Paolo Carlini [Tue, 18 May 2004 15:58:33 +0000 (15:58 +0000)]
[multiple changes]
2004-05-18 Paolo Carlini <pcarlini@suse.de>
* include/ext/mt_allocator.h:(__mt_alloc::allocate): Minor
tweaks.
2004-05-18 Dhruv Matani <dhruvbird@gmx.net>
* include/ext/mt_allocator.h:(__mt_alloc::allocate): Re-write
allocation loop which removes blocks from the global free list
from O(N) to O(1) when the required blocks are <= the number
available.
Steven Bosscher [Tue, 18 May 2004 14:43:58 +0000 (14:43 +0000)]
s390.c (s390_expand_movstr, [...]): Do not use expand_start_loop and expand_end_loop, instead build the loop manually.
* config/s390/s390.c (s390_expand_movstr, s390_expand_clrstr,
s390_expand_cmpmem): Do not use expand_start_loop and
expand_end_loop, instead build the loop manually.
PR libstdc++/14340
* include/debug/safe_iterator.h (_Safe_iterator converting
constructor): Only allow declaration to instantiate when the
incoming _Safe_iterator has exactly the right iterator type.
Jeff Law [Tue, 18 May 2004 02:53:55 +0000 (20:53 -0600)]
toplev.h (flag_delete_null_pointer_checks): Move from here to...
* toplev.h (flag_delete_null_pointer_checks): Move from here to...
* flags.h (flag_delete_null_pointer_checks): Here.
* tree-flow.h (cprop_into_successor_phis): Add argument to prototype.
* tree-phinodes.c (resize_phi_node): Initialize PHI_ARG_NONZERO.
(add_phi_arg, remove_phi_arg_num): Similarly.
* tree-ssa-copy.c (cprop_into_successor_phis): Propagate nonzero
property into PHI nodes.
* tree-ssa-dom.c: Remove redundant inclusion of flags.h.
(record_equivalences_from_phis): If all PHI arguments are known to be
nonzero, then the result must be nonzero as well.
(cprop_into_phis): Pass nonzero_vars bitmap to cprop_into_successor_phis (record_equivalences_from_stmt): Check flag_delete_null_pointer_checks
appropriately. Walk the USE-DEF chains and propagate nonzero property
as appropriate.
* tree.h (PHI_ARG_NONZERO): Define.
(phi_arg_d): Add nonzero flag.
Zack Weinberg [Tue, 18 May 2004 01:26:21 +0000 (01:26 +0000)]
Makefile.def, [...]: Remove all mention of libf2c.
top:
* Makefile.def, Makefile.tpl, configure.in: Remove all mention
of libf2c.
* configure, Makefile.in: Regenerate.
contrib:
* gcc_update: Remove gcc/f/intdoc.texi and all libf2c files
from list of files to be touched.
* convert_to_f2c, convert_to_g2c, download_f2c: Delete.
gcc:
* f: Entire directory removed
* c-common.h (CTI_G77_INTEGER_TYPE, CTI_G77_UINTEGER_TYPE)
(CTI_G77_LONGINT_TYPE, CTI_G77_ULONGINT_TYPE)
(g77_integer_type_node, g77_uinteger_type_node)
(g77_longint_type_node, or g77_ulongint_type_node): Delete.
* c-common.c (c_common_nodes_and_builtins): Do not initialize
the above set of variables.
* config/i386/uwin.h: No need to define WIN32_UWIN_TARGET.
* doc/invoke.texi, doc/standards.texi: Remove cross-references
to g77 manual.
gcc/po:
* exgettext (spec_error_string): Do not scan beyond the end of
the string for a close brace. Do not bail out at the first
incidence of %%e.
* gcc.pot: Regenerate.
Frank Ch. Eigler [Mon, 17 May 2004 20:34:19 +0000 (20:34 +0000)]
libmudflap.exp (libmudflap-init): For C++ test cases only...
2004-05-17 Frank Ch. Eigler <fche@redhat.com>
* lib/libmudflap.exp (libmudflap-init): For C++ test cases only,
import some build settings from libstdc++-v3 testsuite_flags.
