From: Richard Biener <rguenther@suse.de>
Date: Wed, 26 Nov 2025 11:49:58 +0000 (+0100)
Subject: Fix loop masked inbranch SIMD clone codegen for AVX512
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=6828f063b59ddae8eef63529ae27224bdfbf1cbc;p=thirdparty%2Fgcc.git

Fix loop masked inbranch SIMD clone codegen for AVX512

The following applies the earlier fix for conditional SIMD clone
calls to the loop masked case.

	* tree-vect-stmts.cc (vectorizable_simd_clone_call): Handle
	AVX512 masking for loop masked SIMD clone call.
---

diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 264475ff091..476a6e570e8 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -4820,9 +4820,8 @@ vectorizable_simd_clone_call (vec_info *vinfo, stmt_vec_info stmt_info,
 
 	  tree masktype = bestn->simdclone->args[mask_i].vector_type;
 	  if (SCALAR_INT_MODE_P (bestn->simdclone->mask_mode))
-	    /* Guess the number of lanes represented by masktype.  */
 	    callee_nelements = exact_div (bestn->simdclone->simdlen,
-					  bestn->simdclone->nargs - nargs);
+					  bestn->simdclone->args[i].linear_step);
 	  else
 	    callee_nelements = TYPE_VECTOR_SUBPARTS (masktype);
 	  o = vector_unroll_factor (nunits, callee_nelements);
@@ -4832,7 +4831,7 @@ 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, masktype, j);
+					     ncopies_in, vectype, j);
 		}
 	      else
 		mask = vect_build_all_ones_mask (vinfo, stmt_info, masktype);