Re: [PATCH 88/89] Use gimple_phi in many more places.

[Jeff Law <law at redhat dot com>]

On 04/21/14 10:57, David Malcolm wrote:
> This patch converts almost every usage of gsi_start_phis to act on a
> gimple_phi_iterator: everywhere (I believe) where the underlying
> statements are then accessed via a gimple_phi_ accessor.
>
> There are some places where the phi nodes then just go through
> generic gimple APIs; in these cases I didn't bother converting
> the iterators.
>
> gcc/
> 	* gimple-ssa-strength-reduction.c (slsr_process_phi): Require a
> 	gimple_phi.
> 	* ipa-inline-analysis.c (predicate_for_phi_result): Likewise.
> 	* tree-emutls.c (lower_emutls_phi_arg): Likewise.
> 	* tree-if-conv.c (predicate_scalar_phi): Likewise.
> 	* tree-into-ssa.c (mark_virtual_phi_result_for_renaming): Likewise.
> 	* tree-into-ssa.h (mark_virtual_phi_result_for_renaming): Likewise.
> 	* tree-phinodes.c (degenerate_phi_result): Likewise.
> 	* tree-phinodes.h (degenerate_phi_result): Likewise.
> 	* tree-ssa.c (verify_use): Likewise.
> 	* tree-switch-conversion.c (array_value_type): Likewise.
>
> 	* graphite-scop-detection.c (same_close_phi_node): Require a pair
> 	of gimple_phi.
> 	(remove_duplicate_close_phi): Require a gimple_phi and a
> 	gimple_phi_iterator.
> 	(make_close_phi_nodes_unique): Convert a local into a
> 	gimple_phi_iterator.
>
> 	* gimple-pretty-print.c (gimple_dump_bb_for_graph): Split iterator
> 	into gimple_stmt_iterator and gimple_phi_iterator, converting local
> 	from gimple to gimple_phi.
> 	* gimple-ssa-strength-reduction.c
> 	(find_candidates_dom_walker::before_dom_children): Likewise.
> 	* ipa-inline-analysis.c (estimate_function_body_sizes): Likewise.
> 	* ipa-split.c (verify_non_ssa_vars): Likewise.
> 	(visit_bb): Likewise.
> 	(split_function): Likewise.
> 	* lto-streamer-out.c (output_function): Likewise.
> 	* sese.c (sese_build_liveouts_bb): Likewise.
> 	* tree-cfg.c (gimple_can_merge_blocks_p): Likewise.
> 	* tree-complex.c (init_dont_simulate_again): Likewise.
> 	* tree-dfa.c (collect_dfa_stats): Likewise.
> 	* tree-eh.c (sink_clobbers): Likewise.
> 	* tree-emutls.c (lower_emutls_function_body): Likewise.
> 	* tree-into-ssa.c (rewrite_dom_walker::before_dom_children):
> 	Likewise.
> 	(rewrite_update_dom_walker::before_dom_children): Likewise.
> 	(prepare_block_for_update): Likewise.
> 	* tree-loop-distribution.c (stmts_from_loop): Likewise.
> 	(generate_loops_for_partition): Likewise.
> 	(destroy_loop): Likewise.
> 	(tree_loop_distribution): Likewise.
> 	* tree-ssa-coalesce.c (build_ssa_conflict_graph): Likewise.
> 	* tree-ssa-copy.c (init_copy_prop): Likewise.
> 	* tree-ssa-copyrename.c (rename_ssa_copies): Likewise.
> 	* tree-ssa-loop-ivcanon.c (propagate_constants_for_unrolling): Likewise.
> 	* tree-ssa-loop-manip.c (find_uses_to_rename_bb): Likewise.
> 	(verify_loop_closed_ssa): Likewise.
> 	* tree-ssa-math-opts.c (execute_cse_reciprocals): Likewise.
> 	* tree-ssa-pre.c (compute_avail): Likewise.
> 	(eliminate_dom_walker::before_dom_children): Likewise.
> 	* tree-ssa-strlen.c (strlen_dom_walker::before_dom_children): Likewise.
> 	* tree-ssa-structalias.c (compute_points_to_sets): Likewise.
> 	(ipa_pta_execute): Likewise.
> 	* tree-ssa-tail-merge.c (same_succ_hash): Likewise.
> 	(release_last_vdef): Likewise.
> 	* tree-ssa.c (verify_ssa): Likewise.
> 	(execute_update_addresses_taken): Likewise.
> 	* tree-stdarg.c (check_all_va_list_escapes): Likewise.
> 	(execute_optimize_stdarg): Likewise.
> 	* tree-switch-conversion.c (build_arrays): Likewise.
> 	* tree-vect-loop-manip.c (rename_variables_in_bb): Likewise.
> 	* tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
> 	(vect_analyze_loop_operations): Likewise.
> 	(vect_transform_loop): Likewise.
> 	* tree-vrp.c (find_assert_locations_1): Likewise.
> 	(vrp_initialize): Likewise.
>
> 	* graphite-sese-to-poly.c (gsi_for_phi_node): Convert psi to a
> 	gimple_phi_iterator.
> 	(rewrite_degenerate_phi): Require a gimple_phi_iterator; strengthen
> 	local "phi" from gimple to gimple-phi.
>
> 	* ipa-split.c (consider_split): Convert local psi to a
> 	gimple_phi_iterator.
> 	* tree-cfg.c (gimple_merge_blocks): Likewise.
>
> 	* tree-eh.c (unsplit_eh): Replace reuse of gsi with a new
> 	gimple_phi_iterator gpi, using it to convert statement from gimple
> 	to a gimple_phi.
> 	(cleanup_empty_eh_merge_phis): Strengthen ophi from gimple to
> 	gimple_phi.
> 	* tree-ssa-dce.c (eliminate_unnecessary_stmts): Replace reuse of
> 	gsi with a new gimple_phi_iterator gpi, using it to convert
> 	expressions from gimple to gimple_phi; add a checked cast to
> 	gimple_phi.
>
> 	* tree-if-conv.c (predicate_all_scalar_phis): Convert local phi_gsi
> 	to be a gimple_phi_iterator; convert "phi" to be a gimple_phi.
> 	* tree-parloops.c (loop_has_vector_phi_nodes): Convert local to be a
> 	gimple_phi_iterator.
> 	* tree-ssa-ccp.c (ccp_initialize): Likewise.
>
> 	* tree-scalar-evolution.c (analyze_initial_condition): Add checked cast
> 	to a gimple_phi.
> 	* tree-ssa.c (insert_debug_temp_for_var_def): Likewise.
> 	* tree-ssa-dom.c (get_rhs_or_phi_arg): Likewise.
Fine once prereqs go in.

This should be the last patch, if I missed anything let me know.

Given that we're waiting until after 4.9.1 before installing this set, I 
expect there will be minor tweaks just due to ongoing development. 
Similarly there may be additions.  As long as they stay in the same 
general form as this set of patches, they're pre-approved.  In all it's 
mostly mechanical/repetitive stuff.

From a reviewer standpoint, the one thing worth mentioning is that 
there were probably more variable renames than were really necessary and 
that introduces a fair amount of clutter when reviewing the patches. 
However, I also know why you made those changes and I can't fault you 
for carrying more information in the variable name to ease long term 
maintenance at the cost of initial reviewer time.

I'm going to rest my eyes now.

Jeff