* .../cfrags.exp, .../c++frags.exp, .../cthfrags.exp: Corresponding
changes to pass test language.
* mf-runtime.c (__mfu_check): Poison the cache with antidote for
quicker mode-nop handling.
H.J. Lu [Mon, 17 May 2004 14:30:18 +0000 (14:30 +0000)]
re PR target/15084 (-O/-O2 generate wrong code on structure modification)
2004-05-17 H.J. Lu <hongjiu.lu@intel.com>
PR target/15084
* config/i386/i386.md (*movsi_insv_1_rex64): Changed to DImode
and renamed to movdi_insv_1_rex64.
(insv): Support SImode for 32bit and DImode for 64bit.
* gigi.h: (adjust_decl_rtl, kept_level_p, set_block): Deleted.
(add_decl_stmt, add_stmt, block_has_vars): New functions.
(gnat_pushlevel, gnat_poplevel): Renamed from pushlevel and poplevel.
* decl.c (elaborate_expression, maybe_pad_type): Call add_decl_stmt
when making a decl.
(gnat_to_gnu_entity): Likewise.
Use add_stmt to update setjmp buffer.
Set TREE_ADDRESSABLE instead of calling put_var_into_stack and
flush_addressof.
No longer call adjust_decl_rtl.
(DECL_INIT_BY_ASSIGN_P): New macro.
(DECL_STMT_VAR): Likewise.
* trans.c (gigi): Call start_block_stmt to make the outermost
BLOCK_STMT.
(gnat_to_code, gnu_to_gnu, tree_transform, process_decls, process_type):
Call start_block_stmt and end_block_stmt temporarily.
Use gnat_expand_stmt instead of expand_expr_stmt.
(add_decl_stmt): New function.
(tree_transform): Call it.
(add_stmt): Also emit initializing assignment for DECL_STMT if needed.
(end_block_stmt): Set type and NULL_STMT.
(gnat_expand_stmt): Make recursize call instead of calling
expand_expr_stmt.
(gnat_expand_stmt, case DECL_STMT): New case.
(set_lineno_from_sloc): Do nothing if global.
(gnu_block_stmt_node, gnu_block_stmt_free_list): New variables.
(start_block_stmt, add_stmt, end_block_stmt): New functions.
(build_block_stmt): Call them.
(gnat_to_code): Don't expand NULL_STMT.
(build_unit_elab): Rename pushlevel and poplevel to gnat_* and change
args.
(tree_transform): Likewise.
(tree_transform, case N_Null_Statement): Return NULL_STMT.
(gnat_expand_stmt, case NULL_STMT): New case.
(gnat_expand_stmt, case IF_STMT): Allow nested IF_STMT to have no
IF_STMT_TRUE.
* utils2.c (gnat_mark_addressable, case VAR_DECL): Do not set
TREE_ADDRESSABLE.
* utils.c (create_var_decl): Do not call expand_decl or
expand_decl_init.
Set TREE_ADDRESSABLE instead of calling gnat_mark_addressable.
Set DECL_INIT_BY_ASSIGN_P when needed and do not generate MODIFY_EXPR
here.
(struct e_stack): Add chain_next to GTY.
(struct binding_level): Deleted.
(struct ada_binding_level): New struct.
(free_block_chain): New.
(global_binding_level, clear_binding_level): Deleted.
(global_bindings_p): Rework to see if no chain.
(kept_level_p, set_block): Deleted.
(gnat_pushlevel): Renamed from pushlevel and extensive reworked to use
new data structure and work directly on BLOCK node.
(gnat_poplevel): Similarly.
(get_decls): Look at BLOCK_VARS.
(insert_block): Work directly on BLOCK node.
(block_has_var): New function.
(pushdecl): Rework for new binding structures.
(gnat_init_decl_processing): Rename and rework calls to pushlevel and
poplevel.
(build_subprog_body): Likewise.
(end_subprog_body): Likewise; also set up BLOCK in DECL_INITIAL.
