Get rid of NOTE_LINE_NUMBER infrastructure
Richard Guenther
richard.guenther@gmail.com
Thu May 17 10:24:00 GMT 2007
On 5/4/07, Jan Hubicka <jh@suse.cz> wrote:
> Hi,
> this patch cleans up the infrastructure for INSN_NOTEs after removal of line
> number notes. In particular:
> 1) Now misnamed NOTE_LINE_NUMBER is now NOTE_KIND (for symmetry with
> REG_NOTE_KIND)
> 2) 4th field of NOTEs is now unused. We need to keep 5th around for
> code label replacement code and it is not big deal since we anyway
> need all INSN containers to be of same size (at least BARRIER is
> artifically bumped up in size)
> 3) I've dropped squeeze_notes that is moving around block notes and
> was dead for a while, since block notes appears very late
> 4) Code for handling line number notes was dropped from print-rtl and firends.
> 5) SET_INSN_DELETED and NOTE_INSN_BASIC_BLOCK_P is now used more consistently
> replacing the unwound sequence
> 6) I've dropped code from sh.c dealing with LOOP notes since it was dead
> since GCC 4.0.0 and no one complained
> 7) emit_note and friends now take the enum insn_note as argument instead
> of line number
> 8) NOTE_INSN_* enum now starts at 0 instead of -100.
> 9) SET_INSN_DELETED is now used more consistenly (it was still present
> in backend code that would not compile with mapped locations)
>
> It is long and mechanical too ;) I've inspected by hand each occurence
> of NOTE_LINE_NUMBER by hand to see if it is still needed, hope I got it
> right.
>
> Bootstrapped/regtested i686-linux, I've also verified that sh-elf builds
> (since there was quite few changes to sh). I am testing ppc-linux and
> ia64-linux.
>
> :ADDPATCH middle-end:
This is ok for mainline.
Thanks,
Richard.
:REVIEWMAIL:
> Honza
>
> * gengtype.c (adjust_field_rtx_def): Use NOTE_KIND instead of
> NOTE_LINE_NUMBER; use NOTE_INSN_BASIC_BLOCK_P when possible.
> * ddg.c (create_ddg): LIkewise.
> * final.c (final): Remove hunk moving line numbernotes around since
> they are no longer present at this stage.
> (final_scan_insn): Use NOTE_KIND instead of
> NOTE_LINE_NUMBER; use NOTE_INSN_BASIC_BLOCK_P when possible.
> (output_asm_label): Likewise.
> * reorg.c (dbr_schedule): Likewise.
> * haifa-sched.c (unlink_other_notes): Likewise.
> * mode-switching.c (optimize_mode_switching): Likewise.
> * graph.c (start_bb): Likewise.
> * rtl.def (NOTE): Update description.
> * jump.c (squeeze_notes): Delete.
> (mark_jump_label): Use NOTE_KIND instead of NOTE_LINE_NUMBER; use
> NOTE_INSN_BASIC_BLOCK_P when possible.
> * ifcvt.c (dead_or_predicable): Remove call of squeeze_notes.
> * dwarf2out.c (gen_label_die): Use NOTE_KIND instead of
> NOTE_LINE_NUMBER; use NOTE_INSN_BASIC_BLOCK_P when possible.
> (dwarf2out_var_location): Likewise.
> * cfgbuild.c (make_edges): Likewise.
> (find_basic_blocks_1): Likewise.
> * function.c (reorder_blocks_1): Likewise.
> (epilogue_done): Likewise.
> (reposition_prologue_and_epilogue_notes): Likewise.
> * print-rtl.c (print_rtx): Likewise; drop code for printing line number
> notes.
> (print_rtl): Likewise.
> (print_rtl_single): Likewise.
> * gcse.c (insert_insn_start_bb): Likewise.
> * alias.c (init_alias_analysis): Likewise.
> * calls.c (fixup_tail_calls): Likewise.
> * except.c (sjlj_emit_function_enter): Likewise.
> * emit-rtl.c (add_insn_after): Likeiwse.
> (emit_label_before): Likewise.
> (emit_label_after): Likewise.
> (emit_note_before, emit_note_after, emit_note): Update
> parameter to be enum insn_note; do not deal with source
> files.
> * cfgcleanup.c (merge_blocks_move_predecessor_nojumps):
> Use NOTE_KIND instead of NOTE_LINE_NUMBER; use NOTE_INSN_BASIC_BLOCK_P
> when possible.
> (merge_blocks_move_successor_nojumps): Simplify now when
> we don't have BLOCK notes.
> (try_optimize_cfg): Likewise.
> * cfglayout.c (skip_insns_after_block): Likewise.
> (record_effective_endpoints): Likewise.
> (duplicate_insn_chain): Likewise.
> * varasm.c (output_constant_pool_1): Likewise.
> * sched-deps.c (sched_analyze): Likewise.
> * rtl.c (NOTE_INSN_MAX_isnt_negative_adjust_NOTE_INSN_BIAS):
> Exterminate.
> (note_insn_name): Simplify now when NOTE_INSN_BIAS is gone.
> * rtl.h (NOTE_SOURCE_LOCATION, NOTE_EXPANDED_LOCATION): Exterminate.
> (SET_INSN_DELETED): Simplify.
> (NOTE_LINE_NUMBER): Exterminate.
> (NOTE_LINE_KIND): New.
> (NOTE_INSN_BASIC_BLOCK_P): Update.
> (enum insn_note): Simplify.
> (GET_NOTE_INSN_NAME) Simplify.
> (emit_note_before, emit_note_after, emit_note): Update prototype.
> (squeeze_notes): Remove.
> * sched-int.h (NOTE_NOT_BB_P): Update.
> * resource.c (mark_target_live_regs): Update.
> * sched-rgn.c (debug_dependencies): Update.
> * sched-vis.c (print_insn): Update.
> * config/alpha/alpha.c (alpha_handle_trap_shadows): Update.
> * config/i386/i386.c (ix86_output_function_epilogue): Update.
> * config/sh/sh.c (sh_adjust_unroll_max): Function dead since gcc 4.0.0.
> (TARGET_ADJUST_UNROLL_MAX): Likewise.
> (split_branches): Update.
> (sh_optimize_target_register_callee_saved): Remove hunk dead since gcc
> 4.0.0.
> (sh_adjust_unroll_max): Exterminate.
> * config/c4x/c4x.c (c4x_reorg): Use SET_INSN_DELETED.
> * config/m68hc11/m68hc11.c (m68hc11_check_z_replacement): Likewise.
> (m68hc11_reorg): Likewise.
> * config/ia64/ia64.c (emit_insn_group_barriers): Update.
> (emit_predicate_relation_info): Update.
> (process_for_unwind_directive): Update.
> * config/rs6000/rs6000.c (rs6000_output_function_epilogue): Update.
> (output_call): Update.
> * config/pa/pa.c (output_lbranch): Update.
> (output_millicode_call): Update.
> (output_call): Update.
> (pa_combine_instructions): Update.
> * config/mips/mips.c (mips16_gp_pseudo_reg): Update.
> * config/bfin/bfin.c (gen_one_bundle): Update.
> * cfgrtl.c (can_delete_note_p): Update.
> (delete_insn): Update.
> (rtl_merge_blocks): Update.
> (commit_one_edge_insertion): Update.
> (rtl_verify_flow_info): Update.
> * stmt.c (expand_case): Do not call squeeze_notes.
More information about the Gcc-patches
mailing list