* [PATCH] predcom: Punt for steps which aren't multiples of access size [PR111683]
@ 2024-03-23 7:48 Jakub Jelinek
2024-03-23 10:15 ` Richard Biener
0 siblings, 1 reply; 2+ messages in thread
From: Jakub Jelinek @ 2024-03-23 7:48 UTC (permalink / raw)
To: Richard Biener; +Cc: gcc-patches
Hi!
On the following testcases, there is no overlap between data references
within a single iteration, but the data references have size which is twice
as large as the step, which means the data references overlap with the next
iteration which predcom doesn't take into account.
As discussed in the PR, even if the reference size is smaller than step,
if step isn't a multiple of the reference size, there could be overlaps with
some other iteration later on.
The initial version of the patch regressed (test still passed, but predcom
didn't optimize anymore) pr71083.c which has a packed char, short structure
and was reading/writing the short 2 bytes in there with step 3.
The following patch deals with that by retrying for COMPONENT_REFs also the
aggregate sizes etc., so that it then compares 3 bytes against step 3.
In make check-gcc/check-g++ this patch I believe affects code generation
for only the 2 new testcases according to statistics I've gathered.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2024-03-23 Jakub Jelinek <jakub@redhat.com>
PR middle-end/111683
* tree-predcom.cc (pcom_worker::suitable_component_p): If has_write
and comp_step is RS_NONZERO, return false if any reference in the
component doesn't have DR_STEP a multiple of access size.
* gcc.dg/pr111683-1.c: New test.
* gcc.dg/pr111683-2.c: New test.
--- gcc/tree-predcom.cc.jj 2024-03-22 09:19:27.700756950 +0100
+++ gcc/tree-predcom.cc 2024-03-22 14:01:21.758978338 +0100
@@ -1102,8 +1102,39 @@ pcom_worker::suitable_component_p (struc
gcc_assert (ok);
first->offset = 0;
- for (i = 1; comp->refs.iterate (i, &a); i++)
+ FOR_EACH_VEC_ELT (comp->refs, i, a)
{
+ if (has_write && comp->comp_step == RS_NONZERO)
+ {
+ /* Punt for non-invariant references where step isn't a multiple
+ of reference size. If step is smaller than reference size,
+ it overlaps the access in next iteration, if step is larger,
+ but not multiple of the access size, there could be overlap
+ in some later iteration. There might be more latent issues
+ about this in predcom or data reference analysis. If the
+ reference is a COMPONENT_REF, also check if step isn't a
+ multiple of the containg aggregate size. See PR111683. */
+ tree ref = DR_REF (a->ref);
+ tree step = DR_STEP (a->ref);
+ if (TREE_CODE (ref) == COMPONENT_REF
+ && DECL_BIT_FIELD (TREE_OPERAND (ref, 1)))
+ ref = TREE_OPERAND (ref, 0);
+ do
+ {
+ tree sz = TYPE_SIZE_UNIT (TREE_TYPE (ref));
+ if (TREE_CODE (sz) != INTEGER_CST)
+ return false;
+ if (wi::multiple_of_p (wi::to_offset (step),
+ wi::to_offset (sz), SIGNED))
+ break;
+ if (TREE_CODE (ref) != COMPONENT_REF)
+ return false;
+ ref = TREE_OPERAND (ref, 0);
+ }
+ while (1);
+ }
+ if (i == 0)
+ continue;
/* Polynomial offsets are no use, since we need to know the
gap between iteration numbers at compile time. */
poly_widest_int offset;
--- gcc/testsuite/gcc.dg/pr111683-1.c.jj 2024-03-22 11:14:29.292908760 +0100
+++ gcc/testsuite/gcc.dg/pr111683-1.c 2024-03-22 11:14:29.292908760 +0100
@@ -0,0 +1,22 @@
+/* PR middle-end/111683 */
