OSDN Git Service

2011-12-02 Richard Guenther <rguenther@suse.de>
[pf3gnuchains/gcc-fork.git] / gcc / ChangeLog
index d77a380..3865341 100644 (file)
@@ -1,3 +1,768 @@
+2011-12-02  Richard Guenther  <rguenther@suse.de>
+
+       PR lto/47259
+       * ipa.c (varpool_externally_visible_p): Register variables
+       are always externally visible.
+
+2011-12-02  Sameera Deshpande  <sameera.deshpande@arm.com>
+
+       * gimple-pretty-print.c (dump_gimple_call): Dereference fn only if
+         non-NULL.
+
+2011-12-02  Martin Jambor  <mjambor@suse.cz>
+
+       PR tree-optimization/50622
+       * tree-sra.c (load_assign_lhs_subreplacements): Force gimple operand
+       if both lacc and racc are grp_partial_lhs.
+
+2011-12-01  Kaz Kojima  <kkojima@gcc.gnu.org>
+
+       PR target/50814.
+       * config/sh/sh.c (expand_ashiftrt): Handle TARGET_SH2A same as
+       TARGET_SH3.
+       (shl_sext_kind): Likewise.
+       * config/sh/sh.h (SH_DYNAMIC_SHIFT_COST): Likewise.
+       * config/sh/sh.md (ashlsi3_sh2a, ashrsi3_sh2a, lshrsi3_sh2a):
+       Remove.
+       (ashlsi3_std): Handle TARGET_SH2A same as TARGET_SH3.
+       (ashlsi3): Likewise.
+       (ashrsi3_d): Likewise.
+       (lshrsi3_d): Likewise.
+       (lshrsi3): Likewise.
+
+2011-12-01   Diego Novillo  <dnovillo@google.com>
+
+       PR bootstrap/51346
+       * ipa-inline.c (can_inline_edge_p): If the edge E has a
+       statement, use the statement's inline indicator instead
+       of E's.
+       Remove consistency check.
+
+2011-12-01   Diego Novillo  <dnovillo@google.com>
+
+       PR bootstrap/51346
+       Revert
+
+       2011-11-29   Diego Novillo  <dnovillo@google.com>
+
+               * gimple.c (gimple_call_set_cannot_inline): Move from gimple.h.
+               Update field call_stmt_cannot_inline_p from call
+               graph edge, if needed.
+               * gimple.h (gimple_call_set_cannot_inline): Move to gimple.c.
+
+2011-12-01  Nathan Sidwell  <nathan@acm.org>
+
+       PR gcov-profile/51113
+       * coverage.c (build_var): Keep coverage variables static.
+
+2011-12-01  Jakub Jelinek  <jakub@redhat.com>
+
+       PR tree-optimization/51356
+       * tree-vect-patterns.c (vect_recog_bool_pattern): Give up if
+       vectype doesn't have VECTOR_MODE_P.
+
+       PR debug/50317
+       * tree-ssa-dce.c (remove_dead_stmt): Add a debug stmt when removing
+       as unnecessary a store to a variable with gimple reg type.
+       * tree-ssa-live.c (remove_unused_locals): Clear TREE_ADDRESSABLE bit
+       on local unreferenced variables.
+       * cfgexpand.c (expand_gimple_basic_block): Don't emit DEBUG_INSNs
+       for !target_for_debug_bind variables.
+
+2011-12-01  Patrick Marlier  <patrick.marlier@gmail.com>
+
+       PR middle-end/51273
+       * cgraph.h (cgraph_call_node_duplication_hooks): Declare.
+       * cgraph.c (cgraph_call_node_duplication_hooks): Make global. 
+       * cgraphunit.c (cgraph_copy_node_for_versioning): Call it.
+
+2011-12-01  Andrew Pinski  <apinski@cavium.com>
+
+       PR lto/51198
+       * tree.c (free_lang_data_in_decl): Clear FIELD_DECL's
+       DECL_INITIAL also.
+
+2011-12-01  Jakub Jelinek  <jakub@redhat.com>
+
+       PR tree-optimization/51246
+       * tree-predcom.c (replace_ref_with): Handle also clobber on the
+       rhs.
+
+       PR rtl-optimization/51014
+       * loop-unroll.c (apply_opt_in_copies): Ignore label DEBUG_INSNs
+       both from bb and orig_bb.
+
+2011-12-01  Joern Rennecke  <joern.rennecke@embecosm.com>
+
+       PR tree-optimization/50802
+       * tree-vrp.c (simplify_conversion_using_ranges): Rewrite test
+       considering what happens to ranges during sign changes and/or
+       intermediate narrowing conversions.
+
+2011-11-30  John David Anglin  <dave.anglin@nrc-cnrc.gc.ca>
+
+       PR middle-end/50283
+       * config/pa/pa.md (in_branch_delay): Disallow frame related insns.
+       (in_nullified_branch_delay): Likewise.
+       (in_call_delay): Likewise.
+
+2011-11-30  Richard Henderson  <rth@redhat.com>
+
+       * predict.c (expr_expected_value_1): Assume compare-and-swap builtin
+       boolean return is true.
