* [PR46761] graphite: fix testing of boundary wrapping
@ 2010-12-13 14:44 Alexander Monakov
2010-12-13 16:36 ` Sebastian Pop
0 siblings, 1 reply; 7+ messages in thread
From: Alexander Monakov @ 2010-12-13 14:44 UTC (permalink / raw)
To: gcc-patches; +Cc: Sebastian Pop
Hi,
As indicated by the testcase, sometimes Graphite would generate wrong region
guards if UB_ONE overflows, but does not become zero (i.e. it is signed).
The patch changes the corresponding test to use TREE_OVERFLOW flag.
Bootstrapped and regtested on x86_64-linux, OK for trunk?
2010-12-13 Alexander Monakov <amonakov@ispras.ru>
PR middle-end/46761
* graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Use
TREE_OVERFLOW_P to test overflow.
testsuite:
* gcc.dg/graphite/pr46761.c: New.
diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
index 4894b52..183dde7 100644
--- a/gcc/graphite-clast-to-gimple.c
+++ b/gcc/graphite-clast-to-gimple.c
@@ -985,7 +985,7 @@ graphite_create_new_loop_guard (sese region, edge entry_edge,
: PLUS_EXPR, type, ub, one);
/* When ub + 1 wraps around, use lb <= ub. */
- if (integer_zerop (ub_one))
+ if (TREE_OVERFLOW_P (ub_one))
cond_expr = fold_build2 (LE_EXPR, boolean_type_node, lb, ub);
else
cond_expr = fold_build2 (LT_EXPR, boolean_type_node, lb, ub_one);
diff --git a/gcc/testsuite/gcc.dg/graphite/pr46761.c b/gcc/testsuite/gcc.dg/graphite/pr46761.c
new file mode 100644
index 0000000..f45398a
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/graphite/pr46761.c
@@ -0,0 +1,22 @@
+/* { dg-do run } */
+/* { dg-options "-O -fgraphite-identity" } */
+
+#define N 128
+
+int
+main ()
+{
+ int arr[N], i, s = 0;
+ for (i = 0; i < N; i++)
+ arr[i] = i;
+
+ for (i = 0; i < N; i++)
+ if (arr[i] != i)
+ __builtin_abort ();
+
+ for (i = 0; i < N; i++)
+ s += arr[i];
+ if (s != (N * (N - 1)) / 2)
+ __builtin_abort ();
+ return 0;
+}
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PR46761] graphite: fix testing of boundary wrapping
2010-12-13 14:44 [PR46761] graphite: fix testing of boundary wrapping Alexander Monakov
@ 2010-12-13 16:36 ` Sebastian Pop
2010-12-13 17:31 ` Richard Guenther
0 siblings, 1 reply; 7+ messages in thread
From: Sebastian Pop @ 2010-12-13 16:36 UTC (permalink / raw)
To: Alexander Monakov; +Cc: gcc-patches
On Mon, Dec 13, 2010 at 07:40, Alexander Monakov <amonakov@ispras.ru> wrote:
> Hi,
>
> As indicated by the testcase, sometimes Graphite would generate wrong region
> guards if UB_ONE overflows, but does not become zero (i.e. it is signed).
> The patch changes the corresponding test to use TREE_OVERFLOW flag.
>
> Bootstrapped and regtested on x86_64-linux, OK for trunk?
>
> 2010-12-13 Alexander Monakov <amonakov@ispras.ru>
>
> PR middle-end/46761
> * graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Use
> TREE_OVERFLOW_P to test overflow.
>
> testsuite:
> * gcc.dg/graphite/pr46761.c: New.
>
Ok. Thanks for fixing this,
Sebastian
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PR46761] graphite: fix testing of boundary wrapping
2010-12-13 16:36 ` Sebastian Pop
@ 2010-12-13 17:31 ` Richard Guenther
2010-12-15 16:20 ` Alexander Monakov
0 siblings, 1 reply; 7+ messages in thread
From: Richard Guenther @ 2010-12-13 17:31 UTC (permalink / raw)
To: Sebastian Pop; +Cc: Alexander Monakov, gcc-patches
2010/12/13 Sebastian Pop <sebpop@gmail.com>:
> On Mon, Dec 13, 2010 at 07:40, Alexander Monakov <amonakov@ispras.ru> wrote:
>> Hi,
>>
>> As indicated by the testcase, sometimes Graphite would generate wrong region
>> guards if UB_ONE overflows, but does not become zero (i.e. it is signed).
