[Patch AArch64] Switch constant pools to separate rodata sections.
Now that PR63304 is fixed and we have an option to address
any part of the memory using adrp / add or adrp / ldr instructions
it makes sense to switch out literal pools into their own
mergeable sections by default.
This would mean that by default we could now start getting
the benefits of constant sharing across the board, potentially
improving code size. The other advantage of doing so, for the
security conscious is that this prevents intermingling of literal
pools with code.
Wilco's kindly done some performance measurements and suggests that
there is not really a performance regression in doing this.
I've looked at the code size for SPEC2k6 today at -Ofast and
in general there is a good code size improvement as expected
by sharing said constants.
Tested on aarch64-none-elf with no regressions and bootstrapped
and regression tested in my tree for a number of days now.
Ilya Enkovich [Thu, 5 Nov 2015 09:46:10 +0000 (09:46 +0000)]
targhooks.c (default_get_mask_mode): Use BLKmode in case target doesn't support required vector mode.
gcc/
2015-10-29 Ilya Enkovich <enkovich.gnu@gmail.com>
* targhooks.c (default_get_mask_mode): Use BLKmode in
case target doesn't support required vector mode.
* stor-layout.c (layout_type); Check for BLKmode.
gcc/cp/
* (cp_parser_oacc_single_int_clause): New function.
(cp_parser_oacc_clause_vector_length): Delete.
(cp_parser_omp_clause_num_gangs): Delete.
(cp_parser_omp_clause_num_workers): Delete.
(cp_parser_oacc_all_clauses): Use cp_parser_oacc_single_int_clause
for num_gangs, num_workers and vector_length.
Aditya Kumar [Wed, 4 Nov 2015 20:59:18 +0000 (20:59 +0000)]
remove parameter_rename_map
This map was used in the transition to the new scop detection: with the new scop
detection, we do not need this map anymore.
* graphite-isl-ast-to-gimple.c (gcc_expression_from_isl_ast_expr_id):
Remove use of parameter_rename_map.
(copy_def): Remove.
(copy_internal_parameters): Remove.
(graphite_regenerate_ast_isl): Remove call to copy_internal_parameters.
* sese.c (new_sese_info): Do not initialize parameter_rename_map.
(free_sese_info): Do not free parameter_rename_map.
(set_rename): Do not use parameter_rename_map.
(rename_uses): Update call to set_rename.
(graphite_copy_stmts_from_block): Do not use parameter_rename_map.
* sese.h (parameter_rename_map_t): Remove.
(struct sese_info_t): Remove field parameter_rename_map.
Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r229783
Aditya Kumar [Wed, 4 Nov 2015 20:59:12 +0000 (20:59 +0000)]
improve debug of codegen
- fix printing of ISL stmt and parameter names
- move dot_scop* functions outside of anonymous namespace.
* graphite-isl-ast-to-gimple.c: Include tree-cfg.h.
(translate_isl_ast_node_user): Add more dumps: call print_loops_bb.
* graphite-scop-detection.c (dot_all_scops_1): Moved out of
anonymous namespace.
* graphite-sese-to-poly.c (ssa_name_version_typesize): Remove.
(isl_id_for_pbb): Use a buffer of size 10.
(isl_id_for_ssa_name): Same.
* sese.c (set_rename): Add more dumps.
Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r229782
Martin Liska [Wed, 4 Nov 2015 16:50:45 +0000 (17:50 +0100)]
Pass manager: add support for termination of pass list
* cgraphunit.c (cgraph_node::expand_thunk): Call
allocate_struct_function before init_function_start.
(cgraph_node::expand): Use push_cfun and pop_cfun.
* config/i386/i386.c (ix86_code_end): Call
allocate_struct_function before init_function_start.
* config/rs6000/rs6000.c (rs6000_code_end): Likewise.
* function.c (init_function_start): Move preamble to all
callers.
* passes.c (do_per_function_toporder): Use push_cfun and pop_cfun.
(execute_one_pass): Handle newly added TODO_discard_function.
(execute_pass_list_1): Terminate if cfun equals to NULL.
(execute_pass_list): Do not push and pop cfun, expect that
cfun is set.
* tree-pass.h (TODO_discard_function): Define.
Tom de Vries [Wed, 4 Nov 2015 14:18:43 +0000 (14:18 +0000)]
Handle recursive restrict in function parameter
2015-11-04 Tom de Vries <tom@codesourcery.com>
PR tree-optimization/67742
* tree-ssa-structalias.c (struct fieldoff): Add restrict_pointed_type
field.
(push_fields_onto_fieldstack): Handle restrict_pointed_type field.
