1 2011-10-25 Ian Lance Taylor <iant@google.com>
3 * tree-eh.c (do_return_redirection): Remove return_value_p
4 parameter. Change all callers.
5 (lower_try_finally_nofallthru): Remove local return_val.
6 (lower_try_finally_onedest): Likewise.
7 (lower_try_finally_copy): Likewise.
8 (lower_try_finally_switch): Likewise.
10 2011-10-25 H.J. Lu <hongjiu.lu@intel.com>
12 * config/i386/mmx.md (*mmx_maskmovq): Replace :SI with :P and
13 remove "&& !TARGET_64BIT"
14 (*mmx_maskmovq_rex): Removed.
16 2011-10-25 Eric Botcazou <ebotcazou@adacore.com>
18 PR rtl-optimization/46603
19 * reload.c (push_reload): In the out case, reload the subreg as well
20 as the reg if it has word mode.
22 2011-10-25 Eric Botcazou <ebotcazou@adacore.com>
24 * dwarf2out.c (add_gnat_descriptive_type_attribute): Temporarily
25 suppress debug info for the parent type.
27 2011-10-25 Eric Botcazou <ebotcazou@adacore.com>
29 * config/ia64/ia64.c (ia64_profile_hook): Fix thinko.
31 2011-10-25 Richard Henderson <rth@redhat.com>
33 * config/i386/sse.md (VEC_EXTRACT_EVENODD_MODE): Remove.
34 (vec_extract_even<mode>, vec_extract_odd<mode>): Remove.
36 * config/rs6000/altivec.md (vec_extract_evenv8hi,
37 vec_extract_evenv16qi, vec_extract_oddv4si,
38 vec_extract_oddv4sf): Remove.
40 * config/spu/spu.md (vec_extract_evenv4si, vec_extract_evenv4sf,
41 vec_extract_evenv8hi, vec_extract_evenv16qi, vec_extract_oddv4si,
42 vec_extract_oddv4sf, vec_extract_oddv8hi, vec_extract_oddv16qi,
43 vec_interleave_highv4sf, vec_interleave_lowv4sf,
44 vec_interleave_highv4si, vec_interleave_lowv4si,
45 vec_interleave_highv8hi, vec_interleave_lowv8hi,
46 vec_interleave_highv16qi, vec_interleave_lowv16qi): Remove.
48 * expr.c (expand_expr_real_2) [VEC_EXTRACT_EVEN_EXPR]: Use binop.
49 [VEC_EXTRACT_ODD_EXPR, VEC_INTERLEAVE_HIGH_EXPR]: Likewise.
50 [VEC_INTERLEAVE_LOW_EXPR]: Likewise.
51 * optabs.c (expand_binop): Implement vec_interleave_high_optab,
52 vec_interleave_low_optab, vec_extract_even_optab,
53 vec_extract_odd_optab with expand_vec_perm.
54 (can_vec_perm_for_code_p): New.
56 * tree-vect-data-refs.c (vect_strided_store_supported): Allow for
57 fallback via can_vec_perm_for_code_p.
58 (vect_strided_load_supported): Likewise.
59 * tree-vect-generic.c (expand_vector_operations_1): Never lower
60 VEC_INTERLEAVE_HIGH_EXPR, VEC_INTERLEAVE_LOW_EXPR,
61 VEC_EXTRACT_EVEN_EXPR, VEC_EXTRACT_ODD_EXPR.
63 * target.def (vec_perm_const_ok): Change parameters to mode and
65 * doc/tm.texi: Rebuild.
66 * config/i386/i386.c (ix86_vectorize_vec_perm_const_ok): Change
67 parameters to mode and array of indicies.
68 * expr.c (expand_expr_real_2) [VEC_PERM_EXPR]: Expand operands here.
69 * optabs.c (can_vec_perm_p): Rename from can_vec_perm_expr_p.
70 Change parameters to mode and array of indicies.
71 (expand_vec_perm_1): Rename from expand_vec_perm_expr_1.
72 (expand_vec_perm): Rename from expand_vec_perm_expr. Change
73 parameters to mode and rtx inputs. Try lowering to QImode
74 vec_perm_const before trying fully variable permutation.
75 * optabs.h: Update decls.
76 * tree-vect-generic.c (lower_vec_perm): Extract array of indices from
77 VECTOR_CST to pass to can_vec_perm_p.
78 * tree-vect-slp.c (vect_get_mask_element): Change mask parameter type
79 from int pointer to unsigned char pointer.
80 (vect_transform_slp_perm_load): Update for change to can_vec_perm_p.
81 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
83 * tree.def (VEC_EXTRACT_EVEN_EXPR): Fix typo in text name.
84 (VEC_EXTRACT_ODD_EXPR, VEC_INTERLEAVE_HIGH_EXPR,
85 VEC_INTERLEAVE_LOW_EXPR): Likewise.
87 2011-10-25 Mike Stump <mikestump@comcast.net>
89 * reload.c (regno_clobbered_p): Fix typo.
91 2011-10-25 Dodji Seketeli <dodji@redhat.com>
93 * input.c (expand_location): Rewrite using
94 linemap_resolve_location and linemap_expand_location. Add a
97 2011-10-25 Jakub Jelinek <jakub@redhat.com>
99 PR tree-optimization/50596
100 * tree-vect-stmts.c (vect_mark_relevant): Only use
101 FOR_EACH_IMM_USE_FAST if lhs is SSA_NAME.
102 (vectorizable_store): If is_pattern_stmt_p look through
103 VIEW_CONVERT_EXPR on lhs.
104 * tree-vect-patterns.c (check_bool_pattern, adjust_bool_pattern):
105 Use unsigned type instead of signed.
106 (vect_recog_bool_pattern): Optimize also stores into bool memory in
107 addition to casts from bool to integral types.
108 (vect_mark_pattern_stmts): If pattern_stmt already has vinfo
109 created, don't create it again.
111 2011-10-25 Kai Tietz <ktietz@redhat.com>
113 * config/i386/i386.c (ix86_frame_pointer_required): Require
114 frame-pointer, if setjmp is used for 32-bit ms-abi.
116 2011-10-24 Michael Meissner <meissner@linux.vnet.ibm.com>
118 * builtins.c (set_builtin_user_assembler_name): Remove extra
119 newline added in October 11th, 2011 change.
121 2011-10-24 David S. Miller <davem@davemloft.net>
123 * config/sparc/little-endian.opt: Delete.
124 * config.gcc: Remove references to config/sparc/little-endian.opt
125 * doc/invoke.texi: Remove documentation of -mlittl-endian on sparc.
126 * config/sparc/linux64.h: Delete references to -mlittle-endian.
127 * config/sparc/netbsd-elf.h: Likewise.
128 * config/sparc/openbsd64.h: Likewise.
129 * config/sparc/sparc.h: Likewise.
130 * config/sparc/sp64-elf.h: Likewise and delete overrides for
131 BYTES_BIG_ENDIAN and WORDS_BIG_ENDIAN.
132 * config/sparc/sparc.c (dump_target_flag_bits): Remove reference
133 to MASK_LITTLE_ENDIAN.
134 * config/sparc/sparc.opt (Mask(LITTLE_ENDIAN)): Delete.
136 * config/sparc/sparc.md: Only use F, G, and C constraints in FP
137 insns. Only use D, Y, and Z constraints in vector insns.
139 * config/sparc/sparc.md (cpu_feature, enabled): New attributes.
140 (*movsi_insn_novis3, *movsi_insn_vis3): Consolidate into one pattern
142 (*movdi_insn_sp32_v9_novis3, *movdi_insn_sp32_v9_vis3): Consolidate
143 into *movdi_insn_sp32.
144 (*movdi_insn_sp64_novis3, *movdi_insn_sp64_vis3): Consolidate into
145 one pattern called *movdi_insn_sp64.
146 (*movsf_insn_novis3, *movsf_insn_vis3, *movsf_insn_no_fpu):
147 Consolidate into one pattern called *movsf_insn.
148 (*movdf_insn_sp32_no_fpu, *movdf_insn_sp32_v9_novis3,
149 *movdf_insn_sp32_v9_vis3, *movdf_insn_sp32_v9_no_fpu): Consolidate
150 into *movdf_insn_sp32.
151 (*movdf_insn_sp64_novis3, *movdf_insn_sp64_vis3,
152 *movdf_insn_sp64_no_fpu): Consolidate into one pattern called
154 (*zero_extendsidi2_insn_sp64_novis3,
155 *zero_extendsidi2_insn_sp64_vis3): Consolidate into one pattern
156 called *zero_extendsidi2_insn_sp64.
157 (*sign_extendsidi2_insn_novis3, *sign_extendsidi2_insn_vis3):
158 Consolidate into one pattern named *sign_extendsidi2_insn.
159 (*mov<VM32:mode>_insn_novis3, *mov<VM32:mode>_insn_vis3):
160 Consolidate into one pattern named *mov<VM32:mode>_insn.
161 (*mov<VM64:mode>_insn_sp64_novis3,
162 *mov<VM64:mode>_insn_sp64_novis3): Consolidate into one pattern
163 named *mov<VM64:mode>_insn_sp64.
164 (*mov<VM64:mode>_insn_sp32_novis3,
165 *mov<VM64:mode>_insn_sp32_vis3): Consolidate into one pattern
166 named *mov<VM64:mode>_insn_sp32.
168 2011-10-24 Andreas Krebbel <Andreas.Krebbel@de.ibm.com>
170 * tree-ssa-strlen.c (get_string_length): Change assertion to STPCPY.
171 (zero_length_string): Change assertion to accept strinfo without
172 length but with stmt instead.
173 Set the endptr pointer also if starting a new chain.
174 (adjust_related_strinfos): Ignore strinfos marked for delayed
176 (handle_builtin_strcpy): Mark earlier strinfo elements also for
177 delayed length computation.
179 2011-10-24 Georg-Johann Lay <avr@gjlay.de>
182 Port from 4.6 branch r180379
183 * doc/invoke.texi (AVR Options): New subsubsection to explain EIND
184 handling and indirect jump/calls on devices > 128k.
186 2011-10-24 Anatoly Sokolov <aesok@post.ru>
187 Georg-Johann Lay <avr@gjlay.de>
190 * doc/extend.texi (Declaring Attributes of Functions):
191 Document OS_main and OS_task attributes.
192 (Specifying Attributes of Variables): Move up
193 subsection "AVR Variable Attributes" as of alphabetical order.
195 2011-10-24 Richard Guenther <rguenther@suse.de>
197 * tree-vect-stmts.c (vect_get_vec_def_for_operand): Convert constants
198 to vector element type.
199 (vectorizable_assignment): Bail out for non-mode-precision operations.
200 (vectorizable_shift): Likewise.
201 (vectorizable_operation): Likewise.
202 (vectorizable_type_demotion): Likewise.
203 (vectorizable_type_promotion): Likewise.
204 (vectorizable_store): Handle non-mode-precision stores.
205 (vectorizable_load): Handle non-mode-precision loads.
206 (get_vectype_for_scalar_type_and_size): Return a vector type
207 for non-mode-precision integers.
208 * tree-vect-loop.c (vectorizable_reduction): Bail out for
209 non-mode-precision reductions.
211 2011-10-24 Julian Brown <julian@codesourcery.com>
213 * config/m68k/m68k.c (notice_update_cc): Tighten condition for
214 setting CC_REVERSED for FP comparisons.
216 2011-10-24 Richard Guenther <rguenther@suse.de>
218 PR tree-optimization/50838
219 * tree-data-ref.c (dr_analyze_indices): Properly canonicalize
220 a MEM_REF base if we change it.
222 2011-10-24 Bernd Schmidt <bernds@codesourcery.com>
225 * rtlanal.c: Swap includes of "hard-reg-set.h" and "rtl.h".
227 PR rtl-optimization/50833
228 * function.c (thread_prologue_and_epilogue_insns): Expect the
229 return insn optimization only if optimize.
231 2011-10-24 Georg-Johann Lay <avr@gjlay.de>
233 * config/avr/avr.c: Break long lines.
234 Define target hooks on the fly if applicable.
235 (TARGET_ASM_FUNCTION_RODATA_SECTION): Remove first definition
237 (targetm): Move definition to end of file.
238 (avr_can_eliminate): Make static on the fly.
239 (avr_frame_pointer_required_p): Ditto.
240 (avr_hard_regno_scratch_ok): Ditto.
241 (avr_builtin_setjmp_frame_value): Make static on the fly.
242 Indent according to coding rules.
243 (avr_case_values_threshold): Ditto.
244 (avr_attribute_table): Move down.
246 2011-10-24 Ira Rosen <ira.rosen@linaro.org>
248 PR tree-optimization/50730
249 * tree-vect-data-refs.c (vect_analyze_data_refs): Stop basic block
250 analysis if encountered unsupported data-ref.
252 2011-10-23 David S. Miller <davem@davemloft.net>
254 * config/sparc/sparc.c (sparc_option_override): Remove -mv8plus
256 * config/sparc/linux64.h (CC1_SPEC): When defaulting to 64-bit,
257 append -mcpu=v9 when -mv8plus is given.
259 * config/sparc/sparc.h (SECONDARY_MEMORY_NEEDED): We can move
260 between float and non-float regs when VIS3.
261 * config/sparc/sparc.c (eligible_for_restore_insn): We can't
262 use a restore when the source is a float register.
263 (sparc_split_regreg_legitimate): When VIS3 allow moves between
264 float and integer regs.
265 (sparc_register_move_cost): Adjust to account for VIS3 moves.
266 (sparc_preferred_reload_class): On 32-bit with VIS3 when moving an
267 integer reg to a class containing EXTRA_FP_REGS, constrain to
269 (sparc_secondary_reload): On 32-bit with VIS3 when moving between
270 float and integer regs we sometimes need a FP_REGS class
271 intermediate move to satisfy the reload. When this happens
272 specify an extra cost of 2.
273 (*movsi_insn): Rename to have "_novis3" suffix and add !VIS3
275 (*movdi_insn_sp32_v9): Likewise.
276 (*movdi_insn_sp64): Likewise.
277 (*movsf_insn): Likewise.
278 (*movdf_insn_sp32_v9): Likewise.
279 (*movdf_insn_sp64): Likewise.
280 (*zero_extendsidi2_insn_sp64): Likewise.
281 (*sign_extendsidi2_insn): Likewise.
282 (*movsi_insn_vis3): New insn.
283 (*movdi_insn_sp32_v9_vis3): New insn.
284 (*movdi_insn_sp64_vis3): New insn.
285 (*movsf_insn_vis3): New insn.
286 (*movdf_insn_sp32_v9_vis3): New insn.
287 (*movdf_insn_sp64_vis3): New insn.
288 (*zero_extendsidi2_insn_sp64_vis3): New insn.
289 (*sign_extendsidi2_insn_vis3): New insn.
290 (TFmode reg/reg split): Make sure both REG operands are float.
291 (*mov<VM32:mode>_insn): Add "_novis3" suffix and !VIS3 guard. Remove
292 easy constant to integer reg alternatives.
293 (*mov<VM64:mode>_insn_sp64): Likewise.
294 (*mov<VM64:mode>_insn_sp32_novis3): Likewise.
295 (*mov<VM32:mode>_insn_vis3): New insn.
296 (*mov<VM64:mode>_insn_sp64_vis3): New insn.
297 (*mov<VM64:mode>_insn_sp32_vis3): New insn.
298 (VM64 reg<-->reg split): New spliiter for 32-bit.
300 * config/sparc/sparc.c (sparc_split_regreg_legitimate): New
302 * config/sparc/sparc-protos.h (sparc_split_regreg_legitimate):
304 * config/sparc/sparc.md (DImode reg/reg split): Use it.
305 (DFmode reg/reg split): Likewise.
307 * config/sparc/sparc.md (*movdi_insn_sp32_v9): Add alternatives for
308 generating fzero and fone instructions.
309 (DImode const_int --> reg splitter): Only trigger for integer regs.
311 * config/sparc/predicates.md (input_operand): Disallow vector
312 constants other than 0 and -1.
313 * config/sparc/sparc.c (sparc_preferred_reload_class): Return
314 NO_REGS for vector constants other than 0 and -1.
316 * config/sparc/sparc.h (SPARC_FIRST_INT_REG, SPARC_LAST_INT_REG,
317 SPARC_INT_REG_P): Define.
318 (HARD_REGNO_NREGS): Use SPARC_INT_REG_P.
319 (REGNO_OK_FOR_INDEX_P): Likewise.
320 * config/sparc/sparc.c (gen_df_reg): Likewise.
321 (eligible_for_return_delay): Likewise.
322 (eligible_for_sibcall_delay): Likewise.
323 (sparc_legitimate_address_p): Likewise.
324 (emit_save_or_restore_regs): Likewise.
325 (registers_ok_for_ldd_peep): Likewise.
326 * config/spac/sparc.md (DI mode splitters): Likewise.
327 (SF mode const splitters): Likewise.
328 (DF mode splitters): Likewise.
329 (32-bit DI mode logical op splitters): Likewise.
331 2011-10-23 Paolo Carlini <paolo.carlini@oracle.com>
335 2011-10-23 Paolo Carlini <paolo.carlini@oracle.com>
338 * doc/invoke.texi ([-Wnarrowing], [-Wc++0x-compat]): Update.
340 2011-10-23 Uros Bizjak <ubizjak@gmail.com>
342 * config/i386/sse.md (avx2_lshl<mode>3): Remove insn pattern.
343 (VI248_256): Remove mode iterator.
344 * config/i386/i386.c (ix86_expand_vec_perm): Use gen_ashlv4di3
345 instead of gen_avx2_lshlv4di3.
346 (bdesc_args): Use CODE_FOR_ashl{v16hi,v8si,v4di}3 instead of
347 CODE_FOR_avx2_lshl{v16hi,v8si,v4di}3.
349 2011-10-23 Uros Bizjak <ubizjak@gmail.com>
351 * config/i386/sse.md (sseintprefix): Rename from gthrfirstp.
352 (<avx_avx2>_maskload<ssemodesuffix><avxsizesuffix>): Delete expander.
353 (<avx_avx2>_maskload<ssemodesuffix><avxsizesuffix>) Merge insn
354 pattern from *avx2_maskload<ssemodesuffix><avxsizesuffix> and
355 *avx_maskload<ssemodesuffix><avxsizesuffix> using V48_AVX mode
356 iterator. Use sseintprefix mode attribute.
357 (<avx_avx2>_maskstore<ssemodesuffix><avxsizesuffix>): Delete expander.
358 (<avx_avx2>_maskstore<ssemodesuffix><avxsizesuffix>) Merge insn
359 pattern from *avx2_maskstore<ssemodesuffix><avxsizesuffix> and
360 *avx_maskstore<ssemodesuffix><avxsizesuffix> using V48_AVX mode
361 iterator. Use sseintprefix mode attribute.
362 (*avx2_gathersi<mode>) Use sseintprefix and ssemodesuffix mode
364 (*avx2_gatherdi<mode>): Ditto.
365 (*avx2_gatherdi<mode>256): Ditto.
366 (VI48_AVX2): Remove mode iterator.
367 (gthrlastfp): Remove mode attribute.
369 2011-10-23 Paolo Carlini <paolo.carlini@oracle.com>
372 * doc/invoke.texi ([-Wnarrowing], [-Wc++0x-compat]): Update.
374 2011-10-23 Tom de Vries <tom@codesourcery.com>
376 PR tree-optimization/50763
377 * tree-ssa-tail-merge.c (same_succ_flush_bb): New function, factored out
379 (same_succ_flush_bbs): Use same_succ_flush_bb.
380 (purge_bbs): Remove argument. Remove calls to same_succ_flush_bbs,
381 release_last_vdef and delete_basic_block.
382 (unlink_virtual_phi): New function.
383 (update_vuses): Add and use vuse1_phi_args argument. Set var to
384 SSA_NAME_VAR of vuse1 or vuse2, and use var. Handle case that def_stmt2
385 is NULL. Use phi result as phi arg in case vuse1 or vuse2 is NULL_TREE.
386 Replace uses of vuse1 if vuse2 is NULL_TREE. Fix code to limit
387 replacement of uses. Propagate phi argument for phis with a single
389 (replace_block_by): Update vops if phi_vuse1 or phi_vuse2 is NULL_TREE.
390 Set vuse1_phi_args if vuse1 is a phi defined in bb1. Add vuse1_phi_args
391 as argument to call to update_vuses. Call release_last_vdef,
392 same_succ_flush_bb, delete_basic_block. Update CDI_DOMINATORS info.
393 (tail_merge_optimize): Remove argument in call to purge_bbs. Remove
394 call to free_dominance_info. Only call calculate_dominance_info once.
396 2011-10-23 Eric Botcazou <ebotcazou@adacore.com>
398 * fold-const.c (invert_tree_comparison): Always invert EQ_EXPR/NE_EXPR.
400 PR tree-optimization/44683
401 * tree-ssa-dom.c (record_edge_info): Record simple equivalences only if
402 we can be sure that there are no signed zeros involved.
404 2011-10-23 Jan Hubicka <jh@suse.cz>
406 * ipa-inline.c (estimate_badness): Scale up and handle overflows.
408 2011-10-23 Uros Bizjak <ubizjak@gmail.com>
411 * config/i386/sse.md (<avx_avx2>_maskload<ssemodesuffix><avxsizesuffix>):
412 Remove (match_dup 0).
413 (*avx2_maskload<ssemodesuffix><avxsizesuffix>): New insn pattern.
414 (*avx_maskload<ssemodesuffix><avxsizesuffix>): Ditto.
415 (*avx2_maskstore<ssemodesuffix><avxsizesuffix>): Ditto.
416 (*avx_maskstore<ssemodesuffix><avxsizesuffix>): Ditto.
417 (*avx2_maskmov<ssemodesuffix><avxsizesuffix>): Remove insn pattern.
418 (*avx_maskmov<ssemodesuffix><avxsizesuffix>): Ditto.
