1 2009-01-07 Jan Sjodin <jan.sjodin@amd.com>
3 PR tree-optimization/38492
4 PR tree-optimization/38498
5 * tree-check.c (operator_is_linear, scev_is_linear_expression): New.
6 * tree-chrec.h (scev_is_linear_expression): Declared.
7 * graphite.c (graphite_cannot_represent_loop_niter): New.
8 (scopdet_basic_block_info): Call graphite_cannot_represent_loop_niter.
9 (graphite_loop_normal_form): Use gcc_assert.
10 (scan_tree_for_params): Use CASE_CONVERT.
11 (phi_node_is_iv, bb_contains_non_iv_scalar_phi_nodes): New.
12 (build_scop_conditions_1): Call bb_contains_non_iv_scalar_phi_nodes.
13 Use gcc_assert. Discard scops that contain unhandled cases.
14 (build_scop_conditions): Return a boolean status for unhandled cases.
15 (strip_mine_profitable_p): Print the loop number, not its depth.
16 (is_interchange_valid): Pass the depth of the loop nest, don't
18 (graphite_trans_bb_block): Same.
19 (graphite_trans_bb_block): Print tentative of loop blocking.
20 (graphite_trans_scop_block): Do not print that the loop has been
22 (graphite_transform_loops): Do not handle scops that contain condition
25 2009-01-07 H.J. Lu <hongjiu.lu@intel.com>
27 AVX Programming Reference (December, 2008)
28 * config/i386/avxintrin.h (_mm256_stream_si256): New.
29 (_mm256_stream_pd): Likewise.
30 (_mm256_stream_ps): Likewise.
32 * config/i386/i386.c (ix86_builtins): Add IX86_BUILTIN_MOVNTDQ256,
33 IX86_BUILTIN_MOVNTPD256 and IX86_BUILTIN_MOVNTPS256.
34 (ix86_special_builtin_type): Add VOID_FTYPE_PV4DI_V4DI.
35 (bdesc_special_args): Add __builtin_ia32_movntdq256,
36 __builtin_ia32_movntpd256 and __builtin_ia32_movntps256.
37 (ix86_init_mmx_sse_builtins): Handle VOID_FTYPE_PV4DI_V4DI.
38 (ix86_expand_special_args_builtin): Likewise.
40 * config/i386/sse.md (AVXMODEDI): New.
41 (avx_movnt<mode>): Likewise.
42 (avx_movnt<mode>): Likewise.
43 (<sse>_movnt<mode>): Remove AVX support.
44 (sse2_movntv2di): Likewise.
46 2009-01-07 Richard Guenther <rguenther@suse.de>
49 * fold-const.c (extract_muldiv): Remove obsolete comment.
50 (fold_plusminus_mult_expr): Undo MINUS_EXPR
51 to PLUS_EXPR canonicalization for the canonicalization.
53 2009-01-07 Gerald Pfeifer <gerald@pfeifer.com>
55 * doc/install.texi (alpha*-dec-osf*): Remove note on 32-bit
56 hosted cross-compilers generating less efficient code.
58 2009-01-06 Richard Sandiford <rdsandiford@googlemail.com>
60 * function.h (rtl_data): Add a dbr_scheduled_p field.
61 * reorg.c (dbr_schedule): Set it.
62 (gate_handle_delay_slots): Check it.
63 * config/mips/mips.c (mips_base_delayed_branch): Delete.
64 (mips_reorg): Check flag_delayed_branch instead of
65 mips_base_delayed_branch.
66 (mips_override_options): Don't set mips_base_delayed_branch
67 or flag_delayed_branch.
69 2009-01-06 Richard Sandiford <rdsandiford@googlemail.com>
71 PR rtl-optimization/38426.
72 * ira.c (ira): Set current_function_is_leaf earlier.
74 2009-01-06 Jakub Jelinek <jakub@redhat.com>
76 PR rtl-optimization/38722
77 * combine.c (try_combine): Don't modify PATTERN (i3) and notes
78 too early, only set a flag and modify after last possible
81 2009-01-06 Janis Johnson <janis187@us.ibm.com>
84 * ginclude/float.h: Rename DECnn_DEN to DECnn_SUBNORMAL_MIN.
85 * real.c (decimal_single_format): Correct values of emin and emax.
86 (decimal_double_format): Ditto.
87 (decimal_quad_format): Ditto.
