1 2008-09-14 Andreas Schwab <schwab@suse.de>
3 * tree-call-cdce.c (check_target_format): Accept Motorola formats.
5 2008-09-14 Jan Hubicka <jh@suse.cz>
7 * invoke.texi (-fconserve-stack): Document.
8 * opts.c (decode_options): Handle conserve_stack.
9 * common.opt (fconvserve_stack): New.
11 2008-09-14 David Edelsohn <edelsohn@gnu.org>
13 * config.host: Add x-aix to host_xmake_file.
14 * config/rs6000/t-aix52: Do not override LDFLAGS.
15 * config/rs6000/x-aix: New file.
17 2008-09-14 Andy Hutchinson <hutchinsonandy@aim.com>
23 * config/avr/avr.c (legitimate_address_p): Fix problem where subreg
24 is not recognized as a valid register usage. Allow REG_X to be used
26 * config/avr/avr.h (LEGITIMIZE_RELOAD_ADDRESS): Remove code that
27 forces a reload when using a base register.
29 2008-09-14 Danny Smith <dannysmith@users.sourceforge.net>
31 * config/i386/cygming-crtend.c (register_frame_ctor): If
32 DEFAULT_USE_CXA_ATEXIT, register __gcc_deregister_frame
34 (deregister_frame_ctor): Rename to ...
35 (deregister_frame_dtor): Use to call __gcc_deregister_frame
36 if !DEFAULT_USE_CXA_ATEXIT.
38 2008-09-13 Jan Hubicka <jh@suse.cz>
40 * cgraph.c: Include value-prof.h
41 * Makefile.in (cgraph.o): Add value-prof.h dependency.
43 2008-09-13 Jan Hubicka <jh@suse.cz>
46 * tree-profile.c (add_abnormal_goto_call_edges): New function.
47 (tree_gen_interval_profiler, tree_gen_pow2_profiler,
48 tree_gen_one_value_profiler, tree_gen_average_profiler,
49 tree_gen_ic_func_profiler, tree_gen_ior_profiler): Use it.
51 2008-09-13 Jan Hubicka <jh@suse.cz>
53 * i386.c (ix86_expand_movmem, ix86_expand_movstr): Avoid processing too
54 many bytes on misalligned blocks.
56 2008-09-13 Ralf Wildenhues <Ralf.Wildenhues@gmx.de>
58 * doc/cfg.texi: Fix some typos.
59 * doc/extend.texi: Likewise.
60 * doc/install.texi: Likewise.
61 * doc/invoke.texi: Likewise.
62 * doc/rtl.texi: Likewise.
63 * doc/tree-ssa.texi: Likewise.
65 2008-09-13 Richard Guenther <rguenther@suse.de>
68 * tree-nested.c (finalize_nesting_tree_1): Generate valid
69 gimple stores for memory rhs.
71 2008-09-13 H.J. Lu <hongjiu.lu@intel.com>
73 PR rtl-optimization/37489
74 * cse.c (fold_rtx): Don't return const_true_rtx for float
75 compare if FLOAT_STORE_FLAG_VALUE is undefined.
77 2008-09-13 Jan Hubicka <jh@suse.cz>
79 PR tree-optimization/37392
80 * tree-inline.c (remap_gimple_stmt): Do not remap return value.
82 2008-09-13 Martin Jambor <mjambor@suse.cz>
83 Jan Hubicka <jh@suse.cz>
85 * cgraph.c (free_edges): New variable.
86 (NEXT_FREE_EDGE): New macro.
87 (cgraph_free_edge): New function.
88 (cgraph_remove_edge): Call cgraph_remove_edge_1.
89 (cgraph_node_remove_callees): Likewise.
90 (cgraph_node_remove_callers): Likewise.
91 (cgraph_create_edge): Reuse edges from the free list. Do not
92 update uid if doing so.
93 (cgraph_remove_*_hook): Add free call.
95 2008-09-13 Richard Sandiford <rdsandiford@googlemail.com>
97 * ira-color.c (conflict_allocno_vec): Delete.
98 (update_cost_queue_elem): New structure.
99 (update_cost_queue): New variable.
100 (update_cost_queue_tail): Likewise.
101 (update_cost_queue_elems): Likewise.
102 (allocno_update_cost_check): Delete.
103 (initiate_cost_update): Allocate update_cost_queue_elems
104 instead of allocno_update_cost_check.
105 (finish_cost_update): Update the free()s accordingly.
106 (start_update_cost): New function.
107 (queue_update_cost): Likewise.
108 (get_next_update_cost): Likewise.
109 (update_copy_costs_1): Inline into...
110 (update_copy_costs): ...here. Use a queue instead of recursive calls.
111 Use cover_class instead of ALLOCNO_COVER_CLASS (another_allocno),
112 once we've established they are equal. Don't allocate update
113 costs if there is nothing to add to them.
114 (update_conflict_hard_regno_costs): Remove ALLOCNO and
115 DIVISOR arguments. Use a queue instead of recursive calls;
116 process all the allocnos in the initial queue, rather than
118 (assign_hard_reg): Use queue_update_cost instead of
119 conflict_allocno_vec. Queue coalesced allocnos instead
120 of calling update_conflict_hard_regno_costs for each one.
121 Just call update_conflict_hard_regno_costs once for the
123 (ira_color): Remove conflict_allocno_vec handling.
125 2008-09-12 Sebastian Pop <sebastian.pop@amd.com>
127 PR tree-optimization/37484
128 * graphite.c (scop_record_loop): Use snprintf instead of sprintf.
129 (save_var_name): Same.
130 (initialize_cloog_names): Same.
131 (initialize_cloog_names): Same.
133 2008-09-12 Sebastian Pop <sebastian.pop@amd.com>
135 * tree-scalar-evolution.c (set_instantiated_value): Set instantiated_below.
137 2008-09-12 Vladimir Makarov <vmakarov@redhat.com>
141 * ira-build.c (common_loop_tree_node_dominator): Remove.
142 (copy_live_ranges_to_removed_store_destinations): New function.
143 (regno_top_level_allocno_map): Move to top level from ...
144 (ira_flattening): ... here. Use
145 copy_live_ranges_to_removed_store_destinations.
147 * ira-emit.c (generate_edge_moves): Fix a comment.
149 2008-09-12 Anatoly Sokolov <aesok@post.ru>
152 * config/avr/avr.md (movsi_lreg_const peephole2): Add match_dup for
153 scratch register after 'set' pattern.
155 2008-09-12 Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
157 * emit-rtl.c (set_reg_attrs_from_value): Fix invalid alignment
158 information passed to mark_reg_pointer.
159 * explow.c (force_reg): Likewise.
161 2008-09-12 Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
163 * config/spu/spu.c (spu_override_options): Default to -mno-safe-hints
164 when building for the celledp architecture.
166 2008-09-12 Richard Guenther <rguenther@suse.de>
168 * tree-vrp.c (simplify_truth_ops_using_ranges): Fix types.
170 2008-09-12 Eric Botcazou <ebotcazou@adacore.com>
172 PR rtl-optimization/37424
173 * ira-color.c (coalesced_pseudo_reg_slot_compare): Untie by comparing
174 the regnos instead of the addresses.
176 2008-09-11 Janis Johnson <janis187@us.ibm.com>
178 * ginclude/float.h (DEC_EVAL_METHOD): Correct the macro name.
180 2008-09-11 Richard Guenther <rguenther@suse.de>
182 * tree-ssa-uncprop.c (associate_equivalences_with_edges): Avoid
183 entering the weaker equivalence recording.
185 * tree-ssa-phiprop.c (propagate_with_phi): Fix tuplification.
187 2008-09-11 Jeff Law <law@redhat.com>
189 * reload1.c (alter_reg): Undo the BYTE_BIG_ENDIAN correction performed
190 by assign_stack_local on the IRA path for stack slot sharing
191 as well as the non-IRA path.
193 2008-09-11 Uros Bizjak <ubizjak@gmail.com>
195 * config/i386/i386.h: Fix whitespace issues.
197 2008-09-11 Martin Jambor <mjambor@suse.cz>
199 * ipa-prop.h (struct ipa_param_flags): Removed.
200 (struct ipa_param_descriptor): New structure.
201 (struct ipa_node_params): ipcp_lattices, param_decls and
202 param_flags moved to ipa_param_description.
203 (ipa_get_ith_param): Renamed to ipa_get_param, changed to access
204 descriptors. Renamed all users.
205 (ipa_is_ith_param_modified): Renamed to ipa_is_param_modified,
206 changed to access descriptors. Renamed all users.
207 (ipa_is_ith_param_called): Renamed to ipa_is_param_called, changed
208 to access descriptors. Renamed all users.
209 * ipa-cp.c (ipcp_init_cloned_node): Call
210 ipa_initialize_node_params instead of ipa_count_formal_params and
211 ipa_create_param_decls_array.
212 (ipcp_analyze_node): Likewise.
213 (ipcp_get_ith_lattice): Renamed to ipcp_get_lattice, changed to access
214 descriptors. Renamed all users.
215 (ipcp_initialize_node_lattices): Remove allocation.
216 * ipa-inline.c (inline_indirect_intraprocedural_analysis): Call
217 ipa_initialize_node_params instead of ipa_count_formal_params and
218 ipa_create_param_decls_array.
219 * ipa-prop.c (ipa_create_param_decls_array): Renamed to
220 ipa_populate_param_decls, made static, added parameter info,
221 renamed mt to node, removed allocation, changed to use
223 (ipa_count_formal_params): Made static, added parameter info,
225 (ipa_initialize_node_params): New function.
226 (ipa_check_stmt_modifications): Changed to use descriptors.
227 (ipa_detect_param_modifications): Removed allocation, changed to
229 (ipa_note_param_call): Changed to use descriptors.
230 (ipa_analyze_params_uses): Removed allocation.
231 (ipa_free_node_params_substructures): Changed to use descriptors.
232 (ipa_edge_duplication_hook): Use the unused attribute.
233 (ipa_node_duplication_hook): Use the unused attribute, changed to
234 use descriptors, changed to duplicate descriptors.
236 * ipa-inline.c (cgraph_mark_inline_edge): New parameter new_edges,
237 changed all callers. Call ipa_propagate_indirect_call_infos if doing
238 indirect inlining. Made static.
239 (cgraph_decide_inlining): Freeing ipa-prop structures after inlining
240 functions called only once.
241 (cgraph_decide_recursive_inlining): Don't call
242 ipa_propagate_indirect_call_infos, pass new_edges to
243 cgraph_mark_inline_edge instead.
244 (cgraph_decide_inlining_of_small_functions): Don't call
245 ipa_propagate_indirect_call_infos, pass new_edges to
246 cgraph_mark_inline_edge instead.
247 (cgraph_decide_inlining): Don't call
248 ipa_propagate_indirect_call_infos.
249 * ipa-prop.c: Check that vectors are allocated.
251 * ipa-inline.c (cgraph_mark_inline_edge): Returns boolean, true
252 iff a new cgraph edges have been created.
253 (cgraph_decide_inlining): New variable redo_always_inline.
254 Flattening and always_inlining loop until callgraph stabilizes.
255 * ipa-prop.c (update_call_notes_after_inlining): Returns boolean,
256 true iff new cgraph edges have been created.
257 (propagate_info_to_inlined_callees): Likewise.
258 (ipa_propagate_indirect_call_infos): Likewise.
260 2008-09-11 Richard Guenther <rguenther@suse.de>
262 * tree-vectorizer.c (slpeel_add_loop_guard): Fix types.
263 (set_prologue_iterations): Likewise.
264 * tree-vect-transform.c (vect_create_addr_base_for_vector_ref):
266 (vect_update_init_of_dr): Likewise.
267 * tree-ssa-forwprop.c (forward_propagate_addr_expr_1): Fix
269 * fold-const.c (fold_unary): Do not generate calculations
272 2008-09-11 Paolo Bonzini <bonzini@gnu.org>
274 * dojump.c (do_jump) [BIT_AND_EXPR]: Move below. Fall through to
275 TRUTH_AND_EXPR for boolean (1-bit precision) expressions.
276 (do_jump) [BIT_IOR_EXPR]: Compile as TRUTH_OR_EXPR.
278 * tree-flow.h (simplify_stmt_using_ranges): Accept a GSI, return a bool.
279 * tree-ssa-propagate.c (substitute_and_fold): Pass a GSI to
280 VRP's simplify_stmt_using_ranges. Do simplify_stmt_using_ranges
281 before finalizing the changes.
282 * tree-vrp.c (extract_range_from_binary_expr): Add limited support
284 (simplify_truth_ops_using_ranges): New.
285 (simplify_div_or_mod_using_ranges, simplify_abs_using_ranges,
286 simplify_cond_using_ranges, simplify_switch_using_ranges): Return
287 whether a simplification was made.
288 (simplify_stmt_using_ranges): Ditto, and accept a GSI. For GS_ASSIGN,
289 use a switch statement and also call simplify_truth_ops_using_ranges.
291 2008-09-11 Jan Hubicka <jh@suse.cz>
293 * ggc-common.c (loc_array): Make static.
294 * dce.c (rest_of_handle_ud_dce): Free worklist.
296 2008-09-11 Jan Hubicka <jh@suse.cz>
298 * cgraph.c (cgraph_release_function_body): Plug memory leak on
299 ipa_transforms_to_apply and CFG; ggc_free gimple df and struct function
300 so we are sure we do not dangle it forever.
301 * function.c (outer_function_chain, find_function_data): Remove.
302 (function_context_stack): New.
303 (push_function_context, pop_function_context): Update.
304 * function.h (struct function): Remove pointer outer.
305 (outer_function_chain, find_function_data): Remove.
306 * stmt.c (force_label_rtx): Remove dead call of find_function_data.
308 2008-09-11 Jan Hubicka <jh@suse.cz>
311 * cgraph.c (cgraph_create_edge): Use !cgraph_edge for sanity check.
313 2008-09-11 Jan Hubicka <jh@suse.cz>
315 * tree-ssa-pre.c (phi_translate_1): Fix memory leak
317 2008-09-11 Jan Hubicka <jh@suse.cz>
319 * tree-ssa.c (redirect_edge_var_map_clear): Fix formatting.
320 (free_var_map_entry): New function.
321 (redirect_edge_var_map_destroy): Use it.
323 2008-09-11 Jan Hubicka <jh@suse.cz>
326 * tree-inline.c (add_lexical_block): Replace with ...
327 (prepend_lexical_block): ... prepend at begginig.
328 (remap_blocks): Use it and reverse later.
329 (expand_call_inline): Use prepend_lexical_block.
331 2008-09-11 Jan Hubicka <jh@suse.cz>
333 * gimplify.c (pop_gimplify_context): Free bind_expr_stack.
335 2008-09-11 Jan Hubicka <jh@suse.cz>
337 * function.c (free_after_compilation): Call insn_locators_free.
338 * cfglayout.c (insn_locators_free): New function.
339 * rtl.h (insn_locators_free): Declare.
341 2008-09-11 Jan Hubicka <jh@suse.cz>
343 * ifcvt.c (check_cond_move_block): Make regs argument pointer to
345 (cond_move_process_if_block): Update call.
347 2008-09-11 Jan Hubicka <jh@suse.cz>
349 * toplev.c (dump_memory_report): Call dump_vec_loc_statistics.
350 * vec.c: Include hashtab.h
351 (vec_descriptor, ptr_hash_entry): New structures.
352 (vec_desc_hash, vec_ptr_map): New static variables.
353 (hash_descriptor, eq_descriptor, hash_ptr, eq_ptr, vec_descriptor,
354 register_overhead, free_overhead, vec_heap_free): New functions.
355 (vec_gc_o_reserve_1): ggc_free when resizing to 0.
356 (vec_heap_o_reserve_1): free when resizing to 0; add statistics.
357 (cmp_statistic, add_statistics, dump_vec_loc_statistics): New functions.
359 2008-09-11 Ira Rosen <irar@il.ibm.com>
361 PR tree-optimization/37474
362 * tree-vect-analyze.c (vect_supported_load_permutation_p): Check the
363 length of load permutation.
365 2008-09-11 Andreas Schwab <schwab@suse.de>
367 * config/m68k/m68k.h (IRA_COVER_CLASSES): Define.
369 2008-09-11 Jakub Jelinek <jakub@redhat.com>
372 * expmed.c (extract_low_bits): Avoid creating invalid subregs.
373 * dse.c (find_shift_sequence): Use extract_low_bits instead of
376 2008-09-11 Ira Rosen <irar@il.ibm.com>
378 * tree-vect-transform.c (vectorizable_store): Use the rhs vector type
381 2008-09-10 Jakub Jelinek <jakub@redhat.com>
384 * gimplify.c (gimplify_body): Call default_rtl_profile.
387 * config/rs6000/rs6000-c.c (rs6000_macro_to_expand): Return NULL
388 instead of tok->val.node if not expanding to something else. Handle
389 intervening CPP_PADDING tokens.
390 (altivec_categorize_keyword): Remove unneeded comparisons.
392 2008-09-10 Richard Guenther <rguenther@suse.de>
394 * tree-ssa-pre.c (phi_translate_1): Fix memory leak.
396 2008-09-10 Richard Guenther <rguenther@suse.de>
399 * tree-inline.c (insert_init_stmt): Make sure to not
400 insert invalid gimple stores.
402 2008-09-10 Sebastian Pop <sebastian.pop@amd.com>
404 PR tree-optimization/37388
405 * toplev.c (process_options): Fail and warn when graphite
406 flags are used, but the compiler has not been configured
407 with graphite libraries.
408 * graphite.c (graphite_transform_loops): Remove printfs
409 to dump_file for the case when graphite is not available.
411 2008-09-10 H.J. Lu <hongjiu.lu@intel.com>
414 * config/i386/i386.c (ix86_expand_vector_init_interleave): Force
415 the even element into register.
416 (ix86_expand_vector_init_general): Don't use
417 ix86_expand_vector_init_interleave on V16QImode and V8HImode
418 if we can't move from GPR to SSE register directly.
420 2008-09-10 Bernd Schmidt <bernd.schmidt@analog.com>
422 * config/bfin/bfin.c (workaround_speculation): Correct algorithm to
423 not lose track of the number of NOPs needed. Number of NOPs needed
424 for sync vs. loads workaround was switched; corrected. Run second
425 pass for all workarounds. No NOPs needed after call insns. Change
426 second pass to use find_next_insn_start and find_load helpers in order
427 to properly detect parallel insns.
428 * config/bfin/bfin.md (cbranch_with_nops): Increase length.
430 2008-09-10 Jan Hubicka <jh@suse.cz>
432 * value-prof.c (gimple_ic): Fix tuplification bug.
433 * sched-deps.c (sched_insns_conditions_mutex_p): Silence unitialized
436 2008-09-10 Jakub Jelinek <jakub@redhat.com>
438 PR tree-optimization/37353
439 * tree-call-cdce.c (cond_dead_built_in_calls): Remove.
440 (shrink_wrap_conditional_dead_built_in_calls): Add calls argument, use
441 calls instead of cond_dead_built_in_calls.
442 (tree_call_cdce): Add cond_dead_built_in_calls automatic variable,
443 initalize the vector only before adding first entry. Use VEC_safe_push
444 instead of VEC_quick_push. Pass cond_dead_built_in_calls to
445 shrink_wrap_conditional_dead_built_in_calls call.
447 2008-09-10 Ira Rosen <irar@il.ibm.com>
449 PR tree-optimization/37385
450 * tree-vect-transform.c (vect_create_data_ref_ptr): Add a new argument,
451 and use it as a vector type if not NULL.
452 (vectorizable_store): Call vect_create_data_ref_ptr with the type of
454 (vect_setup_realignment): Call vect_create_data_ref_ptr with additional
456 (vectorizable_load): Likewise.
458 2008-09-10 Jakub Jelinek <jakub@redhat.com>
460 * config/i386/i386.md (SWI32): New mode iterator.
461 (jcc_fused_1, jcc_fused_2, jcc_fused_3, jcc_fused_4): Use it instead
464 2008-09-09 Vladimir Makarov <vmakarov@redhat.com>
466 PR rtl-optimization/37435
467 * caller-save.c (insert_restore, insert_save): Check the mode by
470 2008-09-09 John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
473 * dwarf2asm.c (dw2_force_const_mem): Use unstripped symbol name for
475 (dw2_output_indirect_constant_1): Use unstripped name in symbol
478 2008-09-09 Sandra Loosemore <sandra@codesourcery.com>
480 * doc/invoke.texi (ARM Options): Correct errors in discussion
481 of -mfloat-abi, -mhard-float, and -msoft-float.
483 2008-09-09 Jakub Jelinek <jakub@redhat.com>
484 Jan Hubicka <jh@suse.cz>
487 * tree-inline.c (copy_bb): Insert stmt into copy_basic_block before
488 calling gimple_regimplify_operands on it. Iterate over all newly
489 added statements, not just the last one.
490 (insert_init_stmt): Insert stmt into seq first, then call
491 gimple_regimplify_operands on it. Don't create new gimplification
492 context, nor find referenced vars.
494 2008-09-09 Jakub Jelinek <jakub@redhat.com>
497 * ipa-prop.h (ipa_propagate_indirect_call_infos): Change last argument
498 to pointer to vector pointer.
499 * ipa-prop.c (ipa_propagate_indirect_call_infos,
500 propagate_info_to_inlined_callees): Likewise.
501 (update_call_notes_after_inlining): Likewise. Push new indirect edge
502 to *new_edges instead of new_edges. Reread IPA_EDGE_REF after
503 ipa_check_create_edge_args.
504 * ipa-inline.c (cgraph_decide_recursive_inlining): Change last argument
505 to pointer to vector pointer.
506 (cgraph_decide_inlining_of_small_function): Adjust
507 cgraph_decide_recursive_inlining and ipa_propagate_indirect_call_infos
511 * config/i386/i386.md (zero_extendqihi2_movzbl): Enable when optimizing
514 PR rtl-optimization/37408
515 * function.c (assign_parm_find_stack_rtl): Set correct MEM_SIZE
518 2008-09-09 Sebastian Pop <sebastian.pop@amd.com>
520 PR tree-optimization/37375
521 * tree-scalar-evolution.c (scev_info_str): Add field instantiated_below.
522 (new_scev_info_str, eq_scev_info, find_var_scev_info,
523 set_scalar_evolution, get_scalar_evolution, get_instantiated_value,
524 set_instantiated_value): Pass instantiated_below.