420 2011-10-23 Ira Rosen <ira.rosen@linaro.org>
422 PR tree-optimization/50819
423 * tree-vectorizer.h (vect_analyze_data_ref_dependences): Remove
425 * tree-vect-loop.c (vect_analyze_loop_2): Update call to
426 vect_analyze_data_ref_dependences.
427 * tree-vect-data-refs.c (vect_analyze_data_ref_dependence): Remove
428 the last argument. Check load-after-store dependence for unknown
429 dependencies in basic blocks.
430 (vect_analyze_data_ref_dependences): Update call to
431 vect_analyze_data_ref_dependences.
432 * tree-vect-patterns.c (vect_recog_widen_shift_pattern): Fix typo.
433 * tree-vect-slp.c (vect_bb_vectorizable_with_dependencies): Remove.
434 (vect_slp_analyze_bb_1): Update call to
435 vect_analyze_data_ref_dependences. Don't call
436 vect_bb_vectorizable_with_dependencies.
438 2011-10-22 David S. Miller <davem@davemloft.net>
440 * config/sparc/sparc.h (SECONDARY_INPUT_RELOAD_CLASS,
441 SECONDARY_OUTPUT_RELOAD_CLASS): Delete.
442 * config/sparc/sparc.c (TARGET_SECONDARY_RELOAD): Redefine.
443 (sparc_secondary_reload): New function.
445 * config/sparc/sparc.h (sparc_costs): Remove extern decl.
446 (struct processor_costs): Move from here..
447 * config/sparc/sparc.c (struct processor_costs): To here.
448 (sparc_costs): Mark static.
450 * config/sparc/sparc.c (short_branch, reg_unused_after): Delete.
451 * config/sparc/sparc-protos.h (short_branch, reg_unused_after):
452 Get rid of declarations.
454 2011-10-21 Paul Brook <paul@codesourcery.com>
456 * config/c6x/c6x.c (c6x_asm_emit_except_personality,
457 c6x_asm_init_sections): New functions.
458 (TARGET_ASM_EMIT_EXCEPT_PERSONALITY, TARGET_ASM_INIT_SECTIONS): Define.
460 2011-10-21 Jakub Jelinek <jakub@redhat.com>
463 * config/i386/i386.c (expand_vec_perm_even_odd_1): Handle
464 V4DImode and V8SImode for !TARGET_AVX2.
466 2011-10-21 Bernd Schmidt <bernds@codesourcery.com>
469 * sched-deps.c (add_dependence): If not doing predication, promote
470 REG_DEP_CONTROL to REG_DEP_ANTI.
472 2011-10-21 Georg-Johann Lay <avr@gjlay.de>
474 * config/avr/avr.h (LEGITIMIZE_RELOAD_ADDRESS): Pass address of X
475 instead of X to avr_legitimize_reload_address.
476 * config/avr/avr-protos.h (avr_legitimize_reload_address): Change
477 first argument's type from rtx to rtx*.
478 * config/avr/avr.c (avr_legitimize_reload_address): Ditto.
479 Pass PX to push_reload instead of &X. Change log messages for
482 2011-10-21 Roland Stigge <stigge@antcom.de>
485 * params.def: Fix typo "compilatoin" -> "compilation".
487 2011-10-21 H.J. Lu <hongjiu.lu@intel.com>
488 Kirill Yukhin <kirill.yukhin@intel.com>
491 * config/i386/driver-i386.c (host_detect_local_cpu): Do cpuid 7 only
492 if max_level allows that.
494 2011-10-21 Bernd Schmidt <bernds@codesourcery.com>
496 * reg-notes.def (DEP_CONTROL): New.
497 * sched-ebb.c (add_deps_for_risky_insns): Add a REG_DEP_CONTROL when
498 not doing speculation.
499 * rtlanal.c (record_hard_reg_sets, find_all_hard_reg_sets,
500 record_hard_reg_uses_1, record_hard_reg_uses): New functions.
501 * function.c (record_hard_reg_sets, record_hard_reg_uses,
502 record_hard_reg_uses_1): Remove; move to rtlanal.c.
503 * lists.c (copy_INSN_LIST, concat_INSN_LIST): New functions.
504 * haifa-sched.c: Swap includes of "rtl.h" and "hard-reg-set.h".
505 (MUST_RECOMPUTE_SPEC_P): New macro.
506 (real_insn_for_shadow): New function.
507 (cond_clobbered_p, recompute_todo_spec, check_clobbered_conditions,
508 toggle_cancelled_flags): New static functions.
509 (schedule_insn): Relax an assert to only check for empty hard back
510 dependencies. Skip cancelled dependencies. Call
511 check_clobbered_conditions.
512 (copy_insn_list): Remove function, renamed moved to lists.c.
513 (save_backtrack_point): Use new spelling copy_INSN_LIST.
514 (unschedule_insns_until): Ensure TODO_SPEC is reset properly.
515 (restore_last_backtrack_point): Likewise. Call toggle_cancelled_flags.
516 (estimate_insn_tick): Ignore cancelled dependencies.
517 (haifa_speculate_insn): Move declaration.
518 (try_ready): Move code into recompute_todo_spec and call it. Tweak
519 some asserts. Ensure predicated patterns are restored if necessary.
520 Dump DEP_CONTROL flag.
521 (haifa_change_pattern): Merge with sched_change_pattern.
522 (sched_change_pattern): Remove function.
523 * sched-deps.c (NON_FLUSH_JUMP_KIND, NON_FLUSH_JUMP): Remove. All
524 uses changed to simply not test NON_FLUSH_JUMP_P.
525 (ds_to_dk, dk_to_ds, dump_dep, ds_to_dt, dump_ds, check_dep): Handle
527 (dep_spec_p): If DO_PREDICATION, REG_DEP_CONTROL is speculative.
528 (reg_pending_control_uses, control_dependency_cache): New static
530 (sched_get_reverse_condition_uncached): New function.
531 (sd_find_dep_between): Remove pointless assert. Look in
532 control_dependency_cache.
533 (ask_dependency_caches, set_dependency_caches, sd_delete_dep,
534 extend_dependency_caches, sched_deps_finish): Handle REG_DEP_CONTROL
535 and control_dependency_cache.
536 (sd_unresolve_dep): Use dep_spec_p.
537 (add_dependence): Now a wrapper around add_dependence_1, handling
538 REG_DEP_CONTROL specially.
539 (flush_pending_lists): Clear pending_jump_insns.
540 (sched_analyze_1): Handle pending_jump_insns like a memory flush.
541 (sched_analyze_2): Unconditionally add to pending memory flushes,
542 keep previous behaviour but apply it to pending_jump_insns instead.
543 (sched_analyze_insn): Defer adding jump reg dependencies using
544 reg_pending_control_uses; add them to the control_uses list. Handle
545 pending_jump_insns and control_uses when adding dependence lists.
546 (deps_analyze_insn): Update INSN_COND_DEPS.
547 (deps_analyze_insn): Add jumps to pending_jump_insns rather than
548 last_pending_memory_flush.
549 (init_deps): Initialize pending_jump_insns.
550 (free_deps): Free control_uses.
551 (remove_from_deps): Remove from pending_jump_insns.
552 (init_deps_global): Allocate reg_pending_control_uses).
553 (finish_deps_global): Free it.
554 (add_dependence_1): Renamed from add_dependence. Handle
556 * rtl.h (record_hard_reg_uses, find_all_hard_reg_sets): Declare.
557 (copy_INSN_LIST, concat_INSN_LIST): Declare.
558 * sched-int.h (struct deps_reg): Add control_uses.
559 (struct deps_desc): Add pending_jump_insns.
560 (struct _haifa_deps_insn_data): Add cond_deps.
561 (struct _haifa_insn_data): Add must_recompute_spec and predicated_pat.
562 (INSN_COND_DEPS, PREDICATED_PAT): New macros.
563 (BITS_PER_DEP_WEAK): Adjust for two extra bits in the word.
564 (DEP_CONTROL): New macro.
565 (DEP_TYPES): Include it.
566 (HARD_DEP): Adjust definition.
567 (DEP_CANCELLED): New macro.
568 (enum SCHED_FLAGS): Add DO_PREDICATION.
569 (sched_get_reverse_condition_uncached, real_insn_for_shadow): Declare.
570 * sched-rgn.c (concat_INSN_LIST): Remove function.
571 (deps_join): Handle pending_jump_insns.
572 (free_pending_lists): Likewise.
573 * config/c6x/c6x.c (c6x_set_sched_flags): Set DO_PREDICATION for final
576 2011-10-21 Georg-Johann Lay <avr@gjlay.de>
579 * config/avr/libgcc.S (__EIND__): New define to 0x3C.
580 (__tablejump__): Consistently use EIND for indirect jump/call.
581 (__tablejump_elpm__): Ditto.
583 2011-10-21 Bernd Schmidt <bernds@codesourcery.com>
585 * config/c6x/c6x.md (attr "op_pattern"): New.
586 (load_sdata_pic, mov<mode>_insn for QIHIM and SISFVM): Set it.
587 * config/c6x/c6x-mult.md.in (mulhi3_VARIANT_, mulhisi3_insn_VARIANT_):
589 * config/c6x/c6x-mult.md: Regenerate.
590 * config/c6x/c6x.c: Include "regrename.h".
591 (unit_req_table): New typedef.
592 (unit_reqs): Use it for the declaration.
593 (unit_req_factor, get_unit_reqs, merge_unit_reqs, unit_req_imbalance,
594 get_unit_operand_masks, try_rename_operands, reshuffle_units): New
596 (count_unit_reqs): New arg reqs. All callers changed. Use
597 get_unit_reqs, and don't merge here.
598 (res_mii): New arg reqs. All callers changed. Rewrite to use a loop
599 using unit_req_factor.
600 (hwloop_optimize): Call reshuffle_units. Call merge_unit_reqs after
602 (c6x_reorg): Add reg notes problem, and call df_analyze.
603 * Makefile.in ($(out_object_file)): Depend on regrename.h.
605 2011-10-21 Kai Tietz <ktietz@redhat.com>
607 * fold-const.c (simple_operand_p_2): Handle integral
608 casts from boolean-operands.
610 2011-10-21 Jan Hubicka <jh@suse.cz>
612 * cgraph.c (dump_cgraph_node): Dump alias flag.
613 * cgraphunit.c (handle_alias_pairs): Handle weakrefs with
615 (get_alias_symbol): New function.
616 (output_weakrefs): Output also weakrefs with no destinatoin.
617 (lto_output_node): Output weakref alias flag when at function boundary.
619 2011-10-21 Andrew Stubbs <ams@codesourcery.com>
622 * config/arm/driver-arm.c (vendors): Make static.
624 2011-10-21 Uros Bizjak <ubizjak@gmail.com>
626 * config/i386/i386.c (ix86_emit_swdivsf): Force b into register.
627 (ix86_emit_swsqrtsf): Force a into register.
629 2011-10-20 Mike Stump <mikestump@comcast.net>
631 * regcprop.c (copyprop_hardreg_forward_1): Update recog_data
632 after validate_change wipes it out.
634 2011-10-20 Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
636 * config/spu/spu.md ("vec_permv16qi"): Reduce selector modulo 32
637 before using the shufb instruction.
639 2011-10-20 Kirill Yukhin <kirill.yukhin@intel.com>
642 * config/i386/i386.md (bmi_bextr_<mode>): Update register/
643 memory operand order.
644 (bmi2_bzhi_<mode>3): Ditto.
645 (bmi2_pdep_<mode>3): Ditto.
646 (bmi2_pext_<mode>3): Ditto.
648 2011-10-20 Richard Henderson <rth@redhat.com>
650 * target.def (vec_perm_const_ok): Rename from builtin_vec_perm_ok.
651 * optabs.c (can_vec_perm_expr_p): Update to match.
652 (expand_vec_perm_expr): Likewise.
653 * config/i386/i386.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Rename
654 from TARGET_VECTORIZE_BUILTIN_VEC_PERM_OK.
655 * doc/tm.texi.in: Likewise.
657 2011-10-20 Sergey Ostanevich <sergos.gnu@gmail.com>
660 * config/i386/i386.c (processor_target_table): Change Atom
661 align_loops_max_skip to 15.
663 2011-10-20 Richard Henderson <rth@redhat.com>
665 * target.def (builtin_vec_perm): Remove.
666 * doc/tm.texi.in (TARGET_VECTORIZE_BUILTIN_VEC_PERM): Remove.
668 * config/i386/i386.c (ix86_expand_vec_perm_builtin): Remove.
669 (IX86_BUILTIN_VEC_PERM_*): Remove.
670 (bdesc_args): Remove vec_perm builtins
671 (ix86_expand_builtin): Likewise.
672 (ix86_expand_vec_perm_const_1): Rename from
673 ix86_expand_vec_perm_builtin_1.
674 (extract_vec_perm_cst): Merge into...
675 (ix86_vectorize_vec_perm_const_ok): ... here. Rename from
676 ix86_vectorize_builtin_vec_perm_ok.
677 (TARGET_VECTORIZE_BUILTIN_VEC_PERM): Remove.
679 * config/rs6000/rs6000.c (rs6000_builtin_vec_perm): Remove.
680 (TARGET_VECTORIZE_BUILTIN_VEC_PERM): Remove.
682 * config/spu/spu.c (spu_builtin_vec_perm): Remove.
683 (TARGET_VECTORIZE_BUILTIN_VEC_PERM): Remove.
685 2011-10-20 Uros Bizjak <ubizjak@gmail.com>
688 * config/i386/i386.h (RECIP_MASK_DEFAULT): New define.
689 * config/i386/i386.op (recip_mask): Initialize with RECIP_MASK_DEFAULT.
690 * doc/invoke.texi (ix86 Options, -mrecip): Document that GCC
691 implements vectorized single float division and vectorized sqrtf(x)
692 with reciprocal sequence with additional Newton-Raphson step with
695 2011-10-20 Dodji Seketeli <dodji@redhat.com>
697 * ggc-zone.c (ggc_internal_alloc_zone_stat): Rename
698 ggc_alloced_size_order_for_request into ggc_round_alloc_size like
699 it was done in ggc-page.c.
702 * doc/cppopts.texi: Use @smallexample/@end smallexample in
703 documentation for -fdebug-cpp instead of @quotation/@end quotation
704 that is not supported by contrib/texi2pod.pl.
706 2011-10-19 Jan Hubicka <jh@suse.cz>
708 * ipa-inline.c (inline_small_functions): Always update all calles after
711 2011-10-19 Jan Hubicka <jh@suse.cz>
714 * ipa-inline.c (inline_small_functions): Fix checking code to not make
715 effect on fibheap stability.
717 2011-10-20 Maxim Kuvyrkov <maxim@codesourcery.com>
719 * config/m68k/t-linux (M68K_MLIB_CPU): Add ColdFire CPUs.
721 2011-10-20 Ramana Radhakrishnan <ramana.radhakrishnan@linaro.org>
724 * config/arm/arm.c (thumb_unexpanded_epilogue): Handle return
727 2011-10-20 Richard Guenther <rguenther@suse.de>
729 * tree-ssa-loop-im.c (stmt_cost): Add WIDEN_*, FMA_EXPR
730 and rotates to the set of expensive operations.
732 2011-10-19 David S. Miller <davem@davemloft.net>
734 * config/sparc/sparc.c (sparc_expand_move): Use can_create_pseudo_p.
735 (sparc_emit_set_const32): Likewise.
736 (sparc_emit_set_const64_longway): Likewise.
737 (sparc_emit_set_const64): Likewise.
738 (sparc_legitimize_pic_address): Likewise.
739 (memory_ok_for_ldd): Likewise.
741 2011-10-20 Dehao Chen <dehao@google.com>
743 * profile.c (compute_branch_probabilities): Compute and dump the
744 overlap between the static estimation and the instrumentation profile.
745 (OVERLAP_BASE): New macro.
746 (compute_frequency_overlap): New function
748 2011-10-19 Jakub Jelinek <jakub@redhat.com>
750 * config/i386/i386.c (expand_vec_perm_vpshufb2_vpermq_even_odd): Use
751 d->op1 instead of d->op0 for the second vpshufb.
752 (expand_vec_perm_even_odd_1): For V8SImode fix vpshufd immediates.
753 (ix86_expand_vec_perm_const): If mask indicates two operands are
754 needed, but both are the same and expanding them as d.op0 == d.op1
755 failed, retry with d.op0 != d.op1.
756 (ix86_expand_vec_perm_builtin): Likewise. Handle sorry printing
757 also for d.nelt == 32.
760 * cfgexpand.c (expand_debug_expr): Handle WIDEN_LSHIFT_EXPR, ignore
763 2011-10-19 Bernd Schmidt <bernds@codesourcery.com>
765 * regrename.h: New file.
766 * regrename.c: Include it. Also include "emit-rtl.h".
767 (struct du_head, struct du_chain, du_head_p DEF_VEC and
768 DEF_VEC_ALLOC_P): Move to regrename.h.
769 (do_replace): Remove declaration.
770 (insn_rr): New variable.
771 (cur_operand): New static variable.
772 (regrename_chain_from_id): Renamed from chain_from_id and no longer
773 static. All callers changed.
774 (record_operand_use): New static function.
775 (scan_rtx_reg): Use it.
776 (find_best_rename_reg): New function, broken out of rename_chains.
777 (rename_chains): Use it. Don't update chain regno and nregs here, ...
778 (regrename_do_replace): ... do it here instead. Renamed from
779 do_replace, and no longer static. All callers changed.
780 (regrename_analyze): No longer static. New arg bb_mask.
781 All callers changed. If bb_mask is nonzero, use it to limit the
782 number of basic blocks we analyze. If we failed to analyze a block,
783 clear insn operand data.
784 (record_out_operands): New arg insn_info. Update cur_operand if it is
786 (build_def_use): If insn_rr is nonnull, pass an insn_info to
787 record_out_operands, and update cur_operand here as well.
788 (regrename_init, regrename_finish): New functions.
789 (regrename_optimize): Use them.
790 * Makefile.in (regrename.o): Adjust dependencies.
792 2011-10-19 Tom de Vries <tom@codesourcery.com>
794 PR tree-optimization/50769
795 * tree-ssa-tail-merge.c (replace_block_by): Calculate phi_vuse2
796 unconditionally. Handle case that phi_vuse2 is not an SSA_NAME. Add
797 dummy argument .MEM to phi when increasing number of arguments of phi by
798 redirecting edges to the block with phi.
800 2011-10-19 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
803 * configure.ac: Save and restore CXXFLAGS around
804 gcc_AC_CHECK_DECLS uses.
805 Check for madvise() declaration with g++ if --enable-build-with-cxx.
806 * configure: Regenerate.
807 * config.in: Regenerate.
808 * ggc-page.c (USING_MADVISE): Also check HAVE_DECL_MADVISE.
810 2011-10-19 Alexandre Oliva <aoliva@redhat.com>
813 * var-tracking.c (loc_exp_dep, onepart_aux): New structs.
814 (variable_part): Replace offset with union.
815 (enum onepart_enum, onepart_enum_t): New.
816 (variable_def): Drop cur_loc_changed, add onepart.
817 (value_chain_def, const_value_chain): Remove.
818 (VAR_PART_OFFSET, VAR_LOC_1PAUX): New macros, with checking.
819 (VAR_LOC_DEP_LST, VAR_LOC_DEP_LSTP): New macros.
820 (VAR_LOC_FROM, VAR_LOC_DEPTH, VAR_LOC_DEP_VEC): Likewise.
821 (value_chain_pool, value_chains): Remove.
822 (dropped_values): New.
823 (struct parm_reg): Only if HAVE_window_save.
824 (vt_stack_adjustments): Don't record register arguments.
826 (dv_onepart_p): Return a onepart_enum_t.
830 (variable_htab_free): Release onepart aux data. Reset flags.
831 (value_chain_htab_hash, value_chain_htab_eq): Remove.
832 (unshare_variable): Use onepart field. Propagate onepart aux
833 data or offset. Drop cur_loc_changed.
834 (val_store): Cope with NULL insn. Rephrase dump output. Check
835 for unsuitable locs. Add FIXME on using cselib locs.
836 (val_reset): Remove FIXME of unfounded concerns.
837 (val_resolve): Check for unsuitable locs. Add FIXME on using
839 (variable_union): Use onepart field, adjust access to offset.
841 (VALUE_CHANGED, DECL_CHANGED): Update doc.
842 (set_dv_changed): Clear NO_LOC_P when changed.
843 (find_loc_in_1pdv): Use onepart field.
844 (intersect_loc_chains): Likewise.
845 (unsuitable_loc): New.
846 (loc_cmp): Keep ENTRY_VALUEs at the end of the loc list.
847 (add_value_chain, add_value_chains): Remove.
848 (add_cselib_value_chains, remove_value_chain): Likewise.
849 (remove_value_chains, remove_cselib_value_chains): Likewise.
850 (canonicalize_loc_order_check): Use onepart. Drop cur_loc_changed.
851 (canonicalize_values_star, canonicalize_vars_star): Use onepart.
852 (variable_merge_over_cur): Likewise. Adjust access to offset.
853 Drop cur_loc_changed.
854 (variable_merge_over_src): Use onepart field.
855 (remove_duplicate_values): Likewise.
856 (variable_post_merge_new_vals): Likewise.
857 (find_mem_expr_in_1pdv): Likewise.
858 (dataflow_set_preserve_mem_locs): Likewise. Drop cur_loc_changed
860 (dataflow_set_remove_mem_locs): Likewise. Use VAR_LOC_FROM.
861 (variable_different_p): Use onepart field. Move onepart test out
863 (argument_reg_set): Drop.
864 (add_uses, add_stores): Preserve but do not record in dynamic
865 tables equivalences for ENTRY_VALUEs and CFA_based addresses.
866 Avoid unsuitable address expressions.
867 (EXPR_DEPTH): Unlimit.
868 (EXPR_USE_DEPTH): Repurpose PARAM_MAX_VARTRACK_EXPR_DEPTH.
869 (prepare_call_arguments): Use DECL_RTL_IF_SET.
