* [PATCH] Fix x86_64 ICE with -fpie -mcmodel=large (PR target/81766)
@ 2017-09-01 11:42 Jakub Jelinek
2017-09-01 16:22 ` Uros Bizjak
0 siblings, 1 reply; 2+ messages in thread
From: Jakub Jelinek @ 2017-09-01 11:42 UTC (permalink / raw)
To: Uros Bizjak; +Cc: gcc-patches
Hi!
As mentioned in the PR, ix86_init_pic_reg for -mcmodel=large PIC creates
invalid RTL. Shrink wrapping managed to work around it by unconditionally
running find_many_sub_basic_blocks that has been invoked even when the
prologue or split prologue actually didn't contain anything, but that isn't
done anymore.
The problem is that we add a label into the sequence that we then insert on
the single succ edge after ENTRY; but this insertion inserts the label after
the NOTE_INSN_BLOCK_BEG, which is invalid, because labels should precede
that.
Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for
trunk?
2017-09-01 Jakub Jelinek <jakub@redhat.com>
PR target/81766
* config/i386/i386.c (ix86_init_large_pic_reg): Return label instead of void.
(ix86_init_pic_reg): Remember label from ix86_init_large_pic_reg, if non-NULL
and preceded by NOTE_INSN_BASIC_BLOCK, swap the note and label.
* gcc.target/i386/pr81766.c: New test.
--- gcc/config/i386/i386.c.jj 2017-08-07 18:50:10.000000000 +0200
+++ gcc/config/i386/i386.c 2017-08-08 16:01:41.917136120 +0200
@@ -8829,7 +8829,7 @@ ix86_use_pseudo_pic_reg (void)
/* Initialize large model PIC register. */
-static void
+static rtx_code_label *
ix86_init_large_pic_reg (unsigned int tmp_regno)
{
rtx_code_label *label;
@@ -8846,6 +8846,7 @@ ix86_init_large_pic_reg (unsigned int tm
emit_insn (gen_set_got_offset_rex64 (tmp_reg, label));
emit_insn (ix86_gen_add3 (pic_offset_table_rtx,
pic_offset_table_rtx, tmp_reg));
+ return label;
}
/* Create and initialize PIC register if required. */
@@ -8854,6 +8855,7 @@ ix86_init_pic_reg (void)
{
edge entry_edge;
rtx_insn *seq;
+ rtx_code_label *label = NULL;
if (!ix86_use_pseudo_pic_reg ())
return;
@@ -8863,7 +8865,7 @@ ix86_init_pic_reg (void)
if (TARGET_64BIT)
{
if (ix86_cmodel == CM_LARGE_PIC)
- ix86_init_large_pic_reg (R11_REG);
+ label = ix86_init_large_pic_reg (R11_REG);
else
emit_insn (gen_set_got_rex64 (pic_offset_table_rtx));
}
@@ -8887,6 +8889,22 @@ ix86_init_pic_reg (void)
entry_edge = single_succ_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun));
insert_insn_on_edge (seq, entry_edge);
commit_one_edge_insertion (entry_edge);
+
+ if (label)
+ {
+ basic_block bb = BLOCK_FOR_INSN (label);
+ rtx_insn *bb_note = PREV_INSN (label);
+ /* If the note preceding the label starts a basic block, and the
+ label is a member of the same basic block, interchange the two. */
+ if (bb_note != NULL_RTX
+ && NOTE_INSN_BASIC_BLOCK_P (bb_note)
+ && bb != NULL
+ && bb == BLOCK_FOR_INSN (bb_note))
+ {
+ reorder_insns_nobb (bb_note, bb_note, label);
+ BB_HEAD (bb) = label;
+ }
+ }
}
/* Initialize a variable CUM of type CUMULATIVE_ARGS
--- gcc/testsuite/gcc.target/i386/pr81766.c.jj 2017-08-08 16:10:04.299459808 +0200
+++ gcc/testsuite/gcc.target/i386/pr81766.c 2017-08-08 16:09:28.000000000 +0200
@@ -0,0 +1,9 @@
+/* PR target/81766 */
+/* { dg-do compile { target { pie && lp64 } } } */
+/* { dg-options "-O2 -fpie -mcmodel=large" } */
+
+int
+main ()
+{
+ return 0;
+}
Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Fix x86_64 ICE with -fpie -mcmodel=large (PR target/81766)
2017-09-01 11:42 [PATCH] Fix x86_64 ICE with -fpie -mcmodel=large (PR target/81766) Jakub Jelinek
@ 2017-09-01 16:22 ` Uros Bizjak
0 siblings, 0 replies; 2+ messages in thread
From: Uros Bizjak @ 2017-09-01 16:22 UTC (permalink / raw)
To: Jakub Jelinek; +Cc: gcc-patches
On Fri, Sep 1, 2017 at 1:42 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> Hi!