* ada-tree.def (DECL_STMT, NULL_STMT): New codes.
* ada-tree.h: (DECL_INIT_BY_ASSIGN_P): New macro.
(DECL_STMT_VAR): Likewise.
2004-05-17 Robert Dewar <dewar@gnat.com>
* restrict.ads, restrict.adb (Process_Restriction_Synonym): New
procedure
* sem_prag.adb (Analyze_Pragma, case Restrictions): Cleanup handling
of restriction synonyums by using
Restrict.Process_Restriction_Synonyms.
* snames.ads, snames.adb: Add entries for Process_Restriction_Synonym
Richard Earnshaw [Sun, 16 May 2004 22:22:49 +0000 (22:22 +0000)]
re PR target/10982 ([arm] poor optimisation of "if (var & const)")
PR target/10982
* arm.md (ne_zeroextractsi): Convert to insn-and-split.
(ne_zeroextractsi_shifted): New pattern.
(ite_ne_zeroextractsi): New pattern.
(ite_ne_zeroextractsi_shifted): New pattern.
* include/ext/mt_allocator.h (__mt_alloc<>::deallocate):
Consistently update __bin._M_free[0].
(__mt_alloc<>::allocate): When __bin._M_first[0] != NULL use
__bin._M_free[0] to simplify the while loop (i.e., the number
of iterations becomes known at the outset).
Tobias Schlüter [Sat, 15 May 2004 20:29:06 +0000 (22:29 +0200)]
re PR fortran/13742 (Not Implemented: initial values for COMMON or EQUIVALENCE)
PR fortran/13742
* decl.c (add_init_expr_to_sym): Verify that COMMON variable is
not initialized in a disallowed fashion.
* match.c (gfc_match_common): Likewise.
(var_element): Verify that variable is not in the blank COMMON,
if it is in a common.
Tobias Schlüter [Sat, 15 May 2004 17:31:32 +0000 (19:31 +0200)]
re PR fortran/13702 (When preprocessing Fortran files (.F, .F90 and .F95) cpp should emit line numbers.)
PR fortran/13702
(Port from g95)
* gfortran.h (gfc_linebuf): New typedef.
(linebuf): Remove.
(gfc_file): Revamped, use new gfc_linebuf.
(locus): Revamped, use new types.
(gfc_current_file): Remove.
(gfc_current_form, gfc_source_file): New global variables.
* match.c (gfc_match_space, gfc_match_strings): Use
gfc_current_form to find source form.
* module.c (gfc_dump_module): Use gfc_source_file when printing
module header.
* error.c (show_locus, show_loci) Use new data structures to print
locus.
* scanner.c (first_file, first_duplicated_file, gfc_current_file):
Remove.
(file_head, current_file, gfc_current_form, line_head, line_tail,
gfc_current_locus1, gfc_source_file): New global variables.
(gfc_scanner_init1): Set new global variables.
(gfc_scanner_done1): Free new data structures.
(gfc_current_locus): Return pointer to gfc_current_locus1.
(gfc_set_locus): Set gfc_current_locus1.
(gfc_at_eof): Set new variables.
(gfc_at_bol, gfc_at_eol, gfc_advance_line, gfc_next_char): Adapt
to new locus structure.
(gfc_check_include): Remove.
(skip_free_comments, skip_fixed_comments): Use gfc_current_locus1.
(gfc_skip_comments): Use gfc_current_form, find locus with
gfc_current_locus1.
(gfc_next_char): Use gfc_current_form.
(gfc_peek_char, gfc_gobble_whitespace): Use gfc_current_locus1.
(load_line): Use gfc_current_form. Recognize ^Z as EOF. Fix
comment formatting.
(get_file): New function.
(preprocessor_line, include_line): New functions.
(load_file): Move down, rewrite to match new data structures.
(gfc_new_file): Rewrite to match new data structures.
* parse.c (next_statement): Remove code which is now useless. Use
gfc_source_form and gfc_source_file where appropriate.
* trans-decl.c (gfc_get_label_decl): adapt to new data structures
when determining locus of frontend code.
* trans-io.c (set_error_locus): Same.