870 (dump_var): Adjust access to offset.
871 (variable_from_dropped, recover_dropped_1paux): New.
872 (variable_was_changed): Drop cur_loc_changed. Use onepart.
873 Preserve onepart aux in empty_var. Recover empty_var and onepart
874 aux from dropped_values.
875 (find_variable_location_part): Special-case onepart. Adjust
877 (set_slot_part): Use onepart. Drop cur_loc_changed. Adjust
878 access to offset. Initialize onepaux. Drop value chains.
879 (delete_slot_part): Drop value chains. Use VAR_LOC_FROM.
880 (VEC (variable, heap), VEC (rtx, stack)): Define.
881 (expand_loc_callback_data): Drop dummy, cur_loc_changed,
882 ignore_cur_loc. Add expanding, pending, depth.
883 (loc_exp_dep_alloc, loc_exp_dep_clear): New.
884 (loc_exp_dep_insert, loc_exp_dep_set): New.
885 (notify_dependents_of_resolved_value): New.
886 (update_depth, vt_expand_var_loc_chain): New.
887 (vt_expand_loc_callback): Revamped.
888 (resolve_expansions_pending_recursion): New.
889 (INIT_ELCD, FINI_ELCD): New.
890 (vt_expand_loc): Use the new macros above. Drop ignore_cur_loc
891 parameter, adjust all callers.
892 (vt_expand_loc_dummy): Drop.
893 (vt_expand_1pvar): New.
894 (emit_note_insn_var_location): Operate on non-debug decls only.
895 Revamp multi-part cur_loc recomputation and one-part expansion.
896 Drop cur_loc_changed. Adjust access to offset.
897 (VEC (variable, heap)): Drop.
898 (changed_variables_stack, changed_values_stack): Drop.
899 (check_changed_vars_0, check_changed_vars_1): Remove.
900 (check_changed_vars_2, check_changed_vars_3): Remove.
901 (values_to_stack, remove_value_from_changed_variables): New.
902 (notify_dependents_of_changed_value, process_changed_values): New.
903 (emit_notes_for_changes): Revamp onepart updates.
904 (emit_notes_for_differences_1): Use onepart. Drop cur_loc_changed
905 and value chains. Propagate onepaux. Recover empty_var and onepaux
907 (emit_notes_for_differences_2): Drop value chains.
908 (emit_notes_in_bb): Adjust.
909 (vt_emit_notes): Drop value chains, changed_variables_stack.
910 Initialize and release dropped_values.
911 (create_entry_value): Revamp.
912 (vt_add_function_parameter): Use new interface.
913 (note_register_arguments): Remove.
914 (vt_initialize): Drop value chains and register arguments.
915 (vt_finalize): Drop value chains. Release windowed_parm_regs only
917 * rtl.h: Document various pass-local uses of RTL flags.
918 * tree.h (DECL_RTL_KNOWN_SET): New.
919 * doc/invoke.texi (param max-vartrack-expr-depth): Update
920 description and default.
922 2011-10-19 Georg-Johann Lay <avr@gjlay.de>
925 * config/avr/avr.md (cc): New alternative out_plus_noclobber.
927 (addhi3): Don't pipe through short; use gen_int_mode instead.
928 Prior to reload, expand to gen_addhi3_clobber.
929 (*addhi3): Use avr_out_plus_noclobber if applicable, use
930 out_plus_noclobber in cc and adjust_len attribute.
931 (addhi3_clobber): 2 new RTL peepholes.
932 (addhi3_clobber): New insn.
933 * config/avr/avr-protos.h: (avr_out_plus_noclobber): New prototype.
934 * config/avr/avr.c (avr_out_plus_noclobber): New function.
935 (notice_update_cc): Handle CC_OUT_PLUS_NOCLOBBER.
936 (avr_out_plus_1): Tweak if only MSB is +/-1 and other bytes are 0.
937 Set cc0 to set_zn for adiw on 16-bit values.
938 (adjust_insn_length): Handle ADJUST_LEN_OUT_PLUS_NOCLOBBER.
939 (expand_epilogue): No need to add 0 to frame_pointer_rtx.
941 2011-10-19 Richard Guenther <rguenther@suse.de>
944 * tree-ssa-forwprop.c (forward_propagate_into_cond): Verify
945 the condition is properly gimple before using it.
946 * tree-eh (stmt_could_throw_1_p): Properly extract the
947 operation type from comparisons.
949 2011-10-19 Roland Stigge <stigge@antcom.de>
952 * plugin.c (add_new_plugin): Fix typo in fatal_error message.
954 2011-10-19 Roland Stigge <stigge@antcom.de>
957 * config/rx/rx.c (rx_print_operand): Fix typo in warning message.
959 2011-10-19 Richard Guenther <rguenther@suse.de>
962 * gimple-fold.c (gimplify_and_update_call_from_tree): Rewrite.
964 2011-10-19 Andrey Belevantsev <abel@ispras.ru>
966 PR rtl-optimization/50340
967 * sel-sched-ir.c (update_target_availability): LHS register
968 availability is not known if the unavailable LHS of the other
969 expression is a different register.
971 2011-10-19 Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
974 * config/spu/spu.c (spu_emit_vector_compare): Support unordered
975 floating-point comparisons.
977 2011-10-19 Jan Hubicka <jh@suse.cz>
979 * cgraphunit.c (handle_alias_pairs): Also handle wekref with
980 destination declared.
981 (output_weakrefs): New function.
982 * varpool.c (varpool_create_variable_alias): Handle external aliases.
984 2011-10-19 Jakub Jelinek <jakub@redhat.com>
986 * dwarf2out.c (loc_descriptor): For SUBREG pass SUBREG_REG's mode as
987 second argument instead of mode.
989 2011-10-18 Jakub Jelinek <jakub@redhat.com>
991 * config/i386/i386.c (ix86_expand_vec_perm): In merge_two use
992 mode SUBREG of operands[0] as target.
993 (valid_perm_using_mode_p): Don't ignore higher bits of d->perm.
994 (expand_vec_pshufb): For V8SImode vmode emit avx2_permvarv8si.
995 (expand_vec_perm_1): Handle identity and some broadcast
997 (expand_vec_perm_interleave2): Handle also 32-byte modes, using
998 vperm2[fi]128 or vpunpck[lh]* followed by single insn permutation.
999 For d->testing_p return true earlier to avoid creating more GC
1001 (expand_vec_perm_vpermq_perm_1): New function.
1002 (expand_vec_perm_vpshufb2_vpermq): For d->testing_p return true
1003 earlier to avoid creating more GC garbage. Fix handling of
1004 V16HImode. Avoid some SUBREGs in SET_DEST.
1005 (expand_vec_perm_broadcast_1): Return false for 32-byte integer
1007 (expand_vec_perm_vpshufb4_vpermq2): New function.
1008 (ix86_expand_vec_perm_builtin_1): Call expand_vec_perm_vpermq_perm_1
1009 and expand_vec_perm_vpshufb4_vpermq2.
1011 2011-10-18 Andrew Stubbs <ams@codesourcery.com>
1013 * config/arm/driver-arm.c (host_detect_local_cpu): Close the file
1016 2011-10-18 Andrew Stubbs <ams@codesourcery.com>
1018 PR tree-optimization/50717
1019 * tree-ssa-math-opts.c (is_widening_mult_p): Remove the 'type'
1020 parameter. Calculate 'type' from stmt.
1021 (convert_mult_to_widen): Update call the is_widening_mult_p.
1022 (convert_plusminus_to_widen): Likewise.
1024 2011-10-18 Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
1026 * config/spu/spu.c (struct machine_function): New data structure.
1027 (spu_init_machine_status): New function.
1028 (spu_option_override): Install it.
1029 (get_pic_reg): Set and use cfun->machine->pic_reg.
1030 (spu_split_immediate): Do not set crtl->uses_pic_offset_table.
1031 (need_to_save_reg): Use cfun->machine->pic_reg instead of
1032 checking crtl->uses_pic_offset_table.
1033 (spu_expand_prologue): Likewise.
1035 2011-10-18 Jakub Jelinek <jakub@redhat.com>
1037 PR tree-optimization/50735
1038 * function.c (gimplify_parameters): Use create_tmp_var instead of
1039 create_tmp_reg. If parm is not TREE_ADDRESSABLE and type is complex
1040 or vector type, set DECL_GIMPLE_REG_P.
1042 2011-10-18 Andrew Stubbs <ams@codesourcery.com>
1044 * config.host (arm*-*-linux*): Add driver-arm.o and x-arm.
1045 * config/arm/arm.opt: Add 'native' processor_type and
1046 arm_arch enum values.
1047 * config/arm/arm.h (host_detect_local_cpu): New prototype.
1048 (EXTRA_SPEC_FUNCTIONS): New define.
1049 (MCPU_MTUNE_NATIVE_SPECS): New define.
1050 (DRIVER_SELF_SPECS): New define.
1051 * config/arm/driver-arm.c: New file.
1052 * config/arm/x-arm: New file.
1053 * doc/invoke.texi (ARM Options): Document -mcpu=native,
1054 -mtune=native and -march=native.
1056 2011-10-18 Alexander Monakov <amonakov@ispras.ru>
1058 PR rtl-optimization/50205
1059 * sel-sched.c (count_occurrences_1): Simplify on the assumption that
1060 p->x is a register. Forbid substitution when the same register is
1061 found in a different mode.
1062 (count_occurrences_equiv): Assert that 'what' is a register.
1064 2011-10-18 Richard Guenther <rguenther@suse.de>
1066 PR tree-optimization/50767
1067 * tree-ssa-pre.c (create_expression_by_pieces): Update the
1070 2011-10-18 Julian Brown <julian@codesourcery.com>
1072 * config/arm/arm.c (arm_block_move_unaligned_straight)
1073 (arm_adjust_block_mem, arm_block_move_unaligned_loop)
1074 (arm_movmemqi_unaligned): New.
1075 (arm_gen_movmemqi): Support unaligned block copies.
1077 2011-10-18 Ira Rosen <ira.rosen@linaro.org>
1079 * doc/md.texi (vec_widen_ushiftl_hi, vec_widen_ushiftl_lo,
1080 vec_widen_sshiftl_hi, vec_widen_sshiftl_lo): Document.
1081 * tree-pretty-print.c (dump_generic_node): Handle WIDEN_LSHIFT_EXPR,
1082 VEC_WIDEN_LSHIFT_HI_EXPR and VEC_WIDEN_LSHIFT_LO_EXPR.
1083 (op_code_prio): Likewise.
1084 (op_symbol_code): Handle WIDEN_LSHIFT_EXPR.
1085 * optabs.c (optab_for_tree_code): Handle
1086 VEC_WIDEN_LSHIFT_HI_EXPR and VEC_WIDEN_LSHIFT_LO_EXPR.
1087 (init-optabs): Initialize optab codes for vec_widen_u/sshiftl_hi/lo.
1088 * optabs.h (enum optab_index): Add OTI_vec_widen_u/sshiftl_hi/lo.
1089 * genopinit.c (optabs): Initialize the new optabs.
1090 * expr.c (expand_expr_real_2): Handle
1091 VEC_WIDEN_LSHIFT_HI_EXPR and VEC_WIDEN_LSHIFT_LO_EXPR.
1092 * gimple-pretty-print.c (dump_binary_rhs): Likewise.
1093 * tree-vectorizer.h (NUM_PATTERNS): Increase to 8.
1094 * tree.def (WIDEN_LSHIFT_EXPR, VEC_WIDEN_LSHIFT_HI_EXPR,
1095 VEC_WIDEN_LSHIFT_LO_EXPR): New.
1096 * cfgexpand.c (expand_debug_expr): Handle new tree codes.
1097 * tree-vect-patterns.c (vect_vect_recog_func_ptrs): Add
1098 vect_recog_widen_shift_pattern.
1099 (vect_handle_widen_mult_by_const): Rename...
1100 (vect_handle_widen_op_by_const): ...to this. Handle shifts.
1101 Add a new argument, update documentation.
1102 (vect_recog_widen_mult_pattern): Assume that only second
1103 operand can be constant. Update call to
1104 vect_handle_widen_op_by_const.
1105 (vect_recog_over_widening_pattern): Fix typo.
1106 (vect_recog_widen_shift_pattern): New.
1107 * tree-vect-stmts.c (vectorizable_type_promotion): Handle
1109 (supportable_widening_operation): Likewise.
1110 * tree-inline.c (estimate_operator_cost): Handle new tree codes.
1111 * tree-vect-generic.c (expand_vector_operations_1): Likewise.
1112 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
1113 * config/arm/neon.md (neon_vec_<US>shiftl_<mode>): New.
1114 (vec_widen_<US>shiftl_lo_<mode>, neon_vec_<US>shiftl_hi_<mode>,
1115 vec_widen_<US>shiftl_hi_<mode>, neon_vec_<US>shift_left_<mode>):
1117 * config/arm/predicates.md (const_neon_scalar_shift_amount_operand):
1119 * config/arm/iterators.md (V_innermode): New.
1120 * tree-vect-slp.c (vect_build_slp_tree): Require same shift operand
1123 2011-10-18 Richard Guenther <rguenther@suse.de>
1125 * tree-ssa-alias.h (struct pt_solution): Remove
1126 vars_contains_restrict member.
1127 (pt_solutions_same_restrict_base): Remove.
1128 (pt_solution_set): Adjust.
1129 * tree-ssa-alias.c (ptr_deref_may_alias_decl_p): Remove
1130 vars_contains_restrict handling.
1131 (dump_points_to_solution): Likewise.
1132 (ptr_derefs_may_alias_p): Do not call pt_solutions_same_restrict_base.
1133 * tree-ssa-structalias.c (struct variable_info): Remove is_restrict_var
1135 (new_var_info): Do not initialize it.
1136 (ipa_escaped_pt): Adjust.
1137 (make_constraint_from_restrict): Make the tag global.
1138 (make_constraint_from_global_restrict): New function.
1139 (make_constraint_from_heapvar): Remove.
1140 (create_variable_info_for): Do not make restrict vars point
1142 (intra_create_variable_infos): Likewise.
1143 (find_what_var_points_to): Remove vars_contains_restrict handling.
1144 (pt_solution_set): Adjust.
1145 (pt_solution_ior_into): Likewise.
1146 (pt_solutions_same_restrict_base): Remove.
1147 (compute_points_to_sets): Do not test is_restrict_var.
1148 * cfgexpand.c (update_alias_info_with_stack_vars): Adjust.
1149 * gimple-pretty-print.c (pp_points_to_solution): Likewise.
1151 2011-10-18 Tom de Vries <tom@codesourcery.com>
1153 PR tree-optimization/50672
1154 * tree-ssa-dce.c (mark_virtual_operand_for_renaming): New function,
1156 (mark_virtual_phi_result_for_renaming): Use
1157 mark_virtual_operand_for_renaming.
1158 * tree-flow.h (mark_virtual_operand_for_renaming): Declare.
1159 * tree-ssa-tail-merge.c (release_last_vdef): New function.
1160 (purge_bbs): Add update_vops parameter. Call release_last_vdef for each
1161 deleted basic block.
1162 (tail_merge_optimize): Add argument to call to purge_bbs.
1164 2011-10-18 Richard Guenther <rguenther@suse.de>
1167 * expr.c (get_object_or_type_alignment): New function.
1168 (expand_assignment): Use it.
1169 (expand_expr_real_1): Likewise.
1171 2011-10-18 Dodji Seketeli <dodji@redhat.com>
1174 * input.c (dump_line_table_statistics): Use long, not size_t.
1176 2011-10-17 Eric Botcazou <ebotcazou@adacore.com>
1178 * config/sparc/sparc.md (in_call_delay): Fix formatting issues.
1180 2011-10-17 Simon Baldwin <simonb@google.com>
1181 Ian Lance Taylor <iant@google.com>
1183 * configure.ac: Add --with-native-system-header-dir. Set and
1184 substitute NATIVE_SYSTEM_HEADER_DIR. Use native_system_header
1185 when setting target_header_dir.
1186 * config.gcc: Always set native_system_header_dir.
1187 (*-*-gnu*): Set native_system_header_dir. Don't use t-gnu.
1188 (i[34567]86-pc-msdosdjgpp*): Set native_system_header_dir. Don't
1190 (i[34567]86-*-mingw* | x86_64-*-mingw*): Set
1191 native_system_header_dir.
1192 (spu-*-elf*): Set native_system_header_dir.
1193 * Makefile.in (NATIVE_SYSTEM_HEADER_DIR): Set to
1194 @NATIVE_SYSTEM_HEADER_DIR@.
1195 (PREPROCESSOR_DEFINES): Define NATIVE_SYSTEM_HEADER_DIR.
1196 * cppdefault.c (STANDARD_INCLUDE_DIR): Don't define.
1197 (NATIVE_SYSTEM_HEADER_COMPONENT): Rename from
1198 STANDARD_INCLUDE_COMPONENT.
1199 (cpp_include_defaults): Don't use SYSTEM_INCLUDE_DIR. Rename
1200 STANDARD_INCLUDE_DIR to NATIVE_SYSTEM_HEADER_DIR.
1201 * system.h: Poison SYSTEM_INCLUDE_DIR, STANDARD_INCLUDE_DIR, and
1202 STANDARD_INCLUDE_COMPONENT.
1203 * config/i386/t-mingw32 (NATIVE_SYSTEM_HEADER_DIR): Remove.
1204 * config/i386/t-mingw-w32: Likewise.
1205 * config/i386/t-mingw-w64: Likewise.
1206 * config/spu/t-spu-elf: Likewise.
1207 * config/i386/t-djgpp: Remove.
1208 * config/t-gnu: Remove.
1209 * config/i386/mingw32.h (STANDARD_INCLUDE_DIR): Don't define.
1210 (NATIVE_SYSTEM_HEADER_COMPONENT): Rename from
1211 STANDARD_INCLUDE_COMPONENT.
1212 * config/i386/djgpp.h (STANDARD_INCLUDE_DIR): Don't define.
1213 * config/spu/spu-elf.h: Likewise.
1214 * config/vms/xm-vms.h: Likewise.
1215 * config/gnu.h: Likewise.
1216 * config/openbsd.h (INCLUDE_DEFAULTS): Change STANDARD_INCLUDE_DIR
1217 and STANDARD_INCLUDE_COMPONENT to NATIVE_SYSTEM_HEADER_DIR and
1218 NATIVE_SYSTME_HEADER_COMPONENT.
1219 * doc/install.texi (Configuration): Document
1220 --with-native-system-header-dir. Mention it in the documentation
1221 for --with-sysroot and --with-build-sysroot.
1222 * doc/tm.texi.in (Driver): Don't document SYSTEM_INCLUDE_DIR or
1223 STANDARD_INCLUDE_DIR. Rename STANDARD_INCLUDE_COMPONENT to
1224 NATIVE_SYSTEM_HEADER_COMPONENT. Rename uses of
1225 STANDARD_INCLUDE_DIR to NATIVE_SYSTEM_HEADER_DIR.
1226 * doc/fragments.texi (Target Fragment): Don't document
1227 NATIVE_SYSTEM_HEADER_DIR.
1228 * configure, doc/tm.texi: Rebuild.
1230 2011-10-17 Richard Henderson <rth@redhat.com>
1232 * config/sparc/sparc.md: Use register_or_zero_operand where rJ
1235 * config/sparc/sparc.md (vec_perm_constv8qi, vec_perm<mode>): New
1237 * config/sparc/sparc.c (sparc_expand_vec_perm_bmask): New function.
1238 * config/sparc/sparc-protos.h (sparc_expand_vec_perm_bmask): Declare.
1240 2011-10-17 David S. Miller <davem@davemloft.net>
1242 * config/sparc/sparc-modes.def: Add single entry vector modes for
1244 * config/sparc/sparc.md (V32, V32I, V64, V64I, V64N8): Delete
1246 (mov<V32:mode>): Revert back to plain SFmode pattern.
1247 (*movsf_insn): Likewise.
1248 (mov<V64:mode>): Revert back to plain DFmode pattern.
1249 (*movdf_insn_sp32): Likewise.
1250 (*movdf_insn_sp32_v9): Likewise.
1251 (*movdf_insn_sp64): Likewise.
1252 (V64 mode splitters) Likewise.
1253 (addsi3): Remove VIS alternatives.
1255 (and<V64I:mode>3): Revert to DImode only pattern.
1256 (and<V64I:mode>3_sp32): Likewise.
1257 (*and<V64I:mode>3_sp64): Likewise.
1258 (and<V32I:mode>3): Likewise.
1259 (*and_not_<V64I:mode>_sp32): Likewise.
1260 (*and_not_<V64I:mode>_sp64): Likewise.
1261 (*and_not_<V32I:mode>): Likewise.
1262 (ior<V64I:mode>3): Likewise.
1263 (*ior<V64I:mode>3_sp32): Likewise.
1264 (*ior<V64I:mode>3_sp64): Likewise.
1265 (ior<V32I:mode>3): Likewise.
1266 (*or_not_<V64I:mode>_sp32): Likewise.
1267 (*or_not_<V64I:mode>_sp64): Likewise.
1268 (*or_not_<V32I:mode>): Likewise.
1269 (xor<V64I:mode>3): Likewise.
1270 (*xor<V64I:mode>3_sp32): Likewise.
1271 (*xor<V64I:mode>3_sp64): Likewise.
1272 (xor<V32I:mode>3): Likewise.
1273 (V64I mode splitters): Likewise.
1274 (*xor_not_<V64I:mode>_sp32): Likewise.
1275 (*xor_not_<V64I:mode>_sp64): Likewise.
1276 (*xor_not_<V32I:mode>): Likewise.
1277 (one_cmpl<V64I:mode>2): Likewise.
1278 (*one_cmpl<V64I:mode>2_sp32): Likewise.
1279 (*one_cmpl<V64I:mode>2_sp64): Likewise.
1280 (one_cmpl<V32I:mode>2): Likewise.
1281 (VM32, VM64, VMALL): New mode iterators.
