/lroundf/llround/llroundf Adhemerval Zanella
2019-06-13 19:30   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 21/28] powerpc: Refactor powerpc32 lround/lroundf/llround/llroundf Adhemerval Zanella
2019-06-24 21:07   ` Gabriel F. T. Gomes
2019-06-25 18:34     ` Adhemerval Zanella
2019-06-25 18:44       ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 22/28] powerpc: Use generic e_expf Adhemerval Zanella
2019-06-26 12:59   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 23/28] benchtests: hypot benchmark Adhemerval Zanella
2019-06-26 19:58   ` Gabriel F. T. Gomes
2019-03-29 13:35 ` [PATCH 24/28] powerpc: hypot refactor and optimization Adhemerval Zanella
2019-04-01 20:14   ` Joseph Myers
2019-04-03  1:08     ` Adhemerval Zanella
2019-06-26 19:59   ` Gabriel F. T. Gomes
2019-07-08 15:37     ` Adhemerval Zanella
2019-07-08 18:30       ` Adhemerval Zanella
2019-03-29 13:35 ` [PATCH 25/28] powerpc: Refactor modf{f} Adhemerval Zanella
2019-07-05  1:23   ` Gabriel F. T. Gomes
2019-07-08 18:30     ` Adhemerval Zanella
2019-03-29 13:35 ` [PATCH 26/28] benchtests: Add logb{f} benchmark Adhemerval Zanella
2019-07-05  1:23   ` Gabriel F. T. Gomes
2019-07-08 18:31     ` Adhemerval Zanella
2019-03-29 13:35 ` [PATCH 27/28] math: Use wordsize-64 version for s_logb Adhemerval Zanella
2019-07-05  1:23   ` Gabriel F. T. Gomes
2019-07-08 18:31     ` Adhemerval Zanella
2019-03-29 13:35 ` [PATCH 28/28] powerpc: refactor logb{f,l} Adhemerval Zanella
2019-07-05  1:24   ` Gabriel F. T. Gomes
2019-07-08 18:31     ` Adhemerval Zanella

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

  List information: https://www.gnu.org/software/libc/involved.html

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

  git send-email \
    --in-reply-to=20190504011031.5pinthodq7fztpov@tereshkova \
    --to=gabriel@inconstante.eti.br \
    --cc=adhemerval.zanella@linaro.org \
    --cc=libc-alpha@sourceware.org \
    /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).