* [PATCH] Improve build_simple_mem_ref_loc (PR middle-end/83185)
@ 2017-11-29 8:24 Jakub Jelinek
2017-11-29 9:47 ` Richard Biener
0 siblings, 1 reply; 2+ messages in thread
From: Jakub Jelinek @ 2017-11-29 8:24 UTC (permalink / raw)
To: Richard Biener; +Cc: gcc-patches
Hi!
This PR is about forwprop propagating:
_17 = __builtin_alloca_with_align (_16, 256);
_18 = _17 + 32;
__builtin___asan_alloca_poison (_18, _8);
_7 = &*_18[4];
__builtin_va_start (_7, 0);
to:
_17 = __builtin_alloca_with_align (_16, 256);
_18 = _17 + 32;
__builtin___asan_alloca_poison (_18, _8);
_7 = &MEM[(struct [0:D.2257][1] *)_17 + 32B][4];
__builtin_va_start (_7, 0);
which is something the verifiers allow and then backend VA_START
handling calling build_simple_mem_ref_loc on the ADDR_EXPR it got.
get_addr_base_and_unit_offset only looks through a MEM_REF if it
has ADDR_EXPR as the first operand, which is not the case here, so nothing
sums up the 96 offset from the [4] ARRAY_REF with the extra 32 from the
MEM_REF.
Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for
trunk?
2017-11-29 Jakub Jelinek <jakub@redhat.com>
PR middle-end/83185
* tree.c (build_simple_mem_ref_loc): Handle
get_addr_base_and_unit_offset returning a MEM_REF.
* gcc.dg/asan/pr83185.c: New test.
--- gcc/tree.c.jj 2017-11-28 12:11:38.000000000 +0100
+++ gcc/tree.c 2017-11-28 17:22:01.800939050 +0100
@@ -4692,7 +4692,13 @@ build_simple_mem_ref_loc (location_t loc
{
ptr = get_addr_base_and_unit_offset (TREE_OPERAND (ptr, 0), &offset);
gcc_assert (ptr);
- ptr = build_fold_addr_expr (ptr);
+ if (TREE_CODE (ptr) == MEM_REF)
+ {
+ offset += mem_ref_offset (ptr).to_short_addr ();
+ ptr = TREE_OPERAND (ptr, 0);
+ }
+ else
+ ptr = build_fold_addr_expr (ptr);
gcc_assert (is_gimple_reg (ptr) || is_gimple_min_invariant (ptr));
}
tem = build2 (MEM_REF, TREE_TYPE (ptype),
--- gcc/testsuite/gcc.dg/asan/pr83185.c.jj 2017-11-28 17:24:15.540329283 +0100
+++ gcc/testsuite/gcc.dg/asan/pr83185.c 2017-11-28 17:24:19.851277394 +0100
@@ -0,0 +1,14 @@
+/* PR middle-end/83185 */
+/* { dg-do compile } */
+
+#include <stdarg.h>
+
+int bar (void);
+
+void
+foo (int i, ...)
+{
+ va_list aps[bar()];
+ va_start (aps[4], i);
+ va_end (aps[4]);
+}
Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Improve build_simple_mem_ref_loc (PR middle-end/83185)
2017-11-29 8:24 [PATCH] Improve build_simple_mem_ref_loc (PR middle-end/83185) Jakub Jelinek
@ 2017-11-29 9:47 ` Richard Biener
0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2017-11-29 9:47 UTC (permalink / raw)
To: Jakub Jelinek; +Cc: gcc-patches
On Wed, 29 Nov 2017, Jakub Jelinek wrote:
> Hi!
>
> This PR is about forwprop propagating:
> _17 = __builtin_alloca_with_align (_16, 256);
> _18 = _17 + 32;
> __builtin___asan_alloca_poison (_18, _8);
> _7 = &*_18[4];
> __builtin_va_start (_7, 0);
> to:
> _17 = __builtin_alloca_with_align (_16, 256);
> _18 = _17 + 32;
> __builtin___asan_alloca_poison (_18, _8);
> _7 = &MEM[(struct [0:D.2257][1] *)_17 + 32B][4];
> __builtin_va_start (_7, 0);
> which is something the verifiers allow and then backend VA_START
> handling calling build_simple_mem_ref_loc on the ADDR_EXPR it got.
> get_addr_base_and_unit_offset only looks through a MEM_REF if it
> has ADDR_EXPR as the first operand, which is not the case here, so nothing
> sums up the 96 offset from the [4] ARRAY_REF with the extra 32 from the
> MEM_REF.
>
> Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for
> trunk?
Ok.
Richard.
> 2017-11-29 Jakub Jelinek <jakub@redhat.com>
>
> PR middle-end/83185
> * tree.c (build_simple_mem_ref_loc): Handle
> get_addr_base_and_unit_offset returning a MEM_REF.
>
> * gcc.dg/asan/pr83185.c: New test.
>
> --- gcc/tree.c.jj 2017-11-28 12:11:38.000000000 +0100
> +++ gcc/tree.c 2017-11-28 17:22:01.800939050 +0100
> @@ -4692,7 +4692,13 @@ build_simple_mem_ref_loc (location_t loc
> {
> ptr = get_addr_base_and_unit_offset (TREE_OPERAND (ptr, 0), &offset);
> gcc_assert (ptr);
> - ptr = build_fold_addr_expr (ptr);
> + if (TREE_CODE (ptr) == MEM_REF)
> + {
> + offset += mem_ref_offset (ptr).to_short_addr ();
> + ptr = TREE_OPERAND (ptr, 0);
> + }
> + else
> + ptr = build_fold_addr_expr (ptr);
> gcc_assert (is_gimple_reg (ptr) || is_gimple_min_invariant (ptr));
> }
> tem = build2 (MEM_REF, TREE_TYPE (ptype),
> --- gcc/testsuite/gcc.dg/asan/pr83185.c.jj 2017-11-28 17:24:15.540329283 +0100
> +++ gcc/testsuite/gcc.dg/asan/pr83185.c 2017-11-28 17:24:19.851277394 +0100
> @@ -0,0 +1,14 @@
> +/* PR middle-end/83185 */
> +/* { dg-do compile } */
> +
> +#include <stdarg.h>
> +
> +int bar (void);
> +
> +void
> +foo (int i, ...)
> +{
> + va_list aps[bar()];
> + va_start (aps[4], i);
> + va_end (aps[4]);
> +}
>
> 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 9:45 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-11-29 8:24 [PATCH] Improve build_simple_mem_ref_loc (PR middle-end/83185) Jakub Jelinek
2017-11-29 9:47 ` 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).