diff options
| author | Richard Biener <rguenther@suse.de> | 2025-09-12 14:15:59 +0200 |
|---|---|---|
| committer | Richard Biener <rguenther@suse.de> | 2025-09-14 11:05:42 +0200 |
| commit | 57c97de978332fa5683393b38a2db77b78bf0e2c (patch) | |
| tree | 3385622ea9b756307e31dc9e1bda5bc1bec66325 | |
| parent | 80f9440b5ecb1c3ab943b5c862f84e06fbeabd89 (diff) | |
| download | gcc-57c97de978332fa5683393b38a2db77b78bf0e2c.zip gcc-57c97de978332fa5683393b38a2db77b78bf0e2c.tar.gz gcc-57c97de978332fa5683393b38a2db77b78bf0e2c.tar.bz2 | |
Avoid VMAT_ELEMENTWISE for negative stride SLP
The following makes us always use VMAT_STRIDED_SLP for negative
stride multi-element accesses. That handles falling back to
single element accesses transparently.
* tree-vect-stmts.cc (get_load_store_type): Use VMAT_STRIDED_SLP
for negative stride accesses when VMAT_CONTIGUOUS_REVERSE
isn't applicable.
| -rw-r--r-- | gcc/tree-vect-stmts.cc | 14 |
1 files changed, 5 insertions, 9 deletions
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc index f78acaf..852bc40 100644 --- a/gcc/tree-vect-stmts.cc +++ b/gcc/tree-vect-stmts.cc @@ -2026,6 +2026,7 @@ get_load_store_type (vec_info *vinfo, stmt_vec_info stmt_info, without permutation. */ if (! SLP_TREE_LOAD_PERMUTATION (slp_node).exists ()) first_dr_info = STMT_VINFO_DR_INFO (SLP_TREE_SCALAR_STMTS (slp_node)[0]); + if (STMT_VINFO_STRIDED_P (first_stmt_info)) /* Try to use consecutive accesses of as many elements as possible, separated by the stride, until we have a complete vector. @@ -2089,15 +2090,10 @@ get_load_store_type (vec_info *vinfo, stmt_vec_info stmt_info, (vinfo, stmt_info, vectype, vls_type, 1, &neg_ldst_offset); else - { - /* Try to use consecutive accesses of DR_GROUP_SIZE elements, - separated by the stride, until we have a complete vector. - Fall back to scalar accesses if that isn't possible. */ - if (multiple_p (nunits, group_size)) - *memory_access_type = VMAT_STRIDED_SLP; - else - *memory_access_type = VMAT_ELEMENTWISE; - } + /* We can fall back to VMAT_STRIDED_SLP since that does + not care whether the stride between the group instances + is positive or negative. */ + *memory_access_type = VMAT_STRIDED_SLP; } else if (cmp == 0 && loop_vinfo) { |