+
+2011-11-30  Jakub Jelinek  <jakub@redhat.com>
+
+       PR rtl-optimization/51044
+       * dse.c (emit_inc_dec_insn_before): Ensure new_insn is an instruction
+       rather than just insn body.
+
+       PR target/50725
+       * function.c (thread_prologue_and_epilogue_insns): If
+       stack_realign_drap, add drap_reg to SET_UP_BY_PROLOGUE
+       regset.
+
+       PR middle-end/51089
+       * tree-eh.c (cleanup_empty_eh_merge_phis): Add check to
+       avoid creating duplicate edges here.
+       (cleanup_empty_eh_unsplit): And remove it in the caller.
+
+2011-11-30  Andrew Pinski  <apinski@cavium.com>
+
+       PR c/51321
+       * c-parser.c (c_parser_postfix_expression): Check groktypename results
+       before looking at the main variant.
+
+2011-11-30  Jakub Jelinek  <jakub@redhat.com>
+
+       PR rtl-optimization/48721
+       * sched-deps.c (sched_analyze_insn): For SIBLING_CALL_P set
+       reg_pending_barrier to TRUE_BARRIER.
+
+       PR rtl-optimization/48721
+       * sched-ebb.c (begin_move_insn): Insert empty unreachable
+       block after BARRIER if insn is followed by it.
+
+2011-11-30  Richard Henderson  <rth@redhat.com>
+
+       * config/sparc/sparc-opts.h (enum sparc_memory_model_type): New.
+       * config/sparc/sparc.opt (mmemory-model=): New option.
+       * doc/invoke.texi (Sparc Options): Document it.
+       * config/sparc/sparc.c (sparc_option_override): Provide default
+       for sparc_memory_model.
+       (sparc_emit_membar_for_model): Omit barrier combinations that are
+       implied by the memory model.
+
+2011-11-30  Richard Henderson  <rth@redhat.com>
+
+       * config/sparc/sync.md (I124MODE): New mode iterator.
+       (memory_barrier): Remove.
+       (atomic_exchangesi): New.
+       (swapsi): Rename from *swapsi.
+       (atomic_test_and_set<I124MODE>): New.
+       (ldstubqi): Rename from *ldstubqi, merge with expander.
+       (ldstub<I24MODE>): Rename from *ldstub<I24MODE>, merge with expander.
+       (sync_lock_test_and_setsi): Remove.
+       (sync_lock_test_and_set<I12MODE>): Remove.
+
+2011-11-30  Richard Henderson  <rth@redhat.com>
+
+       * config/sparc/constraints.md ("w"): New.
+       * config/sparc/predicates.md (mem_noofs_operand): New.
+       * config/sparc/sparc.c (sparc_expand_compare_and_swap_12): Make static.
+       Generate the boolean result of the atomic_compare_exchange.
+       (sparc_expand_compare_and_swap): New.
+       * config/sparc/sync.md (atomic_compare_and_swap<I>): New.
+       (atomic_compare_and_swap<I48MODE>_1): Rename from
+       *sync_compare_and_swap<I48MODE>, use mem_noofs_operand.
+       (*atomic_compare_and_swapdi_v8plus): Rename from
+       *sync_compare_and_swapdi_v8plus, use mem_noofs_operand.
+       (sync_compare_and_swap<I12MODE>): Remove.
+       (sync_compare_and_swap<I48MODE>): Remove.
+
+2011-11-30  Richard Henderson  <rth@redhat.com>
+
+       * config/sparc/predicates.md (register_or_v9_zero_operand): New.
+       * config/sparc/sparc.md (UNSPEC_ATOMIC): New.
+       * config/sparc/sync.md (atomic_load<I>): New.
+       (atomic_loaddi_1, atomic_store<I>, atomic_storedi_1): New.
+
+2011-11-30  Richard Henderson  <rth@redhat.com>
+
+       * config/sparc/predicates.md (zero_or_v7_operand): New.
+       * config/sparc/sparc.c (sparc_emit_membar_for_model): New.
+       * config/sparc/sparc-protos.h: Update.
+       * config/sparc/sync.md (mem_thread_fence): New.
+       (memory_barrier): Use sparc_emit_membar_for_model.
+       (membar, *membar_empty, *membar_storestore, *membar_storeload): New.
+       (*membar_v8): Accept and ignore the membar mask.
+       (*membar): Accept and print the membar mask.
+
+2011-11-30  Matthew Gretton-Dann  <matthew.gretton-dann@arm.com>
+
+       * config/arm/arm.c (arm_issue_rate): Cortex-A15 can triple issue.
+       * config/arm/arm.md (mul64): New attribute.
+       (generic_sched): Cortex-A15 is not scheduled generically.
+       (cortex-a15.md): Include.
+       * config/arm/cortex-a15.md: New machine description.
+       * config/arm/t-arm (MD_INCLUDES): Add cortex-a15.md.
