platform/upstream/gcc.git
5 years agoomp-low.c (lower_omp_scan): Call lower_omp on stmt's body in worksharing loop scans.
Jakub Jelinek [Thu, 4 Jul 2019 04:55:59 +0000 (06:55 +0200)]
omp-low.c (lower_omp_scan): Call lower_omp on stmt's body in worksharing loop scans.

* omp-low.c (lower_omp_scan): Call lower_omp on stmt's body
in worksharing loop scans.

From-SVN: r273038

5 years agore PR tree-optimization/91074 (c-c++-common/gomp/scan-3.c fails with ICE starting...
Jakub Jelinek [Thu, 4 Jul 2019 04:54:52 +0000 (06:54 +0200)]
re PR tree-optimization/91074 (c-c++-common/gomp/scan-3.c fails with ICE starting with r272958)

PR tree-optimization/91074
* omp-low.c (lower_omp_for_scan): Set DECL_GIMPLE_REG_P on cplx
temporary.

From-SVN: r273037

5 years agore PR rtl-optimization/90756 (g++ ICE in convert_move, at expr.c:218 on i686 and...
Jakub Jelinek [Thu, 4 Jul 2019 04:49:22 +0000 (06:49 +0200)]
re PR rtl-optimization/90756 (g++ ICE in convert_move, at expr.c:218 on i686 and s390x)

PR rtl-optimization/90756
* explow.c (promote_ssa_mode): Always use TYPE_MODE, don't bypass it
for VECTOR_TYPE_P.

* gcc.dg/pr90756.c: New test.

From-SVN: r273036

5 years agocompiler: optimize 0,1,2-case select statement
Ian Lance Taylor [Thu, 4 Jul 2019 02:20:37 +0000 (02:20 +0000)]
compiler: optimize 0,1,2-case select statement

    For a select statement with zero-, one-, or two-case with a
    default case, we can generate simpler code instead of calling the
    generic selectgo. A zero-case select is just blocking the
    execution. A one-case select is mostly just executing the case. A
    two-case select with a default case is a non-blocking send or
    receive. We add these special cases for lowering a select
    statement.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184998

From-SVN: r273034

5 years ago[MIPS][Testsuite] Specify msa-fmadd.c abis.
Chenghua Xu [Thu, 4 Jul 2019 01:54:02 +0000 (01:54 +0000)]
[MIPS][Testsuite] Specify msa-fmadd.c abis.

gcc/testsuite/

        * gcc.target/mips/mips-fmadd.c: Rename to ...
        * gcc.target/mips/mips-fmadd-o32.c: ... Here; add abi=32.
        * gcc.target/mips/mips-fmadd-n64.c: New.

From-SVN: r273033

5 years agocompiler: fix indentation of select statement AST dump
Ian Lance Taylor [Thu, 4 Jul 2019 01:48:57 +0000 (01:48 +0000)]
compiler: fix indentation of select statement AST dump

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184997

From-SVN: r273032

5 years agoDaily bump.
GCC Administrator [Thu, 4 Jul 2019 00:16:21 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r273031

5 years agocompiler: set varargs lowered for imported call expressions
Ian Lance Taylor [Wed, 3 Jul 2019 22:27:51 +0000 (22:27 +0000)]
compiler: set varargs lowered for imported call expressions

    Fix compiler buglet: varargs lowering happens before inlinable
    function bodies are written out to export data, so set the "varargs
    lowered" flag on call expressions that we import.

    Fixes golang/go#32922

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184919

From-SVN: r273026

5 years agoPR libstdc++/91067 fix missing exports for filesystem iterators
Jonathan Wakely [Wed, 3 Jul 2019 21:06:25 +0000 (22:06 +0100)]
PR libstdc++/91067 fix missing exports for filesystem iterators

The copy assignment operator for recursive_directory_iterator was not
exported despite being needed. The __shared_ptr default constructors are
not needed when compiling with GCC but Clang requires them for -O1.

PR libstdc++/91067
* acinclude.m4 (libtool_VERSION): Bump to 6:27:0.
* configure: Regenerate.
* config/abi/pre/gnu.ver (GLIBCXX_3.4.27): Add new version. Export
missing symbols.
* testsuite/27_io/filesystem/iterators/91067.cc: New test.
* testsuite/util/testsuite_abi.cc: Add new symbol version.

From-SVN: r273023

5 years ago[AArch64] Remove constraint strings from define_expand constructs
Dennis Zhang [Wed, 3 Jul 2019 19:52:04 +0000 (19:52 +0000)]
[AArch64] Remove constraint strings from define_expand constructs

A number of AArch64 define_expand patterns have specified constraints
for their operands. But