>
> As mentioned in the PR, ix86_init_pic_reg for -mcmodel=large PIC creates
> invalid RTL. Shrink wrapping managed to work around it by unconditionally
> running find_many_sub_basic_blocks that has been invoked even when the
> prologue or split prologue actually didn't contain anything, but that isn't
> done anymore.
>
> The problem is that we add a label into the sequence that we then insert on
> the single succ edge after ENTRY; but this insertion inserts the label after
> the NOTE_INSN_BLOCK_BEG, which is invalid, because labels should precede
> that.
>
> Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for
> trunk?
>
> 2017-09-01 Jakub Jelinek <jakub@redhat.com>
>
> PR target/81766
> * config/i386/i386.c (ix86_init_large_pic_reg): Return label instead of void.
> (ix86_init_pic_reg): Remember label from ix86_init_large_pic_reg, if non-NULL
> and preceded by NOTE_INSN_BASIC_BLOCK, swap the note and label.
>
> * gcc.target/i386/pr81766.c: New test.
OK.
Thanks,
Uros.
> --- gcc/config/i386/i386.c.jj 2017-08-07 18:50:10.000000000 +0200
> +++ gcc/config/i386/i386.c 2017-08-08 16:01:41.917136120 +0200
> @@ -8829,7 +8829,7 @@ ix86_use_pseudo_pic_reg (void)
>
> /* Initialize large model PIC register. */
>
> -static void
> +static rtx_code_label *
> ix86_init_large_pic_reg (unsigned int tmp_regno)
> {
> rtx_code_label *label;
> @@ -8846,6 +8846,7 @@ ix86_init_large_pic_reg (unsigned int tm
> emit_insn (gen_set_got_offset_rex64 (tmp_reg, label));
> emit_insn (ix86_gen_add3 (pic_offset_table_rtx,
> pic_offset_table_rtx, tmp_reg));
> + return label;
> }
>
> /* Create and initialize PIC register if required. */
> @@ -8854,6 +8855,7 @@ ix86_init_pic_reg (void)
> {
> edge entry_edge;
> rtx_insn *seq;
> + rtx_code_label *label = NULL;
>
> if (!ix86_use_pseudo_pic_reg ())
> return;
> @@ -8863,7 +8865,7 @@ ix86_init_pic_reg (void)
> if (TARGET_64BIT)
> {
> if (ix86_cmodel == CM_LARGE_PIC)
> - ix86_init_large_pic_reg (R11_REG);
> + label = ix86_init_large_pic_reg (R11_REG);
> else
> emit_insn (gen_set_got_rex64 (pic_offset_table_rtx));
> }
> @@ -8887,6 +8889,22 @@ ix86_init_pic_reg (void)
> entry_edge = single_succ_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun));
> insert_insn_on_edge (seq, entry_edge);
> commit_one_edge_insertion (entry_edge);
> +
> + if (label)
> + {
> + basic_block bb = BLOCK_FOR_INSN (label);
> + rtx_insn *bb_note = PREV_INSN (label);
> + /* If the note preceding the label starts a basic block, and the
> + label is a member of the same basic block, interchange the two. */
> + if (bb_note != NULL_RTX
> + && NOTE_INSN_BASIC_BLOCK_P (bb_note)
> + && bb != NULL
> + && bb == BLOCK_FOR_INSN (bb_note))
> + {
> + reorder_insns_nobb (bb_note, bb_note, label);
> + BB_HEAD (bb) = label;
> + }
> + }
> }
>
> /* Initialize a variable CUM of type CUMULATIVE_ARGS
> --- gcc/testsuite/gcc.target/i386/pr81766.c.jj 2017-08-08 16:10:04.299459808 +0200
> +++ gcc/testsuite/gcc.target/i386/pr81766.c 2017-08-08 16:09:28.000000000 +0200
> @@ -0,0 +1,9 @@
> +/* PR target/81766 */
> +/* { dg-do compile { target { pie && lp64 } } } */
> +/* { dg-options "-O2 -fpie -mcmodel=large" } */
> +
> +int
> +main ()
> +{
> + return 0;
> +}
>
> Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2017-09-01 16:22 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-09-01 11:42 [PATCH] Fix x86_64 ICE with -fpie -mcmodel=large (PR target/81766) Jakub Jelinek
2017-09-01 16:22 ` Uros Bizjak
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).