* [PATCH] Perform cfg cleanup in cse if needed (PR rtl-optimization/92610)
@ 2019-11-23 2:11 Jakub Jelinek
2019-11-23 6:56 ` Richard Biener
0 siblings, 1 reply; 2+ messages in thread
From: Jakub Jelinek @ 2019-11-23 2:11 UTC (permalink / raw)
To: Jeff Law, Richard Biener, Eric Botcazou; +Cc: gcc-patches
Hi!
cse_main which sets tem has:
if (cse_jumps_altered || recorded_label_ref)
return 2;
else if (cse_cfg_altered)
return 1;
else
return 0;
at the end, but after these two functions call it, they call also
delete_trivially_dead_insns. Even when cse_main didn't need cfg changes,
delete_trivially_dead_insns could remove a trivially dead insn that had
EH edges and so we might still need to remove unreachable blocks.
The following patch implements it. Bootstrapped/regtested on x86_64-linux
and i686-linux, ok for trunk?
2019-11-23 Jakub Jelinek <jakub@redhat.com>
PR rtl-optimization/92610
* cse.c (rest_of_handle_cse2): Call cleanup_cfg (0) also if
cse_cfg_altered is set, even when tem is 0.
(rest_of_handle_cse_after_global_opts): Likewise.
* g++.dg/opt/pr92610.C: New test.
--- gcc/cse.c.jj 2019-11-19 22:27:01.960059340 +0100
+++ gcc/cse.c 2019-11-22 15:51:34.639806155 +0100
@@ -7701,7 +7701,7 @@ rest_of_handle_cse2 (void)
cse_cfg_altered |= cleanup_cfg (CLEANUP_CFG_CHANGED);
timevar_pop (TV_JUMP);
}
- else if (tem == 1)
+ else if (tem == 1 || cse_cfg_altered)
cse_cfg_altered |= cleanup_cfg (0);
cse_not_expected = 1;
@@ -7775,7 +7775,7 @@ rest_of_handle_cse_after_global_opts (vo
cse_cfg_altered |= cleanup_cfg (CLEANUP_CFG_CHANGED);
timevar_pop (TV_JUMP);
}
- else if (tem == 1)
+ else if (tem == 1 || cse_cfg_altered)
cse_cfg_altered |= cleanup_cfg (0);
flag_cse_follow_jumps = save_cfj;
--- gcc/testsuite/g++.dg/opt/pr92610.C.jj 2019-11-22 15:52:46.254734331 +0100
+++ gcc/testsuite/g++.dg/opt/pr92610.C 2019-11-22 15:53:14.991303498 +0100
@@ -0,0 +1,13 @@
+// PR rtl-optimization/92610
+// { dg-do compile }
+// { dg-options "-w -fdelete-dead-exceptions --param=sccvn-max-alias-queries-per-access=0 -fno-dse -fnon-call-exceptions -Os -funroll-loops -ftrapv" }
+
+struct C { int x; ~C () {} };
+
+int
+main ()
+{
+ C *buffer = new C[42];
+ buffer[-3].x = 42;
+ delete [] buffer;
+}
Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Perform cfg cleanup in cse if needed (PR rtl-optimization/92610)
2019-11-23 2:11 [PATCH] Perform cfg cleanup in cse if needed (PR rtl-optimization/92610) Jakub Jelinek
@ 2019-11-23 6:56 ` Richard Biener
0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2019-11-23 6:56 UTC (permalink / raw)
To: Jakub Jelinek, Jeff Law, Eric Botcazou; +Cc: gcc-patches
On November 23, 2019 2:15:36 AM GMT+01:00, Jakub Jelinek <jakub@redhat.com> wrote:
>Hi!
>
>cse_main which sets tem has:
> if (cse_jumps_altered || recorded_label_ref)
> return 2;
> else if (cse_cfg_altered)
> return 1;
> else
> return 0;
>at the end, but after these two functions call it, they call also
>delete_trivially_dead_insns. Even when cse_main didn't need cfg
>changes,
>delete_trivially_dead_insns could remove a trivially dead insn that had
>EH edges and so we might still need to remove unreachable blocks.
>
>The following patch implements it. Bootstrapped/regtested on
>x86_64-linux
>and i686-linux, ok for trunk?
Ok.
Richard.
>2019-11-23 Jakub Jelinek <jakub@redhat.com>
>
> PR rtl-optimization/92610
> * cse.c (rest_of_handle_cse2): Call cleanup_cfg (0) also if
> cse_cfg_altered is set, even when tem is 0.
> (rest_of_handle_cse_after_global_opts): Likewise.
>
> * g++.dg/opt/pr92610.C: New test.
>
>--- gcc/cse.c.jj 2019-11-19 22:27:01.960059340 +0100
>+++ gcc/cse.c 2019-11-22 15:51:34.639806155 +0100
>@@ -7701,7 +7701,7 @@ rest_of_handle_cse2 (void)
> cse_cfg_altered |= cleanup_cfg (CLEANUP_CFG_CHANGED);
> timevar_pop (TV_JUMP);
> }
>- else if (tem == 1)
>+ else if (tem == 1 || cse_cfg_altered)
> cse_cfg_altered |= cleanup_cfg (0);
>
> cse_not_expected = 1;
>@@ -7775,7 +7775,7 @@ rest_of_handle_cse_after_global_opts (vo
> cse_cfg_altered |= cleanup_cfg (CLEANUP_CFG_CHANGED);
> timevar_pop (TV_JUMP);
> }
>- else if (tem == 1)
>+ else if (tem == 1 || cse_cfg_altered)
> cse_cfg_altered |= cleanup_cfg (0);
>
> flag_cse_follow_jumps = save_cfj;
>--- gcc/testsuite/g++.dg/opt/pr92610.C.jj 2019-11-22 15:52:46.254734331
>+0100
>+++ gcc/testsuite/g++.dg/opt/pr92610.C 2019-11-22 15:53:14.991303498
>+0100
>@@ -0,0 +1,13 @@
>+// PR rtl-optimization/92610
>+// { dg-do compile }
>+// { dg-options "-w -fdelete-dead-exceptions
>--param=sccvn-max-alias-queries-per-access=0 -fno-dse
>-fnon-call-exceptions -Os -funroll-loops -ftrapv" }
>+
>+struct C { int x; ~C () {} };
>+
>+int
>+main ()
>+{
>+ C *buffer = new C[42];
>+ buffer[-3].x = 42;
>+ delete [] buffer;
>+}
>
> Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2019-11-23 6:54 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-11-23 2:11 [PATCH] Perform cfg cleanup in cse if needed (PR rtl-optimization/92610) Jakub Jelinek
2019-11-23 6:56 ` 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).