+
+2011-11-30  Matthew Gretton-Dann  <matthew.gretton-dann@arm.com>
+
+       * config/arm/t-arm (MD_INCLUDES): Ensure all md files are listed.
+
+2011-11-30  Iain Sandoe  <iains@gcc.gnu.org>
+
+       * doc/tm.texi.in (NEXT_OBJC_RUNTIME): Update macro description.
+       * doc/tm.texi: Regenerate.
+       * flags.h (flag_next_runtime): Remove references.
+       * toplev.c: Likewise.
+       * defaults.h (NEXT_OBJC_RUNTIME): Provide default.
+       * gccspec.c (lang_specific_driver): Adjust use of NEXT_OBJC_RUNTIME.
+       * config/darwin.c (darwin_override_options): Provide default
+       Objective-C abi settings and target conflict checks.
+       * config/darwin.h (SUBTARGET_C_COMMON_OVERRIDE_OPTIONS): Remove unused
+       code. Adjust indenting.
+
+2011-11-29  Joseph Myers  <joseph@codesourcery.com>
+
+       * config/sh/sh.h (enum reg_class, REG_CLASS_NAMES)
+       (REG_CLASS_CONTENTS): Add NON_SP_REGS.
+       (REGCLASS_HAS_GENERAL_REG): Handle NON_SP_REGS.
+       * config/sh/constraints.md (u): New constraint.
+
+2011-11-29  Oleg Endo  <oleg.endo@t-online.de>
+
+       PR target/51337
+       * config/sh/sh.c (sh_secondary_reload): Add case when FPUL
+       register is being loaded from a pseudo in memory.
+
+2011-11-29  DJ Delorie  <dj@redhat.com>
+
+       * config.gcc (rl78-*-elf): New case.
+       * doc/extend.texi: Add RL78 documentation.
+       * doc/invoke.texi: Likewise.
+       * doc/md.texi: Likewise.
+       * doc/contrib.texi: Add RL78.
+       * doc/install.texi: Add rl78-*-elf.
+       * config/rl78: New directory for the Renesas RL78.
+
+2011-11-29  Jakub Jelinek  <jakub@redhat.com>
+
+       PR tree-optimization/51247
+       * tree-vrp.c (extract_range_from_assert): For signed 1-bit precision
+       types instead of adding 1 subtract -1 and instead of subtracting 1
+       add -1 to avoid overflows.
+
+2011-11-29  Andrew MacLeod  <amacleod@redhat.com>
+
+       PR target/50123
+       * optabs.c (maybe_optimize_fetch_op): New.  Look for more optimal
+       instructions for a FECTH_OP or OP_FECTH sequence.
+       (expand_atomic_fetch_op): Call maybe_optimize_fetch_op.
+
+2011-11-29  Uros Bizjak  <ubizjak@gmail.com>
+
+       * config/i386/sync.md (UNSPEC_LDA, UNSPEC_STA): New unspecs.
+       (movdi_via_fpu): Remove.
+       (loaddi_via_fpu): New insn pattern.
+       (storedi_via_fpu): Ditto.
+       (atomic_loaddi_fpu): Use loaddi_via_fpu and storedi_via_fpu.
+       (atomic_storedi_fpu): Ditto.
+       * reg-stack.c (get_true_reg): Handle UNSPEC_LDA.
+       (subst_stack_regs_pat): Handle UNSPEC_STA.
+
+2011-11-29  Uros Bizjak  <ubizjak@gmail.com>
+
+       * config/i386/i386.md (*floathi<mode>2_i387_with_temp): Do not
+       allocate scratch memory for alternative 0.
+
+2011-11-29  Michael Meissner  <meissner@linux.vnet.ibm.com>
+
+       * config/rs6000/rs6000-builtins.def: Completely rewrite builtin
+       handling to centralize more of the builtins in this file.  Change
+       some builtin enumerations to be more consistant.  Use a new mask
+       to hold the current builtins, including SPE and PAIRED builtins
+       which no longer are set via target_flags masks.  Add
+       -mdebug=builtin debug support.  For power machines, define all
+       Altivec and VSX buitins when the compiler starts, but don't allow
+       the use of a builtin unless the appropriate switch is used, or
+       #pragma GCC target is used to change the options.  If the user
+       uses #pragma GCC target, update the appropriate hardware macros.
+       * common/config/rs6000/rs6000-common.c (rs6000_handle_option): Ditto.
+       * config/rs6000/rs6000.opt (rs6000_builtin_mask): Ditto.
+       * config/rs6000/rs6000-c.c (rs6000_macro_to_expand): Ditto.
+       (rs6000_define_or_undefine_macro): Ditto.
+       (rs6000_target_modify_macros): Ditto.
+       (rs6000_cpu_cpp_builtins): Ditto.
+       (altivec_overloaded_builtins): Ditto.
+       (altivec_build_resolved_builtin): Ditto.
+       * config/rs6000/rs6000.c (rs6000_target_modify_macros_ptr): Ditto.
+       (rs6000_builtin_info): Ditto.
+       (def_builtin): Ditto.
+       (enable_mask_for_builtins): Ditto.
