| Message ID | patch-17865-tamar@arm.com |
|---|---|
| State | New |
| Headers | show |
| Series | middle-end: don't pass loop_vinfo to vect_set_loop_condition during prolog peeling [PR111866] | expand |
On Fri, 20 Oct 2023, Tamar Christina wrote: > Hi All, > > During the refactoring I had passed loop_vinfo on to vect_set_loop_condition > during prolog peeling. This parameter is unused in most cases except for in > vect_set_loop_condition_partial_vectors where it's behaviour depends on whether > loop_vinfo is NULL or not. Apparently this code expect it to be NULL and it > reads the structures from a different location. > > This fixes the failing testcase which was not using the lens values determined > earlier in vectorizable_store because it was looking it up in the given > loop_vinfo instead. > > Bootstrapped Regtested on aarch64-none-linux-gnu, powerpc64le-unknown-linux-gnu, > x86_64--nonelinux-gnu and no issues. PowerPC test failures are fixed as well. > > Ok for master? Eh, I remember PR110571 seeing this. But OK. Thanks, Richard. > Thanks, > Tamar > > gcc/ChangeLog: > > PR tree-optimization/PR111866 > * tree-vect-loop-manip.cc (vect_do_peeling): Pass null as vinfo to > vect_set_loop_condition during prolog peeling. > > --- inline copy of patch -- > diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc > index db1d4f867ead5c6079cda3ff0d0870234d11e39d..9c2551261e33738feac71721d4f0160033907260 100644 > --- a/gcc/tree-vect-loop-manip.cc > +++ b/gcc/tree-vect-loop-manip.cc > @@ -3235,7 +3235,7 @@ vect_do_peeling (loop_vec_info loop_vinfo, tree niters, tree nitersm1, > > /* Update the number of iterations for prolog loop. */ > tree step_prolog = build_one_cst (TREE_TYPE (niters_prolog)); > - vect_set_loop_condition (prolog, prolog_e, loop_vinfo, niters_prolog, > + vect_set_loop_condition (prolog, prolog_e, NULL, niters_prolog, > step_prolog, NULL_TREE, false); > > /* Skip the prolog loop. */ > > > > >
--- a/gcc/tree-vect-loop-manip.cc +++ b/gcc/tree-vect-loop-manip.cc @@ -3235,7 +3235,7 @@ vect_do_peeling (loop_vec_info loop_vinfo, tree niters, tree nitersm1, /* Update the number of iterations for prolog loop. */ tree step_prolog = build_one_cst (TREE_TYPE (niters_prolog)); - vect_set_loop_condition (prolog, prolog_e, loop_vinfo, niters_prolog, + vect_set_loop_condition (prolog, prolog_e, NULL, niters_prolog, step_prolog, NULL_TREE, false); /* Skip the prolog loop. */
Hi All, During the refactoring I had passed loop_vinfo on to vect_set_loop_condition during prolog peeling. This parameter is unused in most cases except for in vect_set_loop_condition_partial_vectors where it's behaviour depends on whether loop_vinfo is NULL or not. Apparently this code expect it to be NULL and it reads the structures from a different location. This fixes the failing testcase which was not using the lens values determined earlier in vectorizable_store because it was looking it up in the given loop_vinfo instead. Bootstrapped Regtested on aarch64-none-linux-gnu, powerpc64le-unknown-linux-gnu, x86_64--nonelinux-gnu and no issues. PowerPC test failures are fixed as well. Ok for master? Thanks, Tamar gcc/ChangeLog: PR tree-optimization/PR111866 * tree-vect-loop-manip.cc (vect_do_peeling): Pass null as vinfo to vect_set_loop_condition during prolog peeling. --- inline copy of patch -- diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc index db1d4f867ead5c6079cda3ff0d0870234d11e39d..9c2551261e33738feac71721d4f0160033907260 100644 -- diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc index db1d4f867ead5c6079cda3ff0d0870234d11e39d..9c2551261e33738feac71721d4f0160033907260 100644 --- a/gcc/tree-vect-loop-manip.cc +++ b/gcc/tree-vect-loop-manip.cc @@ -3235,7 +3235,7 @@ vect_do_peeling (loop_vec_info loop_vinfo, tree niters, tree nitersm1, /* Update the number of iterations for prolog loop. */ tree step_prolog = build_one_cst (TREE_TYPE (niters_prolog)); - vect_set_loop_condition (prolog, prolog_e, loop_vinfo, niters_prolog, + vect_set_loop_condition (prolog, prolog_e, NULL, niters_prolog, step_prolog, NULL_TREE, false); /* Skip the prolog loop. */