vect: Fix check_reduction_path [PR112374]

As mentioned in the PR, the intent of the r14-5076 changes was that
it doesn't count one of the uses on the use_stmt, but what actually
got implemented is that it does this processing on any op_use_stmt,
even if it is not the use_stmt statement, which means that it
can increase count even on debug stmts (-fcompare-debug failures),
or if there would be some other use stmt with 2+ uses it could count
that as a single use.  Though, because it fails whenever cnt != 1
and I believe use_stmt must be one of the uses, it would probably
fail in the latter case anyway.

The following patch fixes that by doing this extra processing only when
op_use_stmt is use_stmt, and using the normal processing otherwise
(so ignore debug stmts, and increase on any uses on the stmt).

2023-11-17  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/112374
	* tree-vect-loop.cc (check_reduction_path): Perform the cond_fn_p
	special case only if op_use_stmt == use_stmt, use as_a rather than
	dyn_cast in that case.

	* gcc.dg/pr112374-1.c: New test.
	* gcc.dg/pr112374-2.c: New test.
	* g++.dg/opt/pr112374.C: New test.

4 files changed, 79 insertions, 2 deletions

diff --git a/gcc/testsuite/g++.dg/opt/pr112374.C b/gcc/testsuite/g++.dg/opt/pr112374.C
new file mode 100644
index 0000000..15d8a06
--- /dev/null
+++ b/gcc/testsuite/g++.dg/opt/pr112374.C
@@ -0,0 +1,24 @@
+// PR tree-optimization/112374
+// { dg-do compile { target c++11 } }
+// { dg-options "-fcompare-debug -gno-statement-frontiers -O2" }
+// { dg-additional-options "-march=skylake-avx512" { target i?86-*-* x86_64-*-* } }
+// { dg-additional-options "-march=armv9-a" { target aarch64*-*-* } }
+
+struct t
+{
+  long coef[1];
+  t(const unsigned long &a) : coef{(long)a} {};
+  t(const t &a);
+};
+extern void gen_int_mode(t, int);
+struct expand_vec_perm_d {
+  unsigned char perm[64];
+  int vmode;
+  unsigned char nelt;
+};
+void expand_vec_perm_blend(struct expand_vec_perm_d *d) {
+  unsigned long mask = 0;
+  for (unsigned i = 0; i < 4; ++i)
+    mask |= (d->perm[i] >= 4 ? 3 : 0) << (i * 2);
+  gen_int_mode(mask, 0);
+}
diff --git a/gcc/testsuite/gcc.dg/pr112374-1.c b/gcc/testsuite/gcc.dg/pr112374-1.c
new file mode 100644
index 0000000..7fbd67a
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr112374-1.c
@@ -0,0 +1,20 @@
+/* PR tree-optimization/112374 */
+/* { dg-do compile } */
+/* { dg-options "-fcompare-debug -gno-statement-frontiers -O2 -w" } */
+/* { dg-additional-options "-march=skylake-avx512" { target i?86-*-* x86_64-*-* } } */
+/* { dg-additional-options "-march=armv9-a" { target aarch64*-*-* } } */
+
+void foo (int, int);
+struct S { char s[4]; };
+int a, b, c;
+
+void
+bar ()
+{
+  struct S d;
+  long e = 0;
+  for (c = 0; c < 4; ++c)
+    e |= (d.s[c] ? 3 : 0) << c;
+  if (e)
+    foo (a, b);
+}
diff --git a/gcc/testsuite/gcc.dg/pr112374-2.c b/gcc/testsuite/gcc.dg/pr112374-2.c
new file mode 100644
index 0000000..1598c49
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr112374-2.c
@@ -0,0 +1,33 @@
+/* PR tree-optimization/112374 */
+/* { dg-do compile } */
+/* { dg-options "-fcompare-debug -gno-statement-frontiers -O2" } */
+/* { dg-additional-options "-march=skylake-avx512" { target i?86-*-* x86_64-*-* } } */
+/* { dg-additional-options "-march=armv9-a" { target aarch64*-*-* } } */
+
+void foo (int, int);
+struct S { char s[64]; } *p;
+char a, b;
+unsigned char c;
+int d, e;
+
+void
+bar (void)
+{
+  unsigned i;
+  long j = 0;
+  for (i = 0; i < b; ++i)
+    j |= (p->s[i] ? 3 : 0) << i;
+  if (p->s[i + 1])
+  lab:
+    for (;;)
+      ;
+  for (i = 0; i < 4; ++i)
+    j |= p->s[i] << i;
+  for (; i; i += 2)
+    if (c + 1 != a)
+      goto lab;
+  for (; i < 8; ++i)
+    j |= p->s[i] >= 6;
+  if (j)
+    foo (d, e);
+}
diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
index 3f59139..e8b8be5 100644
--- a/gcc/tree-vect-loop.cc
+++ b/gcc/tree-vect-loop.cc
@@ -4118,9 +4118,9 @@ pop:
 	/* In case of a COND_OP (mask, op1, op2, op1) reduction we might have
 	   op1 twice (once as definition, once as else) in the same operation.
 	   Allow this.  */
-	  if (cond_fn_p)
+	  if (cond_fn_p && op_use_stmt == use_stmt)
 	    {
-	      gcall *call = dyn_cast<gcall *> (use_stmt);
+	      gcall *call = as_a<gcall *> (use_stmt);
 	      unsigned else_pos
 		= internal_fn_else_index (internal_fn (op.code));