+       (DEBUG_FMT_X): Ditto.
+       (rs6000_debug_reg_global): Ditto.
+       (rs6000_builtin_mask_calculate): Ditto.
+       (rs6000_option_override_internal): Ditto.
+       (rs6000_builtin_conversion): Ditto.
+       (rs6000_builtin_vectorized_function): Ditto.
+       (bdesc_3arg): Ditto.
+       (bdesc_dst): Ditto.
+       (bdesc_2arg): Ditto.
+       (builtin_description_predicates): Ditto.
+       (bdesc_altivec_preds): Ditto.
+       (bdesc_spe_predicates): Ditto.
+       (bdesc_spe_evsel): Ditto.
+       (bdesc_paired_preds): Ditto.
+       (bdesc_abs): Ditto.
+       (bdesc_1arg): Ditto.
+       (rs6000_overloaded_builtin_p): Ditto.
+       (rs6000_expand_unop_builtin): Ditto.
+       (bdesc_2arg_spe): Ditto.
+       (spe_expand_builtin): Ditto.
+       (rs6000_invalid_builtin): Ditto.
+       (rs6000_expand_builtin): Ditto.
+       (rs6000_init_builtins): Ditto.
+       (spe_init_builtins): Ditto.
+       (paired_init_builtins): Ditto.
+       (altivec_init_builtins): Ditto.
+       (builtin_function_type): Ditto.
+       (rs6000_common_init_builtins): Ditto.
+       (rs6000_builtin_reciprocal): Ditto.
+       (rs6000_builtin_mask_names): Ditto.
+       (rs6000_pragma_target_parse): Ditto.
+       (rs6000_function_specific_print): Ditto.
+       * config/rs6000/rs6000.h (MASK_DEBUG_BUILTIN): Ditto.
+       (MASK_DEBUG_ALL): Ditto.
+       (TARGET_DEBUG_BUILTIN): Ditto.
+       (TARGET_EXTRA_BUILTINS): Ditto.
+       (REGISTER_TARGET_PRAGMAS): Ditto.
+       (enum rs6000_btc): Ditto.
+       (RS6000_BTC_*): Ditto.
+       (RS6000_BTM_*): Ditto.
+       (enum rs6000_builtins): Ditto.
+       * config/rs6000/rs6000-protos.h (rs6000_overloaded_builtin_p): Ditto.
+       (rs6000_target_modify_macros): Ditto.
+       (rs6000_target_modify_macros_ptr): Ditto.
+
+       * config/rs6000/vector.md (unsfloat<VEC_int><mode>2): Use the
+       standard name for converting vector unsigned values to floating
+       point.  Delete the TARGET_VECTORIZE_BUILTIN_CONVERSION hook, which
+       is handled by machine independent code.
+       * config/rs6000/rs6000.c (TARGET_VECTORIZE_BUILTIN_CONVERSION): Ditto.
+       (rs6000_builtin_conversion): Ditto.
+
+2011-11-29  Dodji Seketeli  <dodji@redhat.com>
+
+       * tree-diagnostic (struct loc_t): Rename into struct loc_map_pair.
+       (maybe_unwind_expanded_macro_loc): Adjust.
+
+2011-11-29  David Edelsohn  <dje.gcc@gmail.com>
+           Aldy Hernandez  <aldyh@redhat.com>
+
+       * trans-mem.c (ipa_tm_create_version_alias): Mangle new_decl
+       if DECL_ONE_ONLY.
+       (ipa_tm_create_version): Same.
+
+2011-11-29  Bernd Schmidt  <bernds@codesourcery.com>
+
+       * haifa-sched.c (recompute_todo_spec): Simplify and correct the
+       code checking for a clobber of a condition register when deciding
+       whether to predicate.
+
+2011-11-29   Diego Novillo  <dnovillo@google.com>
+
+       * gimple.c (gimple_call_set_cannot_inline): Move from gimple.h.
+       Update field call_stmt_cannot_inline_p from call graph edge, if needed.
+       * gimple.h (gimple_call_set_cannot_inline): Move to gimple.c.
+
+2011-11-29  Sameera Deshpande  <sameera.deshpande@arm.com>
+
+       * config/arm/arm.md (arm_movdi): Update NEG_POOL_RANGE.
+       (movdf_soft_insn): Likewise.
+       * config/arm/fpa.md (thumb2_movdf_fpa): Likewise.
+       * config/arm/neon.md (neon_mov<mode>): Likewise.
+       * config/arm/vfp.md (movdi_vfp): Likewise.
+       (movdi_vfp_cortexa8): Likewise.
+       (movdf_vfp): Likewise.
+
+2011-11-29  Jakub Jelinek  <jakub@redhat.com>
+
+       PR middle-end/50074
+       * expr.c (expand_expr_addr_expr_1): Don't call force_operand for
+       EXPAND_SUM modifier.
+
+2011-11-29  Ira Rosen  <ira.rosen@linaro.org>
+
+       PR tree-optimization/51301
+       * tree-vect-patterns.c (vect_recog_over_widening_pattern): Check that
+       the last statement doesn't convert to a bigger type than the original
+       type of the computation.
