[gcc(refs/users/rguenth/heads/vect-force-slp)] Avoid bogus SLP outer loop vectorization

2024-05-13 Thread Richard Biener via Gcc-cvs
https://gcc.gnu.org/g:4652b8bdf79f6ba3a86e085b7ce13d23057c57f6

commit 4652b8bdf79f6ba3a86e085b7ce13d23057c57f6
Author: Richard Biener 
Date:   Wed Mar 6 15:13:05 2024 +0100

Avoid bogus SLP outer loop vectorization

This fixes the check for multiple types which go wrong I think
because of bogus pointer IV increments when there are multiple
copies of vector stmts in the inner loop.

* tree-vect-stmts.cc (vectorizable_load): Avoid outer loop
SLP vectorization with multi-copy vector stmts in the inner
loop.
(vectorizable_store): Likewise.

Diff:
---
 gcc/tree-vect-stmts.cc | 7 +--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 414c1fce38db..840ff8a3406a 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -8195,7 +8195,9 @@ vectorizable_store (vec_info *vinfo,
   gcc_assert (ncopies >= 1);
 
   /* FORNOW.  This restriction should be relaxed.  */
-  if (loop && nested_in_vect_loop_p (loop, stmt_info) && ncopies > 1)
+  if (loop
+  && nested_in_vect_loop_p (loop, stmt_info)
+  && (ncopies > 1 || (slp && SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node) > 1)))
 {
   if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
@@ -9941,7 +9943,8 @@ vectorizable_load (vec_info *vinfo,
   gcc_assert (ncopies >= 1);
 
   /* FORNOW. This restriction should be relaxed.  */
-  if (nested_in_vect_loop && ncopies > 1)
+  if (nested_in_vect_loop
+  && (ncopies > 1 || (slp && SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node) > 1)))
 {
   if (dump_enabled_p ())
 dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,


[gcc(refs/users/rguenth/heads/vect-force-slp)] Avoid bogus SLP outer loop vectorization

2024-03-06 Thread Richard Biener via Gcc-cvs
https://gcc.gnu.org/g:d1b89155e4bc4f42a06a36b64208216e5d37e779

commit d1b89155e4bc4f42a06a36b64208216e5d37e779
Author: Richard Biener 
Date:   Wed Mar 6 15:13:05 2024 +0100

Avoid bogus SLP outer loop vectorization

This fixes the check for multiple types which go wrong I think
because of bogus pointer IV increments when there are multiple
copies of vector stmts in the inner loop.

* tree-vect-stmts.cc (vectorizable_load): Avoid outer loop
SLP vectorization with multi-copy vector stmts in the inner
loop.
(vectorizable_store): Likewise.

Diff:
---
 gcc/tree-vect-stmts.cc | 7 +--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 8766ef220c2..2bee550f4a9 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -8176,7 +8176,9 @@ vectorizable_store (vec_info *vinfo,
   gcc_assert (ncopies >= 1);
 
   /* FORNOW.  This restriction should be relaxed.  */
-  if (loop && nested_in_vect_loop_p (loop, stmt_info) && ncopies > 1)
+  if (loop
+  && nested_in_vect_loop_p (loop, stmt_info)
+  && (ncopies > 1 || (slp && SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node) > 1)))
 {
   if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
@@ -9917,7 +9919,8 @@ vectorizable_load (vec_info *vinfo,
   gcc_assert (ncopies >= 1);
 
   /* FORNOW. This restriction should be relaxed.  */
-  if (nested_in_vect_loop && ncopies > 1)
+  if (nested_in_vect_loop
+  && (ncopies > 1 || (slp && SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node) > 1)))
 {
   if (dump_enabled_p ())
 dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,