Nathan Froyd [Thu, 21 Aug 2008 15:37:05 +0000 (15:37 +0000)]
libgomp.exp (libgomp_init): Only set things that depend on blddir if blddir exists.
* testsuite/libgomp.exp (libgomp_init): Only set things that
depend on blddir if blddir exists.
(libgomp_target_compile): Likewise.
* testsuite/libgomp.c++/c++.exp: Likewise.
* testsuite/libgomp.fortran/fortran.exp: Likewise.
Jan Hubicka [Thu, 21 Aug 2008 15:26:59 +0000 (17:26 +0200)]
cgraph.c (first_cgraph_function_insertion_hook): New variable.
* cgraph.c (first_cgraph_function_insertion_hook): New variable.
(cgraph_add_function_insertion_hook, cgraph_remove_function_insertion_hook,
cgraph_call_function_insertion_hooks): New functions.
* cgraph.h (cgraph_add_function_insertion_hook, cgraph_remove_function_insertion_hook,
cgraph_call_function_insertion_hooks): Declare.
* ipa-reference.c (function_insertion_hook_holder): New variable.
(check_operand, look_for_address_of): When checking late, do not care
about module bitmaps.
(add_new_function): New function.
(generate_summary): Register hooks; zero module bitmaps.
(propagate): Unregister hooks.
* ipa-pure-const.c (function_insertion_hook_holder): New variable.
(add_new_function): New function.
(generate_summary): Register hook.
(propagate): Remove hook.
* ipa-cp.c (ipcp_need_redirect_p): Fix to not be constant 0.
* tree-pass.h (pass_ipa_cp): Make ipa_opt_pass.
* ipa-cp.c (ipcp_update_cloned_node): New function.
(build_const_val): Handle functions correctly; bring type logic
into sync with tree-inline.c
(ipcp_init_stage): Take care of computing stuff needed by
indirect inlining; update clones.
(ipcp_generate_summary): Break out of ipcp_driver.
(ipcp_driver): Do only execution and transformation.
(pass_ipa_cp): Make IPA_PASS.
* tree-ssa-ccp.c (fold_stmt_r): Check type before trying to fold
offset to address.
* ipa-inline.c (inline_indirect_intraprocedural_analysis): When doing
ipcp, some info is already available.
* ipa-prop.c (ipa_count_arguments): Grow edge lists as needed.
* tree-inline.c (remap_ssa_name): Unshare expression.
Richard Guenther [Thu, 21 Aug 2008 15:18:58 +0000 (15:18 +0000)]
tree-ssa-pre.c (insert_into_preds_of_block): Before inserting a PHI ask VN if it is already available.
2008-08-21 Richard Guenther <rguenther@suse.de>
* tree-ssa-pre.c (insert_into_preds_of_block): Before inserting
a PHI ask VN if it is already available.
* tree-ssa-sccvn.h (vn_phi_lookup): Declare.
* tree-ssa-sccvn.c (vn_phi_lookup): Export.
Jan Hubicka [Thu, 21 Aug 2008 12:39:35 +0000 (14:39 +0200)]
Backport from LTO branch:
2008-05-05 Kenneth Zadeck <zadeck@naturalbridge.com>
Jan Hubicka <jh@suse.cz>
* ipa-pure-const.c
(init_state, finish_state, set_function_state, generate_summary):
New functions.
(scan_stmt): Renamed from scan_function. Changed to keep state in
local static vars rather than cgraph aux field.
(propagate): Renamed from static_execute. Changed to keep state in
local static vars rather than cgraph aux field.
(pass_ipa_pure_const): Changed from SIMPLE_IPA_PASS to IPA_PASS.
* tree-pass.h (pass_ipa_pure_const): Turn into IPA_PASS.
* tree-pass.h (pass_ipa_reference): Make into ipa_opt_pass.
* ipa-reference.c (init_function_info, generate_summary,
propagate): New functions.
(analyze_function): Call init_function_info.
(static_execute): Stripped into generate_summary and propagate.
(pass_ipa_reference): Made into ipa_opt_pass.
re PR middle-end/179 (-Wuninitialized missing warning with &var)
2008-08-21 Manuel Lopez-Ibanez <manu@gcc.gnu.org>
PR middle-end/179
* tree-ssa.c (warn_uninit): Do not warn for variables that can be
initialized outside the current module.
(warn_uninitialized_var): Ignore left-hand side when walking the
trees. Ignore address expressions. Examine VUSE operands in gimple
statements with a variable declaration on the right-hand side.
testsuite/
* gcc.dg/uninit-6.c (make_something): Remove XFAIL.
* gcc.dg/uninit-6-O0.c (make_something): Remove XFAIL.