>> The patch changes the corresponding test to use TREE_OVERFLOW flag.
>>
>> Bootstrapped and regtested on x86_64-linux, OK for trunk?
>>
>> 2010-12-13 Alexander Monakov <amonakov@ispras.ru>
>>
>> PR middle-end/46761
>> * graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Use
>> TREE_OVERFLOW_P to test overflow.
>>
>> testsuite:
>> * gcc.dg/graphite/pr46761.c: New.
>>
>
> Ok. Thanks for fixing this,
Hmm, I don't like new uses of TREE_OVERFLOW checking. And for
unsigned types it won't be set anyway.
Richard.
> Sebastian
>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PR46761] graphite: fix testing of boundary wrapping
2010-12-13 17:31 ` Richard Guenther
@ 2010-12-15 16:20 ` Alexander Monakov
2010-12-16 11:57 ` Richard Guenther
0 siblings, 1 reply; 7+ messages in thread
From: Alexander Monakov @ 2010-12-15 16:20 UTC (permalink / raw)
To: Richard Guenther; +Cc: Sebastian Pop, gcc-patches
[-- Attachment #1: Type: TEXT/PLAIN, Size: 3440 bytes --]
On Mon, 13 Dec 2010, Richard Guenther wrote:
> 2010/12/13 Sebastian Pop <sebpop@gmail.com>:
> > On Mon, Dec 13, 2010 at 07:40, Alexander Monakov <amonakov@ispras.ru> wrote:
> >> Hi,
> >>
> >> As indicated by the testcase, sometimes Graphite would generate wrong region
> >> guards if UB_ONE overflows, but does not become zero (i.e. it is signed).
> >> The patch changes the corresponding test to use TREE_OVERFLOW flag.
> >>
> >> Bootstrapped and regtested on x86_64-linux, OK for trunk?
> >>
> >> 2010-12-13  Alexander Monakov  <amonakov@ispras.ru>
> >>
> >> Â Â Â Â PR middle-end/46761
> >> Â Â Â Â * graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Use
> >> Â Â Â Â TREE_OVERFLOW_P to test overflow.
> >>
> >> testsuite:
> >> Â Â Â Â * gcc.dg/graphite/pr46761.c: New.
> >>
> >
> > Ok. Â Thanks for fixing this,
>
> Hmm, I don't like new uses of TREE_OVERFLOW checking. And for
> unsigned types it won't be set anyway.
Thanks. The following patch changes guard generation so that using unadjusted
UB is preferred (if it's an integer constant or an SSA_NAME). If it is
neither, it uses the previous behaviour of bumping UB by one and changing the
test. Bootstrapped and tested with make -k check RUNTESTFLAGS="graphite.exp",
OK for trunk?
2010-12-15 Alexander Monakov <amonakov@ispras.ru>
    PR middle-end/46761
    * graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Prefer
    to use unadjusted UB.
diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
index 4894b52..4725608 100644
--- a/gcc/graphite-clast-to-gimple.c
+++ b/gcc/graphite-clast-to-gimple.c
@@ -975,20 +975,24 @@ graphite_create_new_loop_guard (sese region, edge entry_edge,
newivs_index, params_index);
tree ub = clast_to_gcc_expression (type, stmt->UB, region, newivs,
newivs_index, params_index);
- tree one = POINTER_TYPE_P (type) ? size_one_node
- : fold_convert (type, integer_one_node);
- /* Adding +1 and using LT_EXPR helps with loop latches that have a
- loop iteration count of "PARAMETER - 1". For PARAMETER == 0 this becomes
- 2^{32|64}, and the condition lb <= ub is true, even if we do not want this.
