https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121844
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hmm, it isn't enough to re-order IV creation - we are refering to IP_NORMAL pos
during use rewriting as well:
#1 0x0000000002013101 in stmt_after_ip_normal_pos (loop=0x7ffff6e4a200,
stmt=<gimple_assign 0x7ffff6e42318>)
at /space/rguenther/src/gcc-autopar_devel/gcc/tree-ssa-loop-ivopts.cc:933
#2 0x000000000201325d in stmt_after_increment (loop=0x7ffff6e4a200,
cand=0x618be10, stmt=<gimple_assign 0x7ffff6e42318>)
at /space/rguenther/src/gcc-autopar_devel/gcc/tree-ssa-loop-ivopts.cc:978
#3 0x000000000201c59d in var_at_stmt (loop=0x7ffff6e4a200, cand=0x618be10,
stmt=<gimple_assign 0x7ffff6e42318>)
at /space/rguenther/src/gcc-autopar_devel/gcc/tree-ssa-loop-ivopts.cc:3910
#4 0x000000000201c9f4 in get_computation_aff_1 (data=0x7fffffffd6b0,
at=<gimple_assign 0x7ffff6e42318>, use=0x61829f0,
cand=0x618be10, aff_inv=0x7fffffffd2f0, aff_var=0x7fffffffd0e0, prat=0x0)
at /space/rguenther/src/gcc-autopar_devel/gcc/tree-ssa-loop-ivopts.cc:3975
#5 0x000000000201d033 in get_computation_aff (data=0x7fffffffd6b0,
at=<gimple_assign 0x7ffff6e42318>, use=0x61829f0,
cand=0x618be10, aff=0x7fffffffd2f0) at
/space/rguenther/src/gcc-autopar_devel/gcc/tree-ssa-loop-ivopts.cc:4072
#6 0x000000000202909a in rewrite_use_address (data=0x7fffffffd6b0,
use=0x61829f0, cand=0x618be10)
at /space/rguenther/src/gcc-autopar_devel/gcc/tree-ssa-loop-ivopts.cc:7591
#7 0x00000000020299dc in rewrite_groups (data=0x7fffffffd6b0)
at /space/rguenther/src/gcc-autopar_devel/gcc/tree-ssa-loop-ivopts.cc:7728
#8 0x000000000202b4ff in tree_ssa_iv_optimize_loop (data=0x7fffffffd6b0,
loop=0x7ffff6e4a200, toremove=0x7fffffffd690)
at /space/rguenther/src/gcc-autopar_devel/gcc/tree-ssa-loop-ivopts.cc:8219
I'm testing an alternative.
