OSDN Git Service

* config/i386/i386.c (ix86_target_string): Use ARRAY_SIZE.
[pf3gnuchains/gcc-fork.git] / gcc / ChangeLog
index 2c5a939..03338fc 100644 (file)
@@ -1,3 +1,375 @@
+2009-01-07  Uros Bizjak  <ubizjak@gmail.com>
+
+       * config/i386/i386.c (ix86_target_string): Use ARRAY_SIZE.
+       (ix86_target_string): Ditto.
+       (ix86_valid_target_attribute_inner_p): Ditto.
+
+2009-01-07  Jan Sjodin  <jan.sjodin@amd.com>
+
+       PR tree-optimization/38492
+       PR tree-optimization/38498
+       * tree-check.c (operator_is_linear, scev_is_linear_expression): New.
+       * tree-chrec.h (scev_is_linear_expression): Declared.
+       * graphite.c (graphite_cannot_represent_loop_niter): New.
+       (scopdet_basic_block_info): Call graphite_cannot_represent_loop_niter.
+       (graphite_loop_normal_form): Use gcc_assert.
+       (scan_tree_for_params): Use CASE_CONVERT.
+       (phi_node_is_iv, bb_contains_non_iv_scalar_phi_nodes): New.
+       (build_scop_conditions_1): Call bb_contains_non_iv_scalar_phi_nodes.
+       Use gcc_assert.  Discard scops that contain unhandled cases.
+       (build_scop_conditions): Return a boolean status for unhandled cases.
+       (strip_mine_profitable_p): Print the loop number, not its depth.
+       (is_interchange_valid): Pass the depth of the loop nest, don't
+       recompute it wrongly.
+       (graphite_trans_bb_block): Same.
+       (graphite_trans_bb_block): Print tentative of loop blocking.
+       (graphite_trans_scop_block): Do not print that the loop has been
+       blocked.
+       (graphite_transform_loops): Do not handle scops that contain condition
+       scalar phi nodes.
+
+2009-01-07  H.J. Lu  <hongjiu.lu@intel.com>
+
+       AVX Programming Reference (December, 2008)
+       * config/i386/avxintrin.h (_mm256_stream_si256): New.
+       (_mm256_stream_pd): Likewise.
+       (_mm256_stream_ps): Likewise.
+
+       * config/i386/i386.c (ix86_builtins): Add IX86_BUILTIN_MOVNTDQ256,
+       IX86_BUILTIN_MOVNTPD256 and IX86_BUILTIN_MOVNTPS256.
+       (ix86_special_builtin_type): Add VOID_FTYPE_PV4DI_V4DI.
+       (bdesc_special_args): Add __builtin_ia32_movntdq256,
+       __builtin_ia32_movntpd256 and __builtin_ia32_movntps256.
+       (ix86_init_mmx_sse_builtins): Handle VOID_FTYPE_PV4DI_V4DI.
+       (ix86_expand_special_args_builtin): Likewise.
+
+       * config/i386/sse.md (AVXMODEDI): New.
+       (avx_movnt<mode>): Likewise.
+       (avx_movnt<mode>): Likewise.
+       (<sse>_movnt<mode>): Remove AVX support.
+       (sse2_movntv2di): Likewise.
+
+2009-01-07  Richard Guenther  <rguenther@suse.de>
+
+       PR middle-end/38751
+       * fold-const.c (extract_muldiv): Remove obsolete comment.
+       (fold_plusminus_mult_expr): Undo MINUS_EXPR
+       to PLUS_EXPR canonicalization for the canonicalization.
+
+2009-01-07  Gerald Pfeifer  <gerald@pfeifer.com>
+
+       * doc/install.texi (alpha*-dec-osf*): Remove note on 32-bit
+       hosted cross-compilers generating less efficient code.
+
+2009-01-06  Richard Sandiford  <rdsandiford@googlemail.com>
+
+       * function.h (rtl_data): Add a dbr_scheduled_p field.
+       * reorg.c (dbr_schedule): Set it.