+
+2011-11-28  Richard Henderson  <rth@redhat.com>
+
+       * config/rs6000/rs6000.c (rs6000_expand_atomic_compare_and_swap):
+       Handle overlap between retval and oldval.  Always compute boolval
+       from CR0 EQ value.
+
+2011-11-28  Jakub Jelinek  <jakub@redhat.com>
+
+       PR debug/50317
+       * tree-ssa.c (execute_update_addresses_taken): Remove
+       var ={v} {CLOBBER} stmts instead of rewriting them into
+       var_N ={v} {CLOBBER}.
+
+       PR middle-end/50907
+       * function.c (convert_jumps_to_returns): When redirecting an edge
+       succ to EXIT_BLOCK_PTR, clear EDGE_CROSSING flag.
+
+       PR tree-optimization/50078
+       * tree-ssa-forwprop.c (forward_propagate_addr_expr_1): Copy over
+       TREE_THIS_VOLATILE also from the old to new lhs resp. rhs.
+
+       PR tree-optimization/50682
+       * tree-eh.c (maybe_remove_unreachable_handlers): New function.
+       * tree-flow.h (maybe_remove_unreachable_handlers): New prototype.
+       * tree-optimize.c (execute_cleanup_cfg_post_optimizing): Call it.
+
+2011-11-28  Georg-Johann Lay  <avr@gjlay.de>
+
+       * doc/extend.texi (AVR Built-in Functions): Add documentation for
+       __builtin_avr_map8 and __builtin_avr_map16.
+
+       * config/avr/avr.md: Document new %t and %T asm output codes.
+       (define_c_enum "unspec"): Add UNSPEC_MAP_BITS.
+       (adjust_len): Add map_bits.
+       (map_bitsqi, map_bitshi): New insns.
+       * config/avr/avr-protos.h (avr_out_map_bits): New.
+       * config/avr/avr-protos.c (print_operand): Implement %t and %T.
+       (adjust_insn_length): Handle ADJUST_LEN_MAP_BITS.
+       (avr_double_int_push_digit): New function.
+       (avr_map, avr_revert_map, avr_swap_map, avr_id_map): New functions.
+       (avr_sig_map, avr_map_hamming_byte): New functions.
+       (avr_out_swap_bits, avr_out_revert_bits, avr_move_bits,
+       avr_out_map_bits): New functions.
+       (enum avr_builtin_id): Add AVR_BUILTIN_MAP8, AVR_BUILTIN_MAP16.
+       (avr_init_builtins): Populate __builtin_avr_map8, __builtin_avr_map16.
+       (bdesc_2arg): Add __builtin_avr_map8, __builtin_avr_map16 ...
+       (avr_expand_builtin): ...and expand them.
+       * config/avr/avr-c.c (avr_cpu_cpp_builtins): New built-in defines:
+       __BUILTIN_AVR_MAP8, __BUILTIN_AVR_MAP16.
+
+2011-11-27  Richard Sandiford  <rdsandiford@googlemail.com>
+
+       PR target/51278
+       * config/mips/mips.c (mips_rtx_costs): Handle baddu.
+
+2011-11-27  Richard Sandiford  <rdsandiford@googlemail.com>
+
+       * config/mips/mips.c (mips16_unextended_reference_p): Test for BLKmode.
+
+2011-11-27  Richard Sandiford  <rdsandiford@googlemail.com>
+
+       * config/mips/mips-protos.h (mips_emit_binary): Declare.
+       * config/mips/mips.c (mips_emit_binary): Make global.
+       (mips_set_mips16_mode): Turn off -mfix-r4000 in MIPS16 mode.
+       (mips_conditional_register_usage): Don't treat LO and HI as
+       register operands in MIPS16 mode.
+       (mips_mulsidi3_gen_fn): Use {u,}mulsidi3_{32,64}bit_mips16
+       for MIPS16 code.
+       * config/mips/predicates.md (muldiv_target_operand): New predicate.
+       (move_operand): Allow hilo_operand.
+       * config/mips/mips.md (mul<mode>3): Explicitly specify LO as the
+       target of MIPS16 multiplies, then move it into the target register.
+       (mul<mode>3_internal, *macc2, *msac2): Use muldiv_target_operand.
+       (<u>mulsidi3_32bit_mips16): New expander.
+       (<u>mulsidi3_32bit): Use muldiv_target_operand.
+       (<u>mulsidi3_32bit_r4000): Disable for ISA_HAS_DSP.
+       (<u>mulsidi3_64bit): Require !TARGET_MIPS16.  Split into
+       <u>mulsidi3_64bit_split.
+       (<u>mulsidi3_64bit_mips16): New expander.
+       (<u>mulsidi3_64bit_split): Likewise, using expansions from
+       two previous define_splits.
+       (<u>mulsidi3_64bit_hilo, *muls<u>_di, <u>msubsidi4): Use
+       muldiv_target_operand.
+       (<su>mulsi3_highpart): Use <su>mulsi3_highpart_split for MIPS16 code.