(create_variable_info_for_1): Add and handle handle_param parameter.
Add restrict handling.
(create_variable_info_for): Call create_variable_info_for_1 with extra
arg.
(make_param_constraints): Drop restrict_name parameter. Ignore
vi->only_restrict_pointers.
(intra_create_variable_infos): Call create_variable_info_for_1 with
extra arg. Remove restrict handling. Call make_param_constraints with
one less arg.
* gcc.dg/tree-ssa/restrict-7.c: New test.
* gcc.dg/tree-ssa/restrict-8.c: New test.
I managed to add a new --goarch option to gotest without noticing that
the script already had one. Worse, they set different variables.
Remove the old option in favor of the new one.
Jason Merrill [Tue, 3 Nov 2015 20:51:41 +0000 (15:51 -0500)]
Handle auto parameter packs.
* pt.c (struct find_parameter_pack_data): Add
type_pack_expansion_p field.
(find_parameter_packs_r): Use it to turn 'auto' into a parameter pack.
(uses_parameter_packs, make_pack_expansion)
(check_for_bare_parameter_packs, fixed_parameter_pack_p): Set it.
Martin Sebor [Tue, 3 Nov 2015 18:53:19 +0000 (11:53 -0700)]
re PR c/67882 (surprising offsetof result on an invalid array member without diagnostic)
PR c++-common/67882
* c-family/c-common.h (fold_offsetof_1): Add argument.
* c-family/c-common.c (fold_offsetof_1): Diagnose more invalid
offsetof expressions that reference elements past the end of
an array.
Alexandre Oliva [Tue, 3 Nov 2015 03:57:07 +0000 (03:57 +0000)]
defer mark_addressable calls during expand till the end of expand
for gcc/ChangeLog
* gimple-expr.c: Include hash-set.h and rtl.h.
(mark_addressable_queue): New var.
(mark_addressable): Factor actual marking into...
(mark_addressable_1): ... this. Queue it up during expand.
(mark_addressable_2): New.
(flush_mark_addressable_queue): New.
* gimple-expr.h (flush_mark_addressable_queue): Declare.
* cfgexpand.c: Include gimple-expr.h.
(pass_expand::execute): Flush mark_addressable queue.
Alexandre Oliva [Tue, 3 Nov 2015 00:30:07 +0000 (00:30 +0000)]
[PR68083] don't introduce undefined behavior in ifcombine
The ifcombine pass may move a conditional access to an uninitialized
value before the condition that ensures it is always well-defined,
thus introducing undefined behavior. Stop it from doing so.
for gcc/ChangeLog
PR tree-optimization/68083
* tree-ssa-ifcombine.c: Include tree-ssa.h.
(bb_no_side_effects_p): Test for undefined uses too.
* tree-ssa.c (gimple_uses_undefined_value_p): New.
* tree-ssa.h (gimple_uses_undefined_value_p): Declare.
for gcc/testsuite/ChangeLog
PR tree-optimization/68083
* gcc.dg/torture/pr68083.c: New. From Zhendong Su.
David Malcolm [Mon, 2 Nov 2015 19:47:03 +0000 (19:47 +0000)]
libcpp: Improvements to description of source_location in line-map.h
libcpp/ChangeLog:
* include/line-map.h (source_location): In the table in the
descriptive comment, show UNKNOWN_LOCATION, BUILTINS_LOCATION,
LINE_MAP_MAX_LOCATION_WITH_COLS, LINE_MAP_MAX_SOURCE_LOCATION.
Add notes about ad-hoc values.
Move constant folds for maths functions to new file
The new routines operate on the built-in enum rather than on tree decls.
The idea is to extend this to handle internal functions too, with a
combined enum for both.
The patch also moves fold_fma too, with the same prototype. The long-term
plan is to replace FMA_EXPR with an internal function, for consistency
with the way that things like SQRT will be handled.
Tested on x86_64-linux-gnu, arm-linux-gnueabi and aarch64-linux-gnu.
gcc/
* builtins.h (fold_fma): Move to fold-const-call.h.
* builtins.c: Include fold-const-call.h.
(mathfn_built_in_2): New function, split out from...
(mathfn_built_in_1): ...here.
(do_real_to_int_conversion, fold_const_builtin_pow)
(fold_const_builtin_logb, fold_const_builtin_significand)
(fold_const_builtin_load_exponent, do_mpfr_arg1, do_mpfr_arg2)
(do_mpfr_arg3, do_mpfr_sincos, do_mpfr_bessel_n, do_mpc_arg1): Delete.