* trans.c (gfc_get_backend_locus, gfc_set_backend_locus): Likewise.
* lang-specs.h (@f77-cpp-input, @f95-cpp-input): Remove '-P' from
preprocessor flags.
(all): Add missing initializers.
Alan Modra [Sat, 15 May 2004 12:39:34 +0000 (12:39 +0000)]
rs6000.c (rs6000_va_arg <ABI_V4>): Don't use UNITS_PER_WORD to calculate gpr size.
* config/rs6000/rs6000.c (rs6000_va_arg <ABI_V4>): Don't use
UNITS_PER_WORD to calculate gpr size. Re-instate code to set reg
count to 8 to handle n_reg > 2.
* basic-block.h (life_analysis, delete_noop_moves):
Update prototypes.
* bt-load.c (branch_target_load_optimize): Don't take the
insns stream as an argument. Update the life_analysis calls.
* combine.c (combine_instructions): Update delete_noop_moves
calls.
* flow.c (notice_stack_pointer_modification): Don't take the
insns stream as an argument. Work on the flow graph.
(life_analysis): Likewise.
(delete_noop_moves): Likewise.
* passes.c (rest_of_handle_stack_regs): Update reg_to_stack call.
(rest_of_handle_life): Update life_analysis call.
(rest_of_compilation): Likewise, and also update
branch_target_load_optimize call.
* ra.c (reg_alloc): Update life_analysis call.
* reg-stack.c (reg_to_stack): Likewise. Also, don't take
the insns stream as an argument.
* regrename.c (copyprop_hardreg_forward): Update delete_noop_moves
call.
* rtl.c (branch_target_load_optimize, reg_to_stack): Update
prototypes.
* value-profile.c (branch_prob): Update life_analysis call.
* web.c (web_main): Work on the CFG, not on the insns stream.
Jeff Law [Sat, 15 May 2004 06:21:34 +0000 (00:21 -0600)]
tree-ssa-dom.c (redirect_edges_and_update_ssa_graph): Don't even bother marking bypassed virtuals for out-of-ssa.
* tree-ssa-dom.c (redirect_edges_and_update_ssa_graph): Don't even
bother marking bypassed virtuals for out-of-ssa. Instead merge
bypassed virtuals into vars_to_rename just before into-ssa pass.
Geoffrey Keating [Fri, 14 May 2004 22:33:28 +0000 (22:33 +0000)]
name-lookup.c (struct scope_binding): New.
* name-lookup.c (struct scope_binding): New.
(EMPTY_SCOPE_BINDING): New.
(lookup_using_namespace): Take a scope_binding instead of a
cxx_binding.
(qualified_lookup_using_namespace): Likewise.
(cxx_binding_clear): Delete.
(do_nonmember_using_decl): Use a scope_binding instead of a
cxx_binding.
(lookup_tag): Don't call select_decl.
(ambiguous_decl): Don't return anything (and change callers to match).
Take a scope_binding as the second parameter.
(lookup_namespace_name): Use a scope_binding instead of a
cxx_binding.
(unqualified_namespace_lookup): Likewise.
(lookup_qualified_name): Likewise.
(select_decl): Take a scope_binding instead of a cxx_binding.
Use macros rather than hand-coding tests for type-ness.
Steven Bosscher [Fri, 14 May 2004 18:29:09 +0000 (18:29 +0000)]
re PR rtl-optimization/14472 ([tree-ssa] [regression] no sib calling for enum to int functions)
PR opt/14472
* tree-tailcall.c (process_assignment): Use STRIP_NOPS to
ignore type conversions that do not inhibit tail calling.
(find_tail_calls): Likewise.
Jeff Law [Fri, 14 May 2004 17:51:05 +0000 (11:51 -0600)]
tree-ssa-dom.c (simplify_rhs_and_lookup_avail_expr): Reorganize so that it picks up more opportunities to eliminate ABS expressions...
* tree-ssa-dom.c (simplify_rhs_and_lookup_avail_expr): Reorganize
so that it picks up more opportunities to eliminate ABS expressions
or turn them into negations.
