[gcc r16-5649] Fix OMP SIMD clone mask record/get again
Richard Biener
rguenth@gcc.gnu.org
Thu Nov 27 11:50:46 GMT 2025
https://gcc.gnu.org/g:fd743aab3bc8c521e0e8cd1742a71c2d713e2c76
commit r16-5649-gfd743aab3bc8c521e0e8cd1742a71c2d713e2c76
Author: Richard Biener <rguenther@suse.de>
Date: Thu Nov 27 10:56:43 2025 +0100
Fix OMP SIMD clone mask record/get again
Post-CI checkin detected aarch64 fallout for the last change. AARCH64
has ABI twists that run into a case where an unmasked call when loop
masked allows for a mask that has different shape than that of the
return value which in turn has different type than that of an actual
argument.
While we do not support a mismatch of call mask shape with the
OMP SIMD ABI mask shape when there's no call mask we have no such
restriction.
So the following fixes the record/get of a loop mask in the unmasked
call case, also fixing a latent issue present before. In particular
do not record a random scalar operand as representing the mask.
A testcase is in gcc.target/aarch64/vect-simd-clone-4.c.
* tree-vect-stmts.cc (vectorizable_simd_clone_call): Fix
recording of the mask type again. Adjust placing of
mask arguments for non-masked calls.
Diff:
---
gcc/tree-vect-stmts.cc | 78 ++++++++++++++++++++++++++++++++++++--------------
1 file changed, 56 insertions(+), 22 deletions(-)
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 476a6e570e82..de28316ddc66 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -4047,7 +4047,6 @@ vectorizable_simd_clone_call (vec_info *vinfo, stmt_vec_info stmt_info,
{
tree vec_dest;
tree scalar_dest;
- tree op;
tree vec_oprnd0 = NULL_TREE;
tree vectype;
poly_uint64 nunits;
@@ -4121,6 +4120,7 @@ vectorizable_simd_clone_call (vec_info *vinfo, stmt_vec_info stmt_info,
{
simd_call_arg_info thisarginfo;
affine_iv iv;
+ tree op;
thisarginfo.linear_step = 0;
thisarginfo.align = 0;
@@ -4435,9 +4435,39 @@ vectorizable_simd_clone_call (vec_info *vinfo, stmt_vec_info stmt_info,
case SIMD_CLONE_ARG_TYPE_MASK:
if (loop_vinfo
&& LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo))
- vect_record_loop_mask (loop_vinfo,
- &LOOP_VINFO_MASKS (loop_vinfo),
- ncopies_in, vectype, op);
+ {
+ if (masked_call_offset)
+ /* When there is an explicit mask we require the
+ number of elements to match up. */
+ vect_record_loop_mask (loop_vinfo,
+ &LOOP_VINFO_MASKS (loop_vinfo),
+ ncopies_in, vectype, NULL_TREE);
+ else
+ {
+ /* When there is no explicit mask on the call we have
+ more relaxed requirements. */
+ tree masktype;
+ poly_uint64 callee_nelements;
+ if (SCALAR_INT_MODE_P (bestn->simdclone->mask_mode))
+ {
+ callee_nelements
+ = exact_div (bestn->simdclone->simdlen,
+ bestn->simdclone->args[i].linear_step);
+ masktype = get_related_vectype_for_scalar_type
+ (vinfo->vector_mode, TREE_TYPE (vectype),
+ callee_nelements);
+ }
+ else
+ {
+ masktype = bestn->simdclone->args[i].vector_type;
+ callee_nelements = TYPE_VECTOR_SUBPARTS (masktype);
+ }
+ auto o = vector_unroll_factor (nunits, callee_nelements);