88 * c-cppbuiltin.c (builtin_define_decimal_float_constants): Adjust
89 computation of DECnn_MIN and DECnn_MAX for corrected values of
90 emin and emax. Define __DECnn_SUBNORMAL_MIN__ instead of
91 __DECnn_MIN__, and adjust its computation for the corrected value
94 2009-01-06 Jan Hubicka <jh@suse.cz>
97 * i386.c (ix86_expand_call): Use ARRAY_SIZE
99 2009-01-06 Gerald Pfeifer <gerald@pfeifer.com>
101 * doc/contrib.texi (Contributors): Slightly adjust the end note.
102 Add Robert Clark to the list of testers.
104 2009-01-06 Jan Hubicka <jh@suse.cz>
105 Kai Tietz <kai.tietz@onevision.com>
107 * i386.md (*msabi_syvabi): Add SSE regs clobbers.
108 * i386.c (ix86_expand_call): Add clobbers.
110 2009-01-06 Jan Hubicka <jh@suse.cz>
111 Kai Tietz <kai.tietz@onevision.com>
113 * i386.h (CONDITIONAL_CALL_USAGE): SSE regs are not used for w64 ABI.
114 * i386.c (struct ix86_frame): Add padding0 and nsseregs.
115 (ix86_nsaved_regs): Count only general purpose regs.
116 (ix86_nsaved_sseregs): New.
117 (ix86_compute_frame_layout): Update nsseregs; set preferred alignment
118 to 16 for w64; compute padding and size of sse reg save area.
119 (ix86_emit_save_regs, ix86_emit_save_regs_using_mov): Save only general
121 (ix86_emit_save_sse_regs_using_mov): New.
122 (ix86_expand_prologue): Save SSE regs if needed.
123 (ix86_emit_restore_regs_using_mov): Use only general purpose regs.
124 (ix86_emit_restore_sse_regs_using_mov): New.
125 (ix86_expand_epilogue): Save SSE regs if needed.
127 2009-01-06 Jan Hubicka <jh@suse.cz>
128 Kai Tietz <kai.tietz@onevision.com>
130 * i386.h (ACCUMULATE_OUTGOING_ARGS): Enable for MSABI
131 * i386.c (init_cumulative_args): Disallow calls of MSABI functions when
132 accumulate outgoing args is off.
134 2009-01-06 H.J. Lu <hongjiu.lu@intel.com>
137 * ira-color.c (ira_reuse_stack_slot): Check ENABLE_IRA_CHECKING
138 before using pseudos_have_intersected_live_ranges_p.
140 * ira-int.h (ira_assert): Always define.
142 2009-01-06 H.J. Lu <hongjiu.lu@intel.com>
144 AVX Programming Reference (December, 2008)
145 * config/i386/avxintrin.h (_mm_permute2_pd): Removed.
146 (_mm256_permute2_pd): Likewise.
147 (_mm_permute2_ps): Likewise.
148 (_mm256_permute2_ps): Likewise.
149 * config/i386/i386.md (UNSPEC_VPERMIL2): Likewise.
150 * config/i386/sse.md (avx_vpermil2<mode>3): Likewise.
152 * config/i386/i386.c (ix86_builtins): Remove
153 IX86_BUILTIN_VPERMIL2PD, IX86_BUILTIN_VPERMIL2PS,
154 IX86_BUILTIN_VPERMIL2PD256 and IX86_BUILTIN_VPERMIL2PS256.
155 (ix86_builtin_type): Remove V8SF_FTYPE_V8SF_V8SF_V8SI_INT,
156 V4DF_FTYPE_V4DF_V4DF_V4DI_INT, V4SF_FTYPE_V4SF_V4SF_V4SI_INT
157 and V2DF_FTYPE_V2DF_V2DF_V2DI_INT.
158 (bdesc_args): Remove __builtin_ia32_vpermil2pd,
159 __builtin_ia32_vpermil2ps, __builtin_ia32_vpermil2pd256 and
160 __builtin_ia32_vpermil2ps256.
161 (ix86_init_mmx_sse_builtins): Updated.
162 (ix86_expand_args_builtin): Likewise.
164 2009-01-05 John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
166 * pa.c (output_call): Relocate non-jump insns in the delay slot of
167 long absolute calls when generating PA 2.0 code.
169 2009-01-05 Vladimir Makarov <vmakarov@redhat.com>
171 PR rtl-optimization/38583
172 * params.h (IRA_MAX_CONFLICT_TABLE_SIZE): New macro.
174 * params.def (ira-max-conflict-table-size): New.