+       (<su>mulsi3_highpart_internal): Require !TARGET_MIPS16.
+       Split into <su>mulsi3_highpart_split.
+       (<su>mulsi3_highpart_split): New expander.
+       (<su>muldi3_highpart): Turn into a define_expand.
+       Use <su>muldi3_highpart_split for MIPS16 code.
+       (<su>muldi3_highpart_internal): Renamed from <su>muldi3_highpart.
+       Require !TARGET_MIPS16.  Split into <su>muldi3_highpart_split.
+       (<su>muldi3_highpart_split): New expander.
+       (<u>mulditi3): Explicitly specify LO as the target of MIPS16
+       multiplies, then move it into the target register.
+       (<u>mulditi3_internal, <u>maddsidi4): Use muldiv_target_operand.
+       (divmod<mode>4, udivmod<mode>4): Turn into define_expands.
+       Use <u>divmod<mode>4_split for MIPS16 code, then explicitly
+       move LO into operand 0.
+       (divmod<mode>4_internal, udivmod<mode>4_internal): Renamed
+       from <u>divmod<mode>4.  Use muldiv_target_operand.
+       Require !TARGET_MIPS16.  Split into <u>divmod<mode>4_split.
+       (<u>divmod<mode>4_split): New expander.
+       (<u>divmod<GPR:mode>4_hilo_<HILO:mode>): Use muldiv_target_operand.
+       (mfhi<GPR:mode>_<HILO:mode>): Use hilo_operand.
+
+2011-11-27  Richard Sandiford  <rdsandiford@googlemail.com>
+
+       * hard-reg-set.h (target_hard_regs): Add x_accessible_reg_set
+       and x_operand_reg_set.
+       (accessible_reg_set, operand_reg_set): New macros.
+       * reginfo.c (init_reg_sets): Initialize accessible_reg_set and
+       operand_reg_set.
+       (saved_accessible_reg_set, saved_operand_reg_set): New variables.
+       (save_register_info): Save them.
+       (restore_register_info): Restore them.
+       (init_reg_sets_1): Limit operand_reg_set to accessible_reg_set.
+       Remove NO_REGS registers from operand_reg_set.  Treat members
+       of operand_reg_set as fixed.
+       * recog.c (general_operand): Check operand_reg_set rather than NO_REGS.
+       (register_operand, nonmemory_operand): Likewise.
+       * varasm.c (make_decl_rtl): Always use DECL_MODE as the mode of
+       register variables.  Check accessible_reg_set and operand_reg_set.
+       * config/mips/mips.c (mips_conditional_register_usage): Remove
+       inaccessible register from accessible_reg_set, rather than just
+       making them fixed.
+
+2011-11-27  Gerald Pfeifer  <gerald@pfeifer.com>
+
+       * config/pa/pa-linux.h (TARGET_GAS): Remove comment.
+
+2011-11-26  Richard Henderson  <rth@redhat.com>
+
+       * config/arm/arm.c (arm_init_libfuncs): Call init_sync_libfuncs
+       for Linux.
+
+2011-11-26  Richard Henderson  <rth@redhat.com>
+
+       * optabs.c (expand_atomic_fetch_op): Always return result.
+
+       * optabs.c (init_sync_libfuncs_1): Include max in iteration.
+
+       * omp-low.c (expand_omp_atomic): Assume anything aligned to
+       BIGGEST_ALIGNMENT is aligned.
+
+2011-11-26  Richard Henderson  <rth@redhat.com>
+
+       * config/m68k/m68k.md (UNSPECV_CAS_1, UNSPECV_CAS_2): New.
+       (UNSPECV_TAS_1, UNSPECV_TAS_2): New.
+       (I): New mode iterator.
+       (xz): New mode attribute.
+       * config/m68k/sync.md: New file.
+
+       * config/m68k/linux.h (TARGET_INIT_LIBFUNCS): New.
+       * config/m68k/m68k.c (m68k_init_sync_libfuncs): New.
+
+2011-11-26  Nathan Sidwell  <nathan@acm.org>
+
+       PR gcov-profile/51297
+       * gcov.c (main): Allocate initial names and sources arrays.
+       (find_source): Don't check for null name or source arrays here.
+
+2011-11-26  Jakub Jelinek  <jakub@redhat.com>
+
+       PR rtl-optimization/49912
+       * cfgrtl.c (rtl_verify_flow_info_1): Ignore also EDGE_PRESERVE bit
+       when counting n_branch.
+
+2011-11-26  Uros Bizjak  <ubizjak@gmail.com>
+
+       * config/i386/sync.md (movdi_via_fpu): Add %Z insn suffixes.
+
+2011-11-26  Joern Rennecke  <joern.rennecke@embecosm.com>
+
+       PR middle-end/50074
+       * calls.c (mem_overlaps_already_clobbered_arg_p):
+       Return false if no outgoing arguments have been stored so far.
+
+2011-11-25  Georg-Johann Lay  <avr@gjlay.de>
+
+       * config/avr/avr.c (print_operand): Support code = 'i' for CONST_INT.
