* [PATCH] regalloc: Ignore '^' in early costing [PR114766]
@ 2024-04-29 12:59 Wilco Dijkstra
2024-04-30 20:12 ` Vladimir Makarov
0 siblings, 1 reply; 2+ messages in thread
From: Wilco Dijkstra @ 2024-04-29 12:59 UTC (permalink / raw)
To: GCC Patches; +Cc: vmakarov
According to documentation, '^' should only have an effect during reload.
However ira-costs.cc treats it in the same way as '?' during early costing.
As a result using '^' can accidentally disable valid alternatives and cause
significant regressions (see PR114741). Avoid this by ignoring '^' during
costing.
Passes bootstrap and regress, OK for commit?
gcc:
PR rtl-optimization/114766
* ira-costs.cc (record_reg_classes): Ignore '^' during costing.
---
diff --git a/gcc/ira-costs.cc b/gcc/ira-costs.cc
index c86c5a16563aeefac9d4fa72839bee8d95409f4b..04d2f21b023f3456ba6f8c16c2418d7313965b2f 100644
--- a/gcc/ira-costs.cc
+++ b/gcc/ira-costs.cc
@@ -771,10 +771,6 @@ record_reg_classes (int n_alts, int n_ops, rtx *ops,
c = *++p;
break;
- case '^':
- alt_cost += 2;
- break;
-
case '?':
alt_cost += 2;
break;
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] regalloc: Ignore '^' in early costing [PR114766]
2024-04-29 12:59 [PATCH] regalloc: Ignore '^' in early costing [PR114766] Wilco Dijkstra
@ 2024-04-30 20:12 ` Vladimir Makarov
0 siblings, 0 replies; 2+ messages in thread
From: Vladimir Makarov @ 2024-04-30 20:12 UTC (permalink / raw)
To: Wilco Dijkstra, GCC Patches
On 4/29/24 08:59, Wilco Dijkstra wrote:
> According to documentation, '^' should only have an effect during reload.
> However ira-costs.cc treats it in the same way as '?' during early costing.
> As a result using '^' can accidentally disable valid alternatives and cause
> significant regressions (see PR114741). Avoid this by ignoring '^' during
> costing.
>
> Passes bootstrap and regress, OK for commit?
>
> gcc:
> PR rtl-optimization/114766
> * ira-costs.cc (record_reg_classes): Ignore '^' during costing.
>
Sorry, I can not accept this patch. This constraint is used by other
targets (as I know rs6000, s390, sh, gcn). I suspect changing the
semantics can affect the targets in some undesirable way.
Saying that, I understand that aarch64 needs the new semantics. So I
think we can add a new hint with the required semantics. There are still
few undefined characters for the new hint: '~', '-', '/', and '.' (may
be I missed some),
I propose to use '~' for the new hint. So besides changing code, the
documentation for '^' needs to clarified and documentation for code '~'
should be added.
>
> diff --git a/gcc/ira-costs.cc b/gcc/ira-costs.cc
> index c86c5a16563aeefac9d4fa72839bee8d95409f4b..04d2f21b023f3456ba6f8c16c2418d7313965b2f 100644
> --- a/gcc/ira-costs.cc
> +++ b/gcc/ira-costs.cc
> @@ -771,10 +771,6 @@ record_reg_classes (int n_alts, int n_ops, rtx *ops,
> c = *++p;
> break;
>
> - case '^':
> - alt_cost += 2;
> - break;
> -
> case '?':
> alt_cost += 2;
> break;
>
>
>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2024-04-30 20:13 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-04-29 12:59 [PATCH] regalloc: Ignore '^' in early costing [PR114766] Wilco Dijkstra
2024-04-30 20:12 ` Vladimir Makarov
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).