On Tue, 29 Oct 2019, Richard Sandiford wrote:
> Richard Biener <rguent...@suse.de> writes:
> > Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.
> >
> > Richard.
> >
> > 2019-10-21 Richard Biener <rguent...@suse.de>
> >
> > PR tree-optimization/92162
> > * tree-vect-loop.c (vect_create_epilog_for_reduction): Lookup
> > STMT_VINFO_REDUC_IDX in reduc_info.
> > * tree-vect-stmts.c (vectorizable_condition): Likewise.
> >
> > * gcc.dg/pr92162.c: New testcase.
> >
> > Index: gcc/testsuite/gcc.dg/pr92162.c
> > ===================================================================
> > --- gcc/testsuite/gcc.dg/pr92162.c (nonexistent)
> > +++ gcc/testsuite/gcc.dg/pr92162.c (working copy)
> > @@ -0,0 +1,10 @@
> > +/* { dg-do compile } */
> > +/* { dg-options "-Ofast" } */
> > +
> > +short int s8;
> > +
> > +void __attribute__ ((simd))
> > +gn (void)
> > +{
> > + s8 = 0;
> > +}
> > Index: gcc/tree-vect-loop.c
> > ===================================================================
> > --- gcc/tree-vect-loop.c (revision 277242)
> > +++ gcc/tree-vect-loop.c (working copy)
> > @@ -4259,9 +4259,9 @@ vect_create_epilog_for_reduction (stmt_v
> > (CCOMPARE). The then and else values mirror the main VEC_COND_EXPR:
> > the reduction phi corresponds to NEW_PHI_TREE and the new values
> > correspond to INDEX_BEFORE_INCR. */
> > - gcc_assert (STMT_VINFO_REDUC_IDX (stmt_info) >= 1);
> > + gcc_assert (STMT_VINFO_REDUC_IDX (reduc_info) >= 1);
> > tree index_cond_expr;
> > - if (STMT_VINFO_REDUC_IDX (stmt_info) == 2)
> > + if (STMT_VINFO_REDUC_IDX (reduc_info) == 2)
> > index_cond_expr = build3 (VEC_COND_EXPR, cr_index_vector_type,
> > ccompare, indx_before_incr, new_phi_tree);
> > else
> > Index: gcc/tree-vect-stmts.c
> > ===================================================================
> > --- gcc/tree-vect-stmts.c (revision 277242)
> > +++ gcc/tree-vect-stmts.c (working copy)
> > @@ -9818,7 +9818,7 @@ vectorizable_condition (stmt_vec_info st
> > return false;
> > reduc_info = info_for_reduction (stmt_info);
> > reduction_type = STMT_VINFO_REDUC_TYPE (reduc_info);
> > - reduc_index = STMT_VINFO_REDUC_IDX (stmt_info);
> > + reduc_index = STMT_VINFO_REDUC_IDX (reduc_info);
> > gcc_assert (reduction_type != EXTRACT_LAST_REDUCTION
> > || reduc_index != -1);
> > }
>
> After recent changes, do we still want this last bit? reduc_index
> is only used by EXTRACT_LAST_REDUCTION, which wants to know which
> arm of the COND_EXPR is the reduction phi and which is the "new" value.
> AIUI this what the REDUC_IDX on the COND_EXPR now tells us.
>
> Reverting that fixes ICEs in gcc.target/aarch64/sve/clastb*. Tested on
> aarch64-linux-gnu (with and without SVE) and x86_64-linux-gnu.
Hum, I thought I have done this with r277513 already but appearantly
it somehow got lost.
So, OK.
Thanks,
Richard.
> Thanks,
> Richard
>
>
> 2019-10-29 Richard Sandiford <richard.sandif...@arm.com>
>
> gcc/
> * tree-vect-stmts.c (vectorizable_condition): Get the reduction
> index for the COND_EXPR from stmt_info rather than reduc_info.
>
> Index: gcc/tree-vect-stmts.c
> ===================================================================
> --- gcc/tree-vect-stmts.c 2019-10-29 08:29:12.000000000 +0000
> +++ gcc/tree-vect-stmts.c 2019-10-29 10:01:43.180808257 +0000
> @@ -9903,7 +9903,7 @@ vectorizable_condition (stmt_vec_info st
> return false;
> reduc_info = info_for_reduction (stmt_info);
> reduction_type = STMT_VINFO_REDUC_TYPE (reduc_info);
> - reduc_index = STMT_VINFO_REDUC_IDX (reduc_info);
> + reduc_index = STMT_VINFO_REDUC_IDX (stmt_info);
> gcc_assert (reduction_type != EXTRACT_LAST_REDUCTION
> || reduc_index != -1);
> }