+
+2011-11-24  Andrew MacLeod  <amacleod@redhat.com>
+
+       PR other/51011
+       * tree.h (is_builtin_name): No longer external.
+       * builtins.c (is_builtin_name): Make static.
+       * varasm.c (incorporeal_function_p): __sync and __atomic external calls
+       are not incorporeal and may need asm label processing.
+
+2011-11-25  Georg-Johann Lay  <avr@gjlay.de>
+
+       PR target/50566
+       * config/avr/avr-protos.h (avr_log_t): Add field .builtin.
+       * config/avr/avr-log.c (avr_log_set_avr_log): Initialize it.
+       Don't bypass TARGET_ALL_DEBUG.  Print self-info with ?.
+       (avr_log_vadump): Support %D and %X to print double_int.
+       (avr_double_int_pop_digit): New static function.
+       (avr_dump_double_int_hex): New static function.
+
+2011-11-24  Enkovich Ilya  <ilya.enkovich@intel.com>
+
+       PR target/51287
+       * config/i386/i386.c (distance_non_agu_define_in_bb): Fix insn
+       attr check.
+
+2011-11-24  Andrew MacLeod  <amacleod@redhat.com>
+
+       * optab.c (maybe_emit_atomic_exchange): New.  Try to emit an
+       atomic_exchange pattern.
+       (maybe_emit_sync_lock_test_and_set): New.  Try to emit an exchange
+       using __sync_lock_test_and_set.
+       (maybe_emit_compare_and_swap_exchange_loop): New. Try to emit an
+       exchange using a compare_and_swap loop.
+       (expand_sync_lock_test_and_set): New.  Expand sync_lock_test_and_set.
+       (expand_atomic_test_and_set): New.  Expand test_and_set operation.
+       (expand_atomic_exchange): Use new maybe_emit_* functions.
+       (expand_atomic_store): Use new maybe_emit_* functions.
+       * builtins.c (expand_builtin_sync_lock_test_and_set): Call
+       expand_sync_lock_test_and_set routine.
+       (expand_builtin_atomic_exchange): Remove parameter from call.
+       (expand_builtin_atomic_clear): Use atomic_clear pattern if present.
+       (expand_builtin_atomic_test_and_set): Add target and simply call
+       expand_atomic_test_and_set.
+       (expand_builtin): Add target to expand_builtin_atomic_test_and_set.
+       * expr.h (expand_atomic_exchange): Add parameter.
+       (expand_sync_lock_test_and_set): New prototype.
+       (expand_atomic_test_and_set, expand_atomic_clear): New prototypes.
+
+2011-11-24  H.J. Lu  <hongjiu.lu@intel.com>
+
+       PR target/51134
+       * config/i386/i386.h (processor_costs): Revert revision 181357.
+       * config/i386/i386.c (cost models): Likewise.
+       (core_cost): Likewise.
+       (promote_duplicated_reg): Likewise.
+       (promote_duplicated_reg_to_size): Likewise.
+       (processor_target): Likewise.
+       (expand_set_or_movmem_via_loop_with_iter): Likewise.
+       (expand_set_or_movmem_via_loop): Likewise.
+       (emit_strset): Likewise.
+       (expand_movmem_epilogue): Likewise.
+       (expand_setmem_epilogue): Likewise.
+       (expand_movmem_prologue): Likewise.
+       (expand_setmem_prologue): Likewise.
+       (expand_constant_movmem_prologue): Likewise.
+       (expand_constant_setmem_prologue): Likewise.
+       (decide_alg): Likewise.
+       (decide_alignment): Likewise.
+       (ix86_expand_movmem): Likewise.
+       (ix86_expand_setmem): Likewise.
+       (ix86_slow_unaligned_access): Likewise.
+       * config/i386/i386.md (strset): Likewise.
+       * config/i386/sse.md (vec_dupv4si): Likewise.
+       (vec_dupv2di): Likewise.
+
+2011-11-24  Georg-Johann Lay  <avr@gjlay.de>
+
+       * config/avr/avr.md (*rotlhi2.15): Set length to 4.
+
+2011-11-24  Razya Ladelsky  <razya@il.ibm.com>
+
+       * tree-data-ref.c (initialize_data_dependence_relation): Update
+       comment for the self dependence case.
+       (compute_self_dependence): Remove.
+       * tree-vect-data-refs.c (vect_analyze_data_refs): Remove call to
+       compute_self_dependenc.
+
+2011-11-24  Alan Modra  <amodra@gmail.com>
+
+       * c-decl.c (warn_if_shadowing): Don't warn if shadowed
+       identifier is from system header.
+
+2011-11-23  Jakub Jelinek  <jakub@redhat.com>
+
+       PR target/51261
+       * config/i386/i386.c (decide_alg): Initialize *dynamic_check
+       even if !optimize.
+
+2011-11-23  Vladimir Makarov  <vmakarov@redhat.com>
+
+       PR rtl-optimization/48455
+       * doc/invoke.texi (-fira-region): Document default values.