+       (gate_handle_delay_slots): Check it.
+       * config/mips/mips.c (mips_base_delayed_branch): Delete.
+       (mips_reorg): Check flag_delayed_branch instead of
+       mips_base_delayed_branch.
+       (mips_override_options): Don't set mips_base_delayed_branch
+       or flag_delayed_branch.
+
+2009-01-06  Richard Sandiford  <rdsandiford@googlemail.com>
+
+       PR rtl-optimization/38426.
+       * ira.c (ira): Set current_function_is_leaf earlier.
+
+2009-01-06  Jakub Jelinek  <jakub@redhat.com>
+
+       PR rtl-optimization/38722
+       * combine.c (try_combine): Don't modify PATTERN (i3) and notes
+       too early, only set a flag and modify after last possible
+       undo_all point.
+
+2009-01-06  Janis Johnson  <janis187@us.ibm.com>
+
+       PR c/34252
+       * ginclude/float.h: Rename DECnn_DEN to DECnn_SUBNORMAL_MIN.
+       * real.c (decimal_single_format): Correct values of emin and emax.
+       (decimal_double_format): Ditto.
+       (decimal_quad_format): Ditto.
+       * c-cppbuiltin.c (builtin_define_decimal_float_constants): Adjust
+       computation of DECnn_MIN and DECnn_MAX for corrected values of
+       emin and emax.  Define __DECnn_SUBNORMAL_MIN__ instead of
+       __DECnn_MIN__, and adjust its computation for the corrected value
+       of emin.
+
+2009-01-06  Jan Hubicka  <jh@suse.cz>
+
+       PR target/38744
+       * i386.c (ix86_expand_call): Use ARRAY_SIZE.
+
+2009-01-06  Gerald Pfeifer  <gerald@pfeifer.com>
+
+       * doc/contrib.texi (Contributors): Slightly adjust the end note.
+       Add Robert Clark to the list of testers.
+
+2009-01-06  Jan Hubicka  <jh@suse.cz>
+           Kai Tietz <kai.tietz@onevision.com>
+
+       * i386.md (*msabi_syvabi): Add SSE regs clobbers.
+       * i386.c (ix86_expand_call): Add clobbers.
+
+2009-01-06  Jan Hubicka  <jh@suse.cz>
+           Kai Tietz <kai.tietz@onevision.com>
+
+       * i386.h (CONDITIONAL_CALL_USAGE): SSE regs are not used for w64 ABI.
+       * i386.c (struct ix86_frame): Add padding0 and nsseregs.
+       (ix86_nsaved_regs): Count only general purpose regs.
+       (ix86_nsaved_sseregs): New.
+       (ix86_compute_frame_layout): Update nsseregs; set preferred alignment
+       to 16 for w64; compute padding and size of sse reg save area.
+       (ix86_emit_save_regs, ix86_emit_save_regs_using_mov): Save only
+       general purpose regs.
+       (ix86_emit_save_sse_regs_using_mov): New.
+       (ix86_expand_prologue): Save SSE regs if needed.
+       (ix86_emit_restore_regs_using_mov): Use only general purpose regs.
+       (ix86_emit_restore_sse_regs_using_mov): New.
+       (ix86_expand_epilogue): Save SSE regs if needed.
+
+2009-01-06  Jan Hubicka  <jh@suse.cz>
+           Kai Tietz <kai.tietz@onevision.com>
+
+       * i386.h (ACCUMULATE_OUTGOING_ARGS): Enable for MSABI
+       * i386.c (init_cumulative_args): Disallow calls of MSABI functions
+       when accumulate outgoing args is off.
+
+2009-01-06  H.J. Lu  <hongjiu.lu@intel.com>
+
+       PR bootstrap/38742
+       * ira-color.c (ira_reuse_stack_slot): Check ENABLE_IRA_CHECKING
+       before using pseudos_have_intersected_live_ranges_p.
+
+       * ira-int.h (ira_assert): Always define.
