[Bug tree-optimization/121059] [15/16 regression] ICE when building imagemagick-7.1.1-47 (vect_get_loop_mask, at tree-vect-loop.cc:10960) since r15-5383-g8f68d9cb7897df
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121059 --- Comment #16 from GCC Commits --- The master branch has been updated by Richard Biener : https://gcc.gnu.org/g:71be87055548cf942c7bc56d10ffd479db8569e4 commit r16-2259-g71be87055548cf942c7bc56d10ffd479db8569e4 Author: Richard Biener Date: Mon Jul 14 14:09:28 2025 +0200 tree-optimization/121059 - fixup loop mask query When we opportunistically mask an operand of a AND with an already available loop mask we need to query that set with the correct number of masks we expect. PR tree-optimization/121059 * tree-vect-stmts.cc (vectorizable_operation): Query scalar_cond_masked_set with the correct number of masks. * gcc.dg/vect/pr121059.c: New testcase. Co-Authored-By: Richard Sandiford
[Bug tree-optimization/121059] [15/16 regression] ICE when building imagemagick-7.1.1-47 (vect_get_loop_mask, at tree-vect-loop.cc:10960) since r15-5383-g8f68d9cb7897df
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121059 Richard Biener changed: What|Removed |Added Summary|[15 regression] ICE when|[15/16 regression] ICE when |building|building |imagemagick-7.1.1-47|imagemagick-7.1.1-47 |(vect_get_loop_mask, at |(vect_get_loop_mask, at |tree-vect-loop.cc:10960)|tree-vect-loop.cc:10960) |since |since |r15-5383-g8f68d9cb7897df|r15-5383-g8f68d9cb7897df Known to work|16.0| --- Comment #15 from Richard Biener --- Fix was reverted.
[Bug tree-optimization/121059] [15/16 regression] ICE when building imagemagick-7.1.1-47 (vect_get_loop_mask, at tree-vect-loop.cc:10960) since r15-5383-g8f68d9cb7897df
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121059 --- Comment #12 from GCC Commits --- The master branch has been updated by Richard Biener : https://gcc.gnu.org/g:66346b6d800fc4baae876e0fe4e932401bcc85fa commit r16-2236-g66346b6d800fc4baae876e0fe4e932401bcc85fa Author: Richard Biener Date: Mon Jul 14 14:09:28 2025 +0200 tree-optimization/121059 - record loop mask when required 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. 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.
[Bug tree-optimization/121059] [15/16 regression] ICE when building imagemagick-7.1.1-47 (vect_get_loop_mask, at tree-vect-loop.cc:10960) since r15-5383-g8f68d9cb7897df
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121059
--- Comment #11 from Richard Sandiford ---
> if (masked_loop_p
> && code == BIT_AND_EXPR
> && VECTOR_BOOLEAN_TYPE_P (vectype))
> {
> if (loop_vinfo->scalar_cond_masked_set.contains ({ op0, 1 }))
> {
> mask = vect_get_loop_mask (loop_vinfo, gsi, masks,
> vec_num, vectype, i);
I think the bug is the hard-coded 1 for SLP. It should be vec_num instead, to
match the vect_get_loop_len argument. The ICE occurs in a 4-vector operation
rather than a 1-vector operation.
[Bug tree-optimization/121059] [15/16 regression] ICE when building imagemagick-7.1.1-47 (vect_get_loop_mask, at tree-vect-loop.cc:10960) since r15-5383-g8f68d9cb7897df
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121059
--- Comment #10 from Richard Sandiford ---
(In reply to Richard Biener from comment #9)
> vectorizable_operation during transform does
>
> /* When combining two masks check if either of them is elsewhere
> combined with a loop mask, if that's the case we can mark that
> the
> new combined mask doesn't need to be combined with a loop mask.
> */
> if (masked_loop_p
> && code == BIT_AND_EXPR
> && VECTOR_BOOLEAN_TYPE_P (vectype))
> {
> if (loop_vinfo->scalar_cond_masked_set.contains ({ op0, 1 }))
> {
> mask = vect_get_loop_mask (loop_vinfo, gsi, masks,
> vec_num, vectype, i);
>
> but that's not reflected by analysis, which misses to record a loop mask
> for !mask_out_inactive operations. So the fix is as simple as the following,
> but this might put us to using masks? There is no good way to do this
> I guess. The scalar_cond_masked_set optimization does not have a
> corresponding
> len operation. I'm not sure what we can do here?