(fold_builtin_sincos): Use fold_const_call to handle constants.
(fold_builtin_1, fold_builtin_2, fold_builtin_3): Add explicit
checks for ERROR_MARK. Use fold_const_call to handle constant
folds for math functions.
(fold_fma): Move to fold-const-call.c.
* fold-const.c: Include fold-const-call.h.
* Makefile.in (OBJS): Add fold-const-call.o.
(PLUGIN_HEADERS): Add fold-const-call.h.
* realmpfr.h (real_from_mpfr): Allow the format to be specified
directly.
* realmpfr.c (real_from_mpfr): Likewise.
* fold-const-call.h, fold-const-call.c: New files.
re PR go/67968 (go1: internal compiler error: in write_specific_type_functions, at go/gofrontend/types.cc:1812)
PR go/67968
compiler: Traverse types of call expressions.
https://gcc.gnu.org/PR67968 provides a test case that causes a gccgo
crash on valid code. The compiler failed to build the hash and equality
functions required for a type descriptor. The descriptor is for an
unnamed type that is being returned by a function imported from a
different package. The unnamed type is being implicitly converted to an
interface type by a return statement. The fix is to ensure that the
type of a call expression is always traversed.
Test case sent out for the master testsuite as
https://golang.org/cl/16532 .
Ian Lance Taylor [Sat, 31 Oct 2015 23:48:19 +0000 (23:48 +0000)]
runtime: If no split stacks, allocate stacks using mmap on 64-bit systems.
When not using split stacks, libgo allocate large stacks for each
goroutine. On a 64-bit system, libgo allocates a maximum of 128G for
the Go heap, and allocates 4M for each stack. When the stacks are
allocated from the Go heap, the result is that a program can only create
32K goroutines, which is not enough for an active Go server. This patch
changes libgo to allocate the stacks using mmap directly, rather than
allocating them out of the Go heap. This change is only done for 64-bit
systems when not using split stacks. When using split stacks, the
stacks are allocated using mmap directly anyhow. On a 32-bit system,
there is no maximum size for the Go heap, or, rather, the maximum size
is the available address space anyhow.
This one for TARGET_TOC && TARGET_MINIMAL_TOC. Without it, r30 is not
saved in the prologue for functions that do not use r30, but the register
is set later in the prologue anyhow. This made all java tests fail.
* config/rs6000/rs6000.c (rs6000_reg_live_or_pic_offset_p): Use the
same condition for testing whether RS6000_PIC_OFFSET_TABLE_REGNUM is
live as for using it elsewhere, for TARGET_MINIMAL_TOC.
Jason Merrill [Sat, 31 Oct 2015 16:19:55 +0000 (12:19 -0400)]
Streamline for_each_template_parm.
* pt.c (for_each_template_parm_r): Use WALK_SUBTREE.
Return a meaningful value rather than error_mark_node.
(for_each_template_parm): Return a tree.
(uses_template_parms_level): Return bool.
* cp-tree.h: Adjust.
Tom de Vries [Sat, 31 Oct 2015 08:08:39 +0000 (08:08 +0000)]
Add initial constraints in create_function_info_for
2015-10-31 Tom de Vries <tom@codesourcery.com>
* tree-ssa-structalias.c (ipa_pta_execute): Add extra arg to call to
create_function_info_for. Dump constraints generated during
create_function_info_for. Move intra_create_variable_infos call and
function-return-values-escape bit to ...
(create_function_info_for): ... here, and merge
intra_create_variable_infos call with argument loop. Add and handle
nonlocal_p parameter.
Tom de Vries [Sat, 31 Oct 2015 08:08:28 +0000 (08:08 +0000)]
Improve readability and structure of create_function_info_for
2015-10-31 Tom de Vries <tom@codesourcery.com>
* tree-ssa-structalias.c (create_function_info_for): Make sure prev_vi
updating is alap, and seperated from preceding code. Make sure
insert_vi_for_tree is seperated from surrounding code.
gfortran.h (gfc_omp_namespace): Add locus where member.
gcc/fortran/
* gfortran.h (gfc_omp_namespace): Add locus where member.
* openmp.c (gfc_match_omp_variable_list): Set where for each list
item found.
(resolve_omp_clauses): Remove where argument and use the where
gfc_omp_namespace member when reporting errors.
(resolve_omp_do): Update call to resolve_omp_clauses.
(resolve_oacc_loop): Likewise.
(gfc_resolve_oacc_directive): Likewise.
(gfc_resolve_omp_directive): Likewise.
(gfc_resolve_omp_declare_simd): Likewise.
projects / thirdparty / gcc.git / log