* gcc.dg/uninit-B.c (baz): Remove XFAIL.
* gcc.dg/uninit-B-2.c: New.
* gcc.dg/uninit-B-O0-2.c: New.
* gcc.dg/uninit-pr19430-O0.c: New.
* gcc.dg/uninit-pr19430.c: New.
* gcc.dg/uninit-pr19430-2.c: New.
Emmanuel Briot [Wed, 20 Aug 2008 16:12:38 +0000 (18:12 +0200)]
2008-08-20 Emmanuel Briot <briot@adacore.com>
* g-catiio.ads, g-catiio.adb:
(Value): Avoid an unnecessary system call to Clock in most cases.
This call is only needed when only the time is provided in the string,
and ignored in all other cases. This is more efficient.
Vincent Celier [Wed, 20 Aug 2008 16:11:55 +0000 (18:11 +0200)]
2008-08-20 Vincent Celier <celier@adacore.com>
* ali.adb:
(Scan_ALI): Use Name_Find, not Name_Enter to get the name of a subunit,
as the name may already have been entered in the table by the Project
Manager.
Richard Guenther [Wed, 20 Aug 2008 16:01:59 +0000 (16:01 +0000)]
tree-vrp.c (op_with_constant_singleton_value_range): New function.
2008-08-20 Richard Guenther <rguenther@suse.de>
* tree-vrp.c (op_with_constant_singleton_value_range): New function.
(extract_range_from_binary_expr): Fall back to constant propagation.
(extract_range_from_unary_expr): Likewise.
* gcc.dg/tree-ssa/pr21829.c: Scan optimized and cddce2 dumps
instead of phicprop2. Make sure all is fine after cddce2,
add an XFAILed scan for merging the two remaining ifs.
Arnaud Charlet [Wed, 20 Aug 2008 15:43:11 +0000 (17:43 +0200)]
styleg-c.ads, [...] (Missing_Overriding): new procedure to implement style check that overriding operations are...
2008-08-20 Ed Schonberg <schonberg@adacore.com>
* styleg-c.ads, styleg-c.adb (Missing_Overriding): new procedure to
implement style check that overriding operations are explicitly marked
at such.
* style.ads (Missing_Overriding): new procedure that provides interface
to previous one.
* stylesw.ads, stylesw.adb: New style switch -gnatyO, to enable check
that the declaration or body of overriding operations carries an
explicit overriding indicator.
* sem_ch8.adb
(Analyze_Subprogram_Renaming): if operation is overriding, check whether
explicit indicator should be present.
* sem_ch6.adb (Verify_Overriding_Indicator,
Check_Overriding_Indicator): If operation is overriding, check whether
declaration and/or body of subprogram should be present
Vincent Celier [Wed, 20 Aug 2008 15:42:52 +0000 (17:42 +0200)]
make.adb (Gnatmake_Switch_Found): New Boolean global variable
2008-08-20 Vincent Celier <celier@adacore.com>
* make.adb (Gnatmake_Switch_Found): New Boolean global variable
(Switch_May_Be_Passed_To_The_Compiler): New Boolean global variable
(Add_Switches): New Boolean parameter Unknown_Switches_To_The_Compiler
defaulted to True. Fail when Unknown_Switches_To_The_Compiler is False
and a switch is not recognized by gnatmake.
(Gnatmake): Implement new scheme for gnatmake switches and global
compilation switches.
(Switches_Of): Try successively Switches (<file name>),
Switches ("Ada"), Switches (others) and Default_Switches ("Ada").
Gary Dismukes [Wed, 20 Aug 2008 15:29:44 +0000 (17:29 +0200)]
exp_ch3.adb (Build_Array_Init_Proc): Clarify comment related to creating dummy init proc.
2008-08-20 Gary Dismukes <dismukes@adacore.com>
* exp_ch3.adb (Build_Array_Init_Proc): Clarify comment related to
creating dummy init proc.
(Requires_Init_Proc): Return False in the case No_Default_Initialization
is in force and the type does not have associated default
initialization. Move test of Is_Public (with tests of restrictions
No_Initialize_Scalars and No_Default_Initialization) to end, past tests
for default initialization.
Vincent Celier [Wed, 20 Aug 2008 15:29:33 +0000 (17:29 +0200)]
prj-nmsc.adb (Check_Naming_Schemes): Accept source file names for gprbuild when casing is MixedCase...
2008-08-20 Vincent Celier <celier@adacore.com>
* prj-nmsc.adb (Check_Naming_Schemes): Accept source file names for
gprbuild when casing is MixedCase, whatever the casing of the letters
in the file name.
* exp_ch5.adb (Controlled_Type): New routine.