176 * doc/invoke.texi (ira-max-conflict-table-size): Decribe.
178 * ira.h (ira_conflicts_p): New external definition.
180 * ira-conflicts.c (build_conflict_bit_table): Do not build too big
181 table. Report this. Return result of building.
182 (ira_build_conflicts): Use ira_conflicts_p. Check result of
183 building conflict table.
185 * ira-color.c (fast_allocation): Use num instead of
187 (ira_color): Use ira_conflicts_p.
189 * global.c: Include ira.h.
190 (pseudo_for_reload_consideration_p, build_insn_chain): Use
193 * Makefile.in (global.o): Add ira.h.
195 * ira-build.c (mark_all_loops_for_removal,
196 propagate_some_info_from_allocno): New.
197 (remove_unnecessary_allocnos): Call
198 propagate_some_info_from_allocno.
199 (remove_low_level_allocnos): New.
200 (remove_unnecessary_regions): Add parameter. Call
201 mark_all_loops_for_removal and remove_low_level_allocnos. Pass
202 parameter to remove_unnecessary_regions.
203 (ira_build): Remove all regions but root if the conflict table was
204 not built. Update conflict hard regs for allocnos crossing calls.
206 * ira.c (ira_conflicts_p): New global.
207 (ira): Define and use ira_conflicts_p.
209 * reload1.c (compute_use_by_pseudos, reload, count_pseudo,
210 count_spilled_pseudo, find_reg, alter_reg, finish_spills,
211 emit_input_reload_insns, delete_output_reload): Use
214 2009-01-06 Ben Elliston <bje@au.ibm.com>
216 * gengtype-lex.l (YY_NO_INPUT): Define.
218 2009-01-05 Andrew Pinski <andrew_pinski@playstation.sony.com>
221 * c-common.c (handle_vector_size_attribute): Also reject
224 2009-01-05 Sebastian Pop <sebastian.pop@amd.com>
226 PR tree-optimization/38492
227 * graphite.c (rename_map_elt, debug_rename_elt,
228 debug_rename_map_1, debug_rename_map, new_rename_map_elt,
229 rename_map_elt_info, eq_rename_map_elts,
230 get_new_name_from_old_name, bb_in_sese_p): Moved around.
231 (sese_find_uses_to_rename_use): Renamed sese_build_livein_liveouts_use.
232 (sese_find_uses_to_rename_bb): Renamed sese_build_livein_liveouts_bb.
233 (sese_build_livein_liveouts): New.
234 (new_sese, free_sese): New.
235 (new_scop): Call new_sese.
236 (free_scop): Call free_sese.
237 (rename_variables_from_edge, rename_phis_end_scop): Removed.
238 (register_old_new_names): Renamed register_old_and_new_names.
239 (register_scop_liveout_renames, add_loop_exit_phis,
240 insert_loop_close_phis, struct igp,
241 default_liveout_before_guard, add_guard_exit_phis,
242 insert_guard_phis, copy_renames): New.
243 (translate_clast): Call insert_loop_close_phis and insert_guard_phis.
244 (sese_add_exit_phis_edge): Renamed scop_add_exit_phis_edge.
245 (rewrite_into_sese_closed_ssa): Renamed scop_insert_phis_for_liveouts.
246 (scop_adjust_phis_for_liveouts): New.
247 (gloog): Call scop_adjust_phis_for_liveouts.
249 * graphite.h (struct sese): Documented. Added fields liveout,
251 (SESE_LIVEOUT, SESE_LIVEIN, SESE_LIVEIN_VER, SESE_NUM_VER): New.
252 (new_sese, free_sese, sese_build_livein_liveouts): Declared.
253 (struct scop): Added field liveout_renames.
254 (SCOP_LIVEOUT_RENAMES): New.
256 2009-01-05 Harsha Jagasia <harsha.jagasia@amd.com>
258 PR tree-optimization/38510
259 * graphite.c (recompute_all_dominators): Call mark_irreducible_loops.
260 (translate_clast): Call recompute_all_dominators before
262 (gloog): Call recompute_all_dominators before graphite_verify.
264 2009-01-05 Harsha Jagasia <harsha.jagasia@amd.com>
265 Jan Sjodin <jan.sjodin@amd.com>
267 PR tree-optimization/38500
268 * graphite.c (create_sese_edges): Call fix_loop_structure after
271 2009-01-05 Joel Sherrill <joel.sherrill@oarcorp.com>
273 * config.gcc: Add m32r*-*-rtems*.