>
> diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
> index 4aa69da2218..55002bd0cc2 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))
> + 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)
Yeah, we shouldn't do that. The question is why op0 is in
scalar_cond_masked_set with masked_loop_p true if there's no associated loop
mask.
[Bug tree-optimization/121059] [15/16 regression] ICE when building imagemagick-7.1.1-47 (vect_get_loop_mask, at tree-vect-loop.cc:10960) since r15-5383-g8f68d9cb7897df
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121059
Richard Biener changed:
What|Removed |Added
CC||rsandifo at gcc dot gnu.org
--- Comment #9 from Richard Biener ---
vectorizable_operation during transform does
/* When combining two masks check if either of them is elsewhere
combined with a loop mask, if that's the case we can mark that the
new combined mask doesn't need to be combined with a loop mask.
*/
if (masked_loop_p
&& code == BIT_AND_EXPR
&& VECTOR_BOOLEAN_TYPE_P (vectype))
{
if (loop_vinfo->scalar_cond_masked_set.contains ({ op0, 1 }))
{
mask = vect_get_loop_mask (loop_vinfo, gsi, masks,
vec_num, vectype, i);
but that's not reflected by analysis, which misses to record a loop mask
for !mask_out_inactive operations. So the fix is as simple as the following,
but this might put us to using masks? There is no good way to do this
I guess. The scalar_cond_masked_set optimization does not have a corresponding
len operation. I'm not sure what we can do here?
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 4aa69da2218..55002bd0cc2 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))
+ 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)
[Bug tree-optimization/121059] [15/16 regression] ICE when building imagemagick-7.1.1-47 (vect_get_loop_mask, at tree-vect-loop.cc:10960) since r15-5383-g8f68d9cb7897df
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121059 Sam James changed: What|Removed |Added CC||rdapp at gcc dot gnu.org Summary|[15/16 regression] ICE when |[15/16 regression] ICE when |building|building |imagemagick-7.1.1-47|imagemagick-7.1.1-47 |(vect_get_loop_mask, at |(vect_get_loop_mask, at |tree-vect-loop.cc:10960)|tree-vect-loop.cc:10960) ||since ||r15-5383-g8f68d9cb7897df Keywords|needs-bisection | --- Comment #8 from Sam James --- r15-5383-g8f68d9cb7897df
[Bug tree-optimization/121059] [15/16 regression] ICE when building imagemagick-7.1.1-47 (vect_get_loop_mask, at tree-vect-loop.cc:10960)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121059 Richard Biener changed: What|Removed |Added Status|NEW |ASSIGNED Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot gnu.org Target Milestone|16.0|15.2 Priority|P3 |P2 --- Comment #7 from Richard Biener --- I will have a look somewhen later this week.
[Bug tree-optimization/121059] [15/16 regression] ICE when building imagemagick-7.1.1-47 (vect_get_loop_mask, at tree-vect-loop.cc:10960)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121059 --- Comment #6 from rguenther at suse dot de --- On Mon, 14 Jul 2025, pinskia at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121059 > > Andrew Pinski changed: > >What|Removed |Added > > Ever confirmed|0 |1 > Known to work||14.3.0 > Status|UNCONFIRMED |NEW > Known to fail||15.1.0 >Last reconfirmed||2025-07-14 >Keywords||needs-bisection > > --- Comment #5 from Andrew Pinski --- > (In reply to Sam James from comment #4) > > (In reply to Sam James from comment #2) > > > Created attachment 61857 [details] > > > reduced.i > > > > > > bisecting > > > > r16-2088-ge9079e4f43d135 but given my luck with going back further in the > > other bug, I won't bother here > > `-O3 -march=x86-64-v4 --param vect-partial-vector-usage=1` ICEs for GCC 15.1.0 > also. But works in GCC 14.3.0. It's good we get some more coverage for masked epilogue vect now, but yeah - this was expected to some extent ...