(Expand_N_Extended_Return_Statement): When generating a move of the
final list in extended return statements, check the type of the
function and in the case of double expanded return statements, the type
of the returned object.
(Expand_Simple_Function_Return): Perform an interface conversion when
the type of the returned object is an interface and the context is an
extended return statement.
* exp_ch4.adb (Expand_N_And_Then, Expand_N_Or_Else): Improve constant
folding. We were folding things like "False and then ...", but not
"X and then ..." where X is a constant whose value is known at compile
time.
Thomas Quinot [Wed, 20 Aug 2008 14:27:01 +0000 (16:27 +0200)]
exp_strm.adb (Build_Elementary_Input_Call, [...]): Fix incorrect condition in circuitry that selects the stream attribute routines...
2008-08-20 Thomas Quinot <quinot@adacore.com>
* exp_strm.adb (Build_Elementary_Input_Call,
Build_Elementary_Write_Call): Fix incorrect condition in circuitry that
selects the stream attribute routines for long float types.
Arnaud Charlet [Wed, 20 Aug 2008 13:24:40 +0000 (15:24 +0200)]
2008-08-20 Gary Dismukes <dismukes@adacore.com>
* exp_ch11.adb:
(Expand_Exception_Handlers): Call Make_Exception_Handler instead of
Make_Implicit_Exception_Handler when rewriting an exception handler with
a choice parameter, and pass the handler's Sloc instead of that of the
handled sequence of statements. Make_Implicit_Exception_Handler sets the
Sloc to No_Location (unless debugging generated code), which we don't
want for the case of a user handler.
H.J. Lu [Wed, 20 Aug 2008 13:22:30 +0000 (06:22 -0700)]
re PR target/37169 (Inefficent code for _mm_cvtsi64_si128)
gcc/
2008-08-20 H.J. Lu <hongjiu.lu@intel.com>
PR target/37169
* config/i386/i386.c (ix86_expand_vector_init_one_nonzero): In
V2DI mode, for SSE4.1, use movq instead of vector set if the
second element is zero and inter-unit moves are OK.
Richard Guenther [Wed, 20 Aug 2008 12:59:10 +0000 (12:59 +0000)]
passes.c (init_optimization_passes): Move the second forwprop pass before alias computation.
2008-08-20 Richard Guenther <rguenther@suse.de>
* passes.c (init_optimization_passes): Move the second
forwprop pass before alias computation. Remove the second
DCE pass. Remove the first dominator and phi copy/const
prop passes.
Jose Ruiz [Wed, 20 Aug 2008 12:36:24 +0000 (14:36 +0200)]
errno.c (__get_errno for MaRTE): Use the MaRTE function pthread_errno to get access to the per-task errno variable.
2008-08-20 Jose Ruiz <ruiz@adacore.com>
* errno.c (__get_errno for MaRTE): Use the MaRTE function pthread_errno
to get access to the per-task errno variable.
(__set_errno for MaRTE): Do not redefine this function here since it is
already defined in MaRTE.
Ed Schonberg [Wed, 20 Aug 2008 11:02:51 +0000 (11:02 +0000)]
sem_aggr.adb, [...] (Valid_Ancestor): Resolve confusion between partial and full views of an ancestor of the context...
2008-08-20 Ed Schonberg <schonberg@adacore.com>
* sem_aggr.adb, sem_type.adb, exp_ch9.ads, einfo.ads,
exp_ch6.adb, exp_aggr.adb (Valid_Ancestor): Resolve
confusion between partial and full views of an ancestor of the context
type when the parent is a private extension declared in a parent unit,
and full views are available for the context type.
Nick Clifton [Wed, 20 Aug 2008 10:23:08 +0000 (10:23 +0000)]
configure.ac (HAVE_GAS_CFI_DIRECTIVE): Always test for assembler support of this feature.
* configure.ac (HAVE_GAS_CFI_DIRECTIVE): Always test for assembler
support of this feature. Do not assume that a sufficiently new
assembler will support the feature regardless of the target type.
(HAVE_GAS_CFI_PERSONALITY_DIRECTIVE): Likewise.
* configure: Regenerate.
Richard Guenther [Wed, 20 Aug 2008 08:28:17 +0000 (08:28 +0000)]
tree-vrp.c (found_in_subgraph): Remove.
2008-08-20 Richard Guenther <rguenther@suse.de>
* tree-vrp.c (found_in_subgraph): Remove.
(live): New global static.
(live_on_edge): New function.
(blocks_visited): Remove.
(register_edge_assert_for_2): Use live_on_edge.
(find_conditional_asserts): Remove code dealing with
found_in_subgraph. Do not walk the CFG.