+/* { dg-do run } */
+/* { dg-options "-O2" } */
+
+long long b[6] = { 3, 4, 5, 6, 7, 8 }, c[16];
+long long d[9] = { 3, 7, 12, 18, 22, 26, 21, 15, 8 };
+typedef long long U __attribute__ ((vector_size(16), may_alias, aligned(1)));
+typedef long long V __attribute__ ((vector_size(16), may_alias));
+
+int
+main ()
+{
+ for (int f = 0; f < 6; f++)
+ {
+ *(U *) &c[f] = *(U *) &c[f] + (V) { b[f], b[f] };
+ *(U *) &c[f + 2] = *(U *) &c[f + 2] + (V) { b[f], b[f] };
+ }
+ for (int f = 0; f < 9; f++)
+ if (c[f] != d[f])
+ __builtin_abort ();
+ return 0;
+}
--- gcc/testsuite/gcc.dg/pr111683-2.c.jj 2024-03-22 11:14:29.292908760 +0100
+++ gcc/testsuite/gcc.dg/pr111683-2.c 2024-03-22 11:14:29.292908760 +0100
@@ -0,0 +1,27 @@
+/* PR middle-end/111683 */
+/* { dg-do run } */
+/* { dg-options "-O2" } */
+
+int b[6] = { 3, 4, 5, 6, 7, 8 }, c[12];
+int d[16] = { 0, 1, 3, 6, 10, 14, 12, 9, 5, 0, 0, 0 };
+
+int
+main ()
+{
+ int i;
+ if (sizeof (int) * 2 != sizeof (long long))
+ return 0;
+ for (i = 0; i < 6; i++)
+ {
+ long long a;
+ __builtin_memcpy (&a, &c[i], sizeof (a));
+ a += (((long long) i) << (sizeof (int) * __CHAR_BIT__)) + i;
+ __builtin_memcpy (&c[i], &a, sizeof (a));
+ __builtin_memcpy (&a, &c[i + 2], sizeof (a));
+ a += (((long long) i) << (sizeof (int) * __CHAR_BIT__)) + i;
+ __builtin_memcpy (&c[i + 2], &a, sizeof (a));
+ }
+ if (__builtin_memcmp (&c[0], &d[0], sizeof (c)))
+ __builtin_abort ();
+ return 0;
+}
Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] predcom: Punt for steps which aren't multiples of access size [PR111683]
2024-03-23 7:48 [PATCH] predcom: Punt for steps which aren't multiples of access size [PR111683] Jakub Jelinek
@ 2024-03-23 10:15 ` Richard Biener
0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2024-03-23 10:15 UTC (permalink / raw)
To: Jakub Jelinek; +Cc: Richard Biener, gcc-patches
> Am 23.03.2024 um 08:49 schrieb Jakub Jelinek <jakub@redhat.com>:
>
> Hi!
>
> On the following testcases, there is no overlap between data references
> within a single iteration, but the data references have size which is twice
> as large as the step, which means the data references overlap with the next
> iteration which predcom doesn't take into account.
> As discussed in the PR, even if the reference size is smaller than step,
> if step isn't a multiple of the reference size, there could be overlaps with
> some other iteration later on.
> The initial version of the patch regressed (test still passed, but predcom
> didn't optimize anymore) pr71083.c which has a packed char, short structure
> and was reading/writing the short 2 bytes in there with step 3.
> The following patch deals with that by retrying for COMPONENT_REFs also the
> aggregate sizes etc., so that it then compares 3 bytes against step 3.
> In make check-gcc/check-g++ this patch I believe affects code generation
> for only the 2 new testcases according to statistics I've gathered.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
Ok
Thanks,
Richard
> 2024-03-23 Jakub Jelinek <jakub@redhat.com>
>
> PR middle-end/111683
> * tree-predcom.cc (pcom_worker::suitable_component_p): If has_write
> and comp_step is RS_NONZERO, return false if any reference in the
> component doesn't have DR_STEP a multiple of access size.
>
> * gcc.dg/pr111683-1.c: New test.
> * gcc.dg/pr111683-2.c: New test.