525 (analyze_scalar_evolution_1, analyze_scalar_evolution): Update calls to
527 (instantiate_scev_1, instantiate_scev): Pass a basic block above which
528 the definitions are not instantiated.
530 * tree-scalar-evolution.h (instantiate_scev): Update declaration.
531 (block_before_loop): New.
532 * tree-data-ref.c (dr_analyze_indices): Update uses of instantiate_scev.
533 * graphite.c (block_before_scop): New.
534 (loop_affine_expr, stmt_simple_for_scop_p, harmful_stmt_in_bb): Pass a
535 basic block, not a loop for determining the parameters.
536 (scopdet_edge_info, build_scops_1): Do not pass outermost loop in the
538 (idx_record_params, find_params_in_bb, find_scop_parameters,
539 build_loop_iteration_domains, add_conditions_to_domain): Update calls
542 * Makefile.in (cfgloopmanip.o): Add missing dependency on TREE_FLOW_H.
544 2008-09-09 Vladimir Makarov <vmakarov@redhat.com>
548 * ira-build.c (ira_create_allocno): Setup frequency to 0.
550 * ira-color.c (update_conflict_hard_regno_costs): Remove assert.
551 Check zero freq and increase if necessary.
553 2008-09-09 Vladimir Makarov <vmakarov@redhat.com>
555 * ira-conflicts.c (process_regs_for_copy): Check that the hard
556 regno is in the right range. Add comments.
558 2008-09-09 Rainer Orth <ro@TechFak.Uni-Bielefeld.DE>
560 * Makefile.in (mips-tfile.o-warn): Don't error out on mips-tfile.c
562 * mips-tfile.c (copy_object): Cast alloca result to int *.
563 * mips-tdump.c (print_symbol): Cast xmalloc return values to
565 Rename class to sclass.
566 (read_tfile): Cast read_seek return values to proper types.
567 Cast xcalloc return value to proper type.
569 2008-09-09 Rainer Orth <ro@TechFak.Uni-Bielefeld.DE>
571 * config/sol2.h (REGISTER_TARGET_PRAGMAS): Move ...
572 * config/i386/sol2.h (REGISTER_SUBTARGET_PRAGMAS): ... here.
573 * config/sparc/sol2.h (REGISTER_TARGET_PRAGMAS): ... and here.
575 2008-09-09 Jan Hubicka <jh@suse.cz>
577 * profile.c (is_edge_inconsistent): Add debug output; ignore
578 negative count on fake edges.
579 (is_inconsistent): Add debug output.
581 2008-09-09 Andrey Belevantsev <abel@ispras.ru>
583 * haifa-sched.c (advance_one_cycle): Do not print '\n' before printing
585 (choose_ready): When first insn is chosen from the ready list, also
588 2008-09-09 Andrey Belevantsev <abel@ispras.ru>
590 PR rtl-optimization/37360
591 * haifa-sched.c (max_issue): Do not assert that we never issue more
592 insns than issue_rate. Add comment.
594 2008-09-09 Richard Guenther <rguenther@suse.de>
596 * tree-cfg.c (verify_types_in_gimple_op): Remove.
597 (verify_types_in_gimple_call): Rename to ...
598 (verify_gimple_call): ... this. Enhance.
599 (verify_types_in_gimple_cond): Remove.
600 (verify_gimple_comparison): New function ...
601 (verify_types_in_gimple_assign): ... split out from here.
602 (verify_types_in_gimple_return): Rename to ...
603 (verify_gimple_return): ... this. Enhance.
604 (verify_types_in_gimple_switch): Rename to ...
605 (verify_gimple_switch): ... this. Enhance.
606 (verify_gimple_goto): New function.
607 (verify_types_in_gimple_phi): Rename to ...
608 (verify_gimple_phi): ... this. Enhance.
609 (verify_types_in_gimple_stmt): Adjust calls to helper functions.
610 Fold in single-statement cases from verify_types_in_gimple_seq_2.
611 (verify_types_in_gimple_seq_2): Remove cases handled in
612 verify_types_in_gimple_stmt.
614 2008-09-09 Bernd Schmidt <bernd.schmidt@analog.com>
616 * config/bfin/bfin.c (n_regs_to_save): New static variable.
617 (push_multiple_operation, pop_multiple_operation): Set it.
618 (workaround_rts_anomaly): New function.
619 (workaround_speculation): New function, broken out of bfin_reorg.
620 (bfin_reorg): Call the new functions.
622 2008-09-09 Richard Guenther <rguenther@suse.de>
626 * gimplify.c (gimplify_conversion): Change conversions of
627 non-register type to VIEW_CONVERT_EXPRs.
628 (gimplify_addr_expr): If we need to make the operand
629 addressable make sure to use a properly initialized
630 temporary for that so it gets a valid gimple store.
632 2008-09-09 Aldy Hernandez <aldyh@redhat.com>
634 * function.h (struct function): Add function_start_locus.
635 * cfgexpand.c (gimple_expand_cfg): Use it.
636 * c-parser.c (c_parser_declaration_or_fndef): Set it.
638 2008-09-09 Richard Guenther <rguenther@suse.de>
640 PR tree-optimization/37433
641 * tree-ssa-ccp.c (ccp_fold): Properly guard folding of
644 2008-09-09 Richard Guenther <rguenther@suse.de>
646 PR tree-optimization/37387
647 * tree-ssa-ifcombine.c (ifcombine_iforif): Convert the name
648 and bits to a common type.
650 2008-09-09 Nick Clifton <nickc@redhat.com>
652 * config/v850/v850.md (return): Restore frame size restriction.
654 2008-09-09 Paolo Bonzini <bonzini@gnu.org>
656 * c-common.c (c_expand_decl): Remove.
657 * c-common.h (anon_aggr_type_p): Remove prototype.
658 (DECL_ANON_UNION_ELEMS): Remove.
659 * c-objc-common.h (LANG_HOOKS_EXPAND_DECL): Remove.
660 * c-decl.c (anon_aggr_type_p): Remove.
661 * langhooks-def.h (LANG_HOOKS_EXPAND_DECL): Remove.
662 (LANG_HOOKS_INITIALIZER): Remove LANG_HOOKS_EXPAND_DECL.
663 * langhooks.h (struct lang_hooks): Remove expand_decl.
664 * langhooks.c (lhd_expand_decl): Remove.
665 * stmt.c (expand_anon_union_decl): Remove.
666 * tree.h (expand_anon_union_decl): Remove prototype.
668 2008-08-09 Andy Hutchinson <hutchinsonandy@aim.com>
671 * config/avr/avr.c (avr_reorg): Create RTL for reversed compare with
673 * config/avr/avr.md (QISI) : Define mode iterator.
674 (negated_tst<mode>) : Redefine as split using mode macro.
675 (reversed_tstqi): Define insn as reversed compare with zero.
676 (reversed_tsthi): Ditto.
677 (reversed_tstsi): Ditto.
679 2008-09-08 Jakub Jelinek <jakub@redhat.com>
682 * tree-inline.c (copy_bb): When replacing a gimple_call_va_arg_pack_p
683 call stmt by new_call, clear gimple_bb on stmt after gsi_replace.
686 * predict.c (optimize_function_for_size_p): Don't segfault if
688 * fold-const.c (LOGICAL_OP_NON_SHORT_CIRCUIT, fold_truthop,
689 tree_swap_operands_p): Don't test cfun != NULL before calling
690 optimize_function_for_s*_p.
692 2008-09-08 Eric Botcazou <ebotcazou@adacore.com>
694 * ira-color.c (ira_reuse_stack_slot): Set slot_num on success at the
697 2008-09-08 M R Swami Reddy <MR.Swami.Reddy@nsc.com>
699 * crx/crx.h (IRA_COVER_CLASSES): Define.
701 2008-09-08 Jakub Jelinek <jakub@redhat.com>
704 * tree-ssa-dom.c (optimize_stmt): Call maybe_clean_or_replace_eh_stmt
705 even when a stmt has been gimple_modified_p, but after fold_stmt is
706 not any longer. Remove unneeded may_have_exposed_new_symbols
709 2008-09-08 Richard Guenther <rguenther@suse.de>
711 PR tree-optimization/37421
712 * tree-ssa-sccvn.c (visit_copy): Make sure to fully
715 2008-09-08 Jakub Jelinek <jakub@redhat.com>
718 * opts.c (common_handle_option): Handle OPT_ftree_store_ccp.
720 2008-09-08 John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
723 * pa-hpux.h (LINK_SPEC): Strip -fwhole-program.
724 * pa-hpux10.h (LINK_SPEC): Likewise.
725 * pa-hpux11.h (LINK_SPEC): Likewise.
727 2008-09-07 John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
729 * pa32-regs.h (IRA_COVER_CLASSES): Define.
730 * pa64-regs.h (IRA_COVER_CLASSES): Define.
732 2008-09-07 Helge Deller <deller@gmx.de>
734 * pa/linux-atomic.c: New file.
735 * pa/t-linux (LIB2FUNCS_STATIC_EXTRA): Define.
736 * pa/t-linux64 (LIB2FUNCS_STATIC_EXTRA): Define.
738 2008-09-07 Richard Guenther <rguenther@suse.de>
739 Ira Rosen <irar@il.ibm.com>
741 PR tree-optimization/36630
742 * tree-vect-transform.c (vect_update_ivs_after_vectorizer):
743 Call STRIP_NOPS before calling evolution_part_in_loop_num.
745 2008-09-07 Dorit Nuzman <dorit@il.ibm.com>
746 Ira Rosen <irar@il.ibm.com>
748 PR tree-optimization/35642
749 * config/rs6000/altivec.md (mulv8hi3): Implement.
751 2008-09-06 Jeff Law <law@redhat.com>
753 * h8300/h8300.h (IRA_COVER_CLASSES): Define.
755 2008-09-06 Jan Hubicka <jh@suse.cz>
757 PR tree-optimization/14703
758 * tree-ssa-live.c (remove_unused_scope_block_p): Remove ignored
760 * passes.c (init_optimization_passes): Recompute inline parameters.
762 2008-09-06 Richard Sandiford <rdsandiford@googlemail.com>
764 * config/mips/mips.c (mips_function_ok_for_sibcall): Check for
767 2008-09-06 Richard Sandiford <rdsandiford@goolemail.com>
768 Peter Fuerst <post@pfrst.de>
770 * doc/invoke.texi: Document -mr10k-cache-barrier=.
771 * doc/extend.texi: Document __builtin_mips_cache.
772 * config/mips/mips-ftypes.def: Add a (VOID, SI, CVPOINTER) entry.
773 * config/mips/mips.opt (mr10k-cache-barrier=): New option.
774 * config/mips/mips.h (TARGET_CPU_CPP_BUILTINS): Define
775 __GCC_HAVE_BUILTIN_MIPS_CACHE.
776 (TARGET_CACHE_BUILTIN, ISA_HAS_CACHE): New macros.
777 * config/mips/mips.c (mips_r10k_cache_barrier_setting): New enum.
778 (set_push_mips_isas): New variable.
779 (mips_r10k_cache_barrier): New variable.
780 (cache): New availability predicate.
781 (mips_builtins): Add an entry for __builtin_mips_cache.
782 (mips_build_cvpointer_type): New function.
783 (MIPS_ATYPE_CVPOINTER): New macro.
784 (mips_prepare_builtin_arg): Only use the insn's mode if the rtx's
786 (r10k_simplified_address_p, r10k_simplify_address)
787 (r10k_uncached_address_p, r10k_safe_address_p)
788 (r10k_needs_protection_p_1, r10k_needs_protection_p_store)
789 (r10k_needs_protection_p_call, r10k_needs_protection_p)
790 (r10k_insert_cache_barriers): New functions.
791 (mips_reorg_process_insns): Delete cache barriers after a
792 branch-likely instruction.
793 (mips_reorg): Call r10k_insert_cache_barriers.
794 (mips_handle_option): Handle OPT_mr10k_cache_barrier_.
795 * config/mips/mips.md (UNSPEC_MIPS_CACHE): New constant.
796 (UNSPEC_R10K_CACHE_BARRIER): Likewise.
797 (mips_cache, r10k_cache_barrier): New define_insns.
799 2008-09-06 Richard Sandiford <rdsandiford@googlemail.com>
801 * ira-int.h (ira_zero_hard_reg_set, ira_one_hard_reg_set): Delete.
802 * ira-color.c (setup_allocno_left_conflicts_num): Use
803 hard_reg_set_equal_p instead of a comparison with
804 ira_zero_hard_reg_set.
805 * ira.c (setup_reg_subclasses): Likewise.
806 (setup_cover_and_important_classes): Likewise.
807 (setup_class_translate): Likewise.
808 (setup_reg_class_intersect_union): Likewise.
809 (ira_zero_hard_reg_set, ira_one_hard_reg_set): Delete.
810 (ira_init_once): Don't initialize them.
812 2008-09-05 Vladimir Makarov <vmakarov@redhat.com>
814 * doc/tm.texi (IRA_COVER_CLASSES): Fix a typo.
816 2008-09-05 Vladimir Makarov <vmakarov@redhat.com>
818 * ira-color.c (ira_fast_allocation): Permit global allocno allocation.
820 2008-09-05 Trevor Smigiel <Trevor_Smigiel@playstation.sony.com>
822 Improved branch hints, safe hints, and scheduling.
824 * haifa-sched.c (sched_emit_insn) : Define.
825 * sched-int.h (sched_emit_insn) : Add prototype.
826 * doc/invoke.texi (-mdual-nops, -mhint-max-nops,
827 -mhint-max-distance -msafe-hints) : Document.
828 * config/spu/spu.c (spu_flag_var_tracking): New.
829 (TARGET_SCHED_INIT_GLOBAL, TARGET_SCHED_INIT,
830 TARGET_SCHED_REORDER, TARGET_SCHED_REORDER2,
831 TARGET_ASM_FILE_START): Define.
832 (TARGET_SCHED_ADJUST_PRIORITY): Remove.
833 (STOP_HINT_P, HINTED_P, SCHED_ON_EVEN_P): Define.
834 (spu_emit_branch_hint): Add blocks argument.
835 (insert_branch_hints, insert_nops): Remove.
836 (pad_bb, insert_hbrp_for_ilb_runout, insert_hbrp, in_spu_reorg,
837 uses_ls_unit, spu_sched_init_global, spu_sched_init,
838 spu_sched_reorder, asm_file_start): New functions.
839 (clock_var, spu_sched_length, pipe0_clock,
840 pipe1_clock, prev_clock_var, prev_priority,
841 spu_ls_first, prev_ls_clock): New static variables.
842 * config/spu/spu.h (TARGET_DEFAULT): Add MASK_SAFE_HINTS.
843 * config/spu.md (iprefetch): Add operand, make it clobber MEM.
844 (nopn_nv): Add a non-volatile version of nop.
845 * config/spu/spu.opt (-mdual-nops, -mhint-max-nops,
846 -mhint-max-distance, -msafe-hints): New options.
848 2008-09-05 Janis Johnson <janis187@us.ibm.com>
849 Samuel Tardieu <sam@rfc1149.net>
851 * opts.c (decode_options): Combine nested if statements.
854 * opts.c (decode_options): Handle more relationships among
855 unit-at-a-time, toplevel-reorder, and section-anchors.
857 2008-09-05 David Daney <ddaney@avtrex.com>
859 * doc/install.texi (--enable-reduced-reflection): Document new option.
861 2008-09-05 Bob Wilson <bob.wilson@acm.org>
863 * config/xtensa/predicates.md (nonimmed_operand, mem_operand): Use
865 (constantpool_operand): New.
866 (move_operand): Disallow sub-word modes for the constant pool.
867 * config/xtensa/xtensa.c (TARGET_SECONDARY_RELOAD): Define.
868 (xtensa_secondary_reload_class): Replace with....
869 (xtensa_secondary_reload): this function. Remove SIGN_EXTEND check.
870 Set icode for sub-word reloads from the constant pool.
871 * config/xtensa/xtensa.h (SECONDARY_INPUT_RELOAD_CLASS): Delete.
872 (SECONDARY_OUTPUT_RELOAD_CLASS): Delete.
873 * config/xtensa/xtensa.md (reload<mode>_literal): New.
874 * config/xtensa/xtensa-protos.h: Update prototypes.
876 2008-09-05 Joseph Myers <joseph@codesourcery.com>
878 * config/mips/mips.h (enum reg_class): Add FRAME_REGS.
879 (REG_CLASS_NAMES): Update.
880 (REG_CLASS_CONTENTS): Update.
881 * config/mips/mips.c (mips_regno_to_class): Use FRAME_REGS instead
882 of ALL_REGS for regs 77 and 78.
883 * function.c (instantiate_virtual_regs_in_insn): Assert that
884 return value of simplify_gen_subreg is not NULL.
886 2008-09-05 Paolo Bonzini <bonzini@gnu.org>
888 * emit-rtl.c (gen_rtvec): Rewrite not using gen_rtvec_v.
889 (gen_rtvec_v): Fix coding standards.
891 2008-09-04 Jan Hubicka <jh@suse.cz>
893 * i386.c (decide_alg): Be more conservative about optimizing for size.
895 2008-09-04 Ian Lance Taylor <iant@google.com>
897 * varasm.c (narrowing_initializer_constant_valid_p): New
899 (initializer_constant_valid_p): Call it.
901 2008-09-04 Jeff Law <law@redhat.com>
903 * fold-const.c (native_encode_real): Fix computation of WORDS.
904 (native_interpret_real): Likewise.
906 2008-09-04 Janis Johnson <janis187@us.ibm.com>
908 * config/rs6000/t-rs6000: Remove target gt-rs6000.h.
910 2008-09-04 Vladimir Makarov <vmakarov@redhat.com>
912 * ira-conflicts.c (process_regs_for_copy): Check insn to check
913 that the cost is already taken into account in ira-costs.c
915 * ira-int.h (ira_debug_copy, ira_debug_copies): New.
917 * ira-build.c (print_copy, print_copies, ira_debug_copy,
918 ira_debug_copies): New.
919 (ira_bulid): Call print_copies.
921 * doc/tm.texi (IRA_COVER_CLASSES): Fix the description.
923 2008-09-04 Samuel Tardieu <sam@rfc1149.net>
926 * config/rs6000/rs6000.c (optimization_options): Remove check of
929 2008-09-04 Adam Nemet <anemet@caviumnetworks.com>
931 * config/mips/mips.h (ISA_HAS_SEQ_SNE): New macro.
932 * config/mips/mips.c (mips_expand_scc): Also expand seq and sne if
933 second operand is a reg_imm10_operand.
934 * config/mips/mips.md (*seq_<GPR:mode><GPR2:mode>_seq,
935 *sne_<GPR:mode><GPR2:mode>_sne): New patterns.
936 (*seq_<GPR:mode><GPR2:mode>): Rename to
937 *seq_zero_<GPR:mode><GPR2:mode>. Don't match if
939 (*seq_<GPR:mode><GPR2:mode>_mips16): Rename to
940 *seq_zero_<GPR:mode><GPR2:mode>_mip16. Don't match if
942 (*sne_<GPR:mode><GPR2:mode>): Rename to
943 *sne_zero_<GPR:mode><GPR2:mode>. Don't match if
946 2008-09-04 Adam Nemet <anemet@caviumnetworks.com>
948 * config/mips/mips.h (ISA_HAS_EXTS): New macro.
949 * config/mips/mips.md (*ashr_trunc<mode>): Name the pattern
950 combining an arithmetic right shift by more than 31 and a
951 trunction. Don't match for out-of-range shift amounts. Set
952 attribute mode to <MODE>.
953 (*lshr32_trunc<mode>): Name the pattern combining a logical right
954 shift by 32 and and a truncation. Set attribute mode to <MODE>.
955 (*<optab>_trunc<mode>_exts): New pattern for truncated right
956 shifts by less than 32.
957 (extv): Change predicate on first operand to accept registers.
958 Change predicate of the other operands from immediate_operand to
959 const_int_operand. Expand exts when source is a register.
960 (extzv): Change predicate of the constant operands from
961 immediate_operand to const_int_operand.
962 (extzv<mode>): Change predicate of the constant operands from
963 immediate_operand to const_int_operand and no constraint. Also
965 (*extzv_trunc<mode>_exts): New pattern.
967 2008-09-04 Adam Nemet <anemet@caviumnetworks.com>
969 * config/mips/mips.h (ISA_HAS_CINS): New macro.
970 * config/mips/mips-protos.h (mask_low_and_shift_p,
971 mask_low_and_shift_len): Declare.
972 * config/mips/mips.c (mask_low_and_shift_p,
973 mask_low_and_shift_len): New functions.
974 (mips_print_operand): Handle new operand prefix "m".
975 * config/mips/mips.md (*cins<mode>): New pattern.
977 2008-09-04 Bernd Schmidt <bernd.schmidt@analog.com>
979 * config/bfin/bfin.c (gen_one_bundle): Don't create new nops when
982 2008-09-04 Richard Sandiford <rdsandiford@googlemail.com>
984 * df-scan.c (df_get_entry_block_def_set): Add STACK_POINTER_REGNUM
985 regardless of epilogue_completed.
987 2008-09-04 Vladimir Makarov <vmakarov@redhat.com>
990 * ira-lives.c (process_bb_node_lives): Check setjmp.
992 2008-09-04 Richard Sandiford <rdsandiford@googlemail.com>
995 * ira-build.c (form_loop_tree): Reverse BB walk.
996 (create_bb_allocnos): Likewise.
997 * ira-lives.c (make_regno_born_and_dead, regs_set): Delete.
998 (mark_reg_store): Rename to...
999 (mark_ref_live): ...this and take a df_ref argument instead of
1000 note_stores arguments. Assert that we have a register.
1001 (mark_reg_clobber): Delete.
1002 (def_conflicts_with_inputs_p): New function.
1003 (mark_reg_conflicts): Delete.
1004 (mark_reg_death): Rename to...
1005 (mark_ref_dead): ...this and take a df_ref argument instead of
1006 a register. Assert that we have a register.
1007 (process_bb_node_lives): Hoist frequency calculation out of
1008 instruction walk. Convert from a forwards scan to a backwards scan.
1009 Use DF_REF_USES and DF_REF_DEFS instead of register notes and
1010 note_stores. Remove EH_RETURN_DATA_REGNO and regs_set handling.
1011 (create_allocno_live_ranges): Don't create regs_set.
1013 2008-09-04 Ian Lance Taylor <iant@google.com>
1015 * rtl.h (LABEL_REF_NONLOCAL_P): Don't check for REG_LABEL_OPERAND
1016 or REG_LABEL_TARGET.
