public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [PATCH] Improve seq_cost (PR middle-end/80929)
@ 2017-11-29  8:19 Jakub Jelinek
  2017-11-29  9:46 ` Richard Biener
  0 siblings, 1 reply; 2+ messages in thread
From: Jakub Jelinek @ 2017-11-29  8:19 UTC (permalink / raw)
  To: Richard Biener, Eric Botcazou, Jeff Law; +Cc: gcc-patches

Hi!

This PR complains that seq_cost counts all non-single_set insns as 1
unconditionally, which is indeed bad.  For some like compare + arithmetics
we even have now code in pattern_cost that handles those reasonably by
default, for others targets have the option through a hook to return
whatever they want.

On the other side, IMNSHO we don't want to count CODE_LABELs, NOTEs,
BARRIERs, DEBUG_INSNs...

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2017-11-29  Jakub Jelinek  <jakub@redhat.com>

	PR middle-end/80929
	* rtlanal.c (seq_cost): For non-single_set insns try to use insn_cost.

--- gcc/rtlanal.c.jj	2017-10-27 14:16:39.000000000 +0200
+++ gcc/rtlanal.c	2017-11-27 13:35:46.321509933 +0100
@@ -5342,7 +5342,13 @@ seq_cost (const rtx_insn *seq, bool spee
       if (set)
         cost += set_rtx_cost (set, speed);
-      else
-        cost++;
+      else if (NONDEBUG_INSN_P (seq))
+	{
+	  int this_cost = insn_cost (CONST_CAST_RTX_INSN (seq), speed);
+	  if (this_cost > 0)
+	    cost += this_cost;
+	  else
+	    cost++;
+	}
     }
 
   return cost;

	Jakub

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH] Improve seq_cost (PR middle-end/80929)
  2017-11-29  8:19 [PATCH] Improve seq_cost (PR middle-end/80929) Jakub Jelinek
@ 2017-11-29  9:46 ` Richard Biener
  0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2017-11-29  9:46 UTC (permalink / raw)
  To: Jakub Jelinek; +Cc: Eric Botcazou, Jeff Law, gcc-patches

On Wed, 29 Nov 2017, Jakub Jelinek wrote:

> Hi!
> 
> This PR complains that seq_cost counts all non-single_set insns as 1
> unconditionally, which is indeed bad.  For some like compare + arithmetics
> we even have now code in pattern_cost that handles those reasonably by
> default, for others targets have the option through a hook to return
> whatever they want.
> 
> On the other side, IMNSHO we don't want to count CODE_LABELs, NOTEs,
> BARRIERs, DEBUG_INSNs...
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

Works for me.

Thanks,
Richard.

> 2017-11-29  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR middle-end/80929
> 	* rtlanal.c (seq_cost): For non-single_set insns try to use insn_cost.
> 
> --- gcc/rtlanal.c.jj	2017-10-27 14:16:39.000000000 +0200
> +++ gcc/rtlanal.c	2017-11-27 13:35:46.321509933 +0100
> @@ -5342,7 +5342,13 @@ seq_cost (const rtx_insn *seq, bool spee
>        if (set)
>          cost += set_rtx_cost (set, speed);
> -      else
> -        cost++;
> +      else if (NONDEBUG_INSN_P (seq))
> +	{
> +	  int this_cost = insn_cost (CONST_CAST_RTX_INSN (seq), speed);
> +	  if (this_cost > 0)
> +	    cost += this_cost;
> +	  else
> +	    cost++;
> +	}
>      }
>  
>    return cost;
> 
> 	Jakub
> 
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE LINUX GmbH, GF: Felix Imendoerffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nuernberg)

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2017-11-29  9:43 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-11-29  8:19 [PATCH] Improve seq_cost (PR middle-end/80929) Jakub Jelinek
2017-11-29  9:46 ` Richard Biener

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).