>
> --- gcc/tree-predcom.cc.jj 2024-03-22 09:19:27.700756950 +0100
> +++ gcc/tree-predcom.cc 2024-03-22 14:01:21.758978338 +0100
> @@ -1102,8 +1102,39 @@ pcom_worker::suitable_component_p (struc
> gcc_assert (ok);
> first->offset = 0;
>
> - for (i = 1; comp->refs.iterate (i, &a); i++)
> + FOR_EACH_VEC_ELT (comp->refs, i, a)
> {
> + if (has_write && comp->comp_step == RS_NONZERO)
> + {
> + /* Punt for non-invariant references where step isn't a multiple
> + of reference size. If step is smaller than reference size,
> + it overlaps the access in next iteration, if step is larger,
> + but not multiple of the access size, there could be overlap
> + in some later iteration. There might be more latent issues
> + about this in predcom or data reference analysis. If the
> + reference is a COMPONENT_REF, also check if step isn't a
> + multiple of the containg aggregate size. See PR111683. */
> + tree ref = DR_REF (a->ref);
> + tree step = DR_STEP (a->ref);
> + if (TREE_CODE (ref) == COMPONENT_REF
> + && DECL_BIT_FIELD (TREE_OPERAND (ref, 1)))
> + ref = TREE_OPERAND (ref, 0);
> + do
> + {
> + tree sz = TYPE_SIZE_UNIT (TREE_TYPE (ref));
> + if (TREE_CODE (sz) != INTEGER_CST)
> + return false;
> + if (wi::multiple_of_p (wi::to_offset (step),
> + wi::to_offset (sz), SIGNED))
> + break;
> + if (TREE_CODE (ref) != COMPONENT_REF)
> + return false;
> + ref = TREE_OPERAND (ref, 0);
> + }
> + while (1);
> + }
> + if (i == 0)
> + continue;
> /* Polynomial offsets are no use, since we need to know the
> gap between iteration numbers at compile time. */
> poly_widest_int offset;
> --- gcc/testsuite/gcc.dg/pr111683-1.c.jj 2024-03-22 11:14:29.292908760 +0100
> +++ gcc/testsuite/gcc.dg/pr111683-1.c 2024-03-22 11:14:29.292908760 +0100
> @@ -0,0 +1,22 @@
> +/* PR middle-end/111683 */
> +/* { dg-do run } */
> +/* { dg-options "-O2" } */
> +
> +long long b[6] = { 3, 4, 5, 6, 7, 8 }, c[16];
> +long long d[9] = { 3, 7, 12, 18, 22, 26, 21, 15, 8 };
> +typedef long long U __attribute__ ((vector_size(16), may_alias, aligned(1)));
> +typedef long long V __attribute__ ((vector_size(16), may_alias));
> +
> +int
> +main ()
> +{
> + for (int f = 0; f < 6; f++)
> + {
> + *(U *) &c[f] = *(U *) &c[f] + (V) { b[f], b[f] };
> + *(U *) &c[f + 2] = *(U *) &c[f + 2] + (V) { b[f], b[f] };
> + }
> + for (int f = 0; f < 9; f++)
> + if (c[f] != d[f])
> + __builtin_abort ();
> + return 0;
> +}
> --- gcc/testsuite/gcc.dg/pr111683-2.c.jj 2024-03-22 11:14:29.292908760 +0100
> +++ gcc/testsuite/gcc.dg/pr111683-2.c 2024-03-22 11:14:29.292908760 +0100
> @@ -0,0 +1,27 @@
> +/* PR middle-end/111683 */
> +/* { dg-do run } */
> +/* { dg-options "-O2" } */
> +
> +int b[6] = { 3, 4, 5, 6, 7, 8 }, c[12];
> +int d[16] = { 0, 1, 3, 6, 10, 14, 12, 9, 5, 0, 0, 0 };
> +
> +int
> +main ()
> +{
> + int i;
> + if (sizeof (int) * 2 != sizeof (long long))
> + return 0;
> + for (i = 0; i < 6; i++)
> + {
> + long long a;
> + __builtin_memcpy (&a, &c[i], sizeof (a));
> + a += (((long long) i) << (sizeof (int) * __CHAR_BIT__)) + i;
> + __builtin_memcpy (&c[i], &a, sizeof (a));
> + __builtin_memcpy (&a, &c[i + 2], sizeof (a));
> + a += (((long long) i) << (sizeof (int) * __CHAR_BIT__)) + i;
> + __builtin_memcpy (&c[i + 2], &a, sizeof (a));
> + }
> + if (__builtin_memcmp (&c[0], &d[0], sizeof (c)))
> + __builtin_abort ();
> + return 0;
> +}
>
> Jakub
>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2024-03-23 10:15 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-03-23 7:48 [PATCH] predcom: Punt for steps which aren't multiples of access size [PR111683] Jakub Jelinek
2024-03-23 10:15 ` 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).