1017 * calls.c (emit_library_call_value_1): Use MEM_P rather than
1018 comparing MODE with MEM.
1019 * gimple.c (gimple_build_predict): Cast END_PREDICTORS before
1020 comparing with GF_PREDICT_TAKEN.
1021 (gimple_get_lhs): Change code to enum gimple_code.
1022 (gimple_set_lhs): Likewise.
1023 * ifcvt.c (noce_process_if_block): Correct GET_MODE to GET_CODE.
1024 * omp-low.c (find_omp_clause): Change kind parameter to enum
1026 * tree-flow.h (find_omp_clause): Update declaration.
1027 * regrename.c (clear_dead_regs): Change kind parameter to enum
1029 * reload1.c (eliminate_regs_1): Use REG_NOTE_KIND rather than
1031 * see.c (see_get_extension_data): Change return type to enum
1032 entry_type. Change UNKNOWN to NOT_RELEVANT, SIGN_EXTEND to
1033 SIGNED_EXTENDED_DEF, ZERO_EXTEND to ZERO_EXTENDED_DEF.
1034 (see_gen_normalized_extension): Change extension_code parameter to
1036 (see_seek_pre_extension_expr): Change extension_code to enum
1038 (see_merge_one_def_extension): Likewise.
1039 (see_handle_relevant_defs): Likewise.
1040 (see_handle_relevant_uses): Likewise.
1041 (see_analyze_one_def): Likewise.
1042 * tree-cfg.c (need_fake_edge_p): Compare gimple code with
1043 GIMPLE_ASM rather than ASM_EXPR.
1044 * tree-ssa-alias.c (is_escape_site): Compare gimple code with
1045 GIMPLE_RETURN rather than RETURN_EXPR.
1046 * tree-ssa-ccp.c (likely_value): Change code to enum gimple_code.
1047 (evaluate_stmt): Likewise.
1048 * tree-vect-analyze.c (vect_analyze_operations): Change relevance
1049 to enum vect_relevant.
1050 (vect_mark_stmts_to_be_vectorized): Change assertion to not
1051 compare gimple codes with tree codes.
1053 2008-09-04 Paul Brook <paul@codesourcery.com>
1055 * config/arm/arm.c (arm_size_rtx_costs): Call cost function.
1057 2008-09-04 Bernd Schmidt <bernd.schmidt@analog.com>
1059 * config/bfin/bfin.c (bfin_function_ok_for_sibcall): Restore a null
1060 pointer check lost in the recent no-unit-at-a-time patch.
1062 2008-09-04 Jan Hubicka <jh@suse.cz>
1065 * tree-switch-conversion.c (check_final_bb): Accept only IP
1068 2008-09-04 Jan Hubicka <jh@suse.cz>
1070 * tree-vrp.c (vrp_evaluate_conditional_warnv_with_ops_using_ranges):
1072 (vrp_evaluate_conditional_warnv_with_ops): ... this one. Add
1073 using_ranges argument.
1074 (vrp_evaluate_conditional): Avoid bogus warning for type range.
1075 (vrp_visit_cond_stmt): Update call of
1076 vrp_evaluate_conditional_warnv_with_ops
1078 2008-09-04 Jan Hubicka <jh@suse.cz>
1080 PR tree-optimization/37345
1081 PR tree-optimization/37358
1082 PR tree-optimization/37357
1083 * tree.c (build_function_type_skip_args): Build distinct type copy;
1085 (build_function_decl_skip_args): Set type of new decl not orig decl;
1086 clear DECL_VINDEX for methods turned into functions.
1088 2008-09-04 Nick Clifton <nickc@redhat.com>
1090 * configure.ac (HAVE_GAS_LCOMM_WITH_ALIGNMENT): New assembler check.
1091 * configure: Regenerate.
1092 * config.in: Regenerate.
1093 * config/i386/bsd.h (ASM_OUTPUT_ALIGNED_LOCAL): Use .lcomm with an
1094 alignment field, if it is supported.
1096 2008-09-04 Kai Tietz <kai.tietz@onevision.com>
1098 * config/i386/t-mingw32 (SHLIB_LC): Change order of import
1101 2008-09-03 David Edelsohn <edelsohn@gnu.org>
1103 * config/rs6000/rs6000.c (processor_target_table): Do not
1104 enable VMX for POWER6.
1106 2008-09-03 Anton Blanchard <anton@samba.org>
1108 * config/rs6000/rs6000.c (rs6000_split_lock_test_and_set): Do not
1109 emit memory barrier before operation.
1111 2008-09-03 John David Anglin <dave.anglin@nrc.cnrc.gc.ca>
1114 * config/pa/pa64-hpux.h (ASM_OUTPUT_EXTERNAL_REAL): Call
1115 default_elf_asm_output_external.
1117 2008-09-03 Jeff Law <law@redhat.com>
1119 * h8300.c (expand_a_shift): Avoid invalid RTL sharing.
1121 2008-09-03 Richard Henderson <rth@redhat.com>
1123 * config/alpha/alpha.c (alpha_split_lock_test_and_set): Move
1124 memory barrier to below the test-and-set.
1125 (alpha_split_lock_test_and_set_12): Likewise.
1127 2008-09-03 Vladimir Makarov <vmakarov@redhat.com>
1131 * ira-int.h (ira_loop_tree_node): Rename mentioned_allocnos to
1134 * ira-color.c (print_loop_title): Use all_allocnos.
1135 (color_pass): Ditto. Don't add border_allocnos. Check that
1136 subloop allocno in the correspdoning bitmap all_allocnos.
1138 * ira-emit.c (change_loop): Use all_allocnos.
1140 * ira-build.c (create_loop_tree_nodes, finish_loop_tree_node):
1142 (ira_create_allocno): Set up all_allocnos bit for the created
1144 (create_cap_allocno): Remove setting mentioned_allocnos.
1145 (create_insn_allocnos): Ditto.
1146 (remove_unnecessary_allocnos): Use all_allocnos.
1147 (check_allocno_creation): Check that allocnos are in the
1148 corresponding bitmap all_allocnos.
1150 2008-09-03 Vladimir Makarov <vmakarov@redhat.com>
1154 * ira-conflicts.c (REG_SUBREG_P, go_through_subreg): New.
1155 (process_regs_for_copy): Process subregs. Refine check when cost
1156 is taken into account in ira-costs.c.
1157 (process_reg_shuffles): Use REG_SUBREG_P.
1158 (add_insn_allocno_copies): Ditto. Ignore modes.
1160 * ira-color.c (conflict_allocno_vec): New.
1161 (COST_HOP_DIVISOR): New macro.
1162 (update_copy_costs_1): Use it.
1163 (update_conflict_hard_regno_costs): New function.
1164 (assign_hard_reg): Use it.
1165 (ira_color): Allocate and free conflict_allocno_vec.
1167 2008-09-03 Vladimir Makarov <vmakarov@redhat.com>
1171 * ira-int.h (ira_sort_insn_chain): Remove.
1173 * ira.c (basic_block_order_nums, chain_insn_order,
1174 chain_freq_compare, chain_bb_compare, ira_sort_insn_chain): Remove.
1175 (ira): Don't call ira_sort_insn_chain.
1177 * reload1.c (reload): Don't call ira_sort_insn_chain.
1179 2008-09-03 Andrew Pinski <andrew_pinski@playstation.sony.com>
1182 * cgraphunit.c (update_call_expr): Remove eh regions from statements
1183 which become non throw.
1184 (cgraph_function_versioning): Also clear DECL_WEAK. Call
1185 update_call_expr after updating the flags on the decl.
1187 2008-09-03 Jan Hubicka <jh@suse.cz>
1189 PR tree-optimization/37315
1190 * cgraph.c (cgraph_create_edge): Use gimple_has_body_p.
1191 * cgraphunit.c (verify_cgraph_node): drop gimple_body check.
1192 (cgraph_analyze_functions): Use node->analyzed
1193 (cgraph_mark_functions_to_output): Likewise.
1194 (cgraph_expand_function): All functions can be released after
1196 (cgraph_optimize): Use gimple_has_body_p.
1197 * ipa-inline.c (cgraph_clone_inlined_nodes): Use analyzed flag.
1198 (cgraph_decide_inlining_incrementally): Likewise.
1199 (inline_transform): Inline transform.
1200 * tree-inline.c (initialize_cfun): Do now shallow copy structure;
1202 (inlinable_function_p): Drop gimple_body check.
1203 (expand_call_inline): Use gimple_has_body_p.
1204 * gimple.c (gimple_has_body_p): New.
1205 * gimple.h (gimple_has_body_p): Add prototype.
1206 * tree-cfg.c (execute_build_cfg): Remove gimple_body.
1207 (dump_function_to_file): Use gimple_has_body_p check.
1209 2008-09-03 Jakub Jelinek <jakub@redhat.com>
1212 * gimple.h (gimple_has_substatements): GIMPLE_OMP_CRITICAL has
1214 * tree-inline.c (remap_gimple_stmt): Handle GIMPLE_OMP_CRITICAL.
1216 2008-09-03 Bernd Schmidt <bernd.schmidt@analog.com>
1218 From Michael Frysinger <michael.frysinger@analog.com>
1219 * config/bfin/bfin.c (bfin_cpus[]): Add 0.1 for bf522, bf523, bf524,
1220 bf525, bf526, bf527, bf542, bf544, bf547, bf548, and bf549. Add 0.2
1223 2008-09-03 Hari Sandanagobalane <hariharan@picochip.com>
1226 * doc/extend.texi: Document picoChip builtin functions.
1227 * doc/invoke.texi: Document picoChip options.
1228 * doc/contrib.texi: Add picoChip contribution.
1229 * doc/md.texi: Document picoChip constraints.
1230 * config.gcc: Add picochip-*-*.
1231 * config/picochip/: Add new port.
1233 2008-09-03 Richard Guenther <rguenther@suse.de>
1235 PR tree-optimization/37328
1236 * tree-sra.c (sra_build_assignment): Gimplify properly.
1237 (generate_copy_inout): Take the correct stmt as definition,
1238 remove bogus assert.
1240 2008-09-03 Bernd Schmidt <bernd.schmidt@analog.com>
1242 * config/bfin/bfin.c (bfin_expand_prologue): Honour no_stack_limit
1245 From Michael Frysinger <michael.frysinger@analog.com>
1246 * config/bfin/bfin.c (bfin_handle_option): Remove BF561 warning.
1248 2008-09-03 Danny Smith <dannysmith@usrs.sourceforge.net>
1250 * gthr-win32.h (CONST_CAST2): Really make sure CONST_CAST2 is defined.
1251 (__gthread_setspecific): Revert 2008-08-31 change to
1252 __GTHREAD_HIDE_W32API case. Apply it to !__GTHREAD_HIDE_W32API case.
1254 2008-09-02 Aldy Hernandez <aldyh@redhat.com>
1256 * diagnostic.c (error_at): New.
1257 * toplev.h (error_at): New prototype.
1258 * c-typeck.c (build_array_ref): Call error_at instead of error.
1259 Pass location to pedwarn.
1261 2008-09-02 Adam Nemet <anemet@caviumnetworks.com>
1263 * sel-sched.c (sel_hard_regno_rename_ok): Mark arguments unused.
1265 2008-09-02 Jakub Jelinek <jakub@redhat.com>
1267 * dwarf2out.c (tree_add_const_value_attribute): Only handle
1268 VAR_DECL and CONST_DECL.
1270 2008-09-02 H.J. Lu <hongjiu.lu@intel.com>
1271 Jakub Jelinek <jakub@redhat.com>
1273 * config/i386/i386.c (X86_64_VARARGS_SIZE): Removed.
1274 (setup_incoming_varargs_64): Assume cum != NULL. Set/check
1275 ix86_varargs_gpr_size and ix86_varargs_fpr_size. Use
1276 ix86_varargs_gpr_size instead of X86_64_REGPARM_MAX.
1277 Don't set ix86_save_varrargs_registers.
1278 (ix86_setup_incoming_varargs): Assume cum != NULL.
1279 (ix86_va_start): Check ix86_varargs_gpr_size and
1280 ix86_varargs_fpr_size instead of cfun->va_list_gpr_size and
1281 cfun->va_list_fpr_size, respectively. Subtract 8*X86_64_REGPARM_MAX
1282 from frame pointer if ix86_varargs_gpr_size == 0.
1283 (ix86_compute_frame_layout): Updated.
1284 * config/i386/i386.h (ix86_save_varrargs_registers): Removed.
1285 (ix86_varargs_gpr_size): Define.
1286 (ix86_varargs_fpr_size): Likewise.
1287 (machine_function): Remove save_varrargs_registers.
1288 Add varargs_gpr_size and varargs_fpr_size.
1290 2008-09-02 Jakub Jelinek <jakub@redhat.com>
1292 * config/alpha/alpha.c (va_list_skip_additions,
1293 alpha_stdarg_optimize_hook, alpha_gimplify_va_arg_1): Tuplify.
1294 (alpha_gimplify_va_arg): Call unshare_expr on second use of
1297 PR tree-optimization/36766
1298 * tree-cfg.c (gimple_purge_all_dead_eh_edges): Do nothing
1299 for already removed basic blocks.
1302 * real.c (real_maxval): Clear a lower bit to make real_maxval
1303 match get_max_float for IBM long double format.
1305 2008-09-02 Andreas Schwab <schwab@suse.de>
1307 * configure.ac: Use m4_quote to quote the expansion of m4_do.
1308 * configure: Regenerate.
1310 2008-09-02 Richard Sandiford <rdsandiford@googlemail.com>
1312 * simplify-rtx.c (simplify_binary_operation_1): Check for CONST,
1313 SYMBOL_REF and LABEL_REF when applying plus_constant, instead of
1314 checking for a non-VOID CONSTANT_P.
1316 2008-09-02 Jan Hubicka <jh@suse.cz>
1318 * predict.c (pass_strip_predict_hints): Avoid bugs dump file.
1319 * expmed.c (expand_divmod): Fix test of smod_pow2_cheap and
1322 2008-09-02 H.J. Lu <hongjiu.lu@intel.com>
1324 * expr.c (emit_group_store): Don't assert stack temp mode size.
1326 2008-09-02 Sebastian Pop <sebastian.pop@amd.com>
1327 Tobias Grosser <grosser@fim.uni-passau.de>
1328 Jan Sjodin <jan.sjodin@amd.com>
1329 Harsha Jagasia <harsha.jagasia@amd.com>
1330 Dwarakanath Rajagopal <dwarak.rajagopal@amd.com>
1331 Konrad Trifunovic <konrad.trifunovic@inria.fr>
1332 Adrien Eliche <aeliche@isty.uvsq.fr>
1334 Merge from graphite branch.
1337 * tree-loop-linear.c (perfect_loop_nest_depth): Export.
1338 * doc/invoke.texi (-floop-block, -floop-interchange,
1339 -floop-strip-mine): Document new flags.
1340 * tree-into-ssa.c (gimple_vec): Moved...
1341 * tree-loop-distribution.c (rdg_component): Moved...
1342 * cfgloopmanip.c: Include tree-flow.h.
1343 (update_dominators_in_loop): New.
1344 (create_empty_if_region_on_edge): New.
1345 (create_empty_loop_on_edge): New.
1346 (loopify): Use update_dominators_in_loop.
1347 * tree-pass.h (pass_graphite_transforms): Declared.
1348 * configure: Regenerate.
1349 * tree-phinodes.c (make_phi_node): Export.
1350 (add_phi_node_to_bb): New, split from create_phi_node.
1351 * tree-chrec.c (for_each_scev_op): New.
1352 * tree-chrec.h (for_each_scev_op): Declared.
1353 * tree-ssa-loop-ivopts.c (get_phi_with_result): New.
1354 (remove_statement): Call get_phi_with_result.
1355 * config.in (HAVE_cloog): Undef.
1356 * gdbinit.in (pgg): New.
1357 * timevar.def (TV_GRAPHITE_TRANSFORMS): New.
1358 * tree-ssa-loop.c (graphite_transforms): New.
1359 (gate_graphite_transforms): New.
1360 (pass_graphite_transforms): New.
1361 * configure.ac (PPLLIBS, PPLINC, CLOOGLIBS, CLOOGINC,
1362 HAVE_cloog): Defined.
1363 * tree-vectorizer.c (rename_variables_in_bb): Export.
1364 * tree-data-ref.c (dr_may_alias_p): Export.
1365 (stmt_simple_memref_p): New.
1366 (find_data_references_in_stmt): Export.
1367 (find_data_references_in_loop): Export.
1368 (create_rdg_edge_for_ddr): Initialize RDGE_RELATION.
1369 (create_rdg_edges_for_scalar): Initialize RDGE_RELATION.
1370 (create_rdg_vertices): Export.
1371 (build_empty_rdg): New.
1372 (build_rdg): Call build_empty_rdg. Free dependence_relations.
1373 * tree-data-ref.h (rdg_component): ... here.
1375 (struct data_reference): Add a field scop.
1377 (find_data_references_in_loop): Declared.
1378 (find_data_references_in_stmt): Declared.
1379 (create_rdg_vertices): Declared.
1380 (dr_may_alias_p): Declared.
1381 (stmt_simple_memref_p): Declared.
1382 (struct rdg_edge): Add a field ddr_p relation.
1383 (build_empty_rdg): Declared.
1384 * lambda.h (lambda_matrix): Declare a VEC of.
1385 (find_induction_var_from_exit_cond): Declared.
1386 (lambda_vector_compare): New.
1387 * common.opt (fgraphite, floop-strip-mine,
1388 floop-interchange, floop-block): New flags.
1389 * lambda-code.c (find_induction_var_from_exit_cond): Export.
1390 * cfgloop.c (is_loop_exit): New.
1391 * cfgloop.h (is_loop_exit): Declared.
1392 (create_empty_if_region_on_edge): Declared.
1393 (create_empty_loop_on_edge): Declared.
1394 * tree-flow.h (add_phi_node_to_bb): Declared.
1395 (make_phi_node): Declared.
1396 (rename_variables_in_bb): Declared.
1397 (perfect_loop_nest_depth): Declared.
1398 (graphite_transform_loops): Declared.
1399 * Makefile.in (cfgloopmanip.o): Depend on TREE_FLOW_H.
1400 (graphite.o-warn): Add -Wno-error.
1401 (PPLLIBS, PPLINC, CLOOGLIBS, CLOOGINC): Declared.
1402 (LIBS): Add GMPLIBS, CLOOGLIBS, PPLLIBS.
1403 (INCLUDES): Add PPLINC, CLOOGINC.
1404 (OBJS-common): Add graphite.o.
1405 (graphite.o): Add rule.
1406 * gimple.h (gimple_vec): ... here.
1407 * tree-cfg.c (print_loops): Start printing at ENTRY_BLOCK_PTR.
1408 * passes.c (init_optimization_passes): Schedule
1409 pass_graphite_transforms.
1411 2008-09-02 Samuel Tardieu <sam@rfc1149.net>
1414 * opts.c (decode_options): Move processing of -fno-unit-at-a-time
1415 and -fno-toplevel-reorder after handle_options.
1417 2008-09-02 Paul Brook <paul@codesourcery.com>
1419 * config/arm/arm.c (arm_tune_cortex_a9): Define.
1420 (arm_override_options): Set arm_tune_cortex_a9.
1421 (arm_rtx_costs_1): Make register shifts more expensive on Cortex-A9.
1422 (arm_issue_rate): Handle cortexa9.
1423 * config/arm/arm.h (arm_tune_cortex_a9): Declare.
1424 * config/arm/arm-cores.def: Add cortex-a9.
1425 * config/arm/arm-tune.md: Regenerate.
1426 * config/arm/arm.md: Include cortex-a9.md.
1427 (generic_sched, generic_vfp): Add cortexa9.
1428 * config/arm/cortex-a9.md: New file.
1430 2008-09-02 Richard Guenther <rguenther@suse.de>
1432 PR tree-optimization/37327
1433 * tree-vrp.c (register_new_assert_for): Make sure to not have
1434 TREE_OVERFLOW set on the bound.
1436 2008-09-02 Alexander Monakov <amonakov@ispras.ru>
1438 Fixed typo in date of the 2008-09-01 MOVE_RATIO commit.
1440 * config/ia64/ia64.opt (msched-ar-data-spec): Default to 1.
1441 * config/ia64/ia64.c (ia64_set_sched_flags): Always initialize
1444 2008-09-02 Victor Kaplansky <victork@il.ibm.com>
1446 * gcc/config/spu/spu.md (divdf3): Removed.
1448 2008-09-02 Jakub Jelinek <jakub@redhat.com>
1450 PR tree-optimization/37095
1451 * cgraph.c (cgraph_node): When creating new cgraph node after
1452 assembler_name_hash has been populated, record it in the hash
1455 2008-09-01 Paul Brook <paul@codesourcery.com>
1457 * doc/invoke.texi: Document -mword-relocations.
1458 * config/arm/uclinux-elf.h (TARGET_DEFAULT_WORD_RELOCATIONS): Define.
1459 * config/arm/symbian.h (TARGET_DEFAULT_WORD_RELOCATIONS): Define.
1460 * config/arm/vxworks.h (TARGET_DEFAULT_WORD_RELOCATIONS): Define.
1461 * config/arm/arm.c (arm_split_constant): Use arm_emit_movpair.
1462 (arm_rtx_costs_1, arm_size_rtx_costs): Handle HIGH and LO_SUM.
1463 (arm_emit_movpair): New function.
1464 (arm_print_operand <c>): Handle sybolic addresses.
1465 * config/arm/arm.h (TARGET_USE_MOVT): Define.
1466 (TARGET_DEFAULT_WORD_RELOCATIONS): Define.
1467 * config/arm/arm-protos.h (arm_emit_movpair): Add prototype.
1468 * config/arm/arm.opt: Add -mword-relocations.
1469 * config/arm/arm.md (movsi): Use arm_emit_movpair.
1470 (arm_movt, arm_movw): New.
1472 2008-09-01 Paul Brook <paul@codesourcery.com>
1474 * config/arm/arm.c (arm_override_options): Set arm_abi earlier.
1475 Allow Interworking on ARMv4 EABI based targets.
1476 * config/arm/bpabi.h (TARGET_FIX_V4BX_SPEC): Define.
1477 (SUBTARGET_EXTRA_ASM_SPEC, LINK_SPEC): Add TARGET_FIX_V4BX_SPEC.
1479 2008-09-01 Paul Brook <paul@codesourcery.com>
1481 * config/arm/thumb2.md (thumb2_negscc): Match the correct operand for
1482 optimized LT0 test. Remove optimization for GT.
