diff options
| author | Alex Coplan <alex.coplan@arm.com> | 2024-03-11 13:09:10 +0000 |
|---|---|---|
| committer | Tamar Christina <tamar.christina@arm.com> | 2025-01-10 21:21:06 +0000 |
| commit | 68326d5d1a593dc0bf098c03aac25916168bc5a9 (patch) | |
| tree | f90c4f27b72fe72aecc792ff61ef792a48012ae6 /gcc/tree-vect-loop-manip.cc | |
| parent | ddcfae1d1dfe5875875c9897f0dda14e342b2534 (diff) | |
| download | gcc-68326d5d1a593dc0bf098c03aac25916168bc5a9.zip gcc-68326d5d1a593dc0bf098c03aac25916168bc5a9.tar.gz gcc-68326d5d1a593dc0bf098c03aac25916168bc5a9.tar.bz2 | |
vect: Force alignment peeling to vectorize more early break loops [PR118211]
This allows us to vectorize more loops with early exits by forcing
peeling for alignment to make sure that we're guaranteed to be able to
safely read an entire vector iteration without crossing a page boundary.
To make this work for VLA architectures we have to allow compile-time
non-constant target alignments. We also have to override the result of
the target's preferred_vector_alignment hook if it isn't a power-of-two
multiple of the TYPE_SIZE of the chosen vector type.
gcc/ChangeLog:
PR tree-optimization/118211
PR tree-optimization/116126
* tree-vect-data-refs.cc (vect_analyze_early_break_dependences):
Set need_peeling_for_alignment flag on read DRs instead of
failing vectorization. Punt on gathers.
(dr_misalignment): Handle non-constant target alignments.
(vect_compute_data_ref_alignment): If need_peeling_for_alignment
flag is set on the DR, then override the target alignment chosen
by the preferred_vector_alignment hook to choose a safe
alignment.
(vect_supportable_dr_alignment): Override
support_vector_misalignment hook if need_peeling_for_alignment
is set on the DR: in this case we must return
dr_unaligned_unsupported in order to force peeling.
* tree-vect-loop-manip.cc (vect_do_peeling): Allow prolog
peeling by a compile-time non-constant amount.
* tree-vectorizer.h (dr_vec_info): Add new flag
need_peeling_for_alignment.
gcc/testsuite/ChangeLog:
PR tree-optimization/118211
PR tree-optimization/116126
* gcc.dg/tree-ssa/cunroll-13.c: Don't vectorize.
* gcc.dg/tree-ssa/cunroll-14.c: Likewise.
* gcc.dg/unroll-6.c: Likewise.
* gcc.dg/tree-ssa/gen-vect-28.c: Likewise.
* gcc.dg/vect/vect-104.c: Expect to vectorize.
* gcc.dg/vect/vect-early-break_108-pr113588.c: Likewise.
* gcc.dg/vect/vect-early-break_109-pr113588.c: Likewise.
* gcc.dg/vect/vect-early-break_110-pr113467.c: Likewise.
* gcc.dg/vect/vect-early-break_3.c: Likewise.
* gcc.dg/vect/vect-early-break_65.c: Likewise.
* gcc.dg/vect/vect-early-break_8.c: Likewise.
* gfortran.dg/vect/vect-5.f90: Likewise.
* gfortran.dg/vect/vect-8.f90: Likewise.
* gcc.dg/vect/vect-switch-search-line-fast.c:
Co-Authored-By: Tamar Christina <tamar.christina@arm.com>
Diffstat (limited to 'gcc/tree-vect-loop-manip.cc')
| -rw-r--r-- | gcc/tree-vect-loop-manip.cc | 6 |
1 files changed, 0 insertions, 6 deletions
diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc index 5d1b70a..15cac0f 100644 --- a/gcc/tree-vect-loop-manip.cc +++ b/gcc/tree-vect-loop-manip.cc @@ -3128,12 +3128,6 @@ vect_do_peeling (loop_vec_info loop_vinfo, tree niters, tree nitersm1, int estimated_vf; int prolog_peeling = 0; bool vect_epilogues = loop_vinfo->epilogue_vinfo != NULL; - /* We currently do not support prolog peeling if the target alignment is not - known at compile time. 'vect_gen_prolog_loop_niters' depends on the - target alignment being constant. */ - dr_vec_info *dr_info = LOOP_VINFO_UNALIGNED_DR (loop_vinfo); - if (dr_info && !DR_TARGET_ALIGNMENT (dr_info).is_constant ()) - return NULL; if (!vect_use_loop_mask_for_alignment_p (loop_vinfo)) prolog_peeling = LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo); |