274 * config/m32r/rtems.h: New file.
276 2009-01-05 Ben Elliston <bje@au.ibm.com>
278 * Makefile.in (.po.gmo): Use mkinstalldirs, not test -d || mkdir.
280 (po/gcc.pot): Likewise.
282 2009-01-04 David S. Miller <davem@davemloft.net>
284 * config/sparc/sparc.h (SECONDARY_MEMORY_NEEDED_RTX): Delete.
285 (STARTING_FRAME_OFFSET): Always set to zero.
287 2009-01-04 Richard Sandiford <rdsandiford@googlemail.com>
289 * tree.def (LSHIFT_EXPR, RSHIFT_EXPR): Add commentary.
290 * tree-cfg.c (verify_gimple_assign_binary): Allow shifts of
291 fixed-point types, and vectors of the same.
293 2009-01-04 Richard Sandiford <rdsandiford@googlemail.com>
295 * config/mips/sync.md (*mb_barrier): Rename to...
296 (*memory_barrier): ...this.
298 2009-01-04 Jonathan Wakely <jwakely.gcc@gmail.com>
300 * doc/extend.texi (Function Attributes): Move @cindex after @item
301 for 'artificial' and 'flatten'. Fix grammar for 'externally_visible'
302 and put in alphabetical order. Fix 'target' name and put in order.
303 * doc/invoke.texi (-Wstrict-null-sentinel, -fipa-matrix-reorg): Fix
306 2009-01-04 Uros Bizjak <ubizjak@gmail.com>
308 * config/s390/s390.md (UNSPEC_MB): Rename from UNSPECV_MB.
309 (memory_barrier): Expand as unspec instead of unspec_volatile.
310 Remove mem:BLK from insn operands. Use Pmode scratch register.
311 (*memory_barrier): Define as unspec instead of unspec_volatile.
312 Use (match_dup 0) as input operand.
314 * config/sparc/sparc.md (UNSPEC_MEMBAR): Rename from UNSPECV_MEMBAR.
315 * config/sparc/sync.md (memory_barrier): Expand as unspec instead of
316 unspec_volatile. Remove mem:BLK from insn operands. Use Pmode
317 scratch register. Remove operand 1.
318 (*stbar): Define as unspec instead of unspec_volatile.
319 Use (match_dup 0) as input operand, remove (const_int 8).
320 (*membar): Define as unspec instead of unspec_volatile.
321 Use (match_dup 0) as input operand, remove input operand 2.
323 * config/xtensa/xtensa.md (UNSPEC_MEMW): Rename from UNSPECV_MEMW.
324 (memory_barrier): Expand as unspec instead of unspec_volatile.
325 Remove mem:BLK from insn operands. Use Pmode scratch register.
326 (*memory_barrier): Define as unspec instead of unspec_volatile.
327 Use (match_dup 0) as input operand.
329 * config/ia64/sync.md (memory_barrier): Redefine as expander pattern.
330 Remove mem:BLK from insn operands. Use Pmode scratch register.
331 Set volatile flag on operand 0.
332 (*memory_barrier): New insn pattern.
334 * config/rs6000/sync.md (memory_barrier): Remove mem:BLK from
336 (*memory_barrier): Use (match_dup 0) as input operand.
338 * config/mips/sync.md (memory_barrier): Redefine as expander pattern.
339 Remove mem:BLK from insn operands. Use Pmode scratch register.
340 Set volatile flag on operand 0.
341 (*memory_barrier): New insn pattern.
343 * config/alpha/sync.md (*memory_barrier): Rename from *mb_internal.
345 2009-01-04 Steven Bosscher <steven@gcc.gnu.org>
348 * function.c (struct temp_slot): Move to the section of the file
349 that deals with temp slots. Remove field 'address'.
350 (temp_slot_address_table): New hash table of address -> temp slot.
351 (struct temp_slot_address_entry): New struct, items for the table.
352 (temp_slot_address_compute_hash, temp_slot_address_hash,
353 temp_slot_address_eq, insert_temp_slot_address): Support functions
355 (find_temp_slot_from_address): Rewrite to use the new hash table.
356 (remove_unused_temp_slot_addresses): Remove addresses of temp
357 slots that have been made available.
358 (remove_unused_temp_slot_addresses_1): Call-back for htab_traverse,
359 worker function for remove_unused_temp_slot_addresses.
360 (assign_stack_temp_for_type): Don't clear the temp slot address list.