1484 2008-09-01 John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
1486 * pa64-hpux.h (LIB_SPEC): Link against librt when building static
1487 openmp applications.
1488 * pa-hpux11.h (LIB_SPEC): Likewise.
1490 * ira.c (setup_eliminable_regset): Only declare i if ELIMINABLE_REGS is
1493 2008-09-01 Jan Hubicka <jh@suse.cz>
1495 * predict.c (maybe_hot_count_p, maybe_hot_edge_p): When profile is
1496 present, use only counts; when guessed use only frequencies.
1497 (estimate_bb_frequencies): Use function_frequency.
1498 * profile.c (compute_branch_probabilities): Exit early when
1499 profile reading failed. Set profile_status.
1500 (branch_prob): Do not set profile_status.
1502 2008-09-01 Aldy Hernandez <aldyh@redhat.com>
1504 * tree.c (protected_set_expr_location): New.
1505 * tree.h (protected_set_expr_location): New prototype.
1506 * c-tree.h (build_array_ref): Add argument.
1507 (parser_build_unary_op): Same.
1508 * c-typeck.c (build_indirect_ref): Handle new location argument.
1509 (build_array_ref): Same.
1510 (parser_build_unary_op): Same.
1511 * gimplify.c (gimplify_asm_expr): Set input_location before calling
1513 * c-omp.c (c_finish_omp_atomic): Pass location when calling
1515 * c-common.c (finish_label_address_expr): Handle new location argument.
1516 * c-common.h (build_indirect_ref): Add argument.
1517 (finish_label_address_expr): Same.
1518 * c-parser.c (c_parser_unary_expression): Pass location to build
1520 (c_parser_postfix_expression): Same.
1522 2008-09-01 Paul Brook <paul@codesourcery.com>
1524 * config/arm/arm.md: Include cortex-r4f.md.
1525 (attr fpu): Update type list.
1526 (attr type): Add fcpys, ffariths, ffarithd, fadds, faddd, fconsts,
1527 fconstd, fcmps and fcmpd.
1528 (attr tune_cortexr4): Define.
1529 (attr generic_sched, attr generic_vfp): Use tune_cortexr4.
1530 * config/arm/vfp.md: Document fcpys, ffariths, ffarithd, fadds, faddd,
1531 fconsts, fconstd, fcmps and fcmpd. Use them in insn patterns.
1532 * config/arm/arm.c (arm_issue_rate): Add cortexr4f.
1533 * config/arm/arm1020e.md (v10_ffarith, v10_farith): Use new insn types.
1534 * config/arm/cortex-a8-neon.md (cortex_a8_vfp_add_sub,
1535 cortex_a8_vfp_farith: Ditto.
1536 * config/arm/vfp11.md (vfp_ffarith, vfp_farith): Ditto.
1537 * config/arm/cortex-r4.md: Use tune_cortexr4.
1538 * config/arm/cortex-r4f.md: New file.
1540 2008-09-01 Richard Guenther <rguenther@suse.de>
1542 PR tree-optimization/37305
1543 * tree-ssa-ccp.c (ccp_fold): Do not set TREE_OVERFLOW on
1544 the result of constant conversions.
1545 (fold_gimple_assign): Likewise.
1547 2008-09-01 Andrey Belevantsev <abel@ispras.ru>
1549 * sel-sched-ir.c (cmp_v_in_regset_pool): Surround with
1550 #ifdef ENABLE_CHECKING.
1552 2008-09-01 Andrey Belevantsev <abel@ispras.ru>
1554 * sel-sched-dump.c: Surround with #ifdef INSN_SCHEDULING.
1555 * sched-int.h (print_insn, print_pattern, print_value): Declare also
1556 when !INSN_SCHEDULING.
1558 2008-09-01 Andreas Schwab <schwab@suse.de>
1560 * config/ia64/ia64.c (TARGET_ADDRESS_COST): Update.
1562 2008-09-01 Paul Brook <paul@codesourcery.com>
1564 * config/arm/arm.c (arm_no_early_mul_dep): Handle multiply-subtract.
1566 2008-09-01 Jan Hubicka <jh@suse.cz>
1568 * config/spu/spu.c (spu_expand_block_move) Update call of MOVE_RATIO.
1570 * expr.c (MOVE_BY_PIECES_P, CLEAR_BY_PIECES_P, SET_BY_PIECES_P):
1572 * expr.h (MOVE_RATIO, CLEAR_RATIO, SET_RATIO): Update.
1573 * gimplify.c (gimplify_init_constructor): Add speed operand.
1574 * tree-sra.c (decide_block_copy): Likewise.
1575 * tree-inline.c (estimate_move_cost): Likewise.
1576 * config/alpha/alpha.h (MOVE_RATIO): Update.
1577 * config/frv/frv.c (MOVE_RATIO): Update.
1578 * config/spu/spu.h (MOVE_RATIO): Update.
1579 * config/sparc/sparc.h (MOVE_RATIO): Update.
1580 * config/i386/i386.h (MOVE_RATIO, CLEAR_RATIO): Update.
1581 * config/m68hc11/m68hc11.h (MOVE_RATIO): Update.
1582 * config/cris/cris.h (MOVE_RATIO): Update.
1583 * config/mn10300/mn10300.h (MOVE_RATIO): Update.
1584 * config/arm/arm.h (MOVE_RATIO): Update.
1585 * config/pa/pa.md: Update uses of MOVE_RATIO
1586 * config/pa/pa.h (MOVE_RATIO): Update.
1587 * config/mips/mips.h (MOVE_RATIO, MOVE_BY_PIECES,
1588 CLEAR_RATIO, SET_RATIO): Update.
1589 * config/h8300/h8300.h (MOVE_RATIO): Update.
1590 * config/v850/v850.h (MOVE_RATIO): Update.
1591 * config/bfin/bfin.h (MOVE_RATIO): Update.
1593 2008-08-31 Andrey Belevantsev <abel@ispras.ru>
1594 Dmitry Melnik <dm@ispras.ru>
1595 Dmitry Zhurikhin <zhur@ispras.ru>
1596 Alexander Monakov <amonakov@ispras.ru>
1597 Maxim Kuvyrkov <maxim@codesourcery.com>
1599 * sel-sched.h, sel-sched-dump.h, sel-sched-ir.h, sel-sched.c,
1600 sel-sched-dump.c, sel-sched-ir.c: New files.
1601 * Makefile.in (OBJS-common): Add selective scheduling object
1603 (sel-sched.o, sel-sched-dump.o, sel-sched-ir.o): New entries.
1604 (SEL_SCHED_IR_H, SEL_SCHED_DUMP_H): New entries.
1605 (sched-vis.o): Add dependency on $(INSN_ATTR_H).
1606 * cfghooks.h (get_cfg_hooks, set_cfg_hooks): New prototypes.
1607 * cfghooks.c (get_cfg_hooks, set_cfg_hooks): New functions.
1608 (make_forwarder_block): Update loop latch if we have redirected
1609 the loop latch edge.
1610 * cfgloop.c (get_loop_body_in_custom_order): New function.
1611 * cfgloop.h (LOOPS_HAVE_FALLTHRU_PREHEADERS): New enum field.
1612 (CP_FALLTHRU_PREHEADERS): Likewise.
1613 (get_loop_body_in_custom_order): Declare.
1614 * cfgloopmanip.c (has_preds_from_loop): New.
1615 (create_preheader): Honor CP_FALLTHRU_PREHEADERS.
1616 Assert that the preheader edge will be fall thru when it is set.
1617 * common.opt (fsel-sched-bookkeeping, fsel-sched-pipelining,
1618 fsel-sched-pipelining-outer-loops, fsel-sched-renaming,
1619 fsel-sched-substitution, fselective-scheduling): New flags.
1620 * cse.c (hash_rtx_cb): New.
1622 * dbgcnt.def (sel_sched_cnt, sel_sched_region_cnt,
1623 sel_sched_insn_cnt): New counters.
1624 * final.c (compute_alignments): Export. Free dominance
1625 info after loop_optimizer_finalize.
1626 * genattr.c (main): Output maximal_insn_latency prototype.
1627 * genautomata.c (output_default_latencies): New.
1628 Factor its code from ...
1629 (output_internal_insn_latency_func): ... here.
1630 (output_internal_maximal_insn_latency_func): New.
1631 (output_maximal_insn_latency_func): New.
1632 * hard-reg-set.h (UHOST_BITS_PER_WIDE_INT): Define unconditionally.
1633 (struct hard_reg_set_iterator): New.
1634 (hard_reg_set_iter_init, hard_reg_set_iter_set,
1635 hard_reg_set_iter_next): New functions.
1636 (EXECUTE_IF_SET_IN_HARD_REG_SET): New macro.
1637 * lists.c (remove_free_INSN_LIST_node,
1638 remove_free_EXPR_LIST_node): New functions.
1639 * loop-init.c (loop_optimizer_init): When
1640 LOOPS_HAVE_FALLTHRU_PREHEADERS, set CP_FALLTHRU_PREHEADERS when
1641 calling create_preheaders.
1642 (loop_optimizer_finalize): Do not verify flow info after reload.
1643 * recog.c (validate_replace_rtx_1): New parameter simplify.
1644 Default it to true. Update all uses. Factor out simplifying
1646 (simplify_while_replacing): ... this new function.
1647 (validate_replace_rtx_part,
1648 validate_replace_rtx_part_nosimplify): New.
1649 * recog.h (validate_replace_rtx_part,
1650 validate_replace_rtx_part_nosimplify): Declare.
1651 * rtl.c (rtx_equal_p_cb): New.
1652 (rtx_equal_p): Use it.
1653 * rtl.h (rtx_equal_p_cb, hash_rtx_cb): Declare.
1654 (remove_free_INSN_LIST_NODE, remove_free_EXPR_LIST_node,
1655 debug_bb_n_slim, debug_bb_slim, print_rtl_slim): Likewise.
1656 * vecprim.h: Add a vector type for unsigned int.
1657 * haifa-sched.c: Include vecprim.h and cfgloop.h.
1658 (issue_rate, sched_verbose_param, note_list, dfa_state_size,
1659 ready_try, cycle_issued_insns, spec_info): Make global.
1660 (readyp): Initialize.
1661 (dfa_lookahead): New global variable.
1662 (old_max_uid, old_last_basic_block): Remove.
1663 (h_i_d): Make it a vector.
1664 (INSN_TICK, INTER_TICK, QUEUE_INDEX, INSN_COST): Make them work
1666 (after_recovery, adding_bb_to_current_region_p):
1667 New variables to handle correct insertion of the recovery code.
1668 (struct ready_list): Move declaration to sched-int.h.
1669 (rgn_n_insns): Removed.
1670 (rtx_vec_t): Move to sched-int.h.
1671 (find_insn_reg_weight): Remove.
1672 (find_insn_reg_weight1): Rename to find_insn_reg_weight.
1673 (haifa_init_h_i_d, haifa_finish_h_i_d):
1674 New functions to initialize / finalize haifa instruction data.
1675 (extend_h_i_d, init_h_i_d): Rewrite.
1676 (unlink_other_notes): Move logic to add_to_note_list. Handle
1677 selective scheduler.
1678 (ready_lastpos, ready_element, ready_sort, reemit_notes,
1679 find_fallthru_edge): Make global, remove static prototypes.
1680 (max_issue): Make global. Add privileged_n and state parameters. Use
1682 (extend_global, extend_all): Removed.
1683 (init_before_recovery): Add new param. Fix the handling of the case
1684 when we insert a recovery code before the EXIT which has a predecessor
1685 with a fallthrough edge to it.
1686 (create_recovery_block): Make global. Rename to
1687 sched_create_recovery_block. Update.
1688 (change_pattern): Rename to sched_change_pattern. Make global.
1689 (speculate_insn): Rename to sched_speculate_insn. Make global.
1690 Split haifa-specific functionality into ...
1691 (haifa_change_pattern): New static function.
1692 (sched_extend_bb): New static function.
1693 (sched_init_bbs): New function.
1694 (current_sched_info): Change type to struct haifa_sched_info.
1695 (insn_cost): Adjust for selective scheduling.
1696 (dep_cost_1): New function. Move logic from ...
1697 (dep_cost): ... here.
1698 (dep_cost): Use dep_cost_1.
1699 (contributes_to_priority_p): Use sched_deps_info instead of
1701 (priority): Adjust to work with selective scheduling. Process the
1702 corner case when all dependencies don't contribute to priority.
1703 (rank_for_schedule): Use ds_weak instead of dep_weak.
1704 (advance_state): New function. Move logic from ...
1705 (advance_one_cycle): ... here.
1706 (add_to_note_list, concat_note_lists): New functions.
1707 (rm_other_notes): Make static. Adjust for selective scheduling.
1708 (remove_notes, restore_other_notes): New functions.
1709 (move_insn): Add two arguments. Update assert. Don't call
1711 (choose_ready): Remove lookahead variable, use dfa_lookahead.
1712 Remove more_issue, max_points. Move the code to initialize
1713 max_lookahead_tries to max_issue.
1714 (schedule_block): Remove rgn_n_insns1 parameter. Don't allocate
1715 ready. Adjust use of move_insn. Call restore_other_notes.
1717 (sched_init, sched_finish): Move Haifa-specific initialization/
1719 (haifa_sched_init, haifa_sched_finish): ... respectively.
1721 (setup_sched_dump): New function.
1722 (haifa_init_only_bb): New static function.
1723 (haifa_speculate_insn): New static function.
1724 (try_ready): Use haifa_* instead of speculate_insn and
1726 (extend_ready, extend_all): Remove.
1727 (sched_extend_ready_list, sched_finish_ready_list): New functions.
1728 (create_check_block_twin, add_to_speculative_block): Use
1729 haifa_insns_init instead of extend_global. Update to use new
1730 initialization functions. Change parameter. Factor out code from
1731 create_check_block_twin to ...
1732 (sched_create_recovery_edges) ... this new function.
1733 (add_block): Remove.
1734 (sched_scan_info): New.
1735 (extend_bb): Use sched_scan_info.
1736 (init_bb, extend_insn, init_insn, init_insns_in_bb, sched_scan): New
1737 static functions for walking through scheduling region.
1738 (sched_luids): New vector variable to replace uid_to_luid.
1739 (luids_extend_insn): New function.
1740 (sched_max_luid): New variable.
1741 (luids_init_insn): New function.
1742 (sched_init_luids, sched_finish_luids): New functions.
1743 (insn_luid): New debug function.
1744 (sched_extend_target): New function.
1745 (haifa_init_insn): New static function.
1746 (sched_init_only_bb): New hook.
1747 (sched_split_block): New hook.
1748 (sched_split_block_1): New function.
1749 (sched_create_empty_bb): New hook.
1750 (sched_create_empty_bb_1): New function.
1751 (common_sched_info, ready): New global variables.
1752 (current_sched_info_var): Remove.
1753 (move_block_after_check): Use common_sched_info.
1754 (haifa_luid_for_non_insn): New static function.
1755 (init_before_recovery): Use haifa_init_only_bb instead of
1757 (increase_insn_priority): New.
1758 * modulo-sched.c: (issue_rate): Remove static declaration.
1759 (sms_sched_info): Change type to haifa_sched_info.
1760 (sms_sched_deps_info, sms_common_sched_info): New variables.
1761 (setup_sched_infos): New.
1762 (sms_schedule): Initialize them. Call haifa_sched_init/finish.
1763 Do not call regstat_free_calls_crossed.
1764 (sms_print_insn): Use const_rtx.
1765 * params.def (PARAM_MAX_PIPELINE_REGION_BLOCKS,
1766 PARAM_MAX_PIPELINE_REGION_INSNS, PARAM_SELSCHED_MAX_LOOKAHEAD,
1767 PARAM_SELSCHED_MAX_SCHED_TIMES, PARAM_SELSCHED_INSNS_TO_RENAME,
1768 PARAM_SCHED_MEM_TRUE_DEP_COST): New.
1769 * sched-deps.c (sched_deps_info): New. Update all relevant uses of
1770 current_sched_info to use it.
1771 (enum reg_pending_barrier_mode): Move to sched-int.h.
1772 (h_d_i_d): New variable. Initialize to NULL.
1773 ({true, output, anti, spec, forward}_dependency_cache): Initialize
1775 (estimate_dep_weak): Remove static declaration.
1776 (sched_has_condition_p): New function. Adjust users of
1777 sched_get_condition to use it instead.
1778 (conditions_mutex_p): Add arguments indicating which conditions are
1780 (sched_get_condition_with_rev): Rename from sched_get_condition. Add
1781 argument to indicate whether returned condition is reversed. Do not
1782 generate new rtx when condition should be reversed; indicate it by
1783 setting new argument instead.
1784 (add_dependence_list_and_free): Add deps parameter.
1785 Update all users. Do not free dependence list when
1786 deps context is readonly.
1787 (add_insn_mem_dependence, flush_pending_lists): Adjust for readonly
1789 (remove_from_dependence_list, remove_from_both_dependence_lists): New.
1790 (remove_from_deps): New. Use the above functions.
1791 (cur_insn, can_start_lhs_rhs_p): New static variables.
1792 (add_or_update_back_dep_1): Initialize present_dep_type.
1793 (haifa_start_insn, haifa_finish_insn, haifa_note_reg_set,
1794 haifa_note_reg_clobber, haifa_note_reg_use, haifa_note_mem_dep,
1795 haifa_note_dep): New functions implementing dependence hooks for
1796 the Haifa scheduler.
1797 (note_reg_use, note_reg_set, note_reg_clobber, note_mem_dep,
1798 note_dep): New functions.
1799 (ds_to_dt, extend_deps_reg_info, maybe_extend_reg_info_p): New
1801 (init_deps): Initialize last_reg_pending_barrier and deps->readonly.
1802 (free_deps): Initialize deps->reg_last.
1803 (sched_analyze_reg, sched_analyze_1, sched_analyze_2,
1804 sched_analyze_insn): Update to use dependency hooks infrastructure
1805 and readonly contexts.
1806 (deps_analyze_insn): New function. Move part of logic from ...
1807 (sched_analyze): ... here. Also move some logic to ...
1808 (deps_start_bb): ... here. New function.
1809 (add_forw_dep, delete_forw_dep): Guard use of INSN_DEP_COUNT with
1811 (sched_deps_init): New function. Move code from ...
1812 (init_dependency_caches): ... here. Remove.
1813 (init_deps_data_vector): New.
1814 (sched_deps_finish): New function. Move code from ...
1815 (free_dependency_caches): ... here. Remove.
1816 (init_deps_global, finish_deps_global): Adjust for use with
1817 selective scheduling.
1818 (get_dep_weak): Move logic to ...
1819 (get_dep_weak_1): New function.
1820 (ds_merge): Move logic to ...
1821 (ds_merge_1): New static function.
1822 (ds_full_merge, ds_max_merge, ds_get_speculation_types): New functions.
1823 (ds_get_max_dep_weak): New function.
1824 * sched-ebb.c (sched_n_insns): Rename to sched_rgn_n_insns.
1825 (n_insns): Rename to rgn_n_insns.
1826 (debug_ebb_dependencies): New function.
1827 (init_ready_list): Use it.
1828 (begin_schedule_ready): Use sched_init_only_bb.
1829 (ebb_print_insn): Indicate when an insn starts a new cycle.
1830 (contributes_to_priority, compute_jump_reg_dependencies,
1831 add_remove_insn, fix_recovery_cfg): Add ebb_ prefix to function names.
1832 (add_block1): Remove to ebb_add_block.
1833 (ebb_sched_deps_info, ebb_common_sched_info): New variables.
1834 (schedule_ebb): Initialize them. Use remove_notes instead of
1835 rm_other_notes. Use haifa_local_init/finish.
1836 (schedule_ebbs): Use haifa_sched_init/finish.
1837 * sched-int.h: Include vecprim.h, remove rtl.h.
1838 (struct ready_list): Delete declaration.
1839 (sched_verbose_param, enum sched_pass_id_t,
1840 bb_vec_t, insn_vec_t, rtx_vec_t): New.
1841 (struct sched_scan_info_def): New structure.
1842 (sched_scan_info, sched_scan, sched_init_bbs,
1843 sched_init_luids, sched_finish_luids, sched_extend_target,
1844 haifa_init_h_i_d, haifa_finish_h_i_d): Declare.
1845 (struct common_sched_info_def): New.
1846 (common_sched_info, haifa_common_sched_info,
1847 sched_emulate_haifa_p): Declare.
1849 (sched_luids): Declare.
1850 (INSN_LUID, LUID_BY_UID, SET_INSN_LUID): Declare.
1851 (sched_max_luid, insn_luid): Declare.
1852 (note_list, remove_notes, restore_other_notes, bb_note): Declare.
1853 (sched_insns_init, sched_insns_finish, xrecalloc, reemit_notes,
1854 print_insn, print_pattern, print_value, haifa_classify_insn,
1855 sel_find_rgns, sel_mark_hard_insn, dfa_state_size, advance_state,
1856 setup_sched_dump, sched_init, sched_finish,
1857 sel_insn_is_speculation_check): Export.
1858 (struct ready_list): Move from haifa-sched.c.
1859 (ready_try, ready, max_issue): Export.
1860 (ebb_compute_jump_reg_dependencies, find_fallthru_edge,
1861 sched_init_only_bb, sched_split_block, sched_split_block_1,
1862 sched_create_empty_bb, sched_create_empty_bb_1,
1863 sched_create_recovery_block, sched_create_recovery_edges): Export.
1864 (enum reg_pending_barrier_mode): Export.
1865 (struct deps): New fields `last_reg_pending_barrier' and `readonly'.
1867 (struct sched_info): Rename to haifa_sched_info. Use const_rtx for
1868 print_insn field. Move add_block and fix_recovery_cfg to
1869 common_sched_info_def. Move compute_jump_reg_dependencies,
1871 (struct sched_deps_info_def): ... this new structure.
1872 (sched_deps_info): Declare.
1873 (struct spec_info_def): Remove weakness_cutoff, add
1874 data_weakness_cutoff and control_weakness_cutoff.
1875 (spec_info): Declare.
1876 (struct _haifa_deps_insn_data): Split from haifa_insn_data. Add
1878 (struct haifa_insn_data): Rename to struct _haifa_insn_data.
1879 (haifa_insn_data_def, haifa_insn_data_t): New typedefs.
1880 (current_sched_info): Change type to struct haifa_sched_info.
1881 (haifa_deps_insn_data_def, haifa_deps_insn_data_t): New typedefs.
