* [PATCH] ifcvt: Do not lower bitfields if we can't analyze dr's [PR107275]
@ 2022-10-18 8:35 Andre Vieira (lists)
2022-10-18 9:01 ` Richard Biener
0 siblings, 1 reply; 2+ messages in thread
From: Andre Vieira (lists) @ 2022-10-18 8:35 UTC (permalink / raw)
To: gcc-patches; +Cc: Richard Sandiford, Richard Biener, Jeff Law
[-- Attachment #1: Type: text/plain, Size: 983 bytes --]
The ifcvt dead code elimination code was not built to deal with inline
assembly, as loops with such would never be if-converted in the past since
we can't do data-reference analysis on them and vectorization would
eventually fail.
For this reason we now also do not lower bitfields if the data-reference
analysis
fails, as we would not end up vectorizing it. As a consequence this
also fixes
this PR as the dead code elimination will not run for such cases and
wrongfully
eliminate inline assembly statements.
Bootstrapped and regression tested on aarch64-none-linux-gnu and
x86_64-pc-linux-gnu.
gcc/ChangeLog:
PR tree-optimization/107275
* tree-if-conv.cc (if_convertible_loop_p_1): Move
find_data_references_in_loop call from here...
(if_convertible_loop_p): And move data-reference vector initialization
from here...
(tree_if_conversion):... to here.
gcc/testsuite/ChangeLog:
* gcc.dg/vect/pr107275.c: New test.
[-- Attachment #2: pr107275.patch --]
[-- Type: text/plain, Size: 3270 bytes --]
diff --git a/gcc/testsuite/gcc.dg/vect/pr107275.c b/gcc/testsuite/gcc.dg/vect/pr107275.c
new file mode 100644
index 0000000000000000000000000000000000000000..16327c46add7f74dd36c1a636c0fb9c84d5232c8
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/pr107275.c
@@ -0,0 +1,15 @@
+/* { dg-do compile } */
+struct st
+{
+ int a : 1;
+};
+
+void
+foo (struct st *s, int n)
+{
+ for (int i = 0; i < n; ++i)
+ {
+ s[i].a = i;
+ __asm__ __volatile__ ("":::"memory");
+ }
+}
diff --git a/gcc/tree-if-conv.cc b/gcc/tree-if-conv.cc
index 01637c5da08d5a2a00a495522fc9a6436a804398..a83b013d2ad3d9066b1c2bc62282ec054483598f 100644
--- a/gcc/tree-if-conv.cc
+++ b/gcc/tree-if-conv.cc
@@ -1416,9 +1416,6 @@ if_convertible_loop_p_1 (class loop *loop, vec<data_reference_p> *refs)
basic_block exit_bb = NULL;
vec<basic_block> region;
- if (find_data_references_in_loop (loop, refs) == chrec_dont_know)
- return false;
-
calculate_dominance_info (CDI_DOMINATORS);
for (i = 0; i < loop->num_nodes; i++)
@@ -1541,12 +1538,11 @@ if_convertible_loop_p_1 (class loop *loop, vec<data_reference_p> *refs)
- if its basic blocks and phi nodes are if convertible. */
static bool
-if_convertible_loop_p (class loop *loop)
+if_convertible_loop_p (class loop *loop, vec<data_reference_p> *refs)
{
edge e;
edge_iterator ei;
bool res = false;
- vec<data_reference_p> refs;
/* Handle only innermost loop. */
if (!loop || loop->inner)
@@ -1578,15 +1574,7 @@ if_convertible_loop_p (class loop *loop)
if (loop_exit_edge_p (loop, e))
return false;
- refs.create (5);
- res = if_convertible_loop_p_1 (loop, &refs);
-
- data_reference_p dr;
- unsigned int i;
- for (i = 0; refs.iterate (i, &dr); i++)
- free (dr->aux);
-
- free_data_refs (refs);
+ res = if_convertible_loop_p_1 (loop, refs);
delete innermost_DR_map;
innermost_DR_map = NULL;
@@ -3499,6 +3487,7 @@ tree_if_conversion (class loop *loop, vec<gimple *> *preds)
auto_vec <gassign *, 4> writes_to_lower;
bitmap exit_bbs;
edge pe;
+ vec<data_reference_p> refs;
again:
rloop = NULL;
@@ -3508,6 +3497,7 @@ tree_if_conversion (class loop *loop, vec<gimple *> *preds)
need_to_predicate = false;
need_to_rewrite_undefined = false;
any_complicated_phi = false;
+ refs.create (5);
/* Apply more aggressive if-conversion when loop or its outer loop were
marked with simd pragma. When that's the case, we try to if-convert
@@ -3537,11 +3527,14 @@ tree_if_conversion (class loop *loop, vec<gimple *> *preds)
goto cleanup;
}
+ if (find_data_references_in_loop (loop, &refs) == chrec_dont_know)
+ goto cleanup;
+
if (loop->num_nodes > 2)
{
need_to_ifcvt = true;
- if (!if_convertible_loop_p (loop) || !dbg_cnt (if_conversion_tree))
+ if (!if_convertible_loop_p (loop, &refs) || !dbg_cnt (if_conversion_tree))
goto cleanup;
if ((need_to_predicate || any_complicated_phi)
@@ -3658,6 +3651,13 @@ tree_if_conversion (class loop *loop, vec<gimple *> *preds)
todo |= TODO_cleanup_cfg;
cleanup:
+ data_reference_p dr;
+ unsigned int i;
+ for (i = 0; refs.iterate (i, &dr); i++)
+ free (dr->aux);
+
+ refs.truncate (0);
+
if (ifc_bbs)
{
unsigned int i;
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] ifcvt: Do not lower bitfields if we can't analyze dr's [PR107275]
2022-10-18 8:35 [PATCH] ifcvt: Do not lower bitfields if we can't analyze dr's [PR107275] Andre Vieira (lists)
@ 2022-10-18 9:01 ` Richard Biener
0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2022-10-18 9:01 UTC (permalink / raw)
To: Andre Vieira (lists); +Cc: gcc-patches, Richard Sandiford, Jeff Law
[-- Attachment #1: Type: text/plain, Size: 1310 bytes --]
On Tue, 18 Oct 2022, Andre Vieira (lists) wrote:
> The ifcvt dead code elimination code was not built to deal with inline
> assembly, as loops with such would never be if-converted in the past since
> we can't do data-reference analysis on them and vectorization would eventually
> fail.
> For this reason we now also do not lower bitfields if the data-reference
> analysis
> fails, as we would not end up vectorizing it. As a consequence this also
> fixes
> this PR as the dead code elimination will not run for such cases and
> wrongfully
> eliminate inline assembly statements.
>
> Bootstrapped and regression tested on aarch64-none-linux-gnu and
> x86_64-pc-linux-gnu.
OK.
Thanks,
Richard.
> gcc/ChangeLog:
>
> PR tree-optimization/107275
> * tree-if-conv.cc (if_convertible_loop_p_1): Move
> find_data_references_in_loop call from here...
> (if_convertible_loop_p): And move data-reference vector initialization
> from here...
> (tree_if_conversion):... to here.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.dg/vect/pr107275.c: New test.
>
--
Richard Biener <rguenther@suse.de>
SUSE Software Solutions Germany GmbH, Frankenstrasse 146, 90461 Nuernberg,
Germany; GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman;
HRB 36809 (AG Nuernberg)
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2022-10-18 9:01 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-10-18 8:35 [PATCH] ifcvt: Do not lower bitfields if we can't analyze dr's [PR107275] Andre Vieira (lists)
2022-10-18 9:01 ` 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).