- However lb < ub + 1 is false, as expected. */
- tree ub_one = fold_build2 (POINTER_TYPE_P (type) ? POINTER_PLUS_EXPR
- : PLUS_EXPR, type, ub, one);
-
- /* When ub + 1 wraps around, use lb <= ub. */
- if (integer_zerop (ub_one))
+ /* When ub is simply a constant or a parameter, use lb <= ub. */
+ if (TREE_CODE (ub) == INTEGER_CST || TREE_CODE (ub) == SSA_NAME)
cond_expr = fold_build2 (LE_EXPR, boolean_type_node, lb, ub);
else
- cond_expr = fold_build2 (LT_EXPR, boolean_type_node, lb, ub_one);
+ {
+ tree one = (POINTER_TYPE_P (type)
+ ? size_one_node
+ : fold_convert (type, integer_one_node));
+ /* Adding +1 and using LT_EXPR helps with loop latches that have a
+ loop iteration count of "PARAMETER - 1". For PARAMETER == 0 this becomes
+ 2^k-1 due to unsigned overflow, and the condition lb <= ub is true,
+ even if we do not want this. However lb < ub + 1 is false, as
+ expected. */
+ tree ub_one = fold_build2 (POINTER_TYPE_P (type) ? POINTER_PLUS_EXPR
+ : PLUS_EXPR, type, ub, one);
+
+ cond_expr = fold_build2 (LT_EXPR, boolean_type_node, lb, ub_one);
+ }
exit_edge = create_empty_if_region_on_edge (entry_edge, cond_expr);
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PR46761] graphite: fix testing of boundary wrapping
2010-12-15 16:20 ` Alexander Monakov
@ 2010-12-16 11:57 ` Richard Guenther
2010-12-16 22:29 ` Alexander Monakov
0 siblings, 1 reply; 7+ messages in thread
From: Richard Guenther @ 2010-12-16 11:57 UTC (permalink / raw)
To: Alexander Monakov; +Cc: Sebastian Pop, gcc-patches
On Wed, Dec 15, 2010 at 4:46 PM, Alexander Monakov <amonakov@ispras.ru> wrote:
>
>
> On Mon, 13 Dec 2010, Richard Guenther wrote:
>
>> 2010/12/13 Sebastian Pop <sebpop@gmail.com>:
>> > On Mon, Dec 13, 2010 at 07:40, Alexander Monakov <amonakov@ispras.ru> wrote:
>> >> Hi,
>> >>
>> >> As indicated by the testcase, sometimes Graphite would generate wrong region
>> >> guards if UB_ONE overflows, but does not become zero (i.e. it is signed).
>> >> The patch changes the corresponding test to use TREE_OVERFLOW flag.
>> >>
>> >> Bootstrapped and regtested on x86_64-linux, OK for trunk?
>> >>
>> >> 2010-12-13 Alexander Monakov <amonakov@ispras.ru>
>> >>
>> >> PR middle-end/46761
>> >> * graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Use
>> >> TREE_OVERFLOW_P to test overflow.
>> >>
>> >> testsuite:
>> >> * gcc.dg/graphite/pr46761.c: New.
>> >>
>> >
>> > Ok. Thanks for fixing this,
>>
>> Hmm, I don't like new uses of TREE_OVERFLOW checking. And for
>> unsigned types it won't be set anyway.
>
> Thanks. The following patch changes guard generation so that using unadjusted
> UB is preferred (if it's an integer constant or an SSA_NAME). If it is
> neither, it uses the previous behaviour of bumping UB by one and changing the
> test. Bootstrapped and tested with make -k check RUNTESTFLAGS="graphite.exp",
> OK for trunk?
That looks better.
Thanks,
Richard.
> 2010-12-15 Alexander Monakov <amonakov@ispras.ru>
>
> PR middle-end/46761
> * graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Prefer
> to use unadjusted UB.
>
> diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
> index 4894b52..4725608 100644
> --- a/gcc/graphite-clast-to-gimple.c
> +++ b/gcc/graphite-clast-to-gimple.c
> @@ -975,20 +975,24 @@ graphite_create_new_loop_guard (sese region, edge entry_edge,
> newivs_index, params_index);
> tree ub = clast_to_gcc_expression (type, stmt->UB, region, newivs,
> newivs_index, params_index);
> - tree one = POINTER_TYPE_P (type) ? size_one_node
> - : fold_convert (type, integer_one_node);
> - /* Adding +1 and using LT_EXPR helps with loop latches that have a
> - loop iteration count of "PARAMETER - 1". For PARAMETER == 0 this becomes
> - 2^{32|64}, and the condition lb <= ub is true, even if we do not want this.
