[Bug middle-end/104026] [12 Regression] ICE in wide_int_to_tree_1, at tree.c:1755 via tree-vect-loop-manip.c:673

[rdapp at linux dot ibm.com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104026

--- Comment #8 from rdapp at linux dot ibm.com ---
I think you're right.  In one of the last iterations of the patch I moved

+  LOOP_VINFO_PARTIAL_LOAD_STORE_BIAS (loop_vinfo) = partial_load_bias;

after the unsupported check.  It is now only set to something meaningful if it
is not unsupported.

The following should help:

diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
index 8748b1a5593..3c87920090b 100644
--- a/gcc/tree-vect-loop.c
+++ b/gcc/tree-vect-loop.c
@@ -1170,6 +1170,9 @@ vect_verify_loop_lens (loop_vec_info loop_vinfo)
   machine_mode len_store_mode = get_len_load_store_mode
     (loop_vinfo->vector_mode, false).require ();

+  LOOP_VINFO_PARTIAL_LOAD_STORE_BIAS (loop_vinfo) =
+    VECT_PARTIAL_BIAS_UNSUPPORTED;
+
   signed char partial_load_bias = internal_len_load_store_bias
     (IFN_LEN_LOAD, len_load_mode);