+
+2009-01-06  H.J. Lu  <hongjiu.lu@intel.com>
+
+       AVX Programming Reference (December, 2008)
+       * config/i386/avxintrin.h (_mm_permute2_pd): Removed.
+       (_mm256_permute2_pd): Likewise.
+       (_mm_permute2_ps): Likewise.
+       (_mm256_permute2_ps): Likewise.
+       * config/i386/i386.md (UNSPEC_VPERMIL2): Likewise.
+       * config/i386/sse.md (avx_vpermil2<mode>3): Likewise.
+
+       * config/i386/i386.c (ix86_builtins): Remove
+       IX86_BUILTIN_VPERMIL2PD, IX86_BUILTIN_VPERMIL2PS,
+       IX86_BUILTIN_VPERMIL2PD256 and IX86_BUILTIN_VPERMIL2PS256.
+       (ix86_builtin_type): Remove V8SF_FTYPE_V8SF_V8SF_V8SI_INT,
+       V4DF_FTYPE_V4DF_V4DF_V4DI_INT, V4SF_FTYPE_V4SF_V4SF_V4SI_INT
+       and V2DF_FTYPE_V2DF_V2DF_V2DI_INT.
+       (bdesc_args): Remove __builtin_ia32_vpermil2pd,
+       __builtin_ia32_vpermil2ps, __builtin_ia32_vpermil2pd256 and
+       __builtin_ia32_vpermil2ps256.
+       (ix86_init_mmx_sse_builtins): Updated.
+       (ix86_expand_args_builtin): Likewise.
+
+2009-01-05  John David Anglin  <dave.anglin@nrc-cnrc.gc.ca>
+
+       * pa.c (output_call): Relocate non-jump insns in the delay slot of
+       long absolute calls when generating PA 2.0 code.
+
+2009-01-05  Vladimir Makarov  <vmakarov@redhat.com>
+
+       PR rtl-optimization/38583
+       * params.h (IRA_MAX_CONFLICT_TABLE_SIZE): New macro.
+
+       * params.def (ira-max-conflict-table-size): New.
+
+       * doc/invoke.texi (ira-max-conflict-table-size): Decribe.
+       
+       * ira.h (ira_conflicts_p): New external definition.
+       
+       * ira-conflicts.c (build_conflict_bit_table): Do not build too big
+       table.  Report this.  Return result of building.
+       (ira_build_conflicts): Use ira_conflicts_p.  Check result of
+       building conflict table.
+
+       * ira-color.c (fast_allocation): Use num instead of ira_allocnos_num.
+       (ira_color): Use ira_conflicts_p.
+       
+       * global.c: Include ira.h.
+       (pseudo_for_reload_consideration_p, build_insn_chain): Use
+       ira_conflicts_p.
+
+       * Makefile.in (global.o): Add ira.h.
+       
+       * ira-build.c (mark_all_loops_for_removal,
+       propagate_some_info_from_allocno): New.
+       (remove_unnecessary_allocnos): Call
+       propagate_some_info_from_allocno.
+       (remove_low_level_allocnos): New.
+       (remove_unnecessary_regions): Add parameter.  Call
+       mark_all_loops_for_removal and remove_low_level_allocnos.  Pass
+       parameter to remove_unnecessary_regions.
+       (ira_build): Remove all regions but root if the conflict table was
+       not built.  Update conflict hard regs for allocnos crossing calls.
+
+       * ira.c (ira_conflicts_p): New global.
+       (ira): Define and use ira_conflicts_p.
+
+       * reload1.c (compute_use_by_pseudos, reload, count_pseudo,
+       count_spilled_pseudo, find_reg, alter_reg, finish_spills,
+       emit_input_reload_insns, delete_output_reload): Use ira_conflicts_p.
+       
+2009-01-06  Ben Elliston  <bje@au.ibm.com>
+
+       * gengtype-lex.l (YY_NO_INPUT): Define.
+
+2009-01-05  Andrew Pinski  <andrew_pinski@playstation.sony.com>
+
+       PR c/34911
+       * c-common.c (handle_vector_size_attribute): Also reject
+       BOOLEAN_TYPE types.