+
+       * flags-types.h (enum ira_region): Add new value
+       IRA_REGION_AUTODETECT.
+
+       * common.opt (fira-region): Set up initial value to
+       IRA_REGION_AUTODETECT.
+
+       * toplev.c (process_options): Set up flag_ira_region depending on
+       -O options.
+
+       * ira.c (ira.c): Remove optimize guard for ira_build.
+
+2011-11-23  Chung-Lin Tang  <cltang@codesourcery.com>
+
+       PR rtl-optimization/50496
+       * jump.c (redirect_jump): Assert fail on nlabel == NULL_RTX
+       only after epilogue is created. Add comments.
+
+2011-11-22  Richard Henderson  <rth@redhat.com>
+
+       * config/ia64/ia64.c (ia64_expand_atomic_op): Add model parameter.
+       Generate the barrier required for the memory model.
+       (rtx_needs_barrier): Handle UNSPEC_FETCHADD_REL, UNSPEC_CMPXCHG_REL.
+       * config/ia64/ia64-protos.h: Update.
+       * config/ia64/ia64.md (UNSPEC_FETCHADD_REL): New.
+       (UNSPEC_CMPXCHG_REL): New.
+       * config/ia64/sync.md (mem_thread_fence): New.
+       (atomic_load<IMODE>, atomic_store<IMODE>): New.
+       (atomic_compare_and_swap<IMODE>): New.
+       (cmpxchg_acq_<I124MODE>, cmpxchg_acq_di): New.
+       (atomic_exchange<IMODE>): New.
+       (xchg_acq_<IMODE>): Rename from sync_lock_test_and_set<IMODE>.
+       (atomic_<FETCHOP><IMODE>, atomic_nand<IMODE>): New.
+       (atomic_fetch_<FETCHOP><IMODE>, atomic_fetch_nand<IMODE>): New.
+       (atomic_<FETCHOP>_fetch<IMODE>, atomic_nand_fetch<IMODE>): New.
+       (fetchadd_rel_<I48MODE>): New.
+       (sync_<FETCHOP><IMODE>, sync_nand<IMODE>): Remove.
+       (sync_old_<FETCHOP><IMODE>, sync_old_nand<IMODE>): Remove.
+       (sync_new_<FETCHOP><IMODE>, sync_new_nand<IMODE>): Remove.
+       (sync_compare_and_swap<IMODE>): Remove.
+       (sync_lock_release<IMODE>): Remove.
+
+       * config/ia64/ia64.md: Use define_c_enum for UNSPEC_*
+       and UNSPECV_* constants.
+
+2011-11-22  Tom de Vries  <tom@codesourcery.com>
+
+       PR rtl-optimization/50764
+       * sched-deps.c (sched_analyze_insn): Make sure frame-related insns are
+       not moved past preceding jump.
+
+2011-11-22  Jeff Law  <law@redhat.com>
+
+       * doc/contrib.texi: Add entry for David Binderman.
+
+2011-11-22  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>
+
+       * postreload.c (reload_combine): Mark reg_state as invalid at
+       volatile insns if there has been a use already.
+
+2011-11-21  Aldy Hernandez  <aldyh@redhat.com>
+
+       * opts.c (finish_options): Do not fail for -fgnu-tm.
+       * gimple-streamer-out.c (output_gimple_stmt): Handle
+       GIMPLE_TRANSACTION.
+       * gimple-streamer-in.c (input_gimple_stmt): Same.
+       * lto-cgraph.c (input_overwrite_node): Read tm_clone bit.
+       (lto_output_node): Write tm_clone bit.
+
+2011-11-22  Ian Lance Taylor  <iant@google.com>
+
+       * doc/install.texi (Configuration): Correct doc of
+       --enable-build-poststage1-with-cxx: it is not experimental.
+
+2011-11-22  Jakub Jelinek  <jakub@redhat.com>
+
+       PR tree-optimization/51074
+       * fold-const.c (fold_binary_loc): Fix up VEC_INTERLEAVE_*_EXPR
+       handling for BYTES_BIG_ENDIAN.
+       * optabs.c (can_vec_perm_for_code_p): Likewise.
+
 2011-11-22  Michael Matz  <matz@suse.de>
 
        PR other/51125
 
        PR c++/51264
        * tree.c (iterative_hash_expr): Remove break after return.
-       
+
 2011-11-22  Iain Sandoe  <iains@gcc.gnu.org>
 
        * target.def (tm_clone_table_section): New hook.
 
 2011-11-17  Richard Henderson  <rth@redhat.com>
 
+       * config/rs6000/rs6000.c (rs6000_expand_atomic_compare_and_swap): Get
+       new pseudo for target after convert_modes.
+       (rs6000_expand_atomic_exchange, rs6000_expand_atomic_op): Likewise.
+
+2011-11-17  Richard Henderson  <rth@redhat.com>
+
        * builtins.c (expand_builtin_mem_thread_fence): Remove.
        (expand_builtin_mem_signal_fence): Remove.
        (expand_builtin_atomic_thread_fence): Use expand_mem_thread_fence.