1282 (vbits, vconstr, vfptype): New mode attributes.
1283 (mov<VMALL:mode>): New expander.
1284 (*mov<VM32:mode>_insn): New insn.
1285 (*mov<VM64:mode>_insn_sp64): New insn.
1286 (*mov<VM64:mode>_insn_sp32): New insn, and associated splitter
1287 specifically for the register to memory case.
1288 (vec_init<mode>): New expander.
1289 (VADDSUB): New mode iterator.
1290 (<plusminus_insn>v2si3, <plusminus_insn>v2hi3): Remove and replace
1292 (<plusminus_insn><mode>3): New consolidated pattern.
1293 (VL): New mode iterator for logical operations.
1294 (vlsuf): New more attribute.
1295 (vlop): New code iterator.
1296 (vlinsn, vlninsn): New code attributes.
1297 (<code><mode>3): New insn to non-negated vector logical ops.
1298 (*not_<code><mode>3): Likewise for negated variants.
1299 (*nand<mode>_vis): New insn.
1300 (vlnotop): New code iterator.
1301 (*<code>_not1<mode>_vis, *<code>_not2<mode>_vis): New insns.
1302 (one_cmpl<mode>2): New insn.
1303 (faligndata<V64I:mode>_vis): Rewrite to use VM64 iterator.
1304 (bshuffle<VM64:mode>_vis): Likewise.
1305 (v<vis3_shift_patname><mode>3): Use GCM mode iterator.
1306 (fp<plusminus_insn>64_vis): Use V1DI mode.
1307 (VASS mode iterator): Use V1SI not SI mode.
1308 * config/sparc/sparc.c (sparc_vis_init_builtins): Account for
1309 single-entry vector mode changes.
1310 (sparc_expand_builtin): Likewise.
1311 (sparc_expand_vector_init): New function.
1312 * config/sparc/sparc-protos.h (sparc_expand_vector_init): Declare.
1314 2011-10-17 Kai Tietz <ktietz@redhat.com>
1316 * fold-const.c (simple_operand_p_2): New function.
1317 (fold_truthop): Rename to
1318 (fold_truth_andor_1): function name.
1319 Additionally remove branching creation for logical and/or.
1320 (fold_truth_andor): Handle branching creation for logical and/or here.
1322 2011-10-17 Andi Kleen <ak@linux.intel.com>
1324 * ggc-page.c (USING_MADVISE): Adjust ifdef to check for USING_MMAP.
1326 2011-10-17 Georg-Johann Lay <avr@gjlay.de>
1328 * config/avr/avr.h (ASSEMBLER_DIALECT): Remove.
1329 * config/avr/avr.md (mcu_have_movw, mcu_mega): Remove attributes.
1330 (adjust_len): Add alternative "call".
1331 (isa, enabled): New insn attributes.
1332 (length): Use match_test with AVR_HAVE_JMP_CALL instead of
1334 (*sbrx_branch<mode>): Ditto.
1335 (*sbrx_and_branch<mode>): Ditto.
1336 (*sbix_branch): Ditto.
1337 (*sbix_branch_bit7): Ditto.
1338 (*sbix_branch_tmp): Ditto.
1339 (*sbix_branch_tmp_bit7): Ditto.
1341 (negsi2): Use attribute "isa" instead of assembler dialect.
1342 (extendhisi2): Ditto.
1343 (call_insn, call_value_insn): Set adjust_len attribute.
1344 (indirect_jump): Indent to coding rules.
1345 (call_prologue_saves): Use isa attribute instead of mcu_mega.
1346 (epilogue_restores): Ditto. Fix setting of SP as described in the
1348 (*indirect_jump): Fusion of *jcindirect_jump, *njcindirect_jump
1349 and *indirect_jump_avr6.
1350 (*tablejump): Fusion of *tablejump_rjmp and *tablejump_lib.
1351 (*jcindirect_jump, *njcindirect_jump, *indirect_jump_avr6): Remove.
1352 (*tablejump_rjmp, *tablejump_lib): Remove.
1353 * config/avr/avr.c (adjust_insn_length): Handle ADJUST_LEN_CALL.
1355 2011-10-17 Paolo Carlini <paolo.carlini@oracle.com>
1358 * doc/invoke.texi ([Wnonnull]): Update.
1360 2011-10-17 Richard Henderson <rth@redhat.com>
1363 * optabs.c (expand_vec_perm_expr): Fix indexing error.
1365 2011-10-17 Sergio Durigan Junior <sergiodj@redhat.com>
1367 * configure.ac: Display `yes' if the SystemTap header has been found.
1368 * configure: Regenerate.
1370 2011-10-08 Andi Kleen <ak@linux.intel.com>
1373 * config.in, configure: Regenerate.
1374 * configure.ac (madvise): Add to AC_CHECK_FUNCS.
1375 * ggc-page.c (USING_MADVISE): Add.
1376 (page_entry): Add discarded field.
1377 (alloc_page): Check for discarded pages.
1378 (release_pages): Add USING_MADVISE branch.
1380 2011-10-17 Richard Guenther <rguenther@suse.de>
1382 PR tree-optimization/50729
1383 * tree-vrp.c (extract_range_from_unary_expr_1): Remove redundant test.
1384 (simplify_conversion_using_ranges): Properly test the
1385 intermediate result.
1387 2011-10-15 Tom Tromey <tromey@redhat.com>
1388 Dodji Seketeli <dodji@redhat.com>
1390 * ggc.h (ggc_round_alloc_size): Declare new public entry point.
1391 * ggc-none.c (ggc_round_alloc_size): New public stub function.
1392 * ggc-page.c (ggc_alloced_size_order_for_request): New static
1393 function. Factorized from ggc_internal_alloc_stat.
1394 (ggc_round_alloc_size): New public function. Uses
1395 ggc_alloced_size_order_for_request.
1396 (ggc_internal_alloc_stat): Use ggc_alloced_size_order_for_request.
1397 * ggc-zone.c (ggc_round_alloc_size): New public function extracted
1398 from ggc_internal_alloc_zone_stat.
1399 (ggc_internal_alloc_zone_stat): Use ggc_round_alloc_size.
1400 * toplev.c (general_init): Initialize
1401 line_table->alloced_size_for_request.
1403 2011-10-15 Tom Tromey <tromey@redhat.com>
1404 Dodji Seketeli <dodji@redhat.com>
1406 * input.c (ONE_K, ONE_M, SCALE, STAT_LABEL, FORMAT_AMOUNT): New macros.
1407 (num_expanded_macros_counter, num_macro_tokens_counter): Declare
1409 (dump_line_table_statistics): Define new function.
1410 * input.h (dump_line_table_statistics): Declare new function.
1411 * toplev.c (dump_memory_report): Call dump_line_table_statistics.
1413 2011-10-15 Tom Tromey <tromey@redhat.com>
1414 Dodji Seketeli <dodji@redhat.com>
1416 * doc/cppopts.texi: Document -fdebug-cpp.
1417 * doc/invoke.texi: Add -fdebug-cpp to the list of preprocessor options.
1419 2011-10-15 Tom Tromey <tromey@redhat.com>
1420 Dodji Seketeli <dodji@redhat.com>
1422 * gcc/diagnostic.h (diagnostic_report_current_module): Add a
1424 * diagnostic.c (diagnostic_report_current_module): Add a location
1425 parameter to the function definition. Use it instead of
1426 input_location. Resolve the virtual location rather than just
1427 looking up its map and risking to touch a resulting macro map.
1428 (default_diagnostic_starter): Pass the relevant diagnostic
1429 location to diagnostic_report_current_module.
1430 * tree-diagnostic.c (maybe_unwind_expanded_macro_loc): New.
1431 (virt_loc_aware_diagnostic_finalizer): Likewise.
1432 (diagnostic_report_current_function): Pass the
1433 relevant location to diagnostic_report_current_module.
1434 * tree-diagnostic.h (virt_loc_aware_diagnostic_finalizer): Declare
1436 * toplev.c (general_init): By default, use the new
1437 virt_loc_aware_diagnostic_finalizer as diagnostic finalizer.
1438 * Makefile.in: Add vec.h dependency to tree-diagnostic.c.
1440 2011-10-15 Tom Tromey <tromey@redhat.com>
1441 Dodji Seketeli <dodji@redhat.com>
1443 * doc/cppopts.texi (-ftrack-macro-expansion): Document new option.
1444 * doc/invoke.texi (-ftrack-macro-expansion): Add this to the list of
1445 preprocessor related options.
1447 2011-10-15 Tom Tromey <tromey@redhat>
1448 Dodji Seketeli <dodji@redhat.com>
1450 * input.h (struct expanded_location): Move to libcpp/line-map.h.
1451 (LOCATION_COLUMN): New accessor
1452 (in_system_header_at): Use linemap_location_in_system_header_p.
1453 * diagnostic.c (diagnostic_report_current_module): Adjust to avoid
1454 touching the internals of struct line_map. Use the public API instead.
1455 (diagnostic_report_diagnostic): Don't use relational operator '<'
1456 on virtual locations. Use linemap_location_before_p instead.
1457 * input.c (expand_location): Adjust to expand to the tokens'
1458 spelling location when macro location tracking is on.
1461 2011-10-08 Andi Kleen <ak@linux.intel.com>
1463 * ggc-page.c (GGC_QUIRE_SIZE): Increase to 512
1465 2011-10-13 Andi Kleen <ak@linux.intel.com>
1467 * toplev.c (compile_file): Rename __gnu_slim_lto to __gnu_lto_slim.
1469 2011-10-16 Ira Rosen <ira.rosen@linaro.org>
1471 PR tree-optimization/50727
1472 * tree-vect-patterns.c (vect_operation_fits_smaller_type): Add
1473 DEF_STMT to the list of statements to be replaced by the
1476 2011-10-16 Eric Botcazou <ebotcazou@adacore.com>
1478 PR rtl-optimization/50615
1479 * combine.c (distribute_notes) <REG_ARGS_SIZE>: Skip if I3 is a no-op.
1481 2011-10-16 Jakub Jelinek <jakub@redhat.com>
1483 PR tree-optimization/50596
1484 * tree-vectorizer.h (NUM_PATTERNS): Increase to 7.
1485 * tree-vect-patterns.c (vect_vect_recog_func_ptrs): Add
1486 vect_recog_bool_pattern.
1487 (check_bool_pattern, adjust_bool_pattern_cast,
1488 adjust_bool_pattern, vect_recog_bool_pattern): New functions.
1490 2011-10-16 Ira Rosen <ira.rosen@linaro.org>
1492 * tree-vect-stmts.c (vectorizable_load): For SLP without permutation
1493 treat the first load of the node as the first element in its
1495 * tree-vect-slp.c (vect_get_and_check_slp_defs): Swap the operands if
1496 necessary and possible.
1497 (vect_build_slp_tree): Add new argument. Allow load groups of any size
1498 in basic blocks. Keep all the loads for further permutation check.
1499 Use the new argument to determine if there is a permutation. Update
1500 the recursive calls.
1501 (vect_supported_load_permutation_p): Allow subchains of interleaving
1502 chains in basic block vectorization.
1503 (vect_analyze_slp_instance): Update the call to vect_build_slp_tree.
1504 Check load permutation based on the new parameter.
1505 (vect_schedule_slp_instance): Don't start from the first element in
1506 interleaving chain unless the loads are permuted.
1508 2011-10-15 Jan Hubicka <jh@suse.cz>
1512 * cgraphunit.c (cgraph_expand_function): Expand thunks and alises
1513 after function body.
1515 2011-10-15 Richard Henderson <rth@redhat.com>
1517 * tree-vect-slp.c: Include langhooks.h.
1518 (vect_create_mask_and_perm): Emit VEC_PERM_EXPR, not a builtin.
1519 (vect_transform_slp_perm_load): Use can_vec_perm_expr_p. Simplify
1520 mask creation for VEC_PERM_EXPR.
1521 * tree-vect-stmts.c (perm_mask_for_reverse): Return the mask,
1523 (reverse_vec_elements): Emit VEC_PERM_EXPR not a builtin.
1524 * Makefile.in (tree-vect-slp.o): Update dependency.
1525 * optabs.c (can_vec_perm_expr_p): Allow NULL as unknown constant.
1527 2011-10-15 Alan Modra <amodra@gmail.com>
1530 * ifcvt.c (dead_or_predicable): Revert accidental commit with
1531 HAVE_simple_return test.
1533 2011-10-15 Alan Modra <amodra@gmail.com>
1535 * ifcvt.c (dead_or_predicable): Disable if-conversion when
1536 doing so is likely to kill a shrink-wrapping opportunity.
1538 PR rtl-optimization/49941
1539 * jump.c (mark_jump_label_1): Set JUMP_LABEL for simple_return jumps.
1541 * rtl.h (set_return_jump_label): Declare.
1542 * function.c (set_return_jump_label): New function, extracted..
1543 (thread_prologue_and_epilogue_insns): ..from here. Use it in
1544 another instance to set return jump_label.
1545 * cfgrtl.c (force_nonfallthru_and_redirect): Use set_return_jump_label.
1546 * reorg.c (find_end_label): Likewise.
1548 2011-10-14 David S. Miller <davem@davemloft.net>
1550 * config/sparc/sol2.h: Protect -m{cpu,tune}=native handling
1551 with a more complete cpp test.
1552 * config/sparc/linux64.h: Likewise.
1553 * config/sparc/linux.h: Likewise.
1554 * config/sparc/sparc.opt (sparc_debug): New target variable.
1555 (mdebug): New target option.
1556 * config/sparc/sparc.h (MASK_DEBUG_OPTIONS, MASK_DEBUG_ALL,
1557 TARGET_DEBUG_OPTIONS): New defines.
1558 * config/sparc/sparc.c (debug_target_flag_bits,
1559 debug_target_flags): New functions.
1560 (sparc_option_override): Add name strings back to cpu_table[].
1561 Parse -mdebug string. When TARGET_DEBUG_OPTIONS is true, print
1562 out the target flags before and after override processing as well
1563 as the selected cpu. If MASK_V8PLUS, make sure that the selected
1566 2011-10-15 Oleg Endo <oleg.endo@t-online.de>
1569 * config/sh/sh.h (ZERO_EXTRACT_ANDMASK): New macro.
1570 * config/sh/sh.c (sh_rtx_costs): Add test instruction case.
1571 * config/sh/sh.md (tstsi_t): Name existing insn. Make inner
1572 and instruction commutative.
1573 (tsthi_t, tstqi_t, tstqi_t_zero, tstsi_t_and_not,
1574 tstsi_t_zero_extract_eq, tstsi_t_zero_extract_xor,
1575 tstsi_t_zero_extract_subreg_xor_little,
1576 tstsi_t_zero_extract_subreg_xor_big): New insns.
1577 (*movsicc_t_false, *movsicc_t_true): Replace space with tab in
1579 (*andsi_compact): Reorder alternatives so that K08 is considered first.
1581 2011-10-14 Eric Botcazou <ebotcazou@adacore.com>
1584 * config/sparc/linux64.h (TARGET_DEFAULT): Only override if the default
1585 processor is at least V9 and TARGET_64BIT_DEFAULT is defined.
1587 2011-10-14 Gerald Pfeifer <gerald@pfeifer.com>
1589 * invoke.texi (AVR Options): Avoid \leq{}.
1591 2011-10-14 Kai Tietz <ktietz@redhat.com>
1593 * gimplify.c (gimplify_expr): Take care that for bitwise-binary
1594 transformation the operands have compatible types.
1596 2011-10-14 Jakub Jelinek <jakub@redhat.com>
1598 * config/i386/sse.md (vec_widen_smult_hi_v8hi,
1599 vec_widen_smult_lo_v8hi, vec_widen_umult_hi_v8hi,
1600 vec_widen_umult_lo_v8hi): Macroize using VI2_AVX2
1601 mode iterator and any_extend code iterator.
1602 (vec_widen_<s>mult_hi_v8si, vec_widen_<s>mult_lo_v8si): New expanders.
1603 (vec_widen_smult_hi_v4si, vec_widen_smult_lo_v4si): Enable
1604 also for TARGET_SSE4_1 using pmuldq insn.
1605 (sdot_prodv8hi): Macroize using VI2_AVX2 iterator.
1606 (sse2_sse4_1): New code attr.
1607 (udot_prodv4si): Macroize using any_extend code iterator.
1608 (<s>dot_prodv8si): New expander.
1610 2011-10-14 Yakovlev Vladimir <vladimir.b.yakovlev@intel.com>
1612 * config/i386/i386.c (atom_cost): Changed cost for loading
1613 QImode using movzbl.
1615 2011-10-14 Michael Meissner <meissner@linux.vnet.ibm.com>
1617 * config/rs6000/rs6000.c (rs6000_init_builtins): Fix typo in my
1618 change on October 11th, 2011.
1620 2011-10-14 Jakub Jelinek <jakub@redhat.com>
1622 * config/i386/sse.md (vec_interleave_high<mode>,
1623 vec_interleave_low<mode>): Add AVX2 expanders for VI_256 modes.
1624 * config/i386/i386.c (expand_vec_perm_interleave3): New function.
1625 (ix86_expand_vec_perm_builtin_1): Call it.
1627 2011-10-14 Georg-Johann Lay <avr@gjlay.de>
1629 Fix thinko from r179765
1630 * config/avr/avr.c (avr_option_override): Don't override
1631 flag_omit_frame_pointer if not actually needed.
1633 2011-10-14 Georg-Johann Lay <avr@gjlay.de>
1636 * doc/invoke.texi (AVR Options): Document -mstrict-X.
1638 * config/avr/avr.opt (-mstrict-X): New option.
1639 (avr_strict_X): New variable reflecting -mstrict-X.
1640 * config/avr/avr.c (avr_reg_ok_for_addr_p): Add parameter
1641 outer_code and pass it down to avr_regno_mode_code_ok_for_base_p.
1642 (avr_legitimate_address_p): Pass outer_code to
1643 avr_reg_ok_for_addr_p and use that function in case PLUS.
1644 (avr_mode_code_base_reg_class): Depend on avr_strict_X.
1645 (avr_regno_mode_code_ok_for_base_p): Ditto, and depend on outer_code.
1646 (avr_option_override): Disable -fcaller-saves if -mstrict-X is on.
1648 2011-10-14 Jakub Jelinek <jakub@redhat.com>
1650 * config/i386/sse.md (neg<mode>2): Use VI_AVX2 iterator instead
1653 * config/i386/sse.md (mulv2di3): Macroize using VI8_AVX2 iterator.
1654 (ashl<mode>3): Use VI248_AVX2 iterator instead of VI248_128.
1655 Use <sseinsnmode> instead of TI in mode attr.
1657 2011-10-14 David Alan Gilbert <david.gilbert@linaro.org>
1659 * config/arm/linux-atomic-64bit.c: New (based on linux-atomic.c).
1660 * config/arm/linux-atomic.c: Change comment to point to 64bit version.
1661 (SYNC_LOCK_RELEASE): Instantiate 64bit version.
1662 * config/arm/t-linux-eabi: Pull in linux-atomic-64bit.c.
1664 2011-10-14 David Alan Gilbert <david.gilbert@linaro.org>
1666 * config/arm/arm.c (arm_output_ldrex): Support ldrexd.
1667 (arm_output_strex): Support strexd.
1668 (arm_output_it): New helper to output it in Thumb2 mode only.
1669 (arm_output_sync_loop): Support DI mode. Change comment to
1670 not support const_int.
1671 (arm_expand_sync): Support DI mode.
1672 * config/arm/arm.h (TARGET_HAVE_LDREXBHD): Split into LDREXBH
1674 * config/arm/iterators.md (NARROW): move from sync.md.
1675 (QHSD): New iterator for all current ARM integer modes.
1676 (SIDI): New iterator for SI and DI modes only.
1677 * config/arm/sync.md (sync_predtab): New mode_attr.
1678 (sync_compare_and_swapsi): Fold into sync_compare_and_swap<mode>.
1679 (sync_lock_test_and_setsi): Fold into sync_lock_test_and_setsi<mode>.
1680 (sync_<sync_optab>si): Fold into sync_<sync_optab><mode>.
1681 (sync_nandsi): Fold into sync_nand<mode>.
1682 (sync_new_<sync_optab>si): Fold into sync_new_<sync_optab><mode>.
1683 (sync_new_nandsi): Fold into sync_new_nand<mode>.
1684 (sync_old_<sync_optab>si): Fold into sync_old_<sync_optab><mode>.
1685 (sync_old_nandsi): Fold into sync_old_nand<mode>.
1686 (sync_compare_and_swap<mode>): Support SI & DI.
1687 (sync_lock_test_and_set<mode>): Likewise.
1688 (sync_<sync_optab><mode>): Likewise.
1689 (sync_nand<mode>): Likewise.
1690 (sync_new_<sync_optab><mode>): Likewise.
1691 (sync_new_nand<mode>): Likewise.
1692 (sync_old_<sync_optab><mode>): Likewise.
1693 (sync_old_nand<mode>): Likewise.
1694 (arm_sync_compare_and_swapsi): Turn into iterator on SI & DI.
1695 (arm_sync_lock_test_and_setsi): Likewise.
1696 (arm_sync_new_<sync_optab>si): Likewise.
1697 (arm_sync_new_nandsi): Likewise.
1698 (arm_sync_old_<sync_optab>si): Likewise.
1699 (arm_sync_old_nandsi): Likewise.
1700 (arm_sync_compare_and_swap<mode> NARROW): use sync_predtab, fix indent.
1701 (arm_sync_lock_test_and_setsi<mode> NARROW): Likewise.
1702 (arm_sync_new_<sync_optab><mode> NARROW): Likewise.
1703 (arm_sync_new_nand<mode> NARROW): Likewise.
1704 (arm_sync_old_<sync_optab><mode> NARROW): Likewise.
1705 (arm_sync_old_nand<mode> NARROW): Likewise.
1707 2011-10-14 David Alan Gilbert <david.gilbert@linaro.org>
1710 * config/arm/arm.c (arm_output_sync_loop): Move label before barrier.