+
+2009-01-05  Sebastian Pop  <sebastian.pop@amd.com>
+
+       PR tree-optimization/38492
+       * graphite.c (rename_map_elt, debug_rename_elt,
+       debug_rename_map_1, debug_rename_map, new_rename_map_elt,
+       rename_map_elt_info, eq_rename_map_elts,
+       get_new_name_from_old_name, bb_in_sese_p): Moved around.
+       (sese_find_uses_to_rename_use): Renamed sese_build_livein_liveouts_use.
+       (sese_find_uses_to_rename_bb): Renamed sese_build_livein_liveouts_bb.
+       (sese_build_livein_liveouts): New.
+       (new_sese, free_sese): New.
+       (new_scop): Call new_sese.
+       (free_scop): Call free_sese.
+       (rename_variables_from_edge, rename_phis_end_scop): Removed.
+       (register_old_new_names): Renamed register_old_and_new_names.
+       (register_scop_liveout_renames, add_loop_exit_phis,
+       insert_loop_close_phis, struct igp,
+       default_liveout_before_guard, add_guard_exit_phis,
+       insert_guard_phis, copy_renames): New.
+       (translate_clast): Call insert_loop_close_phis and insert_guard_phis.
+       (sese_add_exit_phis_edge): Renamed scop_add_exit_phis_edge.
+       (rewrite_into_sese_closed_ssa): Renamed scop_insert_phis_for_liveouts.
+       (scop_adjust_phis_for_liveouts): New.
+       (gloog): Call scop_adjust_phis_for_liveouts.
+
+       * graphite.h (struct sese): Documented.  Added fields liveout,
+       num_ver and livein.
+       (SESE_LIVEOUT, SESE_LIVEIN, SESE_LIVEIN_VER, SESE_NUM_VER): New.
+       (new_sese, free_sese, sese_build_livein_liveouts): Declared.
+       (struct scop): Added field liveout_renames.
+       (SCOP_LIVEOUT_RENAMES): New.
+
+2009-01-05  Harsha Jagasia  <harsha.jagasia@amd.com>
+
+       PR tree-optimization/38510
+       * graphite.c (recompute_all_dominators): Call mark_irreducible_loops.
+       (translate_clast): Call recompute_all_dominators before
+       graphite_verify.
+       (gloog): Call recompute_all_dominators before graphite_verify.
+
+2009-01-05  Harsha Jagasia  <harsha.jagasia@amd.com>
+           Jan Sjodin <jan.sjodin@amd.com>
+
+       PR tree-optimization/38500
+       * graphite.c (create_sese_edges): Call fix_loop_structure after
+       splitting blocks.
+
+2009-01-05  Joel Sherrill <joel.sherrill@oarcorp.com>
+
+       * config.gcc: Add m32r*-*-rtems*.
+       * config/m32r/rtems.h: New file.
+
+2009-01-05  Ben Elliston  <bje@au.ibm.com>
+
+       * Makefile.in (.po.gmo): Use mkinstalldirs, not test -d || mkdir.
+       (.po.pox): Likewise.
+       (po/gcc.pot): Likewise.
+
+2009-01-04  David S. Miller  <davem@davemloft.net>
+
+       * config/sparc/sparc.h (SECONDARY_MEMORY_NEEDED_RTX): Delete.
+       (STARTING_FRAME_OFFSET): Always set to zero.
+
+2009-01-04  Richard Sandiford  <rdsandiford@googlemail.com>
+
+       * tree.def (LSHIFT_EXPR, RSHIFT_EXPR): Add commentary.
+       * tree-cfg.c (verify_gimple_assign_binary): Allow shifts of
+       fixed-point types, and vectors of the same.
+
+2009-01-04  Richard Sandiford  <rdsandiford@googlemail.com>
+
+       * config/mips/sync.md (*mb_barrier): Rename to...
+       (*memory_barrier): ...this.
