On Mon, 14 Jul 2025, Richard Sandiford wrote:
> Richard Biener <rguent...@suse.de> writes:
> > For loop masking we need to mask a mask AND operation with the loop
> > mask. The following makes sure we have a corresponding mask
> > available. There's no good way to distinguish loop masking from
> > len masking here, so assume we have recorded a mask for the operands
> > mask producers.
> >
> > Bootstrap and regtest running on x86_64-unknown-linux-gnu.
> >
> > PR tree-optimization/121059
> > * tree-vect-stmts.cc (vectorizable_operation): Record a
> > loop mask for mask AND operations.
> >
> > * gcc.dg/vect/pr121059.c: New testcase.
>
> Please could you revert this? It's not the right fix. The point of the
> code is to opportunistically reuse loop masks that are needed by other
> operations. It isn't supposed to record new loop masks itself.
Reverted. I'll try to find some cycles in the next day to test the
alternative.
Richard.
> Like I say, replacing:
>
> if (loop_vinfo->scalar_cond_masked_set.contains ({ op0, 1 }))
>
> with:
>
> if (loop_vinfo->scalar_cond_masked_set.contains ({ op0, vec_num
> }))
>
> fixed the bug for me.
>
> Thanks,
> Richard
>
> > ---
> > gcc/testsuite/gcc.dg/vect/pr121059.c | 24 ++++++++++++++++++++++++
> > gcc/tree-vect-stmts.cc | 10 ++++++++++
> > 2 files changed, 34 insertions(+)
> > create mode 100644 gcc/testsuite/gcc.dg/vect/pr121059.c
> >
> > diff --git a/gcc/testsuite/gcc.dg/vect/pr121059.c
> > b/gcc/testsuite/gcc.dg/vect/pr121059.c
> > new file mode 100644
> > index 00000000000..d7f69b4f1f5
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.dg/vect/pr121059.c
> > @@ -0,0 +1,24 @@
> > +/* { dg-do compile } */
> > +/* { dg-additional-options "-O3 --param vect-partial-vector-usage=1" } */
> > +/* { dg-additional-options "-march=x86-64-v4" { target avx512f } } */
> > +
> > +typedef struct {
> > + long left, right, top, bottom;
> > +} MngBox;
> > +typedef struct {
> > + MngBox object_clip[6];
> > + char exists[256], frozen[];
> > +} MngReadInfo;
> > +MngReadInfo mng_info;
> > +
> > +long ReadMNGImage_i;
> > +
> > +void ReadMNGImage(int ReadMNGImage_i)
> > +{
> > + for (; ReadMNGImage_i < 256; ReadMNGImage_i++)
> > + if (mng_info.exists[ReadMNGImage_i] && mng_info.frozen[ReadMNGImage_i])
> > + mng_info.object_clip[ReadMNGImage_i].left =
> > + mng_info.object_clip[ReadMNGImage_i].right =
> > + mng_info.object_clip[ReadMNGImage_i].top =
> > + mng_info.object_clip[ReadMNGImage_i].bottom = 0;
> > +}
> > diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
> > index 4aa69da2218..f0dc4843ca7 100644
> > --- a/gcc/tree-vect-stmts.cc
> > +++ b/gcc/tree-vect-stmts.cc
> > @@ -6978,6 +6978,16 @@ vectorizable_operation (vec_info *vinfo,
> > LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo) = false;
> > }
> > }
> > + else if (loop_vinfo
> > + && LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo)
> > + && code == BIT_AND_EXPR
> > + && VECTOR_BOOLEAN_TYPE_P (vectype)
> > + /* We cannot always record a mask since that will disable
> > + len-based partial vectors, but there should be already
> > + one mask producer stmt which should require loop
> > + masking. */
> > + && !masks->is_empty ())
> > + vect_record_loop_mask (loop_vinfo, masks, vec_num, vectype, NULL);
> >
> > /* Put types on constant and invariant SLP children. */
> > if (!vect_maybe_update_slp_op_vectype (slp_op0, vectype)
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH, Frankenstrasse 146, 90461 Nuernberg,
Germany; GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman;
HRB 36809 (AG Nuernberg)