1712 2011-10-14 David Alan Gilbert <david.gilbert@linaro.org>
1714 * config/arm/arm.h (TARGET_HAVE_DMB_MCR): MCR Not available in Thumb1.
1716 2011-10-14 Paolo Carlini <paolo.carlini@oracle.com>
1718 * doc/invoke.texi ([Wformat-zero-length]): Tidy.
1720 2011-10-14 Jakub Jelinek <jakub@redhat.com>
1722 * gimple.c (walk_stmt_load_store_addr_ops): Call visit_addr also
1723 on COND_EXPR/VEC_COND_EXPR comparison operands if they are ADDR_EXPRs.
1725 2011-10-14 Richard Guenther <rguenther@suse.de>
1727 PR tree-optimization/50723
1728 * ipa-split.c (split_function): Use GSI_CONTINUE_LINKING.
1730 2011-10-14 Nicola Pero <nicola.pero@meta-innovation.com>
1732 * gengtype.c (files_rules): Added rules for objc/objc-map.h and
1735 2011-10-14 Paolo Carlini <paolo.carlini@oracle.com>
1738 * doc/invoke.texi ([Wformat-zero-length]): Update.
1740 2011-10-14 Iain Sandoe <iains@gcc.gnu.org>
1743 * config/darwin.c (darwin_patch_builtin): Adjust argument type. Only
1744 build for powerpc targets. (darwin_patch_builtins): Only build for
1747 2011-10-14 Jakub Jelinek <jakub@redhat.com>
1749 * config/i386/sse.md (*avx_cvtdq2pd256_2): Rename to...
1750 (avx_cvtdq2pd256_2): ... this.
1751 (sseunpackfltmode): New mode attr.
1752 (vec_unpacks_float_hi_v8hi, vec_unpacks_float_lo_v8hi,
1753 vec_unpacku_float_hi_v8hi, vec_unpacku_float_lo_v8hi): Macroize
1754 using VI2_AVX2 iterator.
1755 (vec_unpacku_float_hi_v8si, vec_unpacku_float_lo_v8si): New expanders.
1757 2011-10-13 David S. Miller <davem@davemloft.net>
1759 * config/sparc/sparc.md (plusminus): New code iterator.
1760 (plusminus_insn): New code attr.
1761 (addv2si3, subv2si3, addv4hi3, subv4hi3, addv2hi3, subv2hi3): Merge
1762 using plusminus and plusminus_insn.
1763 (fpadd64_vis, fpsub64_vis): Likewise.
1765 2011-10-13 Richard Henderson <rth@redhat.com>
1767 * doc/md.texi (vec_perm): Document fallback to byte permutation.
1768 * genopinit.c (optabs): Add vec_perm_const.
1769 * optabs.c (can_vec_perm_expr_p): Rename from expand_vec_perm_expr_p.
1770 Reject non-vector modes. Allow fallback to byte permutation.
1771 (expand_vec_perm_expr_1): Split out from ...
1772 (expand_vec_perm_expr): ... here. Allow fallback to byte permutation.
1773 * optabs.h (DOI_vec_perm_const, vec_perm_const_optab): New.
1774 * tree-vect-generic.c (lower_vec_perm): Update for name change.
1776 2011-10-13 Richard Henderson <rth@redhat.com>
1778 * config/rs6000/altivec.md (vec_permv16qi): New pattern.
1780 * config/rs6000/spu.md (vec_permv16qi): New pattern.
1782 * config/i386/i386.c (ix86_expand_vec_perm_const): New.
1783 * config/i386/i386-protos.h: Update.
1784 * config/i386/sse.md (VEC_PERM_CONST): New mode iterator.
1785 (vec_perm_const<VEC_PERM_CONST>): New expander.
1787 * optabs.c (expand_vector_broadcast): New.
1788 (expand_binop): Expand scalar shifts of vectors to vector shifts
1789 of vectors, if the former isn't supported.
1790 * tree-vect-generic.c (expand_vector_operations_1): Don't do that
1791 here; always simplify to scalar shift of vector if possible.
1793 * config/rs6000/rs6000.c (rs6000_expand_vector_init): Fix mode
1794 test for vector splat.
1796 2011-10-13 Jakub Jelinek <jakub@redhat.com>
1798 * config/i386/sse.md (vec_set<mode>): Change V_128 iterator mode to V.
1800 2011-10-13 Jakub Jelinek <jakub@redhat.com>
1801 Richard Guenther <rguenther@suse.de>
1803 * tree-ssa.c (maybe_optimize_var): Drop TREE_ADDRESSABLE from vector
1804 or complex vars even if their DECL_UID is in not_reg_needs bitmap.
1806 2011-10-13 Jakub Jelinek <jakub@redhat.com>
1808 * config/i386/sse.md (reduc_umin_v8hi): New pattern.
1809 * config/i386/i386.c (ix86_build_const_vector): Handle
1810 also V32QI, V16QI, V16HI and V8HI modes.
1811 (emit_reduc_half): New function.
1812 (ix86_expand_reduc): Use phminposuw insn for V8HImode UMIN.
1813 Use emit_reduc_half helper function.
1815 2011-10-13 Lawrence Crowl <crowl@google.com>
1816 Diego Novillo <dnovillo@google.com>
1818 * lto-streamer-in.c (input_struct_function_base): Factor out of ...
1819 (input_function): ... here.
1820 * lto-streamer-out.c (output_struct_function_base): Factor out of ...
1821 (output_function): ... here.
1823 2011-10-13 Gabriel Charette <gchare@google.com>
1824 Diego Novillo <dnovillo@google.com>
1826 * streamer-hooks.h (struct streamer_hooks): Add hooks
1827 input_location and output_location.
1828 * lto-streamer-in.c (lto_input_location): Use
1829 streamer_hooks.input_location, if set.
1830 * lto-streamer-out.c (lto_output_location): Use
1831 streamer_hooks.output_location, if set.
1833 2011-10-13 Eric Botcazou <ebotcazou@adacore.com>
1835 * doc/invoke.texi (SPARC options): Document -mfix-at697f.
1836 * config/sparc/sparc.opt (mfix-at697f): New option.
1837 * config/sparc/sparc.c (TARGET_MACHINE_DEPENDENT_REORG): Define.
1838 (sparc_reorg): New function.
1840 2011-10-13 Richard Guenther <rguenther@suse.de>
1842 PR tree-optimization/50712
1843 * ipa-split.c (split_function): Always re-gimplify parameters
1844 when they are not gimple vals before passing them. Properly
1845 check for type compatibility.
1847 2011-10-13 Tom de Vries <tom@codesourcery.com>
1849 * function.c (gimplify_parameters): Set number of arguments of call to
1850 BUILT_IN_ALLOCA_WITH_ALIGN to 2.
1852 2011-10-13 Tom de Vries <tom@codesourcery.com>
1854 * emit-rtl.c (set_mem_attributes_minus_bitpos): Set MEM_READONLY_P
1855 for static const strings.
1856 * varasm.c (build_constant_desc): Generate the memory location of the
1857 constant using gen_const_mem.
1859 2011-10-13 Richard Guenther <rguenther@suse.de>
1861 PR tree-optimization/50698
1862 * tree-data-ref.c (split_constant_offset_1): Also process
1865 2011-10-12 David S. Miller <davem@davemloft.net>
1867 * config/sparc/sparc.md (UNSPEC_FPMERGE): Delete.
1868 (UNSPEC_MUL16AU, UNSPEC_MUL8, UNSPEC_MUL8SU, UNSPEC_MULDSU): New
1870 (fpmerge_vis): Remove inaccurate comment, represent using vec_select
1872 (vec_interleave_lowv8qi, vec_interleave_highv8qi): New insns.
1873 (fmul8x16_vis, fmul8x16au_vis, fmul8sux16_vis, fmuld8sux16_vis):
1874 Reimplement as unspecs and remove inaccurate comments.
1875 (vis3_shift_patname): New code attr.
1876 (<vis3_shift_insn><vbits>_vis): Rename to
1877 "v<vis3_shift_patname><mode>3".
1878 (vis3_addsub_ss_patname): New code attr.
1879 (<vis3_addsub_ss_insn><vbits>_vis): Rename to
1880 "<vis3_addsub_ss_patname><mode>3".
1881 * config/sparc/sparc.c (sparc_vis_init_builtins): Update to
1882 accommodate pattern name changes.
1884 * config/sparc/sparc.h: Do not force TARGET_VIS3 and TARGET_FMAF
1885 to zero when assembler lacks support for such instructions.
1886 * config/sparc/sparc.c (sparc_option_override): Clear MASK_VIS3
1887 and MASK_FMAF in defaults when assembler lacks necessary support.
1889 2011-10-12 Jakub Jelinek <jakub@redhat.com>
1891 * config/i386/sse.md (vec_unpacks_lo_<mode>,
1892 vec_unpacks_hi_<mode>, vec_unpacku_lo_<mode>,
1893 vec_unpacku_hi_<mode>): Change VI124_128 mode to VI124_AVX2.
1894 * config/i386/i386.c (ix86_expand_sse_unpack): Handle
1895 V32QImode, V16HImode and V8SImode for TARGET_AVX2.
1897 * config/i386/sse.md (vec_avx2): New mode_attr.
1898 (mulv16qi3): Macroize to cover also mulv32qi3 for TARGET_AVX2 into ...
1899 (mul<mode>3): ... this.
1901 * config/i386/i386.md (UNSPEC_VPERMDI): Remove.
1902 * config/i386/i386.c (ix86_expand_vec_perm): Handle
1903 V16QImode and V32QImode for TARGET_AVX2.
1904 (MAX_VECT_LEN): Increase to 32.
1905 (expand_vec_perm_blend): Add support for 32-byte integer
1906 vectors with TARGET_AVX2.
1907 (valid_perm_using_mode_p): New function.
1908 (expand_vec_perm_pshufb): Add support for 32-byte integer
1909 vectors with TARGET_AVX2.
1910 (expand_vec_perm_vpshufb2_vpermq): New function.
1911 (expand_vec_perm_vpshufb2_vpermq_even_odd): New function.
1912 (expand_vec_perm_even_odd_1): Handle 32-byte integer vectors
1914 (ix86_expand_vec_perm_builtin_1): Try expand_vec_perm_vpshufb2_vpermq
1915 and expand_vec_perm_vpshufb2_vpermq_even_odd.
1916 * config/i386/sse.md (VEC_EXTRACT_EVENODD_MODE): Add for TARGET_AVX2
1917 32-byte integer vector modes.
1918 (vec_pack_trunc_<mode>): Use VI248_AVX2 instead of VI248_128.
1919 (avx2_interleave_highv32qi, avx2_interleave_lowv32qi): Remove pasto.
1920 (avx2_pshufdv3, avx2_pshuflwv3, avx2_pshufhwv3): Generate
1922 (avx2_pshufd_1, avx2_pshuflw_1, avx2_pshufhw_1): Don't use
1923 match_dup, instead add 4 new operands and require they have
1924 right cross-lane values.
1925 (avx2_permv4di): Change into define_expand.
1926 (avx2_permv4di_1): New instruction.
1927 (avx2_permv2ti): Use nonimmediate_operand instead of register_operand
1928 for "xm" constrained operand.
1929 (VEC_PERM_AVX2): Add V32QI and V16QI for TARGET_AVX2.
1931 * config/i386/sse.md (avx2_gathersi<mode>,
1932 avx2_gatherdi<mode>, avx2_gatherdi<mode>256): Add clobber of
1933 match_scratch, change memory_operand to register_operand,
1934 add (mem:BLK (scratch)) use.
1935 (*avx2_gathersi<mode>, *avx2_gatherdi<mode>,
1936 *avx2_gatherdi<mode>256): Add clobber of match_scratch,
1937 add earlyclobber to the output operand and match_scratch,
1938 add (mem:BLK (scratch)) use, change the other mem to match_operand.
1939 Use %p6 instead of %c6 in the pattern.
1940 * config/i386/i386.c (ix86_expand_builtin): Adjust for
1941 operand 2 being a Pmode register_operand instead of memory_operand.
1943 2011-10-12 Kai Tietz <ktietz@redhat.com>
1945 * config/i386/i386.md (simple_return): Disable if TARGET_SEH is active.
1947 2011-10-12 Steve Ellcey <sje@cup.hp.com>
1949 * config/ia64/ia64.c (ia64_init_builtins): Fix unbalanced parentheses.
1951 2011-10-12 Richard Guenther <rguenther@suse.de>
1953 * tree-ssa-alias.c (maybe_skip_until): Cache also at the point
1954 of the first store we visit in a basic-block.
1955 (get_continuation_for_phi): Search for a candidate VUSE that
1956 might dominates all others. Do pairwise disambiguation against
1959 2011-10-12 Paul Koning <pkoning@gcc.gnu.org>
1961 PR tree-optimization/50189
1962 * tree-vrp.c (extract_range_from_assert): Use the type of
1963 the variable, not the limit.
1965 2011-10-12 Richard Guenther <rguenther@suse.de>
1967 PR tree-optimization/50700
1968 * tree-object-size.c (addr_object_size): Simplify and treat
1969 MEM_REF bases consistently.
1971 2011-10-12 Bernd Schmidt <bernds@codesourcery.com>
1973 * function.c (prepare_shrink_wrap, bb_active_p): New function.
1974 (thread_prologue_and_epilogue_insns): Use bb_active_p. Call
1975 prepare_shrink_wrap, then recompute bb_active_p for the last block.
1977 2011-10-12 Joseph Myers <joseph@codesourcery.com>
1980 * convert.c (convert_to_integer): Do not narrow operands of
1981 pointer subtraction.
1983 2011-10-12 Nick Clifton <nickc@redhat.com>
1985 * config/arm/arm.h (EMIT_EABI_ATTRIBUTE): New macro. Used to
1986 emit a .eabi_attribute assembler directive, possibly with a
1988 * config/arm/arm.c (arm_file_start): Use the new macro.
1989 * config/arm/arm-c.c (arm_output_c_attributes): Likewise.
1991 2011-10-12 Georg-Johann Lay <avr@gjlay.de>
1994 * config/avr/avr.md (*movqi): Rename to movqi_insn.
1995 (*call_insn): Rename to call_insn.
1996 (*call_value_insn): Rename to call_value_insn.
1997 * config/avr/avr.c (avr_2word_insn_p): New static function.
1998 (jump_over_one_insn_p): Use it.
2000 2011-10-12 Richard Sandiford <richard.sandiford@linaro.org>
2002 * expr.h (copy_blkmode_to_reg): Declare.
2003 * expr.c (copy_blkmode_to_reg): New function.
2004 (expand_assignment): Don't expand register RESULT_DECLs before
2005 the lhs. Use copy_blkmode_to_reg to copy BLKmode values into a
2006 RESULT_DECL register.
2007 (expand_expr_real_1): Handle BLKmode decls when looking for promotion.
2008 * stmt.c (expand_return): Move BLKmode-to-register code into
2009 copy_blkmode_to_reg.
2011 2011-10-11 Eric Botcazou <ebotcazou@adacore.com>
2014 * config/sparc/sparc.md (mov<I:mode>cc): Do not save comparison code.
2015 (mov<F:mode>cc): Likewise.
2017 2011-10-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2019 * tree-ssa-address.c (copy_ref_info): Remove copy of TREE_THIS_NOTRAP.
2021 2011-10-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2023 * tree.h (built_in_decls): Delete old interface with two parallel
2024 arrays to hold standard builtin declarations, and replace it with
2025 a function based interface that can support creating builtins on
2026 the fly in the future. Change all uses, and poison the old
2027 names. Make sure 0 is not a legitimate builtin index.
2028 (implicit_built_in_decls): Ditto.
2029 (built_in_info): Ditto.
2030 (BUILTIN_VALID_P): Ditto.
2031 (builtin_decl_explicit): Ditto.
2032 (builtin_decl_implicit): Ditto.
2033 (set_builtin_decl): Ditto.
2034 (set_builtin_decl_implicit_p): Ditto.
2035 (builtin_decl_explicit_p): Ditto.
2036 (builtin_decl_implicit_p): Ditto.
2037 * tree-complex.c (expand_complex_libcall): Ditto.
2038 * tree-loop-distribution.c (generate_memset_zero): Ditto.
2039 * tree-ssa-strlen.c (get_string_length): Ditto.
2040 (handle_builtin_strcpy): Ditto.
2041 (handle_builtin_strcat): Ditto.
2042 * tree.c (iterative_hash_expr): Ditto.
2043 (local_define_builtin): Ditto.
2044 (build_common_builtin_nodes): Ditto.
2045 * builtins.c (built_in_decls): Ditto.
2046 (implicit_built_in_decls): Ditto.
2047 (built_in_info): Ditto
2048 (expand_builtin_classify_type): Ditto.
2049 (mathfn_built_in_1): Ditto.
2050 (expand_builtin_cexpi): Ditto.
2051 (expand_builtin_mempcpy_args): Ditto.
2052 (expand_builtin_stpcpy): Ditto.
2053 (gimplify_va_arg_expr): Ditto.
2054 (expand_builtin_sync_operation): Ditto.
2055 (build_builtin_expect_predicate): Ditto.
2056 (fold_builtin_memory_op): Ditto.
2057 (fold_builtin_strcpy): Ditto.
2058 (fold_builtin_stpcpy): Ditto.
2059 (fold_builtin_strncpy): Ditto.
2060 (fold_builtin_interclass_mathfn): Ditto.
2061 (fold_builtin_classify): Ditto.
2062 (fold_builtin_2): Ditto.
2063 (fold_builtin_strstr): Ditto.
2064 (fold_builtin_strrchr): Ditto.
2065 (fold_builtin_strpbrk): Ditto.
2066 (fold_builtin_strcat): Ditto.
2067 (fold_builtin_strncat): Ditto.
2068 (fold_builtin_strcspn): Ditto.
2069 (fold_builtin_fputs): Ditto.
2070 (fold_builtin_sprintf): Ditto.
2071 (fold_builtin_snprintf): Ditto.
2072 (expand_builtin_memory_chk): Ditto.
2073 (fold_builtin_memory_chk): Ditto.
2074 (fold_builtin_stxcpy_chk): Ditto.
2075 (fold_builtin_strncpy_chk): Ditto.
2076 (fold_builtin_strcat_chk): Ditto.
2077 (fold_builtin_strncat_chk): Ditto.
2078 (fold_builtin_sprintf_chk_1): Ditto.
2079 (fold_builtin_snprintf_chk_1): Ditto.
2080 (fold_builtin_printf): Ditto.
2081 (fold_builtin_fprintf): Ditto.
2082 (fold_call_stmt): Ditto.
2083 (set_builtin_user_assembler_name): Ditto.
2084 * tree-emutls.c (emutls_common_1): Ditto.
2085 * omp-low.c (scan_omp): Ditto.
2086 (lower_rec_input_clauses): Ditto.
2087 (lower_reduction_clauses): Ditto.
2088 (expand_parallel_call): Ditto.
2089 (expand_task_call): Ditto.
2090 (maybe_catch_exception): Ditto.
2091 (optimize_omp_library_calls): Ditto.
2092 (expand_omp_for_generic): Ditto.
2093 (expand_omp_for_static_nochunk): Ditto.
2094 (expand_omp_for_static_chunk): Ditto.
2095 (expand_omp_sections): Ditto.
2096 (expand_omp_atomic_fetch_op): Ditto.
2097 (expand_omp_atomic_pipeline): Ditto.
2098 (expand_omp_atomic_mutex): Ditto.
2099 (lower_omp_single_simple): Ditto.
2100 (lower_omp_single_copy): Ditto.
2101 (lower_omp_master): Ditto.
2102 (lower_omp_ordered): Ditto.
2103 (lower_omp_critical): Ditto.
2104 * tree-ssa-ccp.c (optimize_stdarg_builtin): Ditto.
2105 * builtins.c (DEF_BUILTIN_STUB): Ditto.
2106 (BUILT_IN_NONE): Ditto.
2107 * tree-ssa-math-opts.c (execute_optimize_bswap): Ditto.
2108 * gimple-low.c (lower_function_body): Ditto.
2109 (lower_builtin_setjmp): Ditto.
2110 * c-decl.c (merge_decls): Ditto.
2111 * tree-eh.c (lower_resx): Ditto.
2112 (lower_resx): Ditto.
2113 (lower_eh_dispatch): Ditto.
2114 * function (gimplify_parameters): Ditto.
2115 * c-typeck.c (build_function_call_vec): Ditto.
2116 * gimplify.c (build_stack_save_restore): Ditto.
2117 (gimplify_vla_decl): Ditto.
2118 (gimplify_modify_expr_to_memcpy): Ditto.
2119 (gimplify_modify_expr_to_memset): Ditto.
2120 (gimplify_variable_sized_compare): Ditto.
2121 (gimplify_function_tree): Ditto.
2122 * calls.c (emit_call_1): Ditto.
2123 * tree-ssa-forprop.c (simplify_builtin_call): Ditto.
2124 * tree-nested.c (convert_nl_goto_reference): Ditto.
2125 (convert_tramp_reference_op): Ditto.
2126 (finalize_nesting_tree_1): Ditto.
2127 * tree-ssa-loop-prefetch.c (issue_prefetch_ref): Ditto.
2128 (tree_ssa_prefetch_arrays): Ditto.
2129 * tree-streamer-in.c (streamer_get_builtin_tree): Ditto.
2130 * system.h (built_in_decls): Ditto.
2131 (implicit_built_in_decls): Ditto.
2132 * tree-vect-generic.c (expand_vector_operations_1): Ditto.
2133 * config/sparc/sparc.c (sparc_gimplify_va_arg): Ditto.
2134 * config/i386/i386.c (ix86_gimplify_va_arg): Ditto.
2135 (ix86_veclibabi_svml): Ditto.
2136 (ix86_veclibabi_acml): Ditto.
2137 * config/vms/vms.c (vms_patch_builtins): Ditto.
2138 * config/ia64/ia64.c (ia64_init_builtins): Ditto.