> - However lb < ub + 1 is false, as expected. */
> - tree ub_one = fold_build2 (POINTER_TYPE_P (type) ? POINTER_PLUS_EXPR
> - : PLUS_EXPR, type, ub, one);
> -
> - /* When ub + 1 wraps around, use lb <= ub. */
> - if (integer_zerop (ub_one))
> + /* When ub is simply a constant or a parameter, use lb <= ub. */
> + if (TREE_CODE (ub) == INTEGER_CST || TREE_CODE (ub) == SSA_NAME)
> cond_expr = fold_build2 (LE_EXPR, boolean_type_node, lb, ub);
> else
> - cond_expr = fold_build2 (LT_EXPR, boolean_type_node, lb, ub_one);
> + {
> + tree one = (POINTER_TYPE_P (type)
> + ? size_one_node
> + : fold_convert (type, integer_one_node));
> + /* Adding +1 and using LT_EXPR helps with loop latches that have a
> + loop iteration count of "PARAMETER - 1". For PARAMETER == 0 this becomes
> + 2^k-1 due to unsigned overflow, and the condition lb <= ub is true,
> + even if we do not want this. However lb < ub + 1 is false, as
> + expected. */
> + tree ub_one = fold_build2 (POINTER_TYPE_P (type) ? POINTER_PLUS_EXPR
> + : PLUS_EXPR, type, ub, one);
> +
> + cond_expr = fold_build2 (LT_EXPR, boolean_type_node, lb, ub_one);
> + }
>
> exit_edge = create_empty_if_region_on_edge (entry_edge, cond_expr);
>
>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PR46761] graphite: fix testing of boundary wrapping
2010-12-16 11:57 ` Richard Guenther
@ 2010-12-16 22:29 ` Alexander Monakov
2010-12-16 23:21 ` Sebastian Pop
0 siblings, 1 reply; 7+ messages in thread
From: Alexander Monakov @ 2010-12-16 22:29 UTC (permalink / raw)
To: Richard Guenther; +Cc: Sebastian Pop, gcc-patches
[-- Attachment #1: Type: TEXT/PLAIN, Size: 713 bytes --]
On Thu, 16 Dec 2010, Richard Guenther wrote:
> > Thanks. Â The following patch changes guard generation so that using unadjusted
> > UB is preferred (if it's an integer constant or an SSA_NAME). Â If it is
> > neither, it uses the previous behaviour of bumping UB by one and changing the
> > test. Â Bootstrapped and tested with make -k check RUNTESTFLAGS="graphite.exp",
> > OK for trunk?
>
> That looks better.
>
> Thanks,
> Richard.
Sebastian, is this patch OK to commit?
> > 2010-12-15  Alexander Monakov  <amonakov@ispras.ru>
> >
> > Â Â Â Â Â PR middle-end/46761
> > Â Â Â Â Â * graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Prefer
> > Â Â Â Â Â to use unadjusted UB.
> >
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PR46761] graphite: fix testing of boundary wrapping
2010-12-16 22:29 ` Alexander Monakov
@ 2010-12-16 23:21 ` Sebastian Pop
0 siblings, 0 replies; 7+ messages in thread
From: Sebastian Pop @ 2010-12-16 23:21 UTC (permalink / raw)
To: Alexander Monakov; +Cc: Richard Guenther, gcc-patches
On Thu, Dec 16, 2010 at 15:50, Alexander Monakov <amonakov@ispras.ru> wrote:
>
>
> On Thu, 16 Dec 2010, Richard Guenther wrote:
>
>> > Thanks. The following patch changes guard generation so that using unadjusted
>> > UB is preferred (if it's an integer constant or an SSA_NAME). If it is
>> > neither, it uses the previous behaviour of bumping UB by one and changing the
>> > test. Bootstrapped and tested with make -k check RUNTESTFLAGS="graphite.exp",
>> > OK for trunk?
>>
>> That looks better.
>>
>> Thanks,
>> Richard.
>
> Sebastian, is this patch OK to commit?
Yes, please commit the patch to trunk. Thanks for fixing this bug.
Sebastian
>
>> > 2010-12-15 Alexander Monakov <amonakov@ispras.ru>
>> >
>> > PR middle-end/46761
>> > * graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Prefer
>> > to use unadjusted UB.
>> >
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2010-12-16 22:47 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2010-12-13 14:44 [PR46761] graphite: fix testing of boundary wrapping Alexander Monakov
2010-12-13 16:36 ` Sebastian Pop
2010-12-13 17:31 ` Richard Guenther
2010-12-15 16:20 ` Alexander Monakov
2010-12-16 11:57 ` Richard Guenther
2010-12-16 22:29 ` Alexander Monakov
2010-12-16 23:21 ` Sebastian Pop
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).