+ vect_record_loop_mask (loop_vinfo,
+ &LOOP_VINFO_MASKS (loop_vinfo),
+ ncopies * o, masktype, NULL_TREE);
+ }
+ }
break;
}
}
@@ -4499,7 +4529,7 @@ vectorizable_simd_clone_call (vec_info *vinfo, stmt_vec_info stmt_info,
{
unsigned int k, l, m, o;
tree atype;
- op = gimple_call_arg (stmt, i + masked_call_offset);
+ tree op = gimple_call_arg (stmt, i + masked_call_offset);
switch (bestn->simdclone->args[i].arg_type)
{
case SIMD_CLONE_ARG_TYPE_VECTOR:
@@ -4818,12 +4848,20 @@ vectorizable_simd_clone_call (vec_info *vinfo, stmt_vec_info stmt_info,
gcc_assert (bestn->simdclone->args[mask_i].arg_type ==
SIMD_CLONE_ARG_TYPE_MASK);
- tree masktype = bestn->simdclone->args[mask_i].vector_type;
+ tree mask_argtype = bestn->simdclone->args[mask_i].vector_type;
+ tree mask_vectype;
if (SCALAR_INT_MODE_P (bestn->simdclone->mask_mode))
- callee_nelements = exact_div (bestn->simdclone->simdlen,
- bestn->simdclone->args[i].linear_step);
+ {
+ callee_nelements = exact_div (bestn->simdclone->simdlen,
+ bestn->simdclone->args[i].linear_step);
+ mask_vectype = get_related_vectype_for_scalar_type
+ (vinfo->vector_mode, TREE_TYPE (vectype), callee_nelements);
+ }
else
- callee_nelements = TYPE_VECTOR_SUBPARTS (masktype);
+ {
+ mask_vectype = mask_argtype;
+ callee_nelements = TYPE_VECTOR_SUBPARTS (mask_vectype);
+ }
o = vector_unroll_factor (nunits, callee_nelements);
for (m = j * o; m < (j + 1) * o; m++)
{
@@ -4831,10 +4869,11 @@ vectorizable_simd_clone_call (vec_info *vinfo, stmt_vec_info stmt_info,
{
vec_loop_masks *loop_masks = &LOOP_VINFO_MASKS (loop_vinfo);
mask = vect_get_loop_mask (loop_vinfo, gsi, loop_masks,
- ncopies_in, vectype, j);
+ ncopies * o, mask_vectype, m);
}
else
- mask = vect_build_all_ones_mask (vinfo, stmt_info, masktype);
+ mask = vect_build_all_ones_mask (vinfo, stmt_info,
+ mask_argtype);
gassign *new_stmt;
if (SCALAR_INT_MODE_P (bestn->simdclone->mask_mode))
@@ -4852,23 +4891,18 @@ vectorizable_simd_clone_call (vec_info *vinfo, stmt_vec_info stmt_info,
mask);
gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT);
/* Then zero-extend to the mask mode. */
- mask = fold_build1 (NOP_EXPR, masktype,
+ mask = fold_build1 (NOP_EXPR, mask_argtype,
gimple_get_lhs (new_stmt));
}
else if (bestn->simdclone->mask_mode == VOIDmode)
- {
- tree one = fold_convert (TREE_TYPE (masktype),
- integer_one_node);
- tree zero = fold_convert (TREE_TYPE (masktype),
- integer_zero_node);
- mask = build3 (VEC_COND_EXPR, masktype, mask,
- build_vector_from_val (masktype, one),
- build_vector_from_val (masktype, zero));
- }
+ mask = build3 (VEC_COND_EXPR, mask_argtype, mask,
+ build_one_cst (mask_argtype),
+ build_zero_cst (mask_argtype));
else
gcc_unreachable ();
- new_stmt = gimple_build_assign (make_ssa_name (masktype), mask);
+ new_stmt = gimple_build_assign (make_ssa_name (mask_argtype),
+ mask);
vect_finish_stmt_generation (vinfo, stmt_info,
new_stmt, gsi);
mask = gimple_assign_lhs (new_stmt);
More information about the Gcc-cvs
mailing list