The `recompute_dominator' function used in the code fragment within
this patch assumes correctness in the rest of the CFG. Consequently,
it is wrong to rely upon it before the subsequent updates are made in
the "Update dominators for multiple exits" loop in the function.
Furthermore, if `loop_exit' == `scalar_exit', the "Update dominators for
multiple exits" logic will already take care of updating the
dominator for `scalar_exit->dest', such that the moved statement is
unnecessary.
gcc/ChangeLog:
PR tree-optimization/123152
* tree-vect-loop-manip.cc
(slpeel_tree_duplicate_loop_to_edge_cfg): Correct order of
dominator update
gcc/testsuite/ChangeLog:
* gcc.dg/vect/vect-uncounted-prolog-peel_2.c: New.
* gcc.dg/vect/vect-uncounted-prolog-peel_3.c: Likewise.
* g++.dg/vect/vect-uncounted-prolog-peel_1.cc: Likewise.
---
.../vect/vect-uncounted-prolog-peel_1.cc | 51 +++++++++++++++++++
.../vect/vect-uncounted-prolog-peel_2.c | 25 +++++++++
.../vect/vect-uncounted-prolog-peel_3.c | 26 ++++++++++
gcc/tree-vect-loop-manip.cc | 10 ++--
4 files changed, 107 insertions(+), 5 deletions(-)
create mode 100644 gcc/testsuite/g++.dg/vect/vect-uncounted-prolog-peel_1.cc
create mode 100644 gcc/testsuite/gcc.dg/vect/vect-uncounted-prolog-peel_2.c
create mode 100644 gcc/testsuite/gcc.dg/vect/vect-uncounted-prolog-peel_3.c
diff --git a/gcc/testsuite/g++.dg/vect/vect-uncounted-prolog-peel_1.cc
b/gcc/testsuite/g++.dg/vect/vect-uncounted-prolog-peel_1.cc
new file mode 100644
index 00000000000..07335a23567
--- /dev/null
+++ b/gcc/testsuite/g++.dg/vect/vect-uncounted-prolog-peel_1.cc
@@ -0,0 +1,51 @@
+/* { dg-add-options vect_early_break } */
+/* { dg-do compile } */
+/* { dg-require-effective-target vect_early_break } */
+/* { dg-require-effective-target vect_int } */
+/* { dg-additional-options "-w" } */
+
+namespace a {
+class b {
+public:
+ b d(long);
+};
+class e : public b {};
+typedef short f;
+typedef struct g *h;
+typedef struct g {
+ f c;
+ e *fp;
+ f j, *k, *l;
+} m;
+static h n(m *, unsigned, unsigned *);
+f o(m *, unsigned);
+void p(short *t) {
+ m *ffile;
+ int q;
+ unsigned r;
+ int i;
+ q = t[i];
+ n(ffile, q, &r);
+}
+h n(m *ffile, unsigned q, unsigned *) {
+ h glyph;
+ for (; glyph;)
+ if (glyph->c)
+ o(ffile, q);
+ int i;
+s:
+ for (i = 0; ffile->j; i++)
+ if (ffile->k[i]) {
+ if (q)
+ ffile->fp->d(ffile->l[i]);
+ break;
+ }
+ if (q < 6) {
+ q += 61440;
+ goto s;
+ }
+}
+} // namespace a
+
+/* { dg-final { scan-tree-dump {note:\s*Alignment of access forced using
peeling.} "vect" } } */
+/* { dg-final { scan-tree-dump {vectorized 1 loops in function} "vect" } } */
diff --git a/gcc/testsuite/gcc.dg/vect/vect-uncounted-prolog-peel_2.c
b/gcc/testsuite/gcc.dg/vect/vect-uncounted-prolog-peel_2.c
new file mode 100644
index 00000000000..a9390c401f0
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/vect-uncounted-prolog-peel_2.c
@@ -0,0 +1,25 @@
+/* { dg-add-options vect_early_break } */
+/* { dg-do compile } */
+/* { dg-require-effective-target vect_early_break } */
+/* { dg-require-effective-target vect_int } */
+
+char b;
+int c(char *d) {
+ int *a = 0;
+ while (*d) {
+ while (*a)
+ if (*a++ == 1)
+ return 1;
+ d++;
+ }
+}
+void e() {
+ c(&b);
+ char *f = &b;
+ while (f[0])
+ ++b;
+}
+
+/* { dg-final { scan-tree-dump {note:\s*Alignment of access forced using
peeling.} "vect" } } */
+/* { dg-final { scan-tree-dump {if \(ivtmp_[0-9_]+ >=
prolog_loop_niters.[0-9_]+\)\n\s*goto} "vect" } } */
+/* { dg-final { scan-tree-dump {vectorized 1 loops in function} "vect" } } */
diff --git a/gcc/testsuite/gcc.dg/vect/vect-uncounted-prolog-peel_3.c
b/gcc/testsuite/gcc.dg/vect/vect-uncounted-prolog-peel_3.c
new file mode 100644
index 00000000000..4d842ebd909
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/vect-uncounted-prolog-peel_3.c
@@ -0,0 +1,26 @@
+/* { dg-add-options vect_early_break } */
+/* { dg-do compile } */
+/* { dg-require-effective-target vect_early_break } */
+/* { dg-require-effective-target vect_int } */
+
+struct {
+ int segments[];
+} seek_to_sample_coarse_f;
+
+int seek_to_sample_coarse_i, seek_to_sample_coarse_f_1;
+
+void seek_to_sample_coarse() {
+ int end_pos = seek_to_sample_coarse_f_1;
+ for (;;) {
+ seek_to_sample_coarse_i = end_pos;
+ for (; end_pos > 0; --end_pos)
+ if (seek_to_sample_coarse_f.segments[end_pos - 1])
+ break;
+ if (end_pos)
+ break;
+ }
+}
+
+/* { dg-final { scan-tree-dump {note:\s*Alignment of access forced using
peeling.} "vect" } } */
+/* { dg-final { scan-tree-dump {if \(ivtmp_[0-9_]+ >=
prolog_loop_niters.[0-9_]+\)\n\s*goto} "vect" } } */
+/* { dg-final { scan-tree-dump {vectorized 1 loops in function} "vect" } } */
diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc
index e2ea0426050..99f69d3376b 100644
--- a/gcc/tree-vect-loop-manip.cc
+++ b/gcc/tree-vect-loop-manip.cc
@@ -1989,11 +1989,6 @@ slpeel_tree_duplicate_loop_to_edge_cfg (class loop
*loop, edge loop_exit,
flush_pending_stmts (new_exit);
set_immediate_dominator (CDI_DOMINATORS, preheader, new_exit->src);
- if (create_main_e)
- set_immediate_dominator (CDI_DOMINATORS, scalar_exit->dest,
- recompute_dominator (CDI_DOMINATORS,
- scalar_exit->dest));
-
/* And remove the non-necessary forwarder again. Keep the other
one so we have a proper pre-header for the loop at the exit edge. */
redirect_edge_pred (single_succ_edge (new_preheader),
@@ -2024,6 +2019,11 @@ slpeel_tree_duplicate_loop_to_edge_cfg (class loop
*loop, edge loop_exit,
}
}
}
+
+ if (create_main_e && (loop_exit != scalar_exit))
+ set_immediate_dominator (CDI_DOMINATORS, scalar_exit->dest,
+ recompute_dominator (CDI_DOMINATORS,
+ scalar_exit->dest));
}
free (new_bbs);
--
2.43.0