361 Add the temp slot address to the address -> temp slot map.
362 (update_temp_slot_address): Update via insert_temp_slot_address.
363 (free_temp_slots): Call remove_unused_temp_slot_addresses.
364 (pop_temp_slots): Likewise.
365 (init_temp_slots): Allocate the address -> temp slot map, or empty
366 the map if it is already allocated.
367 (prepare_function_start): Initialize temp slot processing.
369 2009-01-04 Steven Bosscher <steven@gcc.gnu.org>
372 * cfgexpand.c (estimate_stack_frame_size): Simplify the estimate:
373 Calculate the size of all stack vars assuming no packing of stack
374 vars will happen, replacing a quadratic algorithm with a linear one.
376 2009-01-03 Jakub Jelinek <jakub@redhat.com>
379 * expmed.c (store_bit_field_1): Don't modify op0 if movstrict insn
382 2009-01-03 Diego Novillo <dnovillo@google.com>
384 * doc/contrib.texi: Update contributions.
386 2009-01-03 Jakub Jelinek <jakub@redhat.com>
389 * builtins.c (fold_builtin_memory_op): Give up if either operand
390 is volatile. Set srctype or desttype to non-qualified version
394 * builtins.c (fold_builtin_expect): Only check DECL_WEAK for VAR_DECLs
397 2009-01-02 Kenneth Zadeck <zadeck@naturalbridge.com>
399 PR rtl-optimization/35805
400 * df-problems.c (df_lr_finalize): Add recursive call to resolve lr
401 problem if fast dce is able to remove any instructions.
402 * dce.c (dce_process_block): Fix dump message.
404 2009-01-02 Mark Mitchell <mark@codesourcery.com>
407 * tree-ssa-pre.c (compute_antic): Correct loop bounds.
409 2009-01-02 Jakub Jelinek <jakub@redhat.com>
412 * tree-flow.h (op_code_prio, op_prio): New prototypes.
413 * tree-pretty-print.c (op_code_prio): New function.
414 (op_prio): No longer static. Use op_code_prio.
415 * gimple-pretty-print.c (dump_unary_rhs, dump_binary_rhs):
416 Use op_prio and op_code_prio to determine if () should be
417 printed around operand(s) or not.
419 * gimple-pretty-print.c (dump_unary_rhs, dump_binary_rhs,
420 dump_gimple_call, dump_gimple_switch, dump_gimple_cond,
421 dump_gimple_label, dump_gimple_try, dump_symbols, dump_gimple_phi,
422 dump_gimple_mem_ops, dump_bb_header, dump_bb_end, pp_cfg_jump): Use
423 pp_character instead of pp_string for single letter printing.
425 2009-01-02 Richard Sandiford <rdsandiford@googlemail.com>
427 * doc/extend.texi: Fix '#pragma GCC option' typo.
429 2009-01-02 Richard Guenther <rguenther@suse.de>
431 * doc/install.texi (--enable-checking): Mention different
433 (--enable-stage1-checking): Document.
435 2009-01-01 Andrew Pinski <pinskia@gmail.com>
438 * tree-cfg.c (verify_expr): Add INDIRECT_REF case. Change MODIFY_EXPR
441 2009-01-02 Ben Elliston <bje@au.ibm.com>
443 * config/fp-bit.h (pack_d): Constify argument.
444 * config/fp-bit.c (makenan): Constify return type. Remove casts.
445 (isnan): Constify argument.
449 (_fpadd_parts): Constify return type.
450 (_fpmul_parts): Likewise.
451 (_fpdiv_parts): Likewise.
453 2009-01-01 Jakub Jelinek <jakub@redhat.com>
456 * c-typeck.c (add_pending_init): Add IMPLICIT argument. Only
457 warn about overwriting initializer with side-effects or
458 -Woverride-init if !IMPLICIT.
459 (output_init_element): Likewise. Pass IMPLICIT down to
461 (process_init_element): Add IMPLICIT argument. Pass it down
462 to output_init_element.
463 (push_init_element, pop_init_level, set_designator): Adjust
464 process_init_element callers.
465 (set_nonincremental_init, set_nonincremental_init_from_string):
466 Adjust add_pending_init callers.
467 (output_pending_init_elements): Adjust output_init_element callers.
468 * c-tree.h (process_init_element): Adjust prototype.
469 * c-parser.c (c_parser_initelt, c_parser_initval): Adjust
470 process_init_element callers.