We already take care to not apply loop splitting to IL produced by splitting so we should be able to delay updating SSA and loop-closed SSA that was left broken after loop versioning until after we processed all opportunities.
Bootstrapped and tested on x86_64-unknown-linux-gnu, I've also built SPEC CPU 2017 with and without LTO for extra verification and added a -fopt-info-loop message to one path where it was missing to verify we are actually splitting loops in SPEC (we do). Pushed. 2021-05-06 Richard Biener <rguent...@suse.de> * tree-ssa-loop-split.c (split_loop): Delay updating SSA form. Output an opt-info message. (do_split_loop_on_cond): Likewise. (tree_ssa_split_loops): Update SSA form here. --- gcc/tree-ssa-loop-split.c | 17 +++++++---------- 1 file changed, 7 insertions(+), 10 deletions(-) diff --git a/gcc/tree-ssa-loop-split.c b/gcc/tree-ssa-loop-split.c index b80b6a75e62..3a09bbc39e5 100644 --- a/gcc/tree-ssa-loop-split.c +++ b/gcc/tree-ssa-loop-split.c @@ -589,7 +589,6 @@ split_loop (class loop *loop1) profile_probability::always (), true); gcc_assert (loop2); - update_ssa (TODO_update_ssa); edge new_e = connect_loops (loop1, loop2); connect_loop_phis (loop1, loop2, new_e); @@ -621,14 +620,13 @@ split_loop (class loop *loop1) free_original_copy_tables (); - /* We destroyed LCSSA form above. Eventually we might be able - to fix it on the fly, for now simply punt and use the helper. */ - rewrite_into_loop_closed_ssa_1 (NULL, 0, SSA_OP_USE, loop1); - changed = true; if (dump_file && (dump_flags & TDF_DETAILS)) fprintf (dump_file, ";; Loop split.\n"); + if (dump_enabled_p ()) + dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, guard_stmt, "loop split\n"); + /* Only deal with the first opportunity. */ break; } @@ -1532,8 +1530,6 @@ do_split_loop_on_cond (struct loop *loop1, edge invar_branch) to_loop1->flags |= true_invar ? EDGE_FALSE_VALUE : EDGE_TRUE_VALUE; to_loop2->flags |= true_invar ? EDGE_TRUE_VALUE : EDGE_FALSE_VALUE; - update_ssa (TODO_update_ssa); - /* Due to introduction of a control flow edge from loop1 latch to loop2 pre-header, we should update PHIs in loop2 to reflect this connection between loop1 and loop2. */ @@ -1541,8 +1537,6 @@ do_split_loop_on_cond (struct loop *loop1, edge invar_branch) free_original_copy_tables (); - rewrite_into_loop_closed_ssa_1 (NULL, 0, SSA_OP_USE, loop1); - return true; } @@ -1644,7 +1638,10 @@ tree_ssa_split_loops (void) free_dominance_info (CDI_POST_DOMINATORS); if (changed) - return TODO_cleanup_cfg; + { + rewrite_into_loop_closed_ssa (NULL, TODO_update_ssa); + return TODO_cleanup_cfg; + } return 0; } -- 2.26.2