Andrew Pinski [Fri, 14 May 2004 15:27:37 +0000 (15:27 +0000)]
re PR tree-optimization/14466 (missed PHI optimization (different types))
2004-05-14 Andrew Pinski <pinskia@physics.uc.edu>
PR optimization/14466
* tree-complex.c (make_temp): Remove.
(gimplify_val): Replace make_temp with make_rename_temp
and add NULL as the second argument.
(expand_complex_div_wide): Likewise.
* tree-dfa.c (make_rename_temp): New function.
* tree-flow.h (make_rename_temp): Declare.
* tree-sra.c (make_temp): Remove.
(lookup_scalar): Replace make_temp with make_rename_temp.
(create_scalar_copies): Likewise.
* tree-ssa-phiopt.c (conditional_replacement): When we
get non gimple create a temporary variable to hold the
casted expression.
2004-05-14 Andrew Pinski <pinskia@physics.uc.edu>
PR optimization/14466
* gcc.dg/tree-ssa/20040514-1.c: New test.
* lib-writ.adb, lib-load.adb, lib.ads, lib.adb: Remove Dependent_Unit
flag processing. This was suppressing required dependencies in
No_Run_Time mode and is not needed since the binder does not generate
references for things in libgnat anyway.
* sem_ch3.adb (Access_Type_Declaration): Reorganize code to avoid GCC
warning.
2004-05-14 Thomas Quinot <quinot@act-europe.fr>
* gnat_ugn.texi: Document AIX-specific issue with initialization of
resolver library.
* exp_ch4.adb (Insert_Dereference_Action): Do not generate dereference
action for the case of an actual parameter in an init proc call.
2004-05-14 Ed Schonberg <schonberg@gnat.com>
* sem_ch4.adb (Analyze_Selected_Component): If prefix is a protected
subtype, check visible entities in base type.
* exp_ch7.adb (Clean_Simple_Protected_Objects): Do not generate cleanup
actions if the object is a renaming.
* sem_ch12.adb (Same_Instantiated_Entity): Predicate for
Check_Formal_Package_Instance, to determine more precisely when the
formal and the actual denote the same entity.
2004-05-14 Javier Miranda <miranda@gnat.com>
* par-ch10.adb (P_Context_Clause): Complete documentation on AI-262
* sem_ch10.adb (Analyze_With_Clause): After analyzed, the entity
corresponding to a private_with must be removed from visibility; it
will be made visible later, just before we analyze the private part of
the package.
(Check_Private_Child_Unit): Allow private_with clauses in public
siblings.
(Install_Siblings): Make visible the private entities of private-withed
siblings.
(Install_Withed_Unit): Do not install the private withed unit if we
are compiling a package declaration and the Private_With_OK flag was
not set by the caller. These declarations will be installed later,
just before we analyze the private part of the package.
* sem_ch3.adb (Analyze_Object_Declaration): In case of errors detected
during the evaluation of the expression that initializes the object,
decorate it with the expected type to avoid cascade errors.
Code cleanup.
* sem_ch6.adb (Analyze_Subprogram_Body): If we are compiling a library
subprogram we have to install the private_with clauses after its
specification has been analyzed (as documented in AI-262.TXT).
* sem_ch8.adb (Has_Private_With): New function. Determines if the
current compilation unit has a private with on a given entity.
(Find_Direct_Name): Detect the Beaujolais problem described in
AI-262.TXT
* sem_utils.ads, sem_util.adb (Is_Ancestor_Package): New function. It
provides the functionality of the function Is_Ancestor that was
previously available in sem_ch10. It has been renamed to avoid
overloading.
Tobias Schlüter [Fri, 14 May 2004 13:51:27 +0000 (15:51 +0200)]
re PR fortran/15205 (NEAREST intrinsic returns wrong value in DOUBLE PRECISION)
PR fortran/15205
* iresolve.c (gfc_resolve_nearest): Add new function.
* intrinsic.h: ... declare it here.
* intrinsic.c (add_functions): ... add it as resolving function
for NEAREST.