+
+2009-01-04  Jonathan Wakely  <jwakely.gcc@gmail.com>
+
+       * doc/extend.texi (Function Attributes): Move @cindex after @item
+       for 'artificial' and 'flatten'. Fix grammar for 'externally_visible'
+       and put in alphabetical order. Fix 'target' name and put in order.
+       * doc/invoke.texi (-Wstrict-null-sentinel, -fipa-matrix-reorg): Fix
+       typos.
+
+2009-01-04  Uros Bizjak  <ubizjak@gmail.com>
+
+       * config/s390/s390.md (UNSPEC_MB): Rename from UNSPECV_MB.
+       (memory_barrier): Expand as unspec instead of unspec_volatile.
+       Remove mem:BLK from insn operands.  Use Pmode scratch register.
+       (*memory_barrier): Define as unspec instead of unspec_volatile.
+       Use (match_dup 0) as input operand.
+
+       * config/sparc/sparc.md (UNSPEC_MEMBAR): Rename from UNSPECV_MEMBAR.
+       * config/sparc/sync.md (memory_barrier): Expand as unspec instead of
+       unspec_volatile.  Remove mem:BLK from insn operands.  Use Pmode
+       scratch register.  Remove operand 1.
+       (*stbar): Define as unspec instead of unspec_volatile.
+       Use (match_dup 0) as input operand, remove (const_int 8).
+       (*membar): Define as unspec instead of unspec_volatile.
+       Use (match_dup 0) as input operand, remove input operand 2.
+
+       * config/xtensa/xtensa.md (UNSPEC_MEMW): Rename from UNSPECV_MEMW.
+       (memory_barrier): Expand as unspec instead of unspec_volatile.
+       Remove mem:BLK from insn operands.  Use Pmode scratch register.
+       (*memory_barrier): Define as unspec instead of unspec_volatile.
+       Use (match_dup 0) as input operand.
+
+       * config/ia64/sync.md (memory_barrier): Redefine as expander pattern.
+       Remove mem:BLK from insn operands.  Use Pmode scratch register.
+       Set volatile flag on operand 0.
+       (*memory_barrier): New insn pattern.
+
+       * config/rs6000/sync.md (memory_barrier): Remove mem:BLK from
+       insn operands.
+       (*memory_barrier): Use (match_dup 0) as input operand.
+
+       * config/mips/sync.md (memory_barrier): Redefine as expander pattern.
+       Remove mem:BLK from insn operands.  Use Pmode scratch register.
+       Set volatile flag on operand 0.
+       (*mb_internal): New insn pattern.
+
+       * config/alpha/sync.md (*memory_barrier): Rename from *mb_internal.
+
+2009-01-04  Steven Bosscher  <steven@gcc.gnu.org>
+
+       PR middle-end/38586
+       * function.c (struct temp_slot): Move to the section of the file
+       that deals with temp slots.  Remove field 'address'.
+       (temp_slot_address_table): New hash table of address -> temp slot.
+       (struct temp_slot_address_entry): New struct, items for the table.
+       (temp_slot_address_compute_hash, temp_slot_address_hash,
+       temp_slot_address_eq, insert_temp_slot_address): Support functions
+       for the new table.
+       (find_temp_slot_from_address): Rewrite to use the new hash table.
+       (remove_unused_temp_slot_addresses): Remove addresses of temp
+       slots that have been made available.
+       (remove_unused_temp_slot_addresses_1): Call-back for htab_traverse,
+       worker function for remove_unused_temp_slot_addresses.
+       (assign_stack_temp_for_type): Don't clear the temp slot address list.
+       Add the temp slot address to the address -> temp slot map.
+       (update_temp_slot_address): Update via insert_temp_slot_address.
+       (free_temp_slots): Call remove_unused_temp_slot_addresses.
+       (pop_temp_slots): Likewise.
+       (init_temp_slots): Allocate the address -> temp slot map, or empty
+       the map if it is already allocated.
+       (prepare_function_start): Initialize temp slot processing.
+
 2009-01-04  Steven Bosscher  <steven@gcc.gnu.org>
 
        PR middle-end/38584