Add libgomp.c-c++-common/pr106449-2.c (was: [committed] openmp: Fix up handling of non-rectangular simd loops with pointer type iterators [PR106449])
Commit Message
On 29.07.22 10:03, Jakub Jelinek wrote:
> There were 2 issues visible on this new testcase, one that we didn't have
> special POINTER_TYPE_P handling in a few spots of expand_omp_simd ...
> The other issue was that we put n2 expression directly into a
> comparison in a condition and regimplified that, for the &a[512] case that
> and with gimplification being destructed that unfortunately meant modification
> of original fd->loops[?].n2. Fixed by unsharing the expression.
I created a testcase for the non-simd case – and due to messing up, it failed;
hence, I filled PR middle-end/106467. After fixing the testcase, it passes.
(→ closed PR as invalid).
However, given that the testcase now exists, I think it makes sense to add it :-)
Changes compared to the simd testcase: replaced '(parallel for) simd' by 'for',
removed 'linear', used now 'b' and 'c' instead of storing both ptrs in 'b'.
Side remark: Before GCC 12, GCC complained about "q = p + n" with
"error: initializer expression refers to iteration variable ‘p’".
OK for mainline?
Tobias
-----------------
Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 80634 München; Gesellschaft mit beschränkter Haftung; Geschäftsführer: Thomas Heurung, Frank Thürauf; Sitz der Gesellschaft: München; Registergericht München, HRB 106955
Comments
On Fri, Jul 29, 2022 at 11:48:08AM +0200, Tobias Burnus wrote:
> On 29.07.22 10:03, Jakub Jelinek wrote:
> > There were 2 issues visible on this new testcase, one that we didn't have
> > special POINTER_TYPE_P handling in a few spots of expand_omp_simd ...
> > The other issue was that we put n2 expression directly into a
> > comparison in a condition and regimplified that, for the &a[512] case that
> > and with gimplification being destructed that unfortunately meant modification
> > of original fd->loops[?].n2. Fixed by unsharing the expression.
>
> I created a testcase for the non-simd case – and due to messing up, it failed;
> hence, I filled PR middle-end/106467. After fixing the testcase, it passes.
> (→ closed PR as invalid).
>
> However, given that the testcase now exists, I think it makes sense to add it :-)
>
> Changes compared to the simd testcase: replaced '(parallel for) simd' by 'for',
> removed 'linear', used now 'b' and 'c' instead of storing both ptrs in 'b'.
>
> Side remark: Before GCC 12, GCC complained about "q = p + n" with
> "error: initializer expression refers to iteration variable ‘p’".
>
> OK for mainline?
Ok, thanks.
Jakub
Add libgomp.c-c++-common/pr106449-2.c
This run-time test test pointer-based iteration with collapse,
similar to the '(parallel) simd' test for PR106449 but for 'for'.
libgomp/ChangeLog:
* testsuite/libgomp.c-c++-common/pr106449-2.c: New test.
.../testsuite/libgomp.c-c++-common/pr106449-2.c | 64 ++++++++++++++++++++++
1 file changed, 64 insertions(+)
new file mode 100644
@@ -0,0 +1,64 @@
+/* { dg-do run } */
+
+/* Based on pr106449.c - but using 'for' instead of 'simd'.
+ Cf. PR middle-end/106449 (for pr106449.c) and PR middle-end/106467. */
+
+void
+foo (void)
+{
+ int a[1024], *b[65536], *c[65536];
+ int *p, *q, **r = &b[0], **r2 = &c[0], i;
+ #pragma omp for collapse(2)
+ for (p = &a[0]; p < &a[512]; p++)
+ for (q = p + 64; q < p + 128; q++)
+ {
+ *r++ = p;
+ *r2++ = q;
+ }
+ for (i = 0; i < 32768; i++)
+ if (b[i] != &a[i / 64] || c[i] != &a[(i / 64) + 64 + (i % 64)])
+ __builtin_abort ();
+}
+
+void
+bar (int n, int m)
+{
+ int a[1024], *b[32768], *c[32768];
+ int *p, *q, **r = &b[0], **r2 = &c[0], i;
+ #pragma omp for collapse(2)
+ for (p = &a[0]; p < &a[512]; p++)
+ for (q = p + n; q < p + m; q++)
+ {
+ *r++ = p;
+ *r2++ = q;
+ }
+ for (i = 0; i < 32768; i++)
+ if (b[i] != &a[i / 64] || c[i] != &a[(i / 64) + 64 + (i % 64)])
+ __builtin_abort ();
+}
+
+void
+baz (int n, int m)
+{
+ int a[1024], *b[8192], *c[8192];
+ int *p, *q, **r = &b[0], **r2 = &c[0], i;
+ #pragma omp for collapse(2)
+ for (p = &a[0]; p < &a[512]; p += 4)
+ for (q = p + n; q < p + m; q += 2)
+ {
+ *r++ = p;
+ *r2++ = q;
+ }
+ for (i = 0; i < 4096; i++)
+ if (b[i] != &a[(i / 32) * 4] || c[i] != &a[(i / 32) * 4 + 64 + (i % 32) * 2])
+ __builtin_abort ();
+}
+
+int
+main ()
+{
+ foo ();
+ bar (64, 128);
+ baz (64, 128);
+ return 0;
+}