(find_switch_asserts): Likewise.
(find_assert_locations_1): Renamed from find_assert_locations.
Move finding assert locations for conditional and switch
statements first. Update live bitmap. Do not walk the CFG.
(find_assert_locations): New function.
(insert_range_assertions): Remove entry of CFG walk.
Adjust call to find_assert_locations.
* tree-ssa-pre.c (do_regular_insertion): Ignore critical edges
that only can appear because of fake exit edges but assert we
never try to insert on those.
(fini_pre): Do not remove fake exit edges here...
(execute_pre): ...but here, before committing edge inserts.
2008-08-19 Rafael Espindola <espindola@google.com>
* varasm.c (weak_decls): Move earlier in the file.
(assemble_external): Add weak decls to the weak_decls list.
(declare_weak): Don't add decls to the weak_decls list.
Richard Guenther [Tue, 19 Aug 2008 09:06:54 +0000 (09:06 +0000)]
re PR tree-optimization/35972 (load-PRE missed opportunities without SFTs)
2008-08-19 Richard Guenther <rguenther@suse.de>
PR tree-optimization/35972
PR tree-optimization/23094
* tree-ssa-sccvn.h (vn_reference_lookup_pieces): Add maywalk
parameter.
* tree-ssa-sccvn.c (copy_reference_ops_from_ref): Properly
handle MISALIGNED_INDIRECT_REF.
(get_ref_from_reference_ops): New helper.
(vn_reference_lookup_pieces): Walk the use-def chain using the
alias-oracle if requested.
* tree-ssa-pre.c (phi_translate_1): Do reference lookup with
walking the use-def chain.
(compute_avail): But not here.
(create_component_ref_by_pieces_1): Properly handle
MISALIGNED_INDIRECT_REF.
(do_regular_insertion): Handle fully redundant
expressions after PHI-translation also for SSA_NAME values, not
only constants. Correctly use edoubleprime for that.
Ira Rosen [Tue, 19 Aug 2008 08:31:41 +0000 (08:31 +0000)]
tree-vectorizer.c (supportable_widening_operation): Support multi-step conversion...
* tree-vectorizer.c (supportable_widening_operation): Support
multi-step conversion, return the number of steps in such conversion
and the required intermediate types.
(supportable_narrowing_operation): Likewise.
* tree-vectorizer.h (vect_pow2): New function.
(supportable_widening_operation): Change argument types.
(supportable_narrowing_operation): Likewise.
(vectorizable_type_promotion): Add an argument.
(vectorizable_type_demotion): Likewise.
* tree-vect-analyze.c (vect_analyze_operations): Call
vectorizable_type_promotion and vectorizable_type_demotion with
additional argument.
(vect_get_and_check_slp_defs): Detect patterns.
(vect_build_slp_tree): Add an argument, don't fail in case of multiple
types.
(vect_analyze_slp_instance): Don't fail in case of multiple types. Call
vect_build_slp_tree with correct arguments. Calculate unrolling factor
according to the smallest type in the loop.
(vect_detect_hybrid_slp_stmts): Include statements from patterns.
* tree-vect-patterns.c (vect_recog_widen_mult_pattern): Call
supportable_widening_operation with correct arguments.
* tree-vect-transform.c (vect_get_slp_defs): Allocate output vector
operands lists according to the number of vector statements in left
or right node, if exists.
(vect_gen_widened_results_half): Remove unused argument.
(vectorizable_conversion): Call supportable_widening_operation,
supportable_narrowing_operation, and vect_gen_widened_results_half
with correct arguments.
(vectorizable_assignment): Change documentation, support multiple
types in SLP.
(vectorizable_operation): Likewise.
(vect_get_loop_based_defs): New function.
(vect_create_vectorized_demotion_stmts): Likewise.
(vectorizable_type_demotion): Support loop-aware SLP and general
multi-step conversion. Call vect_get_loop_based_defs and
vect_create_vectorized_demotion_stmts for transformation.
(vect_create_vectorized_promotion_stmts): New function.
(vectorizable_type_promotion): Support loop-aware SLP and general
multi-step conversion. Call vect_create_vectorized_promotion_stmts
for transformation.
(vectorizable_store): Change documentation, support multiple
types in SLP.
(vectorizable_load): Likewise.
(vect_transform_stmt): Pass SLP_NODE to
vectorizable_type_promotion and vectorizable_type_demotion.
(vect_schedule_slp_instance): Move here the calculation of number
of vectorized statements for each node from...
(vect_schedule_slp): ... here.
(vect_transform_loop): Call vect_schedule_slp without the last
argument.
projects / thirdparty / gcc.git / log