2139 * config/rs6000/rs6000.c (rs6000_gimplify_va_arg): Ditto.
2140 (rs6000_builtin_vectorized_libmass): Ditto.
2141 (rs6000_init_builtins): Ditto.
2142 * config/darwin.c (darwin_override_options): Ditto.
2143 (darwin_patch_builtin): Ditto.
2144 (darwin_rename_builtins): Ditto.
2145 * config/pa/pa.c (pa_init_builtins): Ditto.
2147 2011-10-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2149 * tree.h (copy_ref_info): Expose existing function.
2150 * tree-ssa-loop-ivopts.c (copy_ref_info): Move function to...
2151 * tree-ssa-address.c (copy_ref_info): ...here, and remove static token.
2153 2011-10-11 Georg-Johann Lay <avr@gjlay.de>
2155 * config/avr/avr-protos.h (avr_mode_code_base_reg_class): New
2157 (avr_regno_mode_code_ok_for_base_p): New prototype.
2158 * config/avr/avr.h (BASE_REG_CLASS): Remove.
2159 (REGNO_OK_FOR_BASE_P): Remove.
2160 (REG_OK_FOR_BASE_NOSTRICT_P): Remove.
2161 (REG_OK_FOR_BASE_STRICT_P): Remove.
2162 (MODE_CODE_BASE_REG_CLASS): New define.
2163 (REGNO_MODE_CODE_OK_FOR_BASE_P): New define.
2164 * config/avr/avr.c (avr_mode_code_base_reg_class): New function.
2165 (avr_regno_mode_code_ok_for_base_p): New function.
2166 (avr_reg_ok_for_addr_p): New static function.
2167 (avr_legitimate_address_p): Use it. Beautify.
2169 2011-10-11 Georg-Johann Lay <avr@gjlay.de>
2172 * config/avr/avr.md (cc): Add out_plus attribute alternative.
2173 (addsi3): Use it. Adapt avr_out_plus to new prototype. Use
2174 avr_out_plus for all CONST_INT addends.
2175 * config/avr/avr-protos.h (avr_out_plus): Change prototype.
2176 * config/avr/avr.c (notice_update_cc): Call avr_out_plus on
2178 (avr_out_plus_1): Change prototype and report effect on cc0.
2179 (avr_out_plus): Ditto.
2180 (adjust_insn_length): Adapt call to avr_out_plus to new prototype.
2182 2011-10-11 H.J. Lu <hongjiu.lu@intel.com>
2184 * config/i386/i386.c (ix86_expand_special_args_builtin): Remove
2187 2011-10-11 Artjoms Sinkarovs <artyom.shinkaroff@gmail.com>
2189 * doc/invoke.texi: Document new warning.
2190 * common.opt (Wvector-operation-performance): Define new warning.
2191 * tree-vect-generic.c (expand_vector_piecewise): Warn about expanded
2193 (exapnd_vector_parallel): Warn about expanded vector operation.
2194 (lower_vec_shuffle): Warn about expanded vector operation.
2195 * c-typeck.c (c_build_vec_perm_expr): Store correct location
2196 when creating VEC_PERM_EXPR.
2198 2011-10-11 Richard Guenther <rguenther@suse.de>
2200 PR tree-optimization/50204
2201 * tree-ssa-alias.c (get_continuation_for_phi_1): Split out
2202 two argument handling from ...
2203 (get_continuation_for_phi): ... here. Handle arbitrary number
2206 2011-10-11 Richard Sandiford <richard.sandiford@linaro.org>
2208 * modulo-sched.c: Fix comment typo. Mention the possibility
2209 of using scheduling windows of II+1 cycles.
2211 2011-10-11 Tristan Gingold <gingold@adacore.com>
2213 * doc/invoke.texi (C Dialect Options): Document
2214 -fallow-parameterless-variadic-functions.
2215 * c-parser.c (c_parser_parms_list_declarator): Handle it.
2217 2011-10-10 Georg-Johann Lay <avr@gjlay.de>
2219 * config/avr/avr.c (avr_option_override): Set flag_omit_frame_pointer
2220 to 0 if frame pointer is needed for unwinding.
2222 2011-10-10 Uros Bizjak <ubizjak@gmail.com>
2225 * optabs.h (DOI_vec_perm): Rename from OTI_vec_perm. Move from enum
2226 optab_index to enum direct_optab_index.
2227 (vec_perm_optab): Update.
2229 2011-10-10 Anatoly Sokolov <aesok@post.ru>
2231 * config/cris/cris.c (cris_preferred_reload_class): New function.
2232 (TARGET_PREFERRED_RELOAD_CLASS): Define.
2233 * config/cris/cris.h (PREFERRED_RELOAD_CLASS): Remove.
2235 2011-10-10 Georg-Johann Lay <avr@gjlay.de>
2237 * config/avr/avr.md (*tablejump_rjmp): Change insn condition to
2239 (*tablejump_lib): Change insn condition to AVR_HAVE_JMP_CALL.
2240 (*tablejump_enh, *tablejump): Remove insns.
2241 * config/avr/libgcc.S (__tablejump__): Use RET instead of EIND +
2242 EIJMP for indirect jump. Use LPM Z+ where available.
2244 2011-10-10 Richard Henderson <rth@redhat.com>
2246 * doc/md.texi (vec_perm_const): Fix typo in cindex.
2248 2011-10-10 Kirill Yukhin <kirill.yukhin@intel.com>
2249 Yakovlev Vladimir <vladimir.b.yakovlev@intel.com>
2251 * config/i386/sse.md (fma_fnmsub_<mode>): Fix a typo.
2253 2011-10-10 Richard Guenther <rguenther@suse.de>
2256 * gimple-fold.c (gimplify_and_update_call_from_tree): Do not
2257 mark symbols for renaming. Append the VUSE to all statements
2258 that possibly can have one.
2260 2011-10-10 Richard Guenther <rguenther@suse.de>
2262 * ipa-split.c (pass_split_functions): Add verification TODOs.
2263 (pass_feedback_split_functions): Likewise.
2265 2011-10-10 Richard Guenther <rguenther@suse.de>
2268 * fold-const.c (fold_binary_loc): Canonicalize x*x to pow (x, 2)
2269 only when optimizing.
2271 2011-10-10 Nick Clifton <nickc@redhat.com>
2274 * compare-elim.c (find_comparisons_in_bb): Use df_get_live_in and
2275 df_get_live_out instead of accessing the bitmaps directly.
2276 (execute_compare_elim_after_reload): Remove calls to df_set_flags,
2277 df_live_add_problem and df_analyze.
2279 2011-10-10 Michael Matz <matz@suse.de>
2282 * tree-emutls.c (gen_emutls_addr): Call add_referenced_var.
2284 2011-10-10 Richard Sandiford <richard.sandiford@linaro.org>
2286 * modulo-sched.c (ps_reg_move_info): Add num_consecutive_stages.
2287 (SCHED_FIRST_REG_MOVE, SCHED_NREG_MOVES): Delete.
2288 (node_sched_params): Remove first_reg_move and nreg_moves.
2289 (ps_num_consecutive_stages, extend_node_sched_params): New functions.
2290 (update_node_sched_params): Move up file.
2291 (print_node_sched_params): Print the stage. Don't dump info related
2292 to first_reg_move and nreg_moves.
2293 (set_columns_for_row): New function.
2294 (set_columns_for_ps): Move up file and use set_columns_for_row.
2295 (schedule_reg_move): New function.
2296 (schedule_reg_moves): Call extend_node_sched_params and
2297 schedule_reg_move. Extend size of uses bitmap. Initialize
2298 num_consecutive_stages. Return false if a move could not be
2300 (apply_reg_moves): Don't emit moves here.
2301 (permute_partial_schedule): Handle register moves.
2302 (duplicate_insns_of_cycles): Remove for_prolog. Emit moves according
2303 to the same stage-count test as ddg nodes.
2304 (generate_prolog_epilog): Update calls accordingly.
2305 (sms_schedule): Allow move-scheduling to add a new first stage.
2307 2011-10-10 Richard Sandiford <richard.sandiford@linaro.org>
2309 * modulo-sched.c (ps_insn): Adjust comment.
2310 (ps_reg_move_info): New structure.
2311 (partial_schedule): Add reg_moves field.
2312 (SCHED_PARAMS): Use node_sched_param_vec instead of node_sched_params.
2313 (node_sched_params): Turn first_reg_move into an identifier.
2314 (ps_reg_move): New function.
2315 (ps_rtl_insn): Cope with register moves.
2316 (ps_first_note): Adjust comment and assert that the instruction
2317 isn't a register move.
2318 (node_sched_params): Replace with...
2319 (node_sched_param_vec): ...this vector.
2320 (set_node_sched_params): Adjust accordingly.
2321 (print_node_sched_params): Take a partial schedule instead of a ddg.
2322 Use ps_rtl_insn and ps_reg_move.
2323 (generate_reg_moves): Rename to...
2324 (schedule_reg_moves): ...this. Remove rescan parameter. Record each
2325 move in the partial schedule, but don't emit it here. Don't perform
2326 register substitutions here either.
2327 (apply_reg_moves): New function.
2328 (duplicate_insns_of_cycles): Use register indices directly,
2329 rather than finding instructions using PREV_INSN. Use ps_reg_move.
2330 (sms_schedule): Call schedule_reg_moves before committing to
2331 a partial schedule. Try the next ii if the schedule fails.
2332 Use apply_reg_moves instead of generate_reg_moves. Adjust
2333 call to print_node_sched_params. Free node_sched_param_vec
2334 instead of node_sched_params.
2335 (create_partial_schedule): Initialize reg_moves.
2336 (free_partial_schedule): Free reg_moves.
2338 2011-10-10 Richard Sandiford <richard.sandiford@linaro.org>
2340 * modulo-sched.c (ps_insn): Replace node field with an identifier.
2341 (SCHED_ASAP): Replace with..
2342 (NODE_ASAP): ...this macro.
2343 (SCHED_PARAMS): New macro.
2344 (SCHED_TIME, SCHED_FIRST_REG_MOVE, SCHED_NREG_MOVES, SCHED_ROW)
2345 (SCHED_STAGE, SCHED_COLUMN): Redefine using SCHED_PARAMS.
2346 (node_sched_params): Remove asap.
2347 (ps_rtl_insn, ps_first_note): New functions.
2348 (set_node_sched_params): Use XCNEWVEC. Don't copy across the
2350 (print_node_sched_params): Use SCHED_PARAMS and NODE_ASAP.
2351 (generate_reg_moves): Pass ids to the SCHED_* macros.
2352 (update_node_sched_params): Take a ps insn identifier rather than
2353 a node as parameter. Use ps_rtl_insn.
2354 (set_columns_for_ps): Update for above field and SCHED_* macro changes.
2355 (permute_partial_schedule): Use ps_rtl_insn and ps_first_note.
2356 (optimize_sc): Update for above field and SCHED_* macro changes.
2357 Update calls to try_scheduling_node_in_cycle and
2358 update_node_sched_params.
2359 (duplicate_insns_of_cycles): Adjust for above field and SCHED_*
2360 macro changes. Use ps_rtl_insn and ps_first_note.
2361 (sms_schedule): Pass ids to the SCHED_* macros.
2362 (get_sched_window): Adjust for above field and SCHED_* macro changes.
2363 Use NODE_ASAP instead of SCHED_ASAP.
2364 (try_scheduling_node_in_cycle): Remove node parameter. Update
2365 call to ps_add_node_check_conflicts. Pass ids to the SCHED_* macros.
2366 (sms_schedule_by_order): Update call to try_scheduling_node_in_cycle.
2367 (ps_insert_empty_row): Adjust for above field changes.
2368 (compute_split_row): Use ids rather than nodes.
2369 (verify_partial_schedule): Adjust for above field changes.
2370 (print_partial_schedule): Use ps_rtl_insn.
2371 (create_ps_insn): Take an id rather than a node.
2372 (ps_insn_find_column): Adjust for above field changes.
2374 (ps_insn_advance_column): Adjust for above field changes.
2375 (add_node_to_ps): Remove node parameter. Update call to
2377 (ps_has_conflicts): Use ps_rtl_insn.
2378 (ps_add_node_check_conflicts): Replace node parameter than an id.
2380 2011-10-10 Richard Sandiford <richard.sandiford@linaro.org>
2382 * modulo-sched.c (undo_replace_buff_elem): Delete.
2383 (generate_reg_moves): Don't build and return an undo list.
2384 (free_undo_replace_buff): Delete.
2385 (sms_schedule): Adjust call to generate_reg_moves.
2386 Don't call free_undo_replace_buff.
2388 2011-10-10 Matthias Klose <doko@ubuntu.com>
2390 * common/config/m32c: Remove empty directory.
2392 2011-10-10 Georg-Johann Lay <avr@gjlay.de>
2394 * config/avr/avr.md (*mulhi3_enh): Treat squaring smarter.
2396 2011-10-10 Georg-Johann Lay <avr@gjlay.de>
2399 * config/avr/avr-mcus.def (AVR_MCU): Set .data_section_start of
2400 atmega164a to 0x100.
2402 2011-10-09 Eric Botcazou <ebotcazou@adacore.com>
2404 * tree-ssa-forwprop.c (forward_propagate_into_comparison): Do not make
2405 the replacement if the conversion to the LHS type is not useless.
2407 2011-10-09 Ira Rosen <ira.rosen@linaro.org>
2409 PR tree-optimization/50635
2410 * tree-vect-patterns.c (vect_handle_widen_mult_by_const): Add
2411 DEF_STMT to the list of statements to be replaced by the
2413 (vect_handle_widen_mult_by_const): Don't check TYPE_OUT.
2415 2011-10-09 Anatoly Sokolov <aesok@post.ru>
2417 * system.h: Commit forgotten hunk in previous patch.
2418 (OUTPUT_ADDR_CONST_EXTRA): Poison.
2420 2011-10-08 Nicola Pero <nicola.pero@meta-innovation.com>
2423 * doc/objc.texi (Garbage Collection): Updated example to protect
2424 +initialize against execution in subclasses.
2426 2011-10-07 Richard Henderson <rth@redhat.com>
2428 * doc/extend.texi (__builtin_shuffle): Improve the description to
2429 include the modulus of the selector. Mention OpenCL.
2430 * doc/md.texi (vec_perm, vec_perm_const): Document named patterns.
2432 * tree.def (VEC_PERM_EXPR): Rename from VEC_SHUFFLE_EXPR.
2433 * genopinit.c (optabs): Rename vshuffle to vec_perm.
2434 * c-typeck.c (c_build_vec_perm_expr): Rename from
2435 c_build_vec_shuffle_expr. Update for name changes.
2436 * optabs.c (expand_vec_perm_expr_p): Rename from
2437 expand_vec_shuffle_expr_p.
2438 (expand_vec_perm_expr): Rename from expand_vec_shuffle_expr.
2439 * optabs.h (OTI_vec_perm): Rename from DOI_vshuffle.
2440 (vec_perm_optab): Rename from vshuffle_optab.
2441 * expr.c, gimple-pretty-print.c, gimple.c, gimplify.c,
2442 c-tree.h, c-parser.c, tree-cfg.c, tree-inline.c, tree-pretty-print.c,
2443 tree-ssa-operands.c, tree-vect-generic.c: Update for name changes.
2445 * config/i386/i386.c (ix86_expand_vec_perm): Rename from
2446 ix86_expand_vshuffle.
2447 * config/i386/i386-protos.h: Update.
2448 * config/i386/sse.md (VEC_PERM_AVX2): Rename from VSHUFFLE_AVX2.
2449 (vec_perm<VEC_PERM_AVX2>): Rename from vshuffle<VSHUFFLE_AVX2>.
2451 2011-10-07 Richard Henderson <rth@redhat.com>
2453 * config/i386/predicates.md (avx2_pblendw_operand): New.
2454 * config/i386/sse.md (sse4_1_pblendw): Un-macroize.
2455 (avx2_pblendw, *avx2_pblendw): New expander and insn.
2457 2011-10-07 Richard Henderson <rth@redhat.com>
2459 * config/i386/i386.c (bdesc_args): Update code for
2460 __builtin_ia32_palignr256. Change type of __builtin_ia32_pslldqi256,
2461 and __builtin_ia32_psrldqi256 to V4DI_FTYPE_V4DI_INT_CONVERT.
2462 (ix86_expand_args_builtin): Handle V4DI_FTYPE_V4DI_INT_CONVERT.
2463 * config/i386/sse.md (mode iterator V16): Add V2TI.
2464 (mode iterator SSESCALARMODE): Use V2TI not V4DI.
2465 (mode attr ssse3_avx2): Add V2TI.
2466 (avx2_lshrqv4di3, avx2_lshlqv4di3): Remove.
2468 2011-10-07 David S. Miller <davem@davemloft.net>
2471 * configure.ac: Add .register directives to VIS3 test.
2472 * configure: Regenerate.
2474 2011-10-07 Richard Henderson <rth@redhat.com>
2476 * config.gcc (x86_64-*): Add core-avx-i, core-avx2 for with_cpu.
2478 2011-10-07 Richard Henderson <rth@redhat.com>
2481 * fold-const.c (fold_checksum_tree): Remove out-of-date assert.
2483 2011-10-07 Andrew Stubbs <ams@codesourcery.com>
2485 * config/arm/predicates.md (shift_amount_operand): Remove constant
2487 (shift_operator): Check range of constants for all shift operators.
2489 2011-10-07 Andreas Krebbel <Andreas.Krebbel@de.ibm.com>
2491 * config/s390/s390.c (s390_emit_tls_call_insn): Remove assertion.
2492 Load GOT pointer for non-pic builds.
2493 (s390_load_got): Replace pic_offset_table_rtx with hardcoded r12.
2494 (s390_emit_call): Likewise.
2496 2011-10-07 Tom de Vries <tom@codesourcery.com>
2499 * tree.c (build_common_builtin_nodes): Add local_define_builtin for
2500 BUILT_IN_ALLOCA_WITH_ALIGN. Mark that BUILT_IN_ALLOCA_WITH_ALIGN can
2502 * builtins.c (expand_builtin_alloca): Handle BUILT_IN_ALLOCA_WITH_ALIGN
2503 arglist. Set align for BUILT_IN_ALLOCA_WITH_ALIGN.
2504 (expand_builtin): Handle BUILT_IN_ALLOCA_WITH_ALIGN.
2505 (is_inexpensive_builtin): Handle BUILT_IN_ALLOCA_WITH_ALIGN.
2506 * tree-ssa-ccp.c (evaluate_stmt): Set align for
2507 BUILT_IN_ALLOCA_WITH_ALIGN.
2508 (fold_builtin_alloca_for_var): Rename to ...
2509 (fold_builtin_alloca_with_align): Set DECL_ALIGN from 2nd
2510 BUILT_IN_ALLOCA_WITH_ALIGN argument.
2511 (ccp_fold_stmt): Try folding BUILT_IN_ALLOCA_WITH_ALIGN using
2512 fold_builtin_alloca_with_align.
2513 (optimize_stack_restore): Handle BUILT_IN_ALLOCA_WITH_ALIGN.
2514 * builtins.def (BUILT_IN_ALLOCA_WITH_ALIGN): Declare using
2516 * ipa-pure-const.c (special_builtin_state): Handle
2517 BUILT_IN_ALLOCA_WITH_ALIGN.
2518 * tree-ssa-alias.c (ref_maybe_used_by_call_p_1)
2519 (call_may_clobber_ref_p_1): Same.
2520 * function.c (gimplify_parameters): Lower vla to
2521 BUILT_IN_ALLOCA_WITH_ALIGN.
2522 * gimplify.c (gimplify_vla_decl): Same.
2523 * cfgexpand.c (expand_call_stmt): Handle BUILT_IN_ALLOCA_WITH_ALIGN.
2524 * tree-mudflap.c (mf_xform_statements): Same.
2525 * tree-ssa-dce.c (mark_stmt_if_obviously_necessary)
2526 (mark_all_reaching_defs_necessary_1, propagate_necessity): Same.
2527 * varasm.c (incorporeal_function_p): Same.
2528 * tree-object-size.c (alloc_object_size): Same.
2529 * gimple.c (gimple_build_call_from_tree): Same.
2531 2011-10-07 Bernd Schmidt <bernds@codesourcery.com>
2533 * function.c (frame_required_for_rtx): Remove function.
2534 (requires_stack_frame_p): New arg set_up_by_prologue. All callers
2535 changed. Compute a set of mentioned registers and compare against
2536 the new arg rather than calling frame_required_for_rtx.
2537 (thread_prologue_and_epilogue_insns): Compute the set_up_by_prologue
2538 reg set. Convert the unconverted_simple_returns mechanism to store
2539 jump insns rather than their basic blocks. Also check the
2540 orig_entry_edge destination for new blocks.
2542 2011-10-07 Jakub Jelinek <jakub@redhat.com>
2544 PR tree-optimization/50650
2545 * tree-vect-patterns.c (vect_recog_mixed_size_cond_pattern): Don't
2546 call vect_is_simple_cond here, instead fail if cond_expr isn't
2547 COMPARISON_CLASS_P or if get_vectype_for_scalar_type returns NULL
2548 for cond_expr's first operand.
2549 * tree-vect-stmts.c (vect_is_simple_cond): Static again.
2550 * tree-vectorizer.h (vect_is_simple_cond): Remove prototype.
2552 2011-10-07 Andreas Krebbel <Andreas.Krebbel@de.ibm.com>
2554 * config/s390/s390.md (DWH, dwh): New mode macros.
2555 ("umulsidi3"): Extend to support "umulditi3" as well.
2557 2011-10-07 Uros Bizjak <ubizjak@gmail.com>
2558 H.J. Lu <hongjiu.lu@intel.com>
2561 * config/i386/i386.c (ix86_fixup_binary_operands): Force src2 of
2562 integer PLUS RTX to a register to improve address combine.
2564 2011-10-06 Richard Henderson <rth@redhat.com>
2566 * combine-stack-adjust.c (maybe_move_args_size_note): Add after
2567 parameter; use it to decide whether to merge two notes.