1882 (h_d_i_d): New variable.
1883 (HDID): New accessor macro.
1884 (h_i_d): Change type to VEC (haifa_insn_data_def, heap) *.
1885 (HID): New accessor macro. Rewrite h_i_d accessor macros through HID
1887 (IS_SPECULATION_CHECK_P): Update for selective scheduler.
1888 (enum SCHED_FLAGS): Update for selective scheduler.
1889 (enum SPEC_SCHED_FLAGS): New flag SEL_SCHED_SPEC_DONT_CHECK_CONTROL.
1890 (init_dependency_caches, free_dependency_caches): Delete declarations.
1891 (deps_analyze_insn, remove_from_deps, get_dep_weak_1,
1892 estimate_dep_weak, ds_full_merge, ds_max_merge, ds_weak,
1893 ds_get_speculation_types, ds_get_max_dep_weak, sched_deps_init,
1894 sched_deps_finish, haifa_note_reg_set, haifa_note_reg_use,
1895 haifa_note_reg_clobber, maybe_extend_reg_info_p, deps_start_bb,
1897 (rm_other_notes): Delete declaration.
1898 (schedule_block): Remove one argument.
1899 (cycle_issued_insns, issue_rate, dfa_lookahead, ready_sort,
1900 ready_element, ready_lastpos, sched_extend_ready_list,
1901 sched_finish_ready_list, sched_change_pattern, sched_speculate_insn,
1902 concat_note_lists): Export.
1903 (struct region): Move from sched-rgn.h.
1904 (nr_regions, rgn_table, rgn_bb_table, block_to_bb, containing_rgn,
1905 RGN_NR_BLOCKS, RGN_BLOCKS, RGN_DONT_CALC_DEPS, RGN_HAS_REAL_EBB,
1906 BLOCK_TO_BB, CONTAINING_RGN): Export.
1907 (ebb_head, BB_TO_BLOCK, EBB_FIRST_BB, EBB_LAST_BB, INSN_BB): Likewise.
1908 (current_nr_blocks, current_blocks, target_bb): Likewise.
1909 (dep_cost_1, sched_is_disabled_for_current_region_p, sched_rgn_init,
1910 sched_rgn_finish, rgn_setup_region, sched_rgn_compute_dependencies,
1911 sched_rgn_local_init, extend_regions,
1912 rgn_make_new_region_out_of_new_block, compute_priorities,
1913 debug_rgn_dependencies, free_rgn_deps, contributes_to_priority,
1914 extend_rgns, deps_join rgn_setup_common_sched_info,
1915 rgn_setup_sched_infos, debug_regions, debug_region, dump_region_dot,
1916 dump_region_dot_file, haifa_sched_init, haifa_sched_finish): Export.
1917 (get_rgn_sched_max_insns_priority, sel_add_to_insn_priority,
1918 increase_insn_priority): Likewise.
1919 * sched-rgn.c: Include sel-sched.h.
1920 (ref_counts): New static variable. Use it ...
1921 (INSN_REF_COUNT): ... here. Rewrite and move closer to uses.
1922 (FED_BY_SPEC_LOAD, IS_LOAD_INSN): Rewrite to use HID accessor macro.
1923 (sched_is_disabled_for_current_region_p): Delete static declaration.
1924 (struct region): Move to sched-int.h.
1925 (nr_regions, rgn_table, rgn_bb_table, block_to_bb, containing_rgn,
1926 ebb_head): Define and initialize.
1927 (RGN_NR_BLOCKS, RGN_BLOCKS, RGN_DONT_CALC_DEPS, RGN_HAS_REAL_EBB,
1928 BLOCK_TO_BB, CONTAINING_RGN, debug_regions, extend_regions,
1929 BB_TO_BLOCK, EBB_FIRST_BB, EBB_LAST_BB): Move to
1931 (find_single_block_region): Add new argument to indicate that EBB
1932 regions should be constructed.
1933 (debug_live): Delete declaration.
1934 (current_nr_blocks, current_blocks, target_bb):
1935 Remove static qualifiers.
1936 (compute_dom_prob_ps, check_live, update_live, set_spec_fed): Delete
1938 (init_regions): Delete declaration.
1939 (debug_region, bb_in_region_p, dump_region_dot_file, dump_region_dot,
1940 rgn_estimate_number_of_insns): New.
1941 (too_large): Use estimate_number_of_insns.
1942 (haifa_find_rgns): New. Move the code from ...
1943 (find_rgns): ... here. Call either sel_find_rgns or haifa_find_rgns.
1944 (free_trg_info): New.
1945 (compute_trg_info): Allocate candidate tables here instead of ...
1946 (init_ready_list): ... here.
1947 (rgn_print_insn): Use const_rtx.
1948 (contributes_to_priority, extend_regions): Delete static declaration.
1949 (add_remove_insn, fix_recovery_cfg): Add rgn_ to function names.
1950 (add_block1): Rename to rgn_add_block.
1951 (debug_rgn_dependencies): Delete static qualifier.
1952 (new_ready): Use sched_deps_info. Simplify.
1953 (rgn_common_sched_info, rgn_const_sched_deps_info,
1954 rgn_const_sel_sched_deps_info, rgn_sched_deps_info, rgn_sched_info):
1956 (region_sched_info): Rename to rgn_const_sched_info.
1957 (deps_join): New, extracted from ...
1958 (propagate_deps): ... here.
1959 (compute_block_dependences, debug_dependencies): Update for selective
1961 (free_rgn_deps, compute_priorities): New functions.
1962 (sched_rgn_init, sched_rgn_finish, rgn_setup_region,
1963 sched_rgn_compute_dependencies): New functions.
1964 (schedule_region): Use them.
1965 (sched_rgn_local_init, sched_rgn_local_free, sched_rgn_local_finish,
1966 rgn_setup_common_sched_info, rgn_setup_sched_infos):
1968 (schedule_insns): Call new functions that were split out.
1969 (rgn_make_new_region_out_of_new_block): New.
1970 (get_rgn_sched_max_insns_priority): New.
1971 (rest_of_handle_sched, rest_of_handle_sched2): Call selective
1972 scheduling when appropriate.
1973 * sched-vis.c: Include insn-attr.h.
1974 (print_value, print_pattern): Make global.
1975 (print_rtl_slim, debug_bb_slim, debug_bb_n_slim): New functions.
1976 * target-def.h (TARGET_SCHED_ADJUST_COST_2,
1977 TARGET_SCHED_ALLOC_SCHED_CONTEXT, TARGET_SCHED_INIT_SCHED_CONTEXT,
1978 TARGET_SCHED_SET_SCHED_CONTEXT, TARGET_SCHED_CLEAR_SCHED_CONTEXT,
1979 TARGET_SCHED_FREE_SCHED_CONTEXT, TARGET_SCHED_GET_INSN_CHECKED_DS,
1980 TARGET_SCHED_GET_INSN_SPEC_DS, TARGET_SCHED_SKIP_RTX_P): New target
1981 hooks. Initialize them to 0.
1982 (TARGET_SCHED_GEN_CHECK): Rename to TARGET_SCHED_GEN_SPEC_CHECK.
1983 * target.h (struct gcc_target): Add them. Rename gen_check field to
1985 * flags.h (sel_sched_switch_set): Declare.
1986 * opts.c (sel_sched_switch_set): New variable.
1987 (decode_options): Unset flag_sel_sched_pipelining_outer_loops if
1988 pipelining is disabled from command line.
1989 (common_handle_option): Record whether selective scheduling is
1990 requested from command line.
1991 * doc/invoke.texi: Document new flags and parameters.
1992 * doc/tm.texi: Document new target hooks.
1993 * config/ia64/ia64.c (TARGET_SCHED_GEN_SPEC_CHECK): Define to
1995 (dfa_state_size): Do not declare locally.
1996 * config/ia64/ia64.opt (msched-ar-data-spec): Default to 0.
1997 * config/rs6000/rs6000.c (rs6000_init_sched_context,
1998 rs6000_alloc_sched_context, rs6000_set_sched_context,
1999 rs6000_free_sched_context): New functions.
2000 (struct _rs6000_sched_context): New.
2001 (rs6000_sched_reorder2): Do not modify INSN_PRIORITY for selective
2003 (rs6000_sched_finish): Do not run for selective scheduling.
2005 2008-08-31 Jan Hubicka <jh@suse.cz>
2007 * frv.c (frv_rtx_costs): Update forward declaration.
2008 * spu.c (spu_rtx_costs): Likewise.
2009 * pdp11.c: Include df.h
2010 (pdp11_output_function_epilogue): Use df_set_regs_ever_live.
2011 * m68hc11.c (m68hc11_gen_highpart): Fix call of gen_rtx_SUBREG.
2012 (m68hc11_rtx_costs_1): Fix call of rtx_cost.
2013 * iq2000.c (iq2000_address_cost): Add speed argument.
2014 (iq2000_rtx_costs): Likewise.
2015 * mn10300.c (mn10300_address_cost_1): Remove speed argument;
2016 update call of mn10300_address_cost.
2017 * mcore.c: Include df.h
2018 (mcore_rtx_costs): Update prototype.
2019 * score3.c: Include df.h
2020 (score3_rtx_costs): Remove speed argument.
2021 * score7.c: Include df.h
2022 (score7_address_cost): Remove speed argument.
2023 * score-protos.h (score_address_cost): Update prototype.
2024 * score.c: Include df.h
2025 (score_rtx_costs): Update call of costs functions.
2026 * v850.c (v850_rtx_costs): Add bool argument.
2028 2008-08-31 Hans-Peter Nilsson <hp@axis.com>
2030 * config/cris/cris.c (cris_rtx_costs): Correct call to rtx_costs.
2032 2008-08-31 Aaron W. LaFramboise <aaronavay62@aaronwl.com>
2034 * gthr-win32.h (__gthread_setspecific): Use CONST_CAST2.
2035 * config/i386/gthr-win32.c (__gthread_setspecific): Same.
2037 2008-08-31 Aaron W. LaFramboise <aaronavay62@aaronwl.com>
2039 * mkmap-flat.awk: Add option pe_dll.
2040 * config/i386/t-cygming (SHLIB_LINK): Support building libgcc_s.
2041 * config/i386/t-cygwin (SHLIB_LC): Add.
2042 * config/i386/t-mingw32 (SHLIB_LC): Add.
2043 * config/i386/mingw32.h (SHARED_LIBGCC_UNDEFS_SPEC): New.
2044 (SUBTARGET_EXTRA_SPECS): Use SHARED_LIBGCC_UNDEFS_SPEC.
2045 (LINK_SPEC): Support libgcc_s.
2046 (LIBGCC_SPEC): Support libgcc_s.
2047 (LIBGCC_SONAME): New.
2049 2008-08-31 Jan Hubicka <jh@suse.cz>
2051 * predict.c (maybe_hot_bb_p, maybe_hot_edge_p): Previous commit
2052 mistakely had old version of patch.
2054 * ipa-cp.c (ipcp_need_original_clone_p): Remove.
2055 (ipcp_estimate_growth): New.
2056 (ipcp_insert_stage): Use ipcp_estimate_growth.
2057 * profile.c (branch_prob): When reading failed, do not consider
2060 2008-08-31 Jan Hubicka <jh@suse.cz>
2062 * tree-ssa-loop-unswitch.c (tree_unswitch_single_loop): Check that
2063 loop is optimized for speed.
2065 2008-08-31 Richard Guenther <rguenther@suse.de>
2068 * fold-const.c (fold_binary): Retain conversions in folding
2071 2008-08-31 Jan Hubicka <jh@suse.cz>
2073 * postreload-gcse.c (eliminate_partially_redundant_loads):
2074 Use optimize_bb_for_size_p.
2075 * predict.c (maybe_hot_frequency_p): Make inline.
2076 (maybe_hot_count_p): Break out from ...
2077 (maybe_hot_bb_p): ... this one.
2078 (maybe_hot_edge_p): Simplify.
2079 * basic-block.h (probably_cold_bb_p): Remove.
2081 2008-08-31 Jakub Jelinek <jakub@redhat.com>
2084 * config/rs6000/rs6000-protos.h (const_vector_elt_as_int): Add
2086 * config/rs6000/rs6000.c (const_vector_elt_as_int): No longer static.
2087 * config/rs6000/altivec.md (easy_vector_constant_add_self splitter):
2088 Also split V4SFmode.
2089 * config/rs6000/predicates.md (easy_vector_constant_add_self): Handle
2092 2008-08-31 Jan Hubicka <jh@suse.cz>
2094 * predict.c (PROB_VERY_LIKELY): Make small enough so things
2096 * predict.def (PRED_NORETURN_CALL, PRED_COLD_CALL): Use it.
2098 2008-08-31 Jakub Jelinek <jakub@redhat.com>
2101 * dwarf2out.c (gen_namespace_die): For DECL_EXTERNAL modules don't
2104 * dwarf2out.c (native_encode_initializer): Subtract min_index for
2105 non-range array index. Handle VIEW_CONVERT_EXPR and NON_LVALUE_EXPR.
2107 2008-08-31 Jan Hubicka <jh@suse.cz>
2109 * sparc.h (BRANCH_COST): Fix macro definition.
2110 * avr.c (avr_operand_rtx_cost): Add speed argument.
2111 (avr_rtx_costs): Update calls of avr_operand_rtx_cost.
2113 2008-08-31 Richard Guenther <rguenther@suse.de>
2115 * tree-cfg.c (verify_types_in_gimple_assign): Dump mismatched
2116 types instead of operands. Re-instantiate verifying of
2117 binary expression operands.
2119 2008-08-30 Jan Hubicka <jh@suse.cz>
2121 * fwprop.c (should_replace_address): Add speed attribute.
2122 (PR_OPTIMIZE_FOR_SPEED): New flag.
2123 (propagate_rtx_1): Use it.
2124 (propagate_rtx): Set it.
2125 (try_fwprop_subst): Update call of rtx_costs.
2126 (forward_propagate_and_simplify): LIkewise.
2127 * hooks.c (hook_int_rtx_bool_0): New
2128 (hook_bool_rtx_int_int_intp_false): Replace by ...
2129 (hook_bool_rtx_int_int_intp_bool_false): .. thisone.
2130 * hooks.h (hook_int_rtx_bool_0): New
2131 (hook_bool_rtx_int_int_intp_false): Replace by ...
2132 (hook_bool_rtx_int_int_intp_bool_false): .. thisone.
2133 * optabs.c (avoid_expensive_constant): UPdate call of rtx_cost.
2134 (prepare_cmp_insn): UPdate call of rtx_cost.
2135 * postreload.c (reload_cse_simplify_set): Update call of rtx_cost.
2136 (reload_cse_simplify_operands): Update call of rtx_cost.
2137 (reload_cse_move2add): call of rtx_cost.
2138 * target.h (struct gcc_target): Update rtx_costs and address_costs.
2139 * rtlanal.c (rtx_cost): Add speed argument.
2140 (address_cost): Add speed argument
2141 (default_address_cost): Likewise.
2142 (insn_rtx_cost): Likewise.
2143 * cfgloopanal.c (seq_cost): Add speed argument.
2144 (target_reg_cost, target_spill_cost): Turn to array.
2145 (init_set_costs): Update for speed.
2146 (estimate_reg_pressure_cost): Add speed argument.
2147 * auto-inc-dec.c (attempt_change): Update call of rtx_cost.
2148 * dojump.c (prefer_and_bit_test): UPdate call of rtx_cost.
2149 * tree-ssa-loop-ivopts.c (struct ivopts_data): New field speed.
2150 (seq_cost): Add speed argument.
2151 (computation_cost): Add speed arugment.
2152 (add_cost, multiply_by_const, get_address_cost): add speed argument.
2153 (force_expr_to_var_cost): Update for profile info.
2154 (force_var_cost): Likewise.
2155 (split_address_cost): Likewise.
2156 (ptr_difference_cost): Likewise.
2157 (difference_cost): Likewise.
2158 (get_computation_cost_at): Likewise.
2159 (determine_iv_cost): Likewise.
2160 (ivopts_global_cost_for_size): Likewise.
2161 (rewrite_use_address): Likewise.
2162 (tree_ssa_iv_optimize_loop): Initialize speed field.
2163 * cse.c (optimize_this_for_speed_p): New static var.
2164 (notreg_cost): Update call of rtx_cost.
2165 (cse_extended_basic_block): set optimize_this_for_speed_p.
2166 * ifcvt.c (cheap_bb_rtx_cost_p): Update call of rtx_cost.
2167 (noce_try_cmove_arith): Likewise.
2168 (noce_try_sign_mask): LIkewise.
2169 * expr.c (compress_float_constant): Update rtx_cost calls.
2170 * tree-ssa-address.c (most_expensive_mult_to_index): Add speed argument.
2171 (addr_to_parts): Likewise.
2172 (create_mem_ref): Likewise.
2173 * dse.c (find_shift_sequence): Add speed argument.
2174 (replace_read): Update call.
2175 * calls.c (precompute_register_parameters): Update call of rtx_cost.
2176 * expmed.c (sdiv_pow2_cheap, smod_pow2_cheap, zero_cost, add_cost,
2177 * neg_cost, shift_cost, shiftadd_cost,
2178 shiftsub_cost, mul_cost, sdiv_cost, udiv_cost ,mul_widen_cost,
2179 mul_highpart_cost): Increase dimension.
2180 (init_expmed): Initialize for both size and speed.
2181 (expand_shift): Use profile.
2182 (synth_mult): Use profile.
2183 (choose_mult_variant): Use profile.
2184 (expand_mult): Use profile.
2185 (expand_mult_highpart_optab): Use profile.
2186 (expand_mult_highpart): Use profile.
2187 (expand_smod_pow2): Use profile.
2188 (expand_divmod): Use profile.
2189 * simplify-rtx.c (simplify_binary_operation_1): Update call of rtx_cost.
2190 * loop-invariant.c (create_new_invariant): Use profile.
2191 (gain_for_invariant): Add speed parameter.
2192 (best_gain_for_invariant): Likewise.
2193 (find_invariants_to_move): Likewise.
2194 (move_single_loop_invariants): Set it.
2195 * target-def.h (TARGET_RTX_COSTS): Use hook.
2196 * rtl.h (rtx_cost, address_cost, insn_rtx_cost): Update prototpe.
2197 (optimize_insn_for_size_p, optimize_insn_for_speed_p): Declare.
2198 * output.h (default_address_cost): Update prototype.
2199 * combine.c (optimize_this_for_speed_p): New static var.
2200 (combine_validate_cost): Update call of rtx_cost.
2201 (combine_instructions): Set optimize_this_for_speed_p.
2202 (expand_compound_operation): Update call of rtx_cost.
2203 (make_extraction):Update call of rtx_cost.
2204 (force_to_mode):Update call of rtx_cost.
2205 (distribute_and_simplify_rtx):Update call of rtx_cost.
2206 * cfgloop.h (target_reg_cost, target_spill_cost): Turn to array.
2207 (estimate_reg_pressure_cost): Update prototype.
2208 * tree-flow.h (multiply_by_cost, create_mem_ref): Update prototype.
2209 * basic-block.h (optimize_insn_for_size_p, optimize_insn_for_speed_p):
2211 * config/alpha/alpha.c (alpha_rtx_costs): Update.
2212 (alpha_rtx_costs): Update.
2213 * config/frv/frv.c (frv_rtx_costs): Update.
2214 * config/s390/s390.c (s390_rtx_costs): Update.
2215 * config/m32c/m32c.c (m32c_memory_move_cost): Update.
2216 (m32c_rtx_costs): Update.
2217 * config/spu/spu.c (TARGET_ADDRESS_COST): Upate.
2218 (spu_rtx_costs): Update.
2219 * config/sparc/sparc.c (sparc_rtx_costs): Update.
2220 * config/m32r/m32r.c (m32r_rtx_costs): Update.
2221 * config/i386/i386.c (:ix86_address_cost): Update.
2222 (ix86_rtx_costs): Update.
2223 * config/sh/sh.c (sh_rtx_costs, sh_address_cost): Update.
2224 * config/pdp11/pdp11.c (pdp11_rtx_costs): Update.
2225 * config/avr/avr.c (avr_rtx_costs, avr_address_cost): Update.
2226 * config/crx/crx.c (crx_address_cost): Update.
2227 * config/xtensa/xtensa.c (xtensa_rtx_costs): Update.
2228 * config/stormy16/stormy16.c
2229 (xstormy16_address_cost, xstormy16_rtx_costs): Update.
2230 * config/m68hc11/m68hc11.c
2231 (m68hc11_address_cost, m68hc11_rtx_costs): Update.
2232 * config/cris/cris.c (cris_rtx_costs, cris_address_cost): Update.
2233 * config/iq2000/iq2000.c (iq2000_rtx_costs, iq2000_address_cost):
2235 * config/mn10300/mn10300.c (mn10300_address_cost, mn10300_rtx_costs):
2237 * config/ia64/ia64.c (ia64_rtx_costs): Update.
2238 * config/m68k/m68k.c (m68k_rtx_costs): Update.
2239 * config/rs6000/rs6000.c (rs6000_rtx_costs): Update.
2240 * config/arc/arc.c (arc_rtx_costs, arc_address_cost): Update.
2241 * config/mcore/mcore.c (TARGET_ADDRESS_COST): Update.
2242 (mcore_rtx_costs): update.
2243 * config/score/score3.c (score3_rtx_costs): Update.
2244 * config/score/score7.c (score7_rtx_costs): Update.
2245 * config/score/score3.h (score3_rtx_costs):Update.
2246 * config/score/score7.h (score7_rtx_costs): Update.
2247 * config/score/score.c (score_rtx_costs): Update.
2248 * config/arm/arm.c (arm_address_cost): Update.
2249 (arm_rtx_costs_1): Update.
2250 (arm_rtx_costs_1): Update.
2251 (arm_size_rtx_costs): Update.
2252 (arm_size_rtx_costs): Update.
2253 (arm_size_rtx_costs): Update.
2254 (arm_xscale_rtx_costs): Update.
2255 (arm_thumb_address_cost): Update.
2256 * config/pa/pa.c (hppa_address_cost): Update.
2257 * config/mips/mips.c (mips_rtx_costs): Update.
2258 * config/vax/vax.c (vax_address_cost): Update.
2259 * config/h8300/h8300.c (h8300_shift_costs): Update.
2260 (h8300_rtx_costs): Update.
2261 * config/v850/v850.c (TARGET_ADDRESS_COST): Update.
2262 (v850_rtx_costs): Update.
2263 * config/mmix/mmix.c (mmix_rtx_costs, mmix_rtx_costs): Update.
2264 * config/bfin/bfin.c
2265 (bfin_address_cost): Update.
2266 (bfin_rtx_costs): Update.
2267 * stmt.c (lshift_cheap_p): Update.
2269 2008-08-30 Andrew Pinski <andrew_pinski@playstation.sony.com>
2272 * expmed.c (extract_bit_field_1): Check the mode size to make
2273 sure the vector modes have the same size.
2275 2008-08-29 Michael Meissner <gnu@the-meissners.org>
2277 * dojump.c (top level): Include basic-block.h to declare
2278 optimize_insn_for_speed_p
2280 * Makefile.in (dodump.h): Add $(BASIC_BLOCK_H) dependency.
2282 * opts.h (CL_SAVE): New option class for marking options that are
2283 target specific options usable in the target attribute.
2284 (CL_MIN_OPTION_CLASS): CL_SAVE is now the minimum option.
2286 * opt-functions.awk (switch_flags): Add CL_SAVE flag so backends
2287 can easily find the target specific options that are safe to use
2288 in the attribute or pragma.
2290 * attribs.c (decl_attributes): Change #pragma GCC option to
2291 #pragma GCC target, and attribute((option(...))) to
2292 attribute((target(...))).
2294 * doc/extend.texi (target attribute): Change from option
2295 attribute. Delete push/pop/reset.
2296 (#pragma GCC target): Change from #pragma GCC option. Delete
2298 (#pragma GCC push_options): Document new pragma.
2299 (#pragma GCC pop_options): Document new pragma.
2300 (#pragma GCC reset_options): Document new pragma.
2302 * targhooks.c (default_target_option_valid_attribute_p): Add
2303 warning about port not supporting target attributes.
2304 (default_target_option_pragma_parse): New function, warn about
2305 #pragma GCC target not being supported.
2307 * targhooks.h (default_target_option_pragma_parse): Add
2310 * tree.h (TI_CURRENT_TARGET_PRAGMA): Rename from
2311 TI_CURRENT_OPTION_PRAGMA.
2312 (current_target_pragma): Rename from current_option_pragma.
2314 * target.h: (struct target_option): Delete booleans for changing
2315 the optimization level on hot/cold functions. Change signature of
2316 pragma_parse hook to take a second tree.
2318 * c-tree.h (c_builtin_function_ext_scope): Add declaration.
2320 * c-decl.c (c_builtin_function_ext_scope): New function, guarantee
2321 that the declaration is done at global scope.
2323 * langhooks.c (add_builtin_function_common): Move most of the code
2324 from add_builtin_function here, calling the hook passed in.
2325 (add_builtin_function): Call add_builtin_function_common with
2326 standard builtin hook.
2327 (add_builtin_function_ext_scope): New function to add builtins to
2330 * langhooks.h (struct lang_hooks): Add builtin_function_ext_scope
2332 (add_builtin_function_ext_scope): Add declaration.
2334 * c-pragma.c (handle_pragma_target): Rename from
2335 handle_pragma_option, #pragma GCC option is now #pragma GCC
2336 target. Move warning about port not supporting target options to
2337 default pragma parse hook. Remove push/pop/reset from this
2339 (handle_pragma_optimize): Remove push/pop/reset from this pragma.
2340 (option_stack): Delete static variable.
2341 (optimize_stack): Ditto.
2342 (optons_stack): New stack of saved target and optimization
2344 (handle_pragma_push_options): New function to handle pushing both
2345 target and optimization options.
2346 (handle_pragma_pop_options): New function to handle popping both
2347 target and optimization options.
2348 (handle_pragma_reset_options): New function to handle resetting
2349 both target and optimization options to their initial state.
2350 (init_pragma): Rename handle_pragma_option to
2351 handle_pragma_target. Add support for push_options, pop_options,
2352 and reset_options pragmas.
2354 * target-def.h (TARGET_OPTION_PRAGMA_PARSE): Change default to
2355 default_target_option_pragma_parse.
2356 (TARGET_OPTION_VALID_ATTRIBUTE_P): Change default to
2357 default_target_option_valid_attribute_p.
2358 (TARGET_OPTION_COLD_ATTRIBUTE_SETS_OPTIMIZATION): Delete.
2359 (TARGET_OPTION_HOT_ATTRIBUTE_SETS_OPTIMIZATION): Ditto.
2360 (TARGET_OPTION_HOOKS): Delete the fields for whether to change
2361 optimization level on hot/cold functions.
2363 * tree-inline.c (tree_can_inline_p): Disable suppressing inlining
2364 if the caller and callee have different optimization levels.
2366 * c-common.c (handle_target_attribute): Rename from
2367 handle_option_attribute, attribute((option(...))) is now
2368 attribute((target(...))). Move warning if the port does not
2369 support target attributes to the default hook.
2370 (handle_hot_attribute): Delete code to change the optimization
2371 level of hot functions.
2372 (handle_cold_attribute): Ditto.
2374 * config/i386/i386-c.c (ix86_pragma_target_parse): Take a second
2375 argument that is the binary tree options to use if there are no
2376 arguments. Call ix86_valid_target_attribute_tree instead of
2377 ix86_valid_option_attribute_tree.
2378 (ix86_pragma_target_parse): Rename from ix86_pragma_option_parse.
2379 (ix86_register_pragmas): Use ix86_pragma_target_parse instead of
2380 ix86_pragma_option_parse.
2382 * config/i386/i386-protos.h (ix86_valid_target_attribute_tree):
2383 Rename from ix86_valid_option_attribute_tree.
2385 * config/i386/i386.c (ix86_add_new_builtins): New function to add
2386 new builtins when the ISA changes.
2387 (ix86_valid_target_attribute_tree): Rename from
2388 ix86_valid_option_attribute_tree. Change callers. If the
2389 function specified optimization options, use those as the starting
2390 point before setting up the target attributes. If the
2391 optimization options were changed in the course of setting the
2392 target attributes, record the new optimization options.
2393 (ix86_valid_target_attribute_tree_inner_p): Rename from
2394 ix86_valid_option_attribute_tree_inner_p. Change callers. Call
2395 ix86_add_new_builtins if the ISA changed.
2396 (ix86_valid_target_attribute_p): Rename from
2397 ix86_valid_option_attribute_p. Change callers.
2398 (enum ix86_builtins): Add IX86_BUILTIN_PCMOV to allow both
2399 __builtin_ia32_pcmov and __builtin_ia32_pcmov_v2di to be declared
2400 as delayed builtin functions.
2401 (struct builtin_isa): New structure to record builtin functions
2402 that should be delayed until the ISA for that function is used.
2403 (ix86_builtins_isa): Change from int to struct to track builtin
2404 functions we want to declare at some point.
2405 (def_builtin): If the front end can delay defining the builtin
2406 functions, don't create builtins for ISAs not part of the default
2408 (def_builtin_const): Ditto.
2409 (bdesc_multi_arg): Declare __builtin_ia32_pcmov and
2410 __builtin_ia32_pcmov_v2di to be different builtin functions.
2411 (ix86_expand_builtin): Changes due to ix86_builtins_isa now being
2412 a structure instead of an int.
2413 (TARGET_OPTION_VALID_ATTRIBUTE_P): Use
2414 ix86_valid_target_attribute_p, not ix86_valid_option_attribute_p.
2415 (TARGET_OPTION_COLD_ATTRIBUTE_SETS_OPTIMIZATION): Delete.
2416 (TARGET_OPTION_HOT_ATTRIBUTE_SETS_OPTIMIZATION): Ditto.
2418 * config/ia64/ia64.h
2419 (TARGET_OPTION_COLD_ATTRIBUTE_SETS_OPTIMIZATION): Delete.
2420 (TARGET_OPTION_HOT_ATTRIBUTE_SETS_OPTIMIZATION): Ditto.
2422 * langhooks-def.h (LANG_HOOKS_BUILTIN_FUNCTION_EXT_SCOPE): New
2423 hook, default to being the same as LANG_HOOKS_BUILTIN_FUNCTION.
2424 (LANG_HOOKS_INITIALIZER): Add
2425 LANG_HOOKS_BUILTIN_FUNCTION_EXT_SCOPE hook.
2427 2008-08-30 Kaz Kojima <kkojima@gcc.gnu.org>
2430 * config/sh/sh.h (LEGITIMIZE_RELOAD_ADDRESS): Generate
2431 the reload address with the index register for SFmode
2432 access with a displacement.
2434 2008-08-30 Jan Hubicka <jh@suse.cz>
2436 * optabs.c (expand_abs_nojump): Update BRANCH_COST call.
2437 * fold-cost.c (LOGICAL_OP_NON_SHORT_CIRCUIT, fold_truthop): Likewise.
2438 * dojump.c (do_jump): Likewise.
2439 * ifcvt.c (MAX_CONDITIONAL_EXECUTE): Likewise.
2440 (note-if_info): Add BRANCH_COST.
2441 (noce_try_store_flag_constants, noce_try_addcc,
2442 noce_try_store_flag_mask, noce_try_cmove_arith, noce_try_cmove_arith,
2443 noce_try_cmove_arith, noce_find_if_block, find_if_case_1,
2444 find_if_case_2): Use computed branch cost.
2445 * expr.h (BRANCH_COST): Update default.
2446 * predict.c (predictable_edge_p): New function.
2447 * expmed.c (expand_smod_pow2, expand_sdiv_pow2, emit_store_flag):
2448 Update BRANCH_COST call.
2449 * basic-block.h (predictable_edge_p): Declare.
2450 * config/alpha/alpha.h (BRANCH_COST): Update.
2451 * config/frv/frv.h (BRANCH_COST): Update.
2452 * config/s390/s390.h (BRANCH_COST): Update.
2453 * config/spu/spu.h (BRANCH_COST): Update.
2454 * config/sparc/sparc.h (BRANCH_COST): Update.
2455 * config/m32r/m32r.h (BRANCH_COST): Update.
2456 * config/i386/i386.h (BRANCH_COST): Update.
2457 * config/i386/i386.c (ix86_expand_int_movcc): Update use of BRANCH_COST.
2458 * config/sh/sh.h (BRANCH_COST): Update.
2459 * config/pdp11/pdp11.h (BRANCH_COST): Update.
2460 * config/avr/avr.h (BRANCH_COST): Update.
2461 * config/crx/crx.h (BRANCH_COST): Update.
2462 * config/xtensa/xtensa.h (BRANCH_COST): Update.
2463 * config/stormy16/stormy16.h (BRANCH_COST): Update.
2464 * config/m68hc11/m68hc11.h (BRANCH_COST): Update.
2465 * config/iq2000/iq2000.h (BRANCH_COST): Update.
2466 * config/ia64/ia64.h (BRANCH_COST): Update.
2467 * config/rs6000/rs6000.h (BRANCH_COST): Update.
2468 * config/arc/arc.h (BRANCH_COST): Update.
2469 * config/score/score.h (BRANCH_COST): Update.
2470 * config/arm/arm.h (BRANCH_COST): Update.
2471 * config/pa/pa.h (BRANCH_COST): Update.
2472 * config/mips/mips.h (BRANCH_COST): Update.
2473 * config/vax/vax.h (BRANCH_COST): Update.
2474 * config/h8300/h8300.h (BRANCH_COST): Update.
2475 * params.def (PARAM_PREDICTABLE_BRANCH_OUTCOME): New.
2476 * doc/invoke.texi (predictable-branch-cost-outcome): Document.
2477 * doc/tm.texi (BRANCH_COST): Update.
2479 2008-08-30 Samuel Tardieu <sam@rfc1149.net>
2482 * config/arm/arm.c (arm_optimization_options): Set
2483 flag_section_anchors to 2 instead of 1 to distinguish it from
2484 -fsection-anchors given explicitely on the command line.
2486 2008-08-30 Richard Sandiford <rdsandiford@googlemail.com>
2488 * recog.c (split_insn): Consider attaching a REG_EQUAL note to the
2489 final insn of a split.
2491 2008-08-30 Jan Hubicka <jh@suse.cz>
2493 * postreload-gcse.c (gate_handle_gcse2): Disable for functions
2494 optimized for speed.
2495 * final.c (compute_alignments): Use optimize_bb_for_size_p.
2496 * tree-call-cdce.c (gate_call_cdce): Use optimize_function_for_speed_p.
2497 * opts.c (flag_predictive_commoning_set, flag_unswitch_loops_set,
2498 flag_gcse_after_reload_set): New static vars.
2499 (common_handle_option): Enable those flags for profile-use.
2500 (decode_options): Remove optimize_size flags that are handled
2501 on higher granuality.
2502 * tree-vectorizer.c (vectorize_loops): Use
2503 optimize_loop_nest_for_speed_p.
2504 * tree-ssa-pre.c (do_pre): Use optimize_function_for_speed_p.
2505 * tree-predcom.c (tree_predictive_commoning): Use
2506 optimize_loop_for_speed_p.
2507 * varasm.c (assemble_start_function): Use optimize_function_for_speed_p.
2508 * bb-reorder.c (rest_of_handle_reorder_blocks): Likewise.
2509 * predict.c (optimize_loop_for_speed_p): Fix walk.
2511 2008-08-30 Jan Hubicka <jh@suse.cz>
2513 * ipa-inline.c (cgraph_estimate_growth): Discover self recursive
2515 (cgraph_decide_inlining_of_small_function): Use edge->count to detect
2516 profile presence locally.
2518 2008-08-29 Joseph Myers <joseph@codesourcery.com>
2521 * tree-vrp.c (find_switch_asserts): Make idx volatile for GCC
2522 versions before 4.0.
2524 2008-08-29 Jan Hubicka <jh@suse.cz>
2526 * tree-inline.c (insert_init_stmt): Insert sequence even when
2529 2008-08-29 Jeff Law <law@redhat.com>
2531 * mn10300.c (mn10300_secondary_reload_class): We need secondary
2532 reloads for AM33-2 if IN is a pseudo with an equivalent memory
2533 location and class is an FP register.
2535 2008-08-29 Jan Hubicka <jh@suse.cz>
2537 * see.c (see_merge_one_def_extension): Silence used uninitialized
2539 * matrix-reorg.c (check_allocation_function): Likewise.
2540 * config/i386/driver-i386.c (detect_caches_amd): Likewise.
2542 2008-08-29 Jakub Jelinek <jakub@redhat.com>
2545 * fold-const.c (fold_binary): In (X | C1) & C2 canonicalization
2546 compute new & and | in type rather than TREE_TYPE (arg0).
2548 * dwarf2out.c (fortran_common): Update comment.
2549 (gen_variable_die): Swap com_die and var_die variables in Fortran
2550 COMMON block handling code.
2552 * dwarf2out.c (descr_info_loc): Handle VAR_DECL.
2554 * dwarf2out.c (gen_const_die): New function.
2555 (size_of_die, value_format, output_die): Output larger
2556 dw_val_class_vec using DW_FORM_block2 or DW_FORM_block4.
2557 (native_encode_initializer): New function.
2558 (tree_add_const_value_attribute): Call it.
2559 (gen_decl_die, dwarf2out_decl): Handle CONST_DECLs if is_fortran ().
2561 * dwarf2out.c (gen_formal_parameter_die, gen_variable_die): For
2562 DECL_BY_REFERENCE decls don't pass TREE_READONLY and
2563 TREE_THIS_VOLATILE to add_type_attribute.
2565 * dwarf2out.c (add_subscript_info): Stop on Fortran TYPE_STRING_FLAG
2567 (gen_array_type_die): Emit DW_TAG_string_type for Fortran character
2570 * dwarf2out.c (loc_by_reference): New function.
2571 (add_location_or_const_value_attribute): Use it.
2574 * dwarf2out.c (gen_variable_die): Represent Fortran COMMON vars
2575 as DW_TAG_variable children of DW_TAG_common_block rather than
2576 DW_TAG_member children. Put DW_AT_external to individual
2577 DW_TAG_variable DIEs, not to DW_TAG_common_block.
2579 * dwarf2out.c (add_bound_info): If lookup_decl_die failed, try
2580 loc_descriptor_from_tree_1.
2584 * debug.h (struct gcc_debug_hooks): Add NAME and CHILD
2585 arguments to imported_module_or_decl.
2586 (debug_nothing_tree_tree): Removed.
2587 (debug_nothing_tree_tree_tree_bool): New prototype.
2588 * debug.c (do_nothing_debug_hooks): Adjust.
2589 (debug_nothing_tree_tree): Removed.
2590 (debug_nothing_tree_tree_tree_bool): New function.
2591 * dwarf2out.c (is_symbol_die): Handle DW_TAG_module.
2592 (gen_variable_die): Put all common vars for the
2593 same COMMON block under one DW_TAG_common_block.
2594 (declare_in_namespace): Return new context_die, for Fortran
2595 return the module DIE instead of adding extra declarations into
2597 (gen_type_die_with_usage): Adjust declare_in_namespace caller.
2598 (gen_namespace_die): If is_fortran (), generate DW_TAG_module
2599 instead of DW_TAG_namespace. If DECL_EXTERNAL is set, add
2601 (dwarf2out_global_decl): Don't skip Fortran global vars.
2602 (gen_decl_die): Likewise. Adjust declare_in_namespace callers.
2603 (dwarf2out_imported_module_or_decl): Add NAME and CHILD arguments.
2604 If NAME is non-NULL, add DW_AT_name. If CHILD is non-NULL, put
2605 DW_TAG_imported_declaration as child of previous
2606 DW_TAG_imported_module.
2607 * dbxout.c (dbx_debug_hooks, xcoff_debug_hooks): Adjust.
2608 * sdbout.c (sdb_debug_hooks): Likewise.
2609 * vmsdbgout.c (vmsdbg_debug_hooks): Likewise.
2611 2008-08-29 Jan Hubicka <jh@suse.cz>
2613 * cgraph.c (cgraph_remove_node): Do not remove nested nodes.
2615 * cgraph.h (cgraph_maybe_hot_edge_p): Declare.
2616 * ipa-cp.c (n_cloning_candidates): New static variable.
2617 (ipcp_print_profile_data, ipcp_function_scale_print): Forward declare.
2618 (ipcp_print_all_lattices): Improve debug output.
2619 (ipcp_cloning_candidate_p): New function.
2620 (ipcp_initialize_node_lattices): Use it.
2621 (ipcp_init_stage): Do only analyzis here; prettier debug output.
2622 (ipcp_propagate_stage): Prettier debug output.
2623 (ipcp_iterate_stage): Initialize latices here; prettier debug output.
2624 (ipcp_print_all_structures): Remove.
2625 (ipcp_need_redirect_p): Test !n_cloning_candidates.
2626 (ipcp_insert_stage): Prettier debug output; call
2627 cgraph_remove_unreachable_nodes before propagating.
2628 (pass_ipa_cp): Schedule function removal pass.
2629 * ipa-inline.c (inline_indirect_intraprocedural_analysis): Better
2631 (cgraph_maybe_hot_edge_p): Move to ...
2632 * predict.c (cgraph_maybe_hot_edge_p) ... here.
2633 * opts.c (flag_ipa_cp_set, flag_ipa_cp_clone_set): New.
2634 (common_handle_option): Set them; enable ipa-cp when profiling.
2635 * ipa-prop.c (ipa_print_node_jump_functions): Prettier output.
2636 (ipa_print_all_jump_functions): Likewise.
2637 (ipa_print_all_tree_maps, ipa_print_node_param_flags): Remove.
2638 (ipa_print_node_params, ipa_print_all_params): New.
2639 * ipa-prop.h (ipa_print_all_tree_maps, ipa_print_node_param_flags,
2640 ipa_print_all_param_flags): Remove.
2641 (ipa_print_node_params, ipa_print_all_params): New.
2643 2008-08-29 Bob Wilson <bob.wilson@acm.org>
2645 * config/xtensa/xtensa.c (xtensa_secondary_reload_class): Revert
2646 change from 2008-04-03.
2647 * config/xtensa/xtensa.h (IRA_COVER_CLASSES): Define.
2649 2008-08-29 Vladimir Makarov <vmakarov@redhat.com>
2652 * ira-color.c (push_allocnos_to_stack): Truncate
2653 removed_splay_allocno_vec.
2655 2008-08-29 Vladimir Makarov <vmakarov@redhat.com>
2658 * ira-color (ira_fast_allocation): Don't assign hard registers to
2661 2008-08-29 Jan Hubicka <jh@suse.cz>
2664 * predict.c (optimize_loop_nest_for_speed_p): Do not ICE
2665 for cold internal loops.
2667 2008-08-29 Richard Guenther <rguenther@suse.de>
2669 * tree-ssa-structalias.c (create_variable_info_for): Do not
2670 create fields for heap vars or vars with a noalias state.
2671 For NO_ALIAS_ANYTHING variables add a self-constraint, not one
2674 2008-08-29 Richard Guenther <rguenther@suse.de>
2676 * common.opt (ftree-store-ccp): Mark as preserved for
2677 backward compatibility.
2678 * doc/invoke.texi (-ftree-store-ccp): Remove documentation.
2679 * tree-pass.h (pass_store_ccp): Remove.
2680 * tree-ssa-propagate.h (struct prop_value_d): Remove mem_ref field.
2681 (first_vdef): Remove declaration.
2682 (get_value_loaded_by): Likewise.
2683 * tree-ssa-ccp.c (do_store_ccp): Remove.
2684 (get_default_value): Simplify as do_store_ccp is always false
2685 now. Do not initialize mem_ref.
2686 (set_value_varying): Likewise.
2687 (canonicalize_float_value): Likewise.
2688 (set_lattice_value): Likewise.
2689 (likely_value): Likewise.
2690 (surely_varying_stmt_p): Likewise.
2691 (ccp_initialize): Likewise.
2692 (ccp_lattice_meet): Likewise.
2693 (ccp_visit_phi_node): Likewise.
2694 (ccp_fold): Likewise.
2695 (evaluate_stmt): Likewise.
2696 (visit_assignment): Likewise.
2697 (ccp_visit_stmt): Likewise.
2698 (execute_ssa_ccp): Fold into ...
2699 (do_ssa_ccp): ... this.
2700 (do_ssa_store_ccp): Remove.
2701 (gate_store_ccp): Likewise.
2702 (pass_store_ccp): Likewise.
2703 * tree-ssa-copy.c (copy_prop_visit_phi_node): Do not
2705 * tree-ssa-propagate.c (first_vdef): Remove.
2706 (get_value_loaded_by): Likewise.
2707 (replace_vuses_in): Likewise.
2708 (substitute_and_fold): Do not call replace_vuses_in.
2709 * opts.c (decode_options): Do not set flag_tree_store_ccp.
2711 2008-08-29 Richard Guenther <rguenther@suse.de>
2714 * tree-ssa-structalias.c (intra_create_variable_infos): Mark
2715 PARAM_NOALIAS tags with is_heapvar.
2716 * tree-ssa-operands.c (access_can_touch_variable): Offset
2717 based tests do not apply for heapvars. Fix offset test.
2719 2008-08-29 Jan Hubicka <jh@suse.cz>
2721 * doc/invoke.texi (-fipa-cp): Enabled by default at -O2/-Os/-O3
2722 (-fipa-cp-clone): Enabled by default at -O3.
2723 * opts.c (decode_options): Enable ipa-cp at -O2, ipa-cp-clone at -O3;
2724 make ipa-cp-clone to imply ipa-cp; disable cloning at -Os.
2726 2008-08-29 Jan Hubicka <jh@suse.cz>
2728 * tree.c (build_function_type_skip_args, build_function_decl_skip_args):
2730 * tree.h (build_function_type_skip_args, build_function_decl_skip_args):
2732 * gimple.c (giple_copy_call_skip_args): New function.
2733 (giple_copy_call_skip_args): Declare.
2735 * cgraph.h (cgraph_function_versioning): Add skip_args arugmnet
2736 * ipa-cp.c (ipcp_node_not_modifiable_p): Rename to ...
2737 (ipcp_node_modifiable_p): ... this one; use tree_versionable_function_p.
2738 (ipcp_create_replace_map): Improve debug output.
2739 (ipcp_need_redirect_p): Return false when not clonning.
2740 (ipcp_update_callgraph): Skip args.
2741 (ipcp_insert_stage): UPdate call of !ipcp_node_modifiable_p;
2743 * cgraphunit.c (cgraph_function_versioning): Add skip_args argument.
2744 (save_inline_function_body): Update call of tree_function_versioning.
2745 * ipa-prop.c (ipa_edge_removal_hook): Do not ICE on unanalyzed nodes.
2746 * tree-inline.c (copy_arguments_for_versioning): Add skip_args argument.
2747 (tree_function_versioning): Likewise.
2748 * tree-inline.h (tree_function_versioning): Update prototype.
2750 2008-08-29 Jan Hubicka <jh@suse.cz>
2752 * loop-unswitch.c (unswitch_single_loop): Use optimize_loop_for_speed_p.
2753 * tree-ssa-threadupdate.c (mark_threaded_blocks):
2754 Use optimize_function_for_size_p.
2755 * tracer.c (ignore_bb_p): Use optimize_bb_for_size_p.
2756 * postreload-gcse.c (eliminate_partially_redundant_load):
2757 Use optimize_bb_for_size_p.
2758 * value-prof.c (gimple_divmod_fixed_value_transform,
2759 gimple_mod_pow2_value_transform, gimple_mod_subtract_transform,
2760 gimple_stringops_transform): Use optimize_bb_for_size_p.
2761 * ipa-cp.c (ipcp_insert_stage): Use optimize_function_for_size_p.
2762 * final.c (compute_alignments): Use optimize_function_for_size_p.
2763 * builtins.c (fold_builtin_cabs): Use optimize_function_for_speed_p.
2764 (fold_builtin_strcpy, fold_builtin_fputs): Use
2765 optimize_function_for_size_p.
2766 * fold-const.c (tree_swap_operands_p): Use optimize_function_for_size_p.
2767 * recog.c (relax_delay_slots): Likewise.
2768 * tree-ssa-math-opts.c (replace_reciprocal):
2769 Use optimize_bb_for_speed_p.
2770 (execute_cse_reciprocals): Use optimize_bb_for_size_p.
2771 * ipa-inline.c (cgraph_decide_recursive_inlining): Use
2772 optimize_function_for_size_p.
2773 (cgraph_decide_inlining_of_small_function): Use
2774 optimize_function_for_size_p.
2775 * global.c (find_reg): Use optimize_function_for_size_p.
2776 * opts.c (decode_options): Do not clear flag_tree_ch,
2777 flag_inline_functions, flag_unswitch_loops, flag_unroll_loops,
2778 flag_unroll_all_loops and flag_prefetch_loop_arrays. Those can
2779 work it out from profile.
2780 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely): Use
2781 optimize_loop_for_speed_p.
2782 * predict.c (optimize_bb_for_size_p, optimize_bb_for_speed_p): Constify
2784 (optimize_loop_nest_for_size_p, optimize_loop_nest_for_speed_p): New.
2785 * tree-parloops.c (parallelize_loops): Use optimize_loop_for_size_p.
2786 * tree-eh.c (decide_copy_try_finally): Use optimize_function_for_size_p.
2787 * local-alloc.c (block_alloc): Pass BB pointer.
2788 (find_free_reg): Add BB pointer, use optimize_bb_for_size_p.
2789 * gcse.c (gcse_main): Use optimize_function_for_size_p.
2790 * loop-unroll.c (decide_unrolling_and_peeling):
2791 Use optimize_loop_for_size_p.
2792 (decide_peel_completely): Likewise.
2793 * tree-vect-analyze.c (vect_mark_for_runtime_alias_test): Use
2794 optimize_loop_for_size_p.
2795 (vect_enhance_data_refs_alignment): Likewise.
2796 * tree-ssa-coalesce.c (coalesce_cost): Add optimize_for_size argument.
2797 (coalesce_cost_bb, coalesce_cost_edge, create_outofssa_var_map):
2799 * cfgcleanup.c (outgoing_edges_match): Use optimize_bb_for_speed_p.
2800 (try_crossjump_bb): Use optimize_bb_for_size_p.
2801 * tree-ssa-loop-prefetch.c (loop_prefetch_arrays): Use
2802 optimize_loop_for_speed_p.
2803 * bb-reorder.c (find_traces_1_round): Likewise.
2804 (copy_bb): Use optimize_bb_for_speed_p.
2805 (duplicate_computed_gotos): Likewise.
2806 * basic-block.h (optimize_loop_nest_for_size_p,
2807 optimize_loop_nest_for_speed_p): New.
2808 * stmt.c (expand_case): Use optimize_insn_for_size_p.
2810 2008-08-29 Tristan Gingold <gingold@adacore.com>
2812 * gcov.c (main): Call expandargv.
2814 2008-08-29 Jan Hubicka <jh@suse.cz>
2816 * tree-ssa-loop-ch.c (should_duplicate_loop_header_p): Check that loop
2818 * predict.c (optimize_bb_for_size_p, optimize_bb_for_speed_p):
2820 (optimize_loop_for_size_p, optimize_loop_for_speed_p): New functions.
2821 * basic-block.h (optimize_bb_for_size_p, optimize_bb_for_speed_p):
2823 (optimize_loop_for_size_p, optimize_loop_for_speed_p): Declare.
2825 2008-08-29 Jan Hubicka <jh@suse.cz>
2827 * tree-pass.h (pass_strip_predict_hints): Declare.
2828 * predict.c (strip_builtin_expect): Rename to ...
2829 (strip_predict_hints): ... this one; strip also GIMPLE_PREDICT.
2830 (tree_bb_level_predictions): Do not remove GIMPLE_PREDICT.
2831 (tree_estimate_probability): Do not strip builtin_expect.
2832 (pass_strip_predict_hints): New pass.
2833 * tree-inline.c (expand_call_inline): When inlining cold function,
2834 predict it as unlikely.
2835 * passes.c (init_optimization_passes): Add pass_strip_predict_hints.
2837 2008-08-29 Richard Guenther <rguenther@suse.de>
2839 PR tree-optimization/37207
2840 * tree-vrp.c (extract_range_from_binary_expr): Also try
2841 to constant fold if only one of the operands is a constant.
2843 2008-08-29 Nick Clifton <nickc@redhat.com>
2845 * config/m32r/m32r.h (LEGITIMATE_CONSTANT_P): Treat LABEL_REFs in
2846 the same way as SYMBOL_REFs.
2848 2008-08-28 Bob Wilson <bob.wilson@acm.org>
2850 * config/xtensa/xtensa.md (<u>mulsidi3): Use a temporary register.
2852 2008-08-28 Adam Nemet <anemet@caviumnetworks.com>
2854 * config/mips/mips.h (ISA_HAS_BBIT): New macro.
2855 * config/mips/mips.md (branch_likely): Remove const. Fix
2857 (define_delay for type "branch"): Change to only apply for branch
2858 with likely variant.
2859 (define_delay for type "branch" and "branch_likely" no). New delay
2861 (equality_op): New code iterator.
2862 (bbv, bbinv): New code attributes.
2863 (*branch_bit<bbv><mode>, *branch_bit<bbv><mode>_inverted): New
2866 2008-08-28 Manuel Lopez-Ibanez <manu@gcc.gnu.org>
2867 Andrew Pinski <pinskia@gcc.gnu.org>
2870 * c-common.c (verify_tree): Fix handling of ADDR_EXPR.
2872 2008-08-28 Paolo Carlini <paolo.carlini@oracle.com>
2874 * gtrh-posix.h: Fix uses of _POSIX_TIMEOUTS per the normal Posix
2875 rule that a symbolic constant must be defined and >= 0 for the
2876 corresponding facility to be present at compile-time.
2877 * gthr-posix.c: Likewise.
2879 2008-08-28 Adam Nemet <anemet@caviumnetworks.com>
2881 * config/mips/mips.h (ISA_HAS_DMUL3): New macro.
2882 * config/mips/mips.md (D): New mode attribute.
2883 (mulsi3, muldi3): Merge it into ...
2884 (mul<mode>3): ... new template. Use _mul3 ending for 3-op patterns.
2885 (muldi3_mul3): New pattern.
2886 (mulsi3_mult3): Rename to mulsi3_mul3.
2888 2008-08-28 Jan Hubicka <jh@suse.cz>
2890 * expmed.c (store_bit_field_1): Be prepared for movstrict expander
2892 * predict.c (always_optimize_for_size_p): Rename to ...
2893 (optimize_function_for_size): ... this one; make extern.
2894 (optimize_function_for_speed_p): New.
2895 (optimize_bb_for_size_p, optimize_bb_for_size_p,
2896 optimize_edge_for_size_p,optimize_edge_for_size_p,
2897 optimize_insn_for_size_p, optimize_insn_for_size_p): Update.
2898 * basic-block.h (optimize_function_for_size_p,
2899 optimize_function_for_speed_p): Declare.
2900 * i386.md (optimize_size checks): Replace them by appropriate
2902 (standard_80387_constant_p, ix86_compute_frame_layout,
2903 ix86_expand_epilogue, ix86_decompose_address,
2904 print_operand, emit_i387_cw_initialization,
2905 inline_memory_move_cost, ix86_pad_returns,
2906 ix86_reorg): Replace optimize_size checks.
2908 2008-08-28 Richard Sandiford <rdsandiford@googlemail.com>
2910 * rtl.h (simplify_subreg_regno): Declare.
2911 * rtlanal.c (simplify_subreg_regno): New function, split out from...
2912 * simplify-rtx.c (simplify_subreg): ...here.
2913 * reload.c (find_reloads): Use simplify_subreg_regno instead of
2914 subreg_offset_representable_p.
2916 2008-08-28 Manuel Lopez-Ibanez <manu@gcc.gnu.org>
2919 * c-typeck.c (convert_for_assignment): Give a note describing what
2920 was passed and what was expected.
2922 2008-08-28 Joey Ye <joey.ye@intel.com>
2924 * doc/extend.texi: Document AVX built-in functions.
2925 * doc/invoke.texi: Document -mavx.
2927 2008-08-28 H.J. Lu <hongjiu.lu@intel.com>
2928 Joey Ye <joey.ye@intel.com>
2929 Xuepeng Guo <xuepeng.guo@intel.com>
2931 * config.gcc (extra_headers): Add gmmintrin.h for x86 and x86-64.
2933 * config/i386/cpuid.h (bit_FMA): New.
2934 (bit_XSAVE): Likewise.
2935 (bit_OSXSAVE): Likewise.
2936 (bit_AVX): Likewise.
2938 * config/i386/gas.h (ASM_OUTPUT_OPCODE): Undefine before
2939 define. Use ASM_OUTPUT_AVX_PREFIX.
2941 * config/i386/gmmintrin.h: New.
2943 * config/i386/i386.c (x86_64_reg_class): Add X86_64_AVX_CLASS.
2944 (OPTION_MASK_ISA_AVX_SET): New.
2945 (OPTION_MASK_ISA_FMA_SET): Likewise.
2946 (OPTION_MASK_ISA_AVX_UNSET): Likewise.
2947 (OPTION_MASK_ISA_FMA_SET): Likewise.
2948 (OPTION_MASK_ISA_SSE4_2_UNSET): Updated.
2949 (ix86_handle_option): Handle OPT_mavx and OPT_mfma.
2950 (pta_flags): Add PTA_AVX and PTA_FMA.
2951 (override_options): Handle PTA_AVX and PTA_FMA.
2952 (init_cumulative_args): Handle warn_avx.
2953 (classify_argument): Return 0 for COImode and OImode. Return
2954 1 and X86_64_AVX_CLASS for 256bit vector types.
2955 (examine_argument): Handle X86_64_AVX_CLASS.
2956 (construct_container): Likewise.
2957 (function_arg_advance_32): Pass OImode and 256bit vector types
2959 (function_arg_advance_64): Take a new argument to indicate if a
2960 parameter is named. Handle 256bit vector types. Return
2961 immediately for unnamed 256bit vector mode parameters.
2962 (function_arg_advance): Updated.
2963 (function_arg_32): Add comments for TImode. Handle OImode
2964 and 256bit vector types.
2965 (function_arg_64): Take a new argument to indicate if a
2966 parameter is named. Handle 256bit vector types. Return NULL
2967 for unnamed 256bit vector mode parameters.
2968 (function_arg): Updated.
2969 (setup_incoming_varargs_64): Support
2970 AVX encoding for *sse_prologue_save_insn.
2971 (ix86_gimplify_va_arg): Handle 256bit vector mode parameters.
2972 (standard_sse_constant_p): Return -2 for all 1s if SSE2 isn't
2973 enabled. For all 1s in 256bit vector modes, return 3 if AVX is
2974 enabled, otherwise return -3.
2975 (standard_sse_constant_opcode): Handle AVX and 256bit vector
2977 (print_reg): Support AVX registers. Handle 'x' and 't'.
2978 Handle 'd' to duplicate the operand.
2979 (print_operand): Likewise. Also support AVX vector compare
2981 (output_387_binary_op): Support AVX.
2982 (output_fp_compare): Likewise.
2983 (ix86_expand_vector_move_misalign): Likewise.
2984 (ix86_attr_length_vex_default): New.
2985 (ix86_builtins): Add IX86_BUILTIN_ADDPD256,
2986 IX86_BUILTIN_ADDPS256, IX86_BUILTIN_ADDSUBPD256,
2987 IX86_BUILTIN_ADDSUBPS256, IX86_BUILTIN_ANDPD256,
2988 IX86_BUILTIN_ANDPS256, IX86_BUILTIN_ANDNPD256,
2989 IX86_BUILTIN_ANDNPS256, IX86_BUILTIN_BLENDPD256,
2990 IX86_BUILTIN_BLENDPS256, IX86_BUILTIN_BLENDVPD256,
2991 IX86_BUILTIN_BLENDVPS256, IX86_BUILTIN_DIVPD256,
2992 IX86_BUILTIN_DIVPS256, IX86_BUILTIN_DPPS256,
2993 IX86_BUILTIN_HADDPD256, IX86_BUILTIN_HADDPS256,
2994 IX86_BUILTIN_HSUBPD256, IX86_BUILTIN_HSUBPS256,
2995 IX86_BUILTIN_MAXPD256, IX86_BUILTIN_MAXPS256,
2996 IX86_BUILTIN_MINPD256, IX86_BUILTIN_MINPS256,
2997 IX86_BUILTIN_MULPD256, IX86_BUILTIN_MULPS256,
2998 IX86_BUILTIN_ORPD256, IX86_BUILTIN_ORPS256,
2999 IX86_BUILTIN_SHUFPD256, IX86_BUILTIN_SHUFPS256,
3000 IX86_BUILTIN_SUBPD256, IX86_BUILTIN_SUBPS256,
3001 IX86_BUILTIN_XORPD256, IX86_BUILTIN_XORPS256,
3002 IX86_BUILTIN_CMPSD, IX86_BUILTIN_CMPSS, IX86_BUILTIN_CMPPD,
3003 IX86_BUILTIN_CMPPS, IX86_BUILTIN_CMPPD256,
3004 IX86_BUILTIN_CMPPS256, IX86_BUILTIN_CVTDQ2PD256,
3005 IX86_BUILTIN_CVTDQ2PS256, IX86_BUILTIN_CVTPD2PS256,
3006 IX86_BUILTIN_CVTPS2DQ256, IX86_BUILTIN_CVTPS2PD256,
3007 IX86_BUILTIN_CVTTPD2DQ256, IX86_BUILTIN_CVTPD2DQ256,
3008 IX86_BUILTIN_CVTTPS2DQ256, IX86_BUILTIN_EXTRACTF128PD256,
3009 IX86_BUILTIN_EXTRACTF128PS256, IX86_BUILTIN_EXTRACTF128SI256,
3010 IX86_BUILTIN_VZEROALL, IX86_BUILTIN_VZEROUPPER,
3011 IX86_BUILTIN_VZEROUPPER_REX64, IX86_BUILTIN_VPERMILVARPD,
3012 IX86_BUILTIN_VPERMILVARPS, IX86_BUILTIN_VPERMILVARPD256,
3013 IX86_BUILTIN_VPERMILVARPS256, IX86_BUILTIN_VPERMILPD,
3014 IX86_BUILTIN_VPERMILPS, IX86_BUILTIN_VPERMILPD256,
3015 IX86_BUILTIN_VPERMILPS256, IX86_BUILTIN_VPERMIL2PD,
3016 IX86_BUILTIN_VPERMIL2PS, IX86_BUILTIN_VPERMIL2PD256,
3017 IX86_BUILTIN_VPERMIL2PS256, IX86_BUILTIN_VPERM2F128PD256,
3018 IX86_BUILTIN_VPERM2F128PS256, IX86_BUILTIN_VPERM2F128SI256,
3019 IX86_BUILTIN_VBROADCASTSS, IX86_BUILTIN_VBROADCASTSD256,
3020 IX86_BUILTIN_VBROADCASTSS256, IX86_BUILTIN_VBROADCASTPD256,
3021 IX86_BUILTIN_VBROADCASTPS256, IX86_BUILTIN_VINSERTF128PD256,
3022 IX86_BUILTIN_VINSERTF128PS256, IX86_BUILTIN_VINSERTF128SI256,
3023 IX86_BUILTIN_LOADUPD256, IX86_BUILTIN_LOADUPS256,
3024 IX86_BUILTIN_STOREUPD256, IX86_BUILTIN_STOREUPS256,
3025 IX86_BUILTIN_LDDQU256, IX86_BUILTIN_LOADDQU256,
3026 IX86_BUILTIN_STOREDQU256, IX86_BUILTIN_MASKLOADPD,
3027 IX86_BUILTIN_MASKLOADPS, IX86_BUILTIN_MASKSTOREPD,
3028 IX86_BUILTIN_MASKSTOREPS, IX86_BUILTIN_MASKLOADPD256,
3029 IX86_BUILTIN_MASKLOADPS256, IX86_BUILTIN_MASKSTOREPD256,
3030 IX86_BUILTIN_MASKSTOREPS256, IX86_BUILTIN_MOVSHDUP256,
3031 IX86_BUILTIN_MOVSLDUP256, IX86_BUILTIN_MOVDDUP256,
3032 IX86_BUILTIN_SQRTPD256, IX86_BUILTIN_SQRTPS256,
3033 IX86_BUILTIN_SQRTPS_NR256, IX86_BUILTIN_RSQRTPS256,
3034 IX86_BUILTIN_RSQRTPS_NR256, IX86_BUILTIN_RCPPS256,
3035 IX86_BUILTIN_ROUNDPD256, IX86_BUILTIN_ROUNDPS256,
3036 IX86_BUILTIN_UNPCKHPD256, IX86_BUILTIN_UNPCKLPD256,
3037 IX86_BUILTIN_UNPCKHPS256, IX86_BUILTIN_UNPCKLPS256,
3038 IX86_BUILTIN_SI256_SI, IX86_BUILTIN_PS256_PS,
3039 IX86_BUILTIN_PD256_PD, IX86_BUILTIN_SI_SI256,
3040 IX86_BUILTIN_PS_PS256, IX86_BUILTIN_PD_PD256,
3041 IX86_BUILTIN_VTESTZPD, IX86_BUILTIN_VTESTCPD,
3042 IX86_BUILTIN_VTESTNZCPD, IX86_BUILTIN_VTESTZPS,
3043 IX86_BUILTIN_VTESTCPS, IX86_BUILTIN_VTESTNZCPS,
3044 IX86_BUILTIN_VTESTZPD256, IX86_BUILTIN_VTESTCPD256,
3045 IX86_BUILTIN_VTESTNZCPD256, IX86_BUILTIN_VTESTZPS256,
3046 IX86_BUILTIN_VTESTCPS256, IX86_BUILTIN_VTESTNZCPS256,
3047 IX86_BUILTIN_PTESTZ256, IX86_BUILTIN_PTESTC256,
3048 IX86_BUILTIN_PTESTNZC256, IX86_BUILTIN_MOVMSKPD256
3049 and IX86_BUILTIN_MOVMSKPS256,
3050 (ix86_special_builtin_type): Add V32QI_FTYPE_PCCHAR,
3051 V8SF_FTYPE_PCV4SF, V8SF_FTYPE_PCFLOAT, V4DF_FTYPE_PCV2DF,
3052 V4DF_FTYPE_PCDOUBLE, V8SF_FTYPE_PCV8SF_V8SF,
3053 V4DF_FTYPE_PCV4DF_V4DF, V4SF_FTYPE_PCV4SF_V4SF,
3054 V2DF_FTYPE_PCV2DF_V2DF, VOID_FTYPE_PCHAR_V32QI,
3055 VOID_FTYPE_PFLOAT_V8SF, VOID_FTYPE_PDOUBLE_V4DF,
3056 VOID_FTYPE_PV8SF_V8SF_V8SF, VOID_FTYPE_PV4DF_V4DF_V4DF,
3057 VOID_FTYPE_PV4SF_V4SF_V4SF and VOID_FTYPE_PV2DF_V2DF_V2DF,
3058 (ix86_builtin_type): Add INT_FTYPE_V8SF_V8SF_PTEST,
3059 INT_FTYPE_V4DI_V4DI_PTEST, INT_FTYPE_V4DF_V4DF_PTEST,
3060 INT_FTYPE_V4SF_V4SF_PTEST, INT_FTYPE_V2DF_V2DF_PTEST,
3061 INT_FTYPE_V8SF, INT_FTYPE_V4DF, V8SI_FTYPE_V8SF, V8SI_FTYPE_V4SI,
3062 V8SF_FTYPE_V8SF, V8SF_FTYPE_V8SI, V8SF_FTYPE_V4SF,
3063 V4SI_FTYPE_V8SI, V4SI_FTYPE_V4DF, V4DF_FTYPE_V4DF,
3064 V4DF_FTYPE_V4SI, V4DF_FTYPE_V4SF, V4DF_FTYPE_V2DF,
3065 V4SF_FTYPE_V4DF, V4SF_FTYPE_V8SF, V2DF_FTYPE_V4DF,
3066 V8SF_FTYPE_V8SF_V8SF, V8SF_FTYPE_V8SF_V8SI,
3067 V4DF_FTYPE_V4DF_V4DF, V4DF_FTYPE_V4DF_V4DI,
3068 V4SF_FTYPE_V4SF_V4SI, V2DF_FTYPE_V2DF_V2DI,
3069 V8SF_FTYPE_V8SF_INT, V4SI_FTYPE_V8SI_INT, V4SF_FTYPE_V8SF_INT,
3070 V2DF_FTYPE_V4DF_INT, V4DF_FTYPE_V4DF_INT,
3071 V8SF_FTYPE_V8SF_V8SF_V8SF, V4DF_FTYPE_V4DF_V4DF_V4DF,
3072 V8SI_FTYPE_V8SI_V8SI_INT, V8SF_FTYPE_V8SF_V8SF_INT,
3073 V4DF_FTYPE_V4DF_V4DF_INT, V4DF_FTYPE_V4DF_V2DF_INT,
3074 V8SF_FTYPE_V8SF_V8SF_V8SI_INT, V4DF_FTYPE_V4DF_V4DF_V4DI_INT,
3075 V4SF_FTYPE_V4SF_V4SF_V4SI_INT and V2DF_FTYPE_V2DF_V2DF_V2DI_INT.
3076 (bdesc_special_args): Add IX86_BUILTIN_VZEROALL,
3077 IX86_BUILTIN_VZEROUPPER. IX86_BUILTIN_VZEROUPPER_REX64,
3078 IX86_BUILTIN_VBROADCASTSS, IX86_BUILTIN_VBROADCASTSD256,
3079 IX86_BUILTIN_VBROADCASTSS256, IX86_BUILTIN_VBROADCASTPD256,
3080 IX86_BUILTIN_VBROADCASTPS256, IX86_BUILTIN_LOADUPD256,
3081 IX86_BUILTIN_LOADUPS256, IX86_BUILTIN_STOREUPD256,
3082 IX86_BUILTIN_STOREUPS256, IX86_BUILTIN_LOADDQU256,
3083 IX86_BUILTIN_STOREDQU256, IX86_BUILTIN_LDDQU256,
3084 IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
3085 IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
3086 IX86_BUILTIN_MASKSTOREPD, IX86_BUILTIN_MASKSTOREPS,
3087 IX86_BUILTIN_MASKSTOREPD256 and IX86_BUILTIN_MASKSTOREPS256.
3088 (ix86_builtins): Add IX86_BUILTIN_ADDPD256,
3089 IX86_BUILTIN_ADDPS256, IX86_BUILTIN_ADDSUBPD256,
3090 IX86_BUILTIN_ADDSUBPS256, IX86_BUILTIN_ANDPD256,
3091 IX86_BUILTIN_ANDPS256, IX86_BUILTIN_ANDNPD256,
3092 IX86_BUILTIN_ANDNPS256, IX86_BUILTIN_DIVPD256,
3093 IX86_BUILTIN_DIVPS256, IX86_BUILTIN_HADDPD256,
3094 IX86_BUILTIN_HSUBPS256, IX86_BUILTIN_HSUBPD256,
3095 IX86_BUILTIN_HADDPS256, IX86_BUILTIN_MAXPD256,
3096 IX86_BUILTIN_MAXPS256, IX86_BUILTIN_MINPD256,
3097 IX86_BUILTIN_MINPS256, IX86_BUILTIN_MULPD256,
3098 IX86_BUILTIN_MULPS256, IX86_BUILTIN_ORPD256,
3099 IX86_BUILTIN_ORPS256, IX86_BUILTIN_SUBPD256,
3100 IX86_BUILTIN_SUBPS256, IX86_BUILTIN_XORPD256,
3101 IX86_BUILTIN_XORPS256, IX86_BUILTIN_VPERMILVARPD,
3102 IX86_BUILTIN_VPERMILVARPS, IX86_BUILTIN_VPERMILVARPD256,
3103 IX86_BUILTIN_VPERMILVARPS256, IX86_BUILTIN_BLENDPD256,
3104 IX86_BUILTIN_BLENDPS256, IX86_BUILTIN_BLENDVPD256,
3105 IX86_BUILTIN_BLENDVPS256, IX86_BUILTIN_DPPS256,
3106 IX86_BUILTIN_SHUFPD256, IX86_BUILTIN_SHUFPS256,
3107 IX86_BUILTIN_CMPSD, IX86_BUILTIN_CMPSS, IX86_BUILTIN_CMPPD,
3109 IX86_BUILTIN_CMPPD256,IX86_BUILTIN_CMPPS256,
3110 IX86_BUILTIN_EXTRACTF128PD256, IX86_BUILTIN_EXTRACTF128PS256,
3111 IX86_BUILTIN_EXTRACTF128SI256, IX86_BUILTIN_CVTDQ2PD256,
3112 IX86_BUILTIN_CVTDQ2PS256, IX86_BUILTIN_CVTPD2PS256,
3113 IX86_BUILTIN_CVTPS2DQ256, IX86_BUILTIN_CVTPS2PD256,
3114 IX86_BUILTIN_CVTTPD2DQ256, IX86_BUILTIN_CVTPD2DQ256,
3115 IX86_BUILTIN_CVTTPS2DQ256, IX86_BUILTIN_VPERM2F128PD256,
3116 IX86_BUILTIN_VPERM2F128PS256, IX86_BUILTIN_VPERM2F128SI256,
3117 IX86_BUILTIN_VPERMILPD, IX86_BUILTIN_VPERMILPS,
3118 IX86_BUILTIN_VPERMILPD256, IX86_BUILTIN_VPERMILPS256,
3119 IX86_BUILTIN_VPERMIL2PD, IX86_BUILTIN_VPERMILPS,
3120 IX86_BUILTIN_VPERMILPD256, IX86_BUILTIN_VPERMILPS256,
3121 IX86_BUILTIN_VPERMIL2PD, IX86_BUILTIN_VPERMIL2PS,
3122 IX86_BUILTIN_VPERMIL2PD256, IX86_BUILTIN_VPERMIL2PS256,
3123 IX86_BUILTIN_VINSERTF128PD256, IX86_BUILTIN_VINSERTF128PS256,
3124 IX86_BUILTIN_VINSERTF128SI256, IX86_BUILTIN_MOVSHDUP256,
3125 IX86_BUILTIN_MOVSLDUP256, IX86_BUILTIN_MOVDDUP256,
3126 IX86_BUILTIN_SQRTPD256, IX86_BUILTIN_SQRTPS256,
3127 IX86_BUILTIN_SQRTPS_NR256, IX86_BUILTIN_RSQRTPS256,
3128 IX86_BUILTIN_RSQRTPS_NR256, IX86_BUILTIN_RCPPS256,
3129 IX86_BUILTIN_ROUNDPD256, IX86_BUILTIN_ROUNDPS256,
3130 IX86_BUILTIN_UNPCKHPD256, IX86_BUILTIN_UNPCKLPD256,
3131 IX86_BUILTIN_UNPCKHPS256, IX86_BUILTIN_UNPCKLPS256,
3132 IX86_BUILTIN_SI256_SI, IX86_BUILTIN_PS256_PS,
3133 IX86_BUILTIN_PD256_PD, IX86_BUILTIN_SI_SI256,
3134 IX86_BUILTIN_PS_PS256, IX86_BUILTIN_PD_PD256,
3135 IX86_BUILTIN_VTESTZPD, IX86_BUILTIN_VTESTCPD,
3136 IX86_BUILTIN_VTESTNZCPD, IX86_BUILTIN_VTESTZPS,
3137 IX86_BUILTIN_VTESTCPS, IX86_BUILTIN_VTESTNZCPS,
3138 IX86_BUILTIN_VTESTZPD256, IX86_BUILTIN_VTESTCPD256,
3139 IX86_BUILTIN_VTESTNZCPD256, IX86_BUILTIN_VTESTZPS256,
3140 IX86_BUILTIN_VTESTCPS256, IX86_BUILTIN_VTESTNZCPS256,
3141 IX86_BUILTIN_PTESTZ256, IX86_BUILTIN_PTESTC256,
3142 IX86_BUILTIN_PTESTNZC256, IX86_BUILTIN_MOVMSKPD256 and
3143 IX86_BUILTIN_MOVMSKPS256.
3144 (ix86_init_mmx_sse_builtins): Support AVX builtins.
3145 (ix86_expand_args_builtin): Likewise.
3146 (ix86_expand_special_args_builtin): Likewise.
3147 (ix86_hard_regno_mode_ok): Handle AVX modes.
3148 (ix86_expand_vector_init_duplicate): Likewise.
3149 (ix86_expand_vector_init_one_nonzero): Likewise.
3150 (ix86_expand_vector_init_one_var): Likewise.
3151 (ix86_expand_vector_init_concat): Likewise.
3152 (ix86_expand_vector_init_general): Likewise.
3153 (ix86_expand_vector_set): Likewise.
3154 (ix86_vector_mode_supported_p): Likewise.
3155 (x86_extended_reg_mentioned_p): Check INSN_P before using
3158 * config/i386/i386-c.c (ix86_target_macros_internal): Handle
3159 OPTION_MASK_ISA_AVX and OPTION_MASK_ISA_FMA.
3161 * config/i386/i386.h (TARGET_AVX): New.
3162 (TARGET_FMA): Likewise.
3163 (TARGET_CPU_CPP_BUILTINS): Handle TARGET_AVX and TARGET_FMA.
3164 (BIGGEST_ALIGNMENT): Set to 256 for TARGET_AVX.
3165 (VALID_AVX256_REG_MODE): New.
3166 (AVX256_VEC_FLOAT_MODE_P): Likewise.
3167 (AVX_FLOAT_MODE_P): Likewise.
3168 (AVX128_VEC_FLOAT_MODE_P): Likewise.
3169 (AVX256_VEC_FLOAT_MODE_P): Likewise.
3170 (AVX_VEC_FLOAT_MODE_P): Likewise.
3171 (ASM_OUTPUT_AVX_PREFIX): Likewise.
3172 (ASM_OUTPUT_OPCODE): Likewise.
3173 (UNITS_PER_SIMD_WORD): Add a FIXME for 32byte vectorizer
3175 (SSE_REG_MODE_P): Allow 256bit vector modes.
3176 (ix86_args): Add a warn_avx field.
3178 * config/i386/i386.md (UNSPEC_PCMP): New.
3179 (UNSPEC_VPERMIL): Likewise.
3180 (UNSPEC_VPERMIL2): Likewise.
3181 (UNSPEC_VPERMIL2F128): Likewise.
3182 (UNSPEC_MASKLOAD): Likewise.
3183 (UNSPEC_MASKSTORE): Likewise.
3184 (UNSPEC_CAST): Likewise.
3185 (UNSPEC_VTESTP): Likewise.
3186 (UNSPECV_VZEROALL): Likewise.
3187 (UNSPECV_VZEROUPPER): Likewise.
3188 (XMM0_REG): Likewise.
3189 (XMM1_REG): Likewise.
3190 (XMM2_REG): Likewise.
3191 (XMM3_REG): Likewise.
3192 (XMM4_REG): Likewise.
3193 (XMM5_REG): Likewise.
3194 (XMM6_REG): Likewise.
3195 (XMM8_REG): Likewise.
3196 (XMM9_REG): Likewise.
3197 (XMM10_REG): Likewise.
3198 (XMM11_REG): Likewise.
3199 (XMM12_REG): Likewise.
3200 (XMM13_REG): Likewise.
3201 (XMM14_REG): Likewise.
3202 (XMM15_REG): Likewise.
3204 (prefix_vex_imm8): Likewise.
3205 (prefix_vex_w): Likewise.
3206 (length_vex): Likewise.
3209 (*avx_ashlti3): Likewise.
3210 (*avx_lshrti3): Likewise.
3211 (*avx_setcc<mode>): Likewise.
3212 (*fop_<mode>_comm_mixed_avx): Likewise.
3213 (*fop_<mode>_comm_avx): Likewise.
3214 (*fop_<mode>_1_mixed_avx): Likewise.
3215 (*fop_<mode>_1_avx): Likewise.
3216 (*avx_<code><mode>3): Likewise.
3217 (*avx_ieee_smin<mode>3): Likewise.
3218 (*avx_ieee_smax<mode>3): Likewise.
3219 (mode): Add OI, V8SF and V4DF.
3220 (length): Support VEX prefix.
3221 (*cmpfp_i_mixed): Set prefix attribute.
3222 (*cmpfp_i_sse): Likewise.
3223 (*cmpfp_iu_mixed): Likewise.
3224 (*cmpfp_iu_sse): Likewise.
3225 (*movsi_1): Support AVX.
3226 (*movdi_2): Likewise.
3227 (*movdi_1_rex64): Likewise.
3228 (*movti_internal): Likewise.
3229 (*movti_rex64): Likewise.
3230 (*movsf_1): Likewise.
3231 (*movdf_nointeger): Likewise.
3232 (*movdf_integer_rex64): Likewise.
3233 (*movtf_internal): Likewise.
3234 (zero_extendsidi2_32): Likewise.
3235 (zero_extendsidi2_rex64): Likewise.
3236 (*extendsfdf2_mixed): Likewise.
3237 (*extendsfdf2_sse): Likewise.
3238 (*truncdfsf_fast_mixed): Likewise.
3239 (*truncdfsf_fast_sse): Likewise.
3240 (*truncdfsf_mixed): Likewise.
3241 (fix_trunc<mode>di_sse): Likewise.
3242 (fix_trunc<mode>si_sse): Likewise.
3243 (*float<SSEMODEI24:mode><MODEF:mode>2_mixed_interunit): Likewise.
3244 (*float<SSEMODEI24:mode><MODEF:mode>2_mixed_nointerunit): Likewise.
3245 (*float<SSEMODEI24:mode><MODEF:mode>2_sse_interunit): Likewise.
3246 (*float<SSEMODEI24:mode><MODEF:mode>2_sse_nointerunit): Likewise.
3247 (*rcpsf2_sse): Likewise.
3248 (*rsqrtsf2_sse): Likewise.
3249 (*sqrt<mode>2_sse): Likewise.
3250 (sse4_1_round<mode>2): Likewise.
3251 (*sse_prologue_save_insn): Disallow REX prefix for AVX.
3252 Support AVX. Set length attribute properly for AVX.
3254 * config/i386/i386-modes.def (VECTOR_MODES (INT, 32)): New.
3255 (VECTOR_MODES (FLOAT, 32)): Likewise.
3256 (VECTOR_MODE (INT, DI, 8)): Likewise.
3257 (VECTOR_MODE (INT, HI, 32)): Likewise.
3258 (VECTOR_MODE (INT, QI, 64)): Likewise.
3259 (VECTOR_MODE (FLOAT, DF, 8)): Likewise.
3260 (VECTOR_MODE (FLOAT, SF, 16)): Likewise.
3261 (VECTOR_MODE (INT, DI, 4)): Removed.
3262 (VECTOR_MODE (INT, SI, 8)): Likewise.
3263 (VECTOR_MODE (INT, HI, 16)): Likewise.
3264 (VECTOR_MODE (INT, QI, 32)): Likewise.
3265 (VECTOR_MODE (FLOAT, SF, 8)): Likewise.
3266 (INT_MODE (OI, 32)): Likewise.
3268 * config/i386/i386.opt (mavx): New.
3271 * config/i386/i386-protos.h (ix86_attr_length_vex_default): New.
3273 * config/i386/mmx.md (*mov<mode>_internal_rex64): Support AVX.
3274 (*mov<mode>_internal_avx): New.
3275 (*movv2sf_internal_rex64_avx): Likewise.
3276 (*movv2sf_internal_avx): Likewise.
3278 * config/i386/predicates.md (const_4_to_5_operand): New.
3279 (const_6_to_7_operand): Likewise.
3280 (const_8_to_11_operand): Likewise.
3281 (const_12_to_15_operand): Likewise.
3282 (avx_comparison_float_operator): Likewise.
3284 * config/i386/sse.md (AVX256MODEI): New.
3285 (AVX256MODE): Likewise.
3286 (AVXMODEQI): Likewise.
3287 (AVXMODE): Likewise.
3288 (AVX256MODEF2P): Likewise.
3289 (AVX256MODE2P): Likewise.
3290 (AVX256MODE4P): Likewise.
3291 (AVX256MODE8P): Likewise.
3292 (AVXMODEF2P): Likewise.
3293 (AVXMODEF4P): Likewise.
3294 (AVXMODEDCVTDQ2PS): Likewise.
3295 (AVXMODEDCVTPS2DQ): Likewise.
3296 (avxvecmode): Likewise.
3297 (avxvecpsmode): Likewise.
3298 (avxhalfvecmode): Likewise.
3299 (avxscalarmode): Likewise.
3300 (avxcvtvecmode): Likewise.
3301 (avxpermvecmode): Likewise.
3302 (avxmodesuffixf2c): Likewise.
3303 (avxmodesuffixp): Likewise.
3304 (avxmodesuffixs): Likewise.
3305 (avxmodesuffix): Likewise.
3306 (vpermilbits): Likewise.
3307 (pinsrbits): Likewise.
3308 (mov<mode>): Likewise.
3309 (*mov<mode>_internal): Likewise.
3310 (push<mode>1): Likewise.
3311 (movmisalign<mode>): Likewise.
3312 (avx_movup<avxmodesuffixf2c><avxmodesuffix>): Likewise.
3313 (avx_movdqu<avxmodesuffix>): Likewise.
3314 (avx_lddqu<avxmodesuffix>): Likewise.
3315 (<plusminus_insn><mode>3): Likewise.
3316 (*avx_<plusminus_insn><mode>3): Likewise.
3317 (*avx_vm<plusminus_insn><mode>3): Likewise.
3318 (mul<mode>3): Likewise.
3319 (*avx_mul<mode>3): Likewise.
3320 (*avx_vmmul<mode>3): Likewise.
3321 (divv8sf3): Likewise.
3322 (divv4df3): Likewise.
3323 (avx_div<mode>3): Likewise.
3324 (*avx_div<mode>3): Likewise.
3325 (*avx_vmdiv<mode>3): Likewise.
3326 (avx_rcpv8sf2): Likewise.
3327 (*avx_vmrcpv4sf2): Likewise.
3328 (sqrtv8sf2): Likewise.
3329 (avx_sqrtv8sf2): Likewise.
3330 (*avx_vmsqrt<mode>2): Likewise.
3331 (rsqrtv8sf2): Likewise.
3332 (avx_rsqrtv8sf2): Likewise.
3333 (*avx_vmrsqrtv4sf2): Likewise.
3334 (<code><mode>3): Likewise.
3335 (*avx_<code><mode>3_finite): Likewise.
3336 (*avx_<code><mode>3): Likewise.
3337 (*avx_vm<code><mode>3): Likewise.
3338 (*avx_ieee_smin<mode>3): Likewise.
3339 (*avx_ieee_smax<mode>3): Likewise.
3340 (avx_addsubv8sf3): Likewise.
3341 (avx_addsubv4df3): Likewise.
3342 (*avx_addsubv4sf3): Likewise.
3343 (*avx_addsubv2df3): Likewise.
3344 (avx_h<plusminus_insn>v4df3): Likewise.
3345 (avx_h<plusminus_insn>v8sf3): Likewise.
3346 (*avx_h<plusminus_insn>v4sf3): Likewise.
3347 (*avx_h<plusminus_insn>v2df3): Likewise.
3348 (avx_cmpp<avxmodesuffixf2c><mode>3): Likewise.
3349 (avx_cmps<ssemodesuffixf2c><mode>3): Likewise.
3350 (*avx_maskcmp<mode>3): Likewise.
3351 (avx_nand<mode>3): Likewise.
3352 (*avx_<code><mode>3): Likewise.
3353 (*avx_nand<mode>3): Likewise.
3354 (*avx_<code><mode>3): Likewise.
3355 (*avx_cvtsi2ss): Likewise.
3356 (*avx_cvtsi2ssq): Likewise.
3357 (*avx_cvtsi2sd): Likewise.
3358 (*avx_cvtsi2sdq): Likewise.
3359 (*avx_cvtsd2ss): Likewise.
3360 (avx_cvtss2sd): Likewise.
3361 (avx_cvtdq2ps<avxmodesuffix>): Likewise.
3362 (avx_cvtps2dq<avxmodesuffix>): Likewise.
3363 (avx_cvttps2dq<avxmodesuffix>): Likewise.
3364 (*avx_cvtsi2sd): Likewise.
3365 (*avx_cvtsi2sdq): Likewise.
3366 (avx_cvtdq2pd256): Likewise.
3367 (avx_cvtpd2dq256): Likewise.
3368 (avx_cvttpd2dq256): Likewise.
3369 (*avx_cvtsd2ss): Likewise.
3370 (*avx_cvtss2sd): Likewise.
3371 (avx_cvtpd2ps256): Likewise.
3372 (avx_cvtps2pd256): Likewise.
3373 (*avx_movhlps): Likewise.
3374 (*avx_movlhps): Likewise.
3375 (avx_unpckhps256): Likewise.
3376 (*avx_unpckhps): Likewise.
3377 (avx_unpcklps256): Likewise.
3378 (*avx_unpcklps): Likewise.
3379 (avx_movshdup256): Likewise.
3380 (avx_movsldup256): Likewise.
3381 (avx_shufps256): Likewise.
3382 (avx_shufps256_1): Likewise.
3383 (*avx_shufps_<mode>): Likewise.
OSDN Git Service