2568 (combine_stack_adjustments_for_block): Use maybe_move_args_size_note
2569 for the deallocation case as well.
2571 2011-10-06 Anatoly Sokolov <aesok@post.ru>
2573 * system.h (OUTPUT_ADDR_CONST_EXTRA): Poison.
2574 * doc/tm.texi.in (OUTPUT_ADDR_CONST_EXTRA): Remove documentation.
2575 * doc/tm.texi: Regenerate.
2576 * target.def (output_addr_const_extra): Use
2577 hook_bool_FILEptr_rtx_false.
2578 * targhooks.c (default_asm_output_addr_const_extra): Remove.
2579 * targhooks.h (default_asm_output_addr_const_extra): Remove.
2580 * hooks.c (hook_bool_FILEptr_rtx_false): New functions.
2581 * hooks.h (hook_bool_FILEptr_rtx_false): Declare.
2583 2011-10-06 David S. Miller <davem@davemloft.net>
2585 * config/sparc/sparc.md (popcount<mode>2, clz<mode>2): Split up into...
2586 (popcountdi2, popcountsi2, clzdi2, clzsi2): Explicit expanders, in the
2587 SI mode 64-bit code gen case explicitly zero-extend and truncate.
2588 (*popcount<mode>_sp64): Split up into...
2589 (*popcountdi_sp64, *popcountsi_64): Explicit instantiations, and in the
2590 SImode case use truncate.
2591 (*clzsi_sp64): Rewrite to use truncate, and let the expander emit the
2592 subtract so the compiler can optimize it.
2593 (SIDI): Remove unused mode iterator.
2595 2011-10-06 Bernd Schmidt <bernds@codesourcery.com>
2597 * function.c (thread_prologue_and_epilogue_insns): Emit split prologue
2598 on the orig_entry_edge. Don't account for it in prologue_clobbered.
2600 2011-10-06 Jakub Jelinek <jakub@redhat.com>
2602 PR tree-optimization/50596
2603 * tree-vectorizer.h (vect_is_simple_cond): New prototype.
2604 (NUM_PATTERNS): Change to 6.
2605 * tree-vect-patterns.c (vect_recog_mixed_size_cond_pattern): New
2607 (vect_vect_recog_func_ptrs): Add vect_recog_mixed_size_cond_pattern.
2608 (vect_mark_pattern_stmts): Don't create stmt_vinfo for def_stmt
2609 if it already has one, and don't set STMT_VINFO_VECTYPE in it
2610 if it is already set.
2611 * tree-vect-stmts.c (vect_mark_stmts_to_be_vectorized): Handle
2612 COND_EXPR in pattern stmts.
2613 (vect_is_simple_cond): No longer static.
2615 2001-10-06 Richard Henderson <rth@redhat.com>
2617 * config/i386/i386.c (ix86_expand_vshuffle): Add AVX2 support.
2618 * config/i386/sse.md (sseshuffint): Remove.
2619 (sseintvecmode): Support V16HI, V8HI, V32QI, V16QI.
2620 (VSHUFFLE_AVX2): New mode iterator.
2621 (vshuffle<mode>): Use it.
2622 (avx_vec_concat<V_256>): Rename from *vec_concat<V_256>_avx.
2624 * config/i386/i386.c (ix86_expand_sse_movcc): Use correct mode
2625 for vector_all_ones_operand.
2626 (ix86_expand_int_vcond): Distinguish between comparison mode
2627 and data mode. Allow them to differ.
2628 (ix86_expand_vshuffle): Don't force data mode to match maskmode.
2630 2001-10-06 Richard Henderson <rth@redhat.com>
2632 * optabs.c (expand_vec_shuffle_expr): Use the proper mode for the
2633 mask operand. Tidy the code.
2635 2011-10-06 Jakub Jelinek <jakub@redhat.com>
2637 * tree-vect-patterns.c (vect_pattern_recog_1): Use
2638 vect_recog_func_ptr typedef for the first argument.
2639 (vect_pattern_recog): Rename vect_recog_func_ptr variable
2640 to vect_recog_func, use vect_recog_func_ptr typedef for it.
2642 PR tree-optimization/49279
2643 * tree-ssa-structalias.c (find_func_aliases): Don't handle
2645 * tree-ssa-forwprop.c (forward_propagate_addr_expr_1): Allow
2646 restrict propagation.
2647 * tree-ssa.c (useless_type_conversion_p): Don't return false
2648 if TYPE_RESTRICT differs.
2650 2011-10-06 Bernd Schmidt <bernds@codesourcery.com>
2652 * function.c (thread_prologue_and_epilogue_insns): Build a vector
2653 of unconverted simple return blocks rather than trying to
2654 recompute them later based on bb_flags bitmap tests.
2656 2011-10-06 Michael Matz <matz@suse.de>
2658 * tree-flow.h (get_var_ann): Don't declare.
2659 * tree-flow-inline.h (get_var_ann): Remove.
2660 (set_is_used): Use var_ann, not get_var_ann.
2661 * tree-dfa.c (add_referenced_var): Inline body of get_var_ann.
2662 * tree-profile.c (gimple_gen_edge_profiler): Call
2663 find_referenced_var_in.
2664 (gimple_gen_interval_profiler): Ditto.
2665 (gimple_gen_pow2_profiler): Ditto.
2666 (gimple_gen_one_value_profiler): Ditto.
2667 (gimple_gen_average_profiler): Ditto.
2668 (gimple_gen_ior_profiler): Ditto.
2669 (gimple_gen_ic_profiler): Ditto plus call add_referenced_var.
2670 (gimple_gen_ic_func_profiler): Call add_referenced_var.
2671 * tree-mudflap.c (execute_mudflap_function_ops): Call
2674 2011-10-06 Jakub Jelinek <jakub@redhat.com>
2676 * tree-vect-patterns.c (vect_handle_widen_mult_by_const): For lhs
2677 don't set SSA_NAME_DEF_STMT that has been already set by
2678 gimple_build_assign_with_ops.
2679 (vect_recog_pow_pattern, vect_recog_widen_sum_pattern,
2680 vect_operation_fits_smaller_type, vect_recog_over_widening_pattern):
2683 * tree.h (avoid_folding_inline_builtin): New prototype.
2684 * builtins.c (avoid_folding_inline_builtin): No longer static.
2685 * gimple-fold.c (gimple_fold_builtin): Give up if
2686 avoid_folding_inline_builtin returns true.
2688 2011-10-06 Richard Guenther <rguenther@suse.de>
2690 * tree-vect-generic.c (vector_element): Look at previous
2693 2011-10-06 David Edelsohn <dje.gcc@gmail.com>
2696 * config/rs6000/aix.h (TARGET_OS_AIX_CPP_BUILTINS): Define
2697 __powerpc__, __PPC__, __unix__.
2699 2011-10-06 Michael Matz <matz@suse.de>
2701 * i386/i386.opt (recip_mask, recip_mask_explicit,
2702 x_recip_mask_explicit): New variables and cl_target member.
2703 (mrecip=): New option.
2704 * i386/i386.h (RECIP_MASK_DIV, RECIP_MASK_SQRT, RECIP_MASK_VEC_DIV,
2705 RECIP_MASK_VEC_SQRT, RECIP_MASK_ALL, RECIP_MASK_NONE): New bitmasks.
2706 (TARGET_RECIP_DIV, TARGET_RECIP_SQRT, TARGET_RECIP_VEC_DIV,
2707 TARGET_RECIP_VEC_SQRT): New tests.
2708 * i386/i386.md (divsf3): Check TARGET_RECIP_DIV.
2709 (sqrt<mode>2): Check TARGET_RECIP_SQRT.
2710 * i386/sse.md (div<mode>3): Check TARGET_RECIP_VEC_DIV.
2711 (sqrt<mode>2): Check TARGET_RECIP_VEC_SQRT.
2712 * i386/i386.c (ix86_option_override_internal): Set recip_mask
2713 for -mrecip and -mrecip=options.
2714 (ix86_function_specific_save): Save recip_mask_explicit.
2715 (ix86_function_specific_restore): Restore recip_mask_explicit.
2717 * doc/invoke.texi (ix86 Options): Document the new option.
2719 2011-10-06 Bernd Schmidt <bernds@codesourcery.com>
2722 * config/arm/arm.md (arm_subsi3_insn): Lose the last alternative.
2724 2011-10-06 Ulrich Weigand <ulrich.weigand@linaro.org>
2727 * config/arm/arm.c (arm_legitimize_reload_address): Recognize
2728 output of a previous pass through legitimize_reload_address.
2729 Do not attempt to optimize addresses if the base register is
2730 equivalent to a constant.
2732 2011-10-06 Andreas Krebbel <Andreas.Krebbel@de.ibm.com>
2734 * function.c (thread_prologue_and_epilogue_insns): Mark
2735 last_bb_active as possibly unused. It is unused for targets which
2736 do neither have "return" nor "simple_return" expanders.
2738 2011-10-06 Richard Guenther <rguenther@suse.de>
2740 * fold-const.c (fold_ternary_loc): Also fold non-constant
2741 vector CONSTRUCTORs. Make more efficient.
2742 * tree-ssa-dom.c (cprop_operand): Don't handle virtual operands.
2743 (cprop_into_stmt): Don't propagate into virtual operands.
2744 (optimize_stmt): Really dump original statement.
2746 2011-10-06 Nick Clifton <nickc@redhat.com>
2748 * config/rx/rx.md (smin3): Revert previous delta.
2750 2011-10-06 Richard Guenther <rguenther@suse.de>
2752 PR tree-optimization/38884
2753 * tree-ssa-sccvn.c (vn_reference_lookup_3): Handle partial
2754 reads from aggregate SSA names.
2756 2011-10-05 Jakub Jelinek <jakub@redhat.com>
2758 * tree-vect-patterns.c (vect_pattern_recog_1): Add stmts_to_replace
2759 argument, truncate it at the beginning instead of allocating there
2760 and freeing at the end.
2761 (vect_pattern_recog): Allocate stmts_to_replace here and free at end,
2762 pass its address to vect_pattern_recog_1.
2764 2011-10-05 David S. Miller <davem@davemloft.net>
2766 * config/sparc/sparc.opt (POPC): New option.
2767 * doc/invoke.texi: Document it.
2768 * config/sparc/sparc.c (sparc_option_override): Enable MASK_POPC by
2769 default on Niagara-2 and later.
2770 * config/sparc/sparc.h (CLZ_DEFINED_VALUE_AT_ZERO): Define.
2771 * config/sparc/sparc.md (SIDI): New mode iterator.
2772 (ffsdi2): Delete commented out pattern and comments.
2773 (popcount<mode>2, clz<mode>2): New expanders.
2774 (*popcount<mode>_sp64, popcountsi_v8plus, popcountdi_v8plus,
2775 *clzdi_sp64, clzdi_v8plus, *clzsi_sp64, clzsi_v8plus): New insns.
2777 2011-10-06 Artjoms Sinkarovs <artyom.shinkaroff@gmail.com>
2780 * c-tree.h (c_expr_t): New typedef for struct c_expr.
2781 (C_EXPR_APPEND): New macro.
2782 * c-parser.c (c_parser_get_builtin_args): Preserve
2783 original_tree_code of c_expr structure.
2784 (c_parser_postfix_expression): Adjust to the new function.
2786 2011-10-05 Bernd Schmidt <bernds@codesourcery.com>
2788 * function.c (thread_prologue_and_epilogue_insns): Don't shrink-wrap
2789 if profiling after the prologue.
2791 2011-10-05 Jakub Jelinek <jakub@redhat.com>
2793 PR tree-optimization/50613
2794 * tree-ssa-strlen.c (find_equal_ptrs): If CASE_CONVERT
2795 operand is ADDR_EXPR, fallthru into ADDR_EXPR handling,
2796 and if it is neither that not SSA_NAME, give up.
2798 2011-10-05 Richard Henderson <rth@redhat.com>
2800 * tree-vect-generic.c (vector_element): Never fail. Use
2801 build_zero_cst. Tidy up type references.
2802 (lower_vec_shuffle): Never fail. Mask shuffle indicies. Reduce
2803 code duplication. Do update_stmt here ...
2804 (expand_vector_operations_1): ... not here.
2806 * config/i386/i386.c (ix86_expand_vshuffle): Never fail. Handle
2807 TARGET_XOP. Fix pshufb constant vector creation. Reduce code
2808 duplication. Handle V2DI without SSE4.1.
2809 * config/i386/i386-protos.h (ix86_expand_vshuffle): Update decl.
2810 * config/i386/i386.md (vshuffle<V_128>): Remove assert for ok.
2812 2011-10-05 Uros Bizjak <ubizjak@gmail.com>
2814 * config/i386/i386.c (distance_non_agu_define): Simplify calculation
2815 of "found". Simplify return value calculation.
2816 (distance_agu_use): Ditto.
2818 2011-10-05 Bernd Schmidt <bernds@codesourcery.com>
2821 * config/i386/i386.c (ix86_add_cfa_restore_note): Omit notes only
2822 if the function was not shrink-wrapped.
2823 (ix86_expand_epilogue): Ensure queued cfa_adjust notes are attached
2825 * function.c (thread_prologue_and_epilogue_insns): Make sure the
2826 shrink_wrapped flag is set even if there is no dump file.
2828 2011-10-05 DJ Delorie <dj@redhat.com>
2829 Nick Clifton <nickc@redhat.com>
2831 * config/rx/rx.opt (mpid): Define.
2832 * config/rx/t-rx (MULTILIB_OPTIONS): Add -mpid
2833 (MULTILIB_DIRNAMES): Add pid.
2834 * config/rx/rx.c (rx_gp_base_regnum_val, rx_pid_base_regnum_val)
2835 (rx_num_interrupt_regs): New variable.
2836 (rx_gp_base_regnum): New function. Returns the number of the
2837 small data area register.
2838 (rx_pid_base_regnum): New function. Returns the number of the pid
2840 (rx_decl_for_addr): New function. Returns the symbolic part of a MEM.
2841 (rx_pid_data_operand): New function. Returns whether an object is
2842 in the position independent data area.
2843 (rx_legitimize_address): New function. Puts undecided PID
2844 objects in the PID data area.
2845 (rx_is_legitimate_address): Add support for PID operands.
2846 (rx_print_operand_address): Likewise.
2847 (rx_print_operand): Likewise.
2848 (rx_maybe_pidify_operand): New function. Determine if an operand
2849 is suitable for PID addressing.
2850 (rx_gen_move_template): Add PID support.
2851 (rx_conditional_register_usage): Likewise.
2852 (rx_option_override): Initialise rx_num_interrupt_regs.
2853 (rx_is_legitimate_constant): Add support for PID constants.
2854 (TARGET_LEGITIMIZE_ADDRESS): Define.
2855 * config/rx/constraints.md (Rpid): Define.
2857 * config/rx/rx.md (UNSPEC_PID_ADDR): Define.
2858 (tablejump): Add PID support.
2860 (mov<>_internal): Likewise.
2861 (addsi3): Convert to an expander. Add PID support.
2862 (pid_addr): New pattern.
2863 * config/rx/rx.h (CPP_SPEC): Define.
2864 (ASM_SPEC): Pass -mpid and -mint-register on to assembler.
2865 (CASE_VECTOR_PC_RELATIVE): Define.
2866 (JUMP_TABLES_IN_TEXT_SECTION): Enable for PID mode.
2867 * config/rx/rx-protos.h (rx_maybe_pidify_operand): Prototype.
2868 * doc/invoke.texi (RX Options): Document -mpid command line option.
2870 2011-10-05 Richard Guenther <rguenther@suse.de>
2872 PR tree-optimization/38885
2873 * tree-ssa-sccvn.c (vn_reference_lookup_3): Handle partial reads
2876 2011-10-05 Bernd Schmidt <bernds@codesourcery.com>
2878 * doc/invoke.texi (-fshrink-wrap): Document.
2879 * opts.c (default_options_table): Add it.
2880 * common.opt (fshrink-wrap): Add.
2881 * function.c (emit_return_into_block): Remove useless declaration.
2882 (record_hard_reg_uses_1, record_hard_reg_uses, frame_required_for_rtx,
2883 requires_stack_frame_p, gen_return_pattern): New static functions.
2884 (emit_return_into_block): New arg simple_p. All callers changed.
2885 Use gen_return_pattern.
2886 (thread_prologue_and_epilogue_insns): Implement shrink-wrapping.
2887 * config/i386/i386.md (return): Expand into a simple_return.
2888 (simple_return): New expander):
2889 (simple_return_internal, simple_return_internal_long,
2890 simple_return_pop_internal_long, simple_return_indirect_internal):
2891 Renamed from return_internal, return_internal_long,
2892 return_pop_internal_long and return_indirect_internal; changed to use
2894 * config/i386/i386.c (ix86_expand_epilogue): Adjust to expand
2896 (ix86_pad_returns): Likewise.
2897 * function.h (struct rtl_data): Add member shrink_wrapped.
2898 * cfgcleanup.c (outgoing_edges_match): If shrink-wrapped, edges that
2899 are not jumps or sibcalls can't be compared.
2901 2011-10-05 Richard Guenther <rguenther@suse.de>
2903 * tree-ssa-sccvn.c (vn_get_expr_for): Handle CONSTRUCTOR of
2905 (simplify_unary_expression): Handle BIT_FIELD_REFs.
2906 (try_to_simplify): Handle BIT_FIELD_REFs.
2908 2011-10-05 Georg-Johann Lay <avr@gjlay.de>
2910 * config/avr/avr-protos.h (avr_out_addto_sp): New prototype.
2911 * config/avr/avr.c (avr_out_addto_sp): New function.
2912 (adjust_insn_length): Handle ADJUST_LEN_ADDTO_SP.
2913 * config/avr/avr.md (adjust_len): Add "addto_sp".
2914 (*movhi_sp): Remove insn.
2915 (*addhi3_sp_R_pc2, *addhi3_sp_R_pc3): Merge to *addhi3_sp_R.
2917 2011-10-05 Richard Guenther <rguenther@suse.de>
2919 * gimple-fold.c (gimple_fold_stmt_to_constant_1): For ternary ops
2920 with an embedded expression valueize and fold that as well.
2921 * tree-ssa-sccvn.c (try_to_simplify): Also allow SSA name
2922 results from gimple_fold_stmt_to_constant_1.
2924 2011-10-05 Nick Clifton <nickc@redhat.com>
2926 * config/rx/rx.md (tablejump): Add missing label.
2927 (adddi3_internal): Mark operand 0 as early-clobbered.
2928 (smaxsi3): Revert previous delta.
2929 (adc_internal): Fix whitespace in generated asm.
2930 (adc_flags): Likewise.
2932 2011-10-05 Andreas Krebbel <Andreas.Krebbel@de.ibm.com>
2934 * expmed.c (expand_mult_highpart_optab): Replace optab_handler
2935 with the new widening_optab_handler.
2937 2011-10-05 Richard Guenther <rguenther@suse.de>
2939 PR tree-optimization/50609
2940 * gimple-fold.c (fold_array_ctor_reference): Also handle
2941 vector typed constructors.
2942 (fold_ctor_reference): Dispatch to fold_array_ctor_reference
2943 for vector typed constructors.
2945 2011-10-05 Uros Bizjak <ubizjak@gmail.com>
2947 * config/i386/i386.c (ix86_emit_binop): New static function.
2948 (ix86_split_lea_for_addr): Use ix86_emit_binop to emit add and shl
2950 (x86_output_mi_thunk): Use ix86_emit_binop to emit add instructions.
2952 2011-10-04 David S. Miller <davem@davemloft.net>
2954 * config/sparc/sparc.md (UNSPEC_FHADD, UNSPEC_FHSUB,
2955 UNSPEC_XMUL): New unspecs.
2956 (muldi3_v8plus): Use output_v8plus_mult.
2957 (*naddsf3, *nadddf3, *nmulsf3, *nmuldf3, *nmuldf3_extend):
2958 New VIS 3.0 combiner patterns.
2959 (fhaddsf_vis, fhadddf_vis, fhsubsf_vis, fhsubdf_vis,
2960 fnhaddsf_vis, fnhaddf_vis, umulxhi_vis, *umulxhi_sp64,
2961 umulxhi_v8plus, xmulx_vis, *xmulx_sp64, xmulx_v8plus,
2962 xmulxhi_vis, *xmulxhi_sp64, xmulxhi_v8plus): New VIS 3.0
2964 * config/sparc/sparc.c (sparc_vis_init_builtins): Emit new builtins.
2965 (output_v8plus_mult): New function.
2966 * config/sparc/sparc-protos.h: Declare it.
2967 * config/sparc/visintrin.h (__vis_fhadds, __vis_fhaddd,
2968 __vis_fhsubs, __vis_fhsubd, __vis_fnhadds, __vis_fnhaddd,
2969 __vis_umulxhi, __vis_xmulx, __vis_xmulxhi): New intrinsics.
2970 * doc/extend.texi: Document new builtins.
2972 2011-10-04 Richard Henderson <rth@redhat.com>
2974 * c-typeck.c (c_build_vec_shuffle_expr): Fix uninitialized variable.
2975 Avoid save_expr unless two_arguments.
2977 2011-10-04 Ozkan Sezer <sezeroz@gmail.com>
2979 * config/i386/mingw-w64.h (CPP_SPEC): Rename _REENTRANCE to _REENTRANT.
2980 * config/i386/mingw32.h (CPP_SPEC): Likewise.
2982 2011-10-04 David S. Miller <davem@davemloft.net>
2984 * config/sparc/driver-sparc.c (cpu_names): Fix string for supersparc
2987 2011-10-04 Jakub Jelinek <jakub@redhat.com>
2989 PR tree-optimization/50604
2990 * builtins.c (fold_builtin_strcpy, fold_builtin_stpcpy,
2991 fold_builtin_strncpy, fold_builtin_stxcpy_chk): Ensure
2992 last argument to memcpy has size_type_node type instead of ssizetype.
2993 * tree-ssa-strlen.c (handle_builtin_memcpy): Use size_type_node
2994 instead of TREE_TYPE (len) as type for newlen.
2996 PR tree-optimization/50522
2997 * tree-ssa-alias.c (ptr_deref_may_alias_decl_p): Don't test
2999 (ptr_derefs_may_alias_p): Call pt_solutions_same_restrict_base
3002 * fold-const.c (fold_unary_loc): Don't optimize
3003 POINTER_PLUS_EXPR casted to TYPE_RESTRICT pointer by
3004 casting the inner pointer if it isn't TYPE_RESTRICT.
3005 * tree-ssa-forwprop.c (forward_propagate_addr_expr_1): Don't through
3006 casts from non-TYPE_RESTRICT pointer to TYPE_RESTRICT pointer.
3008 2011-10-04 Joseph Myers <joseph@codesourcery.com>
3010 * config.gcc (i[34567]86-*-elf*, x86_64-*-elf*): Use i386/t-crtstuff.
3012 2011-10-04 Jan Hubicka <jh@suse.cz>
3014 * lto-streamer.h (lto_input_toplevel_asms): Add order_base parameter.
3015 * lto-streamer-in.c (lto_input_toplevel_asms): Stream in order.
3016 * lto-streamer-out.c (lto_output_toplevel_asms): Stream out order.
3017 * lto-cgraph.c (order_base): New static var.
3018 (lto_output_node): Stream in order.
3019 (lto_output_varpool_node): Stream out order.
3020 (input_node): Stream in order.
3021 (input_varpool_node): Stream out order.
3022 (input_cgraph_1): Initialize order base; update call of
3023 lto_input_toplevel_asms.
3025 2011-10-04 Georg-Johann Lay <avr@gjlay.de>
3028 * config/avr/avr-protos.h (avr_legitimize_reload_address): New
3030 * config/avr/avr.h (LEGITIMIZE_RELOAD_ADDRESS): Copy worker code
3032 * config/avr/avr.c (avr_legitimize_reload_address) ...to this new
3033 function. Log if avr_log.legitimize_reload_address.
3035 2011-10-04 Eric Botcazou <ebotcazou@adacore.com>
3037 * config/sparc/sparc.c (sparc_fold_builtin): Use a sequence of tests.
3039 2011-10-04 Andreas Krebbel <Andreas.Krebbel@de.ibm.com>
3041 * longlong.h (smul_ppmm, sdiv_qrnnd): Add underscores to the local
3042 variables. Fix second operand of DR. Swap inputs for sdiv_qrnnd.
3044 2011-10-03 David S. Miller <davem@davemloft.net>
3046 * config/sparc/sparc.md (fpack16_vis, fpackfix_vis, fpack32_vis): Make
3047 GSR_REG an input operand to UNSPEC instead of a parallel USE.
3048 (faligndata<V64I:mode>_vis): Likewise and use DI mode.
3049 (alignaddrsi_vis, alignaddrdi_vis, alignaddrlsi_vis, alignaddrldi_vis):
3050 Reference GSR_REG in DI mode, simplify convoluted expressions by using
3052 (bshuffle<V64I:mode>_vis): Reference GSR_REG in DI mode.
3054 2011-10-03 Maxim Kuvyrkov <maxim@codesourcery.com>
3056 * tree-eh.c (remove_unreachable_handlers): Obvious cleanup.
3058 2011-10-03 Jakub Jelinek <jakub@redhat.com>
3059 Ian Lance Taylor <iant@google.com>
3061 * godump.c (go_output_typedef): Support printing enum values that
3062 don't fit in a signed HOST_WIDE_INT.
3064 2011-10-03 Anatoly Sokolov <aesok@post.ru>
3066 * config/cris/cris.c (cris_output_addr_const_extra): Make static.
3067 (TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA): Define.
3068 * config/cris/cris.h (OUTPUT_ADDR_CONST_EXTRA): Remove.
3069 * config/cris/cris-protos.h (cris_output_addr_const_extra): Remove.
3071 2011-10-03 Anatoly Sokolov <aesok@post.ru>
3073 * config/m68k/m68k.c (m68k_output_addr_const_extra): Make static.
3074 (TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA): Define.
3075 * config/m68k/m68k.h (OUTPUT_ADDR_CONST_EXTRA): Remove.
3076 * config/m68k/m68k-protos.h (m68k_output_addr_const_extra): Remove.
3078 2011-10-03 Steve Ellcey <sje@cup.hp.com>
3081 * configure.ac (gcc_cv_ld_static_dynamic): Define for *-*-hpux*.
3082 (gcc_cv_ld_static_option): Ditto.
3083 (gcc_cv_ld_dynamic_option): Ditto.
3084 * configure: Regenerate.
3086 2011-10-03 David S. Miller <davem@davemloft.net>
3088 * config/sparc/sparc.md (bmask<P:mode>_vis): Split into explicit 'di'
3089 and 'si' patterns which describe the GSR changes explicitly in the
3090 RTL using zero_extract.
3091 (bshuffle<V64I:mode>_vis): Put the GSR use inside of the unspec.
3093 * config/sparc/sparc.md (bshuffle<V64I:mode>_vis): Don't wrap
3094 GSR_REG in a USE, since it's now a true arg to the UNSPEC.
3096 2011-10-03 Artjoms Sinkarovs <artyom.shinkaroff@gmail.com>
3098 * optabs.c (expand_vec_shuffle_expr_p): New function. Checks
3099 if given expression can be expanded by the target.
3100 (expand_vec_shuffle_expr): New function. Expand VEC_SHUFFLE_EXPR
3101 using target vector instructions.
3102 * optabs.h: New optab vshuffle.
3103 (expand_vec_shuffle_expr_p): New prototype.
3104 (expand_vec_shuffle_expr): New prototype.
3105 (vshuffle_optab): New optab.
3106 * genopinit.c: Adjust to support vecshuffle.
3107 * c-tree.h (c_build_vec_shuffle_expr): New prototype.
3108 * expr.c (expand_expr_real_2): Adjust.
3109 * c-typeck.c: (c_build_vec_shuffle_expr): Build a VEC_SHUFFLE_EXPR
3110 recognizing the cases of two and three arguments.
3111 (convert_arguments) (build_binary_op)
3112 (scalar_to_vector) (build_array_ref): Spurious whitespace.
3113 * gimplify.c (gimplify_exp): Adjusted to support VEC_SHUFFLE_EXPR.
3114 * tree.def: New tree code VEC_SHUFFLE_EXPR.
3115 * tree-inline.c (estimate_operator_cost): Recognize VEC_SHUFFLE_EXPR.
3116 * tree-vect-generic.c (vector_element): New function. Returns an
3117 element of the vector at the given position.
3118 (lower_vec_shuffle): Checks if VEC_SHUFLLE_EXPR is supported
3119 by the backend or expand an expression piecewise.
3120 (expand_vector_operations_1): Adjusted.
3121 (gate_expand_vector_operations_noop): New gate function.
3122 * Makefile.in (tree-vect-generic.o): New include.
3123 * gimple.c (get_gimple_rhs_num_ops): Adjust.
3124 * tree-cfg.c (verify_gimple_assign_trenary): Verify VEC_SHUFFLE_EXPR.
3125 * passes.c: Move veclower down.
3126 * tree-pretty-print.c (dump_generic_node): Recognize
3127 VEC_SHUFFLE_EXPR as valid expression.
3128 * c-parser.c (c_parser_get_builtin_args): Helper function for the
3129 builtins with variable number of arguments.
3130 (c_parser_postfix_expression): Use a new helper function for
3131 RID_CHOOSE_EXPR, RID_BUILTIN_COMPLEX and RID_BUILTIN_SHUFFLE.
3132 * tree-ssa-operands: Adjust.
3133 * c-family/c-common.c: New __builtin_shuffle keyword.
3134 * c-family/c-common.h: New __builtin_shuffle keyword.
3135 * gcc/doc/extend.texi: Adjust.
3137 * gcc/config/i386/sse.md: (sseshuffint) New mode_attr. Correspondence
3138 between the vector and the type of the mask when shuffling.
3139 (vecshuffle<mode>): New expansion.
3140 * gcc/config/i386/i386-protos.h (ix86_expand_vshuffle): New prototype.
3141 * gcc/config/i386/i386.c (ix86_expand_vshuffle): New function.
3142 (ix86_vectorize_builtin_vec_perm_ok): Adjust.
3144 2011-10-03 Jakub Jelinek <jakub@redhat.com>
3146 PR tree-optimization/50587
3147 * tree-ssa-reassoc.c (init_range_entry): Stop iterating when
3148 arg0 is not a SSA_NAME.
3150 2011-10-03 Richard Sandiford <rdsandiford@googlemail.com>
3152 * ipa-inline-analysis.c (MAX_TIME): Update comment.
3154 2011-10-02 Richard Henderson <rth@redhat.com>
3155 David S. Miller <davem@davemloft.net>
3157 * config/sparc/sparc.md (ashlsi3, *ashldi3_sp64): Remove
3158 conditional insn type setting, we always emit a shift.
3159 (*ashlsi3_extend, *lshrsi3_extend0): New patterns.
3160 (*lshrsi3_extend): Rename to *lshrsi3_extend1.
3161 * config/sparc/predicates.md (const_one_operand): Delete.
3163 2011-10-02 Gerald Pfeifer <gerald@pfeifer.com>
3165 * invoke.texi (SPARC Options): Refer to GNU/Linux.
3167 2011-10-02 Richard Sandiford <rdsandiford@googlemail.com>
3169 * config/mips/mips.c (mips_frame_barrier): New function.
3170 (mips_expand_prologue): Call it after allocating stack space.
3171 (mips_deallocate_stack): New function.
3172 (mips_expand_epilogue): Call mips_frame_barrier and
3173 mips_deallocate_stack.
3175 2011-10-02 Richard Sandiford <rdsandiford@googlemail.com>
3178 * config/mips/sync.md (sync_<optab>_12): Allow zero operands.
3179 (sync_old_<optab>_12, sync_new_<optab>_12, sync_nand_12): Likewise.
3180 (sync_old_nand_12, sync_new_nand_12, test_and_set_12): Likewise.
3182 2011-10-02 Jan Hubicka <jh@suse.cz>
3184 * cgraphunit.c (verify_edge_count_and_frequency): Bounds check.
3186 * cgraphunit.c (ipa_passes): Remove unrechable nodes.
3187 * lto-streamer-out.c (produce_symtab): Skip unused extern declarations.
3188 * ipa.c (cgraph_remove_unreachable_nodes): Do not assume that external
3189 functions are reachable when address is taken.
3190 * tree-sra.c (modify_function): Free dominance info.
3192 2011-10-02 Jan Hubicka <jh@suse.cz>
3194 * ipa-inline-analysis.c (inline_summary_alloc): Bounds check.
3196 * ipa-inline-analysis.c (reset_inline_edge_summary): New function.
3197 (reset_inline_summary): New function.
3198 (compute_inline_parameters, inline_node_removal_hook,
3199 inline_edge_removal_hook): Use it.
3200 (inline_free_summary): Reset holders correctly.
3201 (inline_generate_summary): Free summary before computing it.
3203 2011-10-02 Paolo Carlini <paolo.carlini@oracle.com>
3205 PR preprocessor/36819
3206 * incpath.c (merge_include_chains): Call free_path on
3207 heads[QUOTE] and tails[QUOTE].
3209 2011-10-02 Jan Hubicka <jh@suse.cz>
3212 * lto-symtab.c (lto_symtab_resolve_symbols): Do not resolve
3213 when resolution is already availbale from plugin.
3214 (lto_symtab_merge_decls_1): Handle LDPR_PREVAILING_DEF_IRONLY_EXP.
3215 * cgraph.c (ld_plugin_symbol_resolution): Add
3216 prevailing_def_ironly_exp.
3217 * lto-cgraph.c (LDPR_NUM_KNOWN): Update.
3218 * ipa.c (varpool_externally_visible_p): IRONLY variables are never
3220 * varasm.c (resolution_to_local_definition_p): Add
3221 LDPR_PREVAILING_DEF_IRONLY_EXP.
3222 (resolution_local_p): Likewise.
3224 2011-10-01 David S. Miller <davem@davemloft.net>
3226 * config/sparc/sparc.opt (VIS3): New option.
3227 * doc/invoke.texi: Document it.
3228 * config/sparc/sparc.h: Force TARGET_VIS3 to zero if assembler is
3229 not capable of such instructions.
3230 * config/sparc/sparc-c.c (sparc_target_macros): Define __VIS__
3231 to 0x300 when TARGET_VIS3.
3232 * config/sparc/sparc-modes.def: Create 16-byte vector modes.
3233 * config/sparc/sparc.md (UNSPEC_CMASK8, UNSPEC_CMASK16, UNSPEC_CMASK32,
3234 UNSPEC_FCHKSM16, UNSPEC_PDISTN, UNSPC_FUCMP): New unspecs.
3235 (V64N8, VASS): New mode iterators.
3236 (vis3_shift, vis3_addsub_ss): New code iterators.
3237 (vbits, vconstr): New mode attributes.
3238 (vis3_shift_insn, vis3_addsub_ss_insn): New code attributes.
3239 (cmask8<P:mode>_vis, cmask16<P:mode>_vis, cmask32<P:mode>_vis,
3240 fchksm16_vis, <vis3_shift_insn><vbits>_vis, pdistn<mode>_vis,
3241 fmean16_vis, fpadd64_vis, fpsub64_vis,
3242 <vis3_addsub_ss_insn><vbits>_vis, fucmp<code>8<P:mode>_vis): New
3243 VIS 3.0 instruction patterns.
3244 * config/sparc/sparc.c (sparc_option_override): Set MASK_VIS3 by
3245 default when targetting capable cpus. TARGET_VIS3 implies
3246 TARGET_VIS2 and TARGET_VIS, and clear them when TARGET_FPU is disabled.
3247 (sparc_vis_init_builtins): Emit new VIS 3.0 builtins.
3248 (sparc_fold_builtin): Do not eliminate cmask{8,16,32} when result
3250 * config/sparc/visintrin.h (__vis_cmask8, __vis_cmask16,
3251 __vis_cmask32, __vis_fchksm16, __vis_fsll16, __vis_fslas16,
3252 __vis_fsrl16, __vis_fsra16, __vis_fsll32, __vis_fslas32,
3253 __vis_fsrl32, __vis_fsra32, __vis_pdistn, __vis_fmean16,
3254 __vis_fpadd64, __vis_fpsub64, __vis_fpadds16, __vis_fpadds16s,
3255 __vis_fpsubs16, __vis_fpsubs16s, __vis_fpadds32, __vis_fpadds32s,
3256 __vis_fpsubs32, __vis_fpsubs32s, __vis_fucmple8, __vis_fucmpne8,
3257 __vis_fucmpgt8, __vis_fucmpeq8): New VIS 3.0 interfaces.
3258 * doc/extend.texi: Document new VIS 3.0 builtins.
3260 2011-10-01 Eric Botcazou <ebotcazou@adacore.com>
3262 * ira-color.c (assign_hard_reg): Fix typo.
3264 2011-09-30 H.J. Lu <hongjiu.lu@intel.com>
3266 * doc/extend.texi: Add missing ','.
3268 2011-09-30 Bernd Schmidt <bernds@codesourcery.com>
3270 * common/config/c6x/c6x-common.c (c6x_option_optimization_table):
3271 Enable -fmodulo-sched at -O2 and above.
3272 * config/c6x/c6x.md (doloop_end): New expander.
3273 (mvilc, sploop, spkernel, loop_end): New patterns.
3274 (loop_end with memory destination splitter): New.
3275 * config/c6x/c6x.c: Include "hw-doloop.h".
3276 (enum unitreqs): New.
3277 (unit_req_table): New typedef.
3278 (unit_reqs): New static variable.
3279 (unit_req_factor, get_unit_reqs, count_unit_reqs, merge_unit_reqs,
3280 res_mii, split_delayed_nonbranch, undo_split_delayed_nonbranch,
3281 hwloop_pattern_reg, bb_earliest_end_cycle, filter_insns_above,
3282 hwloop_optimize, hwloop_fail, c6x_hwloops): New static functions.
3283 (struct c6x_sched_context): New member last_scheduled_iter0.
3284 (init_sched_state): Initialize it.
3285 (c6x_variable_issue): Update it.
3286 (sploop_max_uid_iter0): New static variable.
3287 (c6x_sched_reorder_1): Be careful about issuing sploop.
3288 (c6x_reorg): Call c6x_hwlooops before the final schedule.
3290 2011-09-30 Georg-Johann Lay <avr@gjlay.de>
3293 * config/avr/avr-protos.h (avr_log_t): New field address_cost.
3294 * config/avr/avr.c (avr_address_cost): Use it.
3295 * config/avr/avr-log.c (avr_log_set_avr_log): Initialize it.
3296 (avr_log_vadump): Unknown %-codes finish printing.
3298 2011-09-30 Jakub Jelinek <jakub@redhat.com>
3301 * gimple-fold.c (fold_stmt_1) <case GIMPLE_ASM>: If
3302 input constraints allow mem and not reg, pass true instead of
3303 false as second argument to maybe_fold_reference.
3305 PR tree-optimization/46309
3306 * fold-const.c (make_range, merge_ranges): Remove prototypes.
3307 (make_range_step): New function.
3308 (make_range): Use it.
3309 * tree.h (make_range_step): New prototypes.
3310 * Makefile.in (tree-ssa-reassoc.o): Depend on $(DIAGNOSTIC_CORE_H).
3311 * tree-ssa-reassoc.c: Include diagnostic-core.h.
3312 (struct range_entry): New type.
3313 (init_range_entry, range_entry_cmp, update_range_test,
3314 optimize_range_tests): New functions.
3315 (reassociate_bb): Call optimize_range_tests.
3317 2011-09-30 Jakub Jelinek <jakub@redhat.com>
3318 Richard Guenther <rguenther@suse.de>
3320 * tree-ssa-structalias.c (find_func_aliases_for_builtin_call): Handle
3321 BUILT_IN_STRDUP and BUILT_IN_STRNDUP.
3322 * tree-ssa-alias.c (call_may_clobber_ref_p_1): Likewise. Fix
3323 handling of BUILT_IN_STRNCAT and BUILT_IN_STRNCAT_CHK.
3324 (ref_maybe_used_by_call_p_1): Fix handling of BUILT_IN_STRCAT,
3325 BUILT_IN_STRNCAT, BUILT_IN_STRCAT_CHK and BUILT_IN_STRNCAT_CHK.
3327 2011-09-30 Jan Beulich <jbeulich@suse.com>
3329 * lto-cgraph.c (output_cgraph): Remove processing of
3330 'cgraph_asm_nodes', call lto_output_toplevel_asms() instead.
3331 (input_cgraph_1): Remove loop calling cgraph_add_asm_node(), call
3332 lto_input_toplevel_asms() instead.
3333 * lto-section-in.c (lto_section_name): Add "asm" entry.
3334 * lto-streamer-in.c (lto_input_toplevel_asms): New.
3335 * lto-streamer-out.c (lto_output_toplevel_asms): New.
3336 * lto-streamer.h (LTO_minor_version): Bump.
3337 (enum lto_section_type): Add LTO_section_asm.
3338 (struct lto_asm_header): New.
3339 (lto_input_toplevel_asms, lto_output_toplevel_asms): Declare.
3340 * tree-streamer.h (streamer_write_string_cst): Declare.
3341 * tree-streamer-out.c (write_string_cst): Rename to
3342 streamer_write_string_cst and make global. Handle incoming string
3344 (streamer_write_tree_header): Adjust call to renamed function.
3346 2011-09-30 Bernd Schmidt <bernds@codesourcery.com>
3348 * haifa-sched.c (modulo_ii, modulo_max_states, modulo_n_insns,
3349 modulo_insns_scheduled, modulo_iter0_max_uid, modulo_backtracks_left,
3350 modulo_last_stage): New static variables.
3351 (set_modulo_params, discard_delay_pairs_above): New functions.
3352 (struct delay_pair): New member stages.
3353 (htab_i2_traverse, htab_i1_traverse): New static functions.
3354 (record_delay_slot_pair): New arg stages. All callers changed.
3356 (pair_delay): Take stages into account.
3357 (add_delay_dependencies): Don't do so for stage pairs.
3358 (struct sched_block_state): New member modulo_epilogue.
3359 (save_backtrack_point): Don't set SHADOW_P for stage pairs.
3360 (unschedule_insns_until): Decrease modulo_insns_scheduled.
3361 Set HARD_DEP without using or.
3362 (resolve_dependencies): New static function.
3363 (prune_ready_list): New arg modulo_epilogue_p. All callers changed.
3364 If it is true, allow only insns with INSN_EXACT_TICK set.
3365 (schedule_block): Return bool, always true for normal scheduling,
3366 true or false depending on modulo scheduling success otherwise.
3367 Add bookkeeping for modulo scheduling, and call resolve_dependencies
3368 on everything left over after a modulo schedule.
3369 (haifa_sched_init): Remove check_cfg call. Clear modulo_ii.
3370 * sched-int.h (schedule_block, record_delay_slot_pair): Adjust
3372 (set_modulo_params, discard_delay_pairs_above): Declare.
3373 * params.def (PARAM_MAX_MODULO_BACKTRACK_ATTEMPS): New.
3374 * doc/invoke.texi (--param): Document it.
3376 * sched-ebb.c (schedule_ebb): No longer static. Remove declaration.
3377 New arg modulo_scheduling. All callers changed. Move note handling
3378 code here from schedule_ebbs.
3379 (schedule_ebbs_finish, schedule_ebbs_init): New functions, broken
3380 out of schedule_ebbs.
3381 (schedule_ebbs): Call them. Remove note handling code moved to
3383 * sched-int.h (schedule_ebb, schedule_ebbs_init,
3384 schedule_ebbs_finish): Declare.
3386 2011-09-30 Richard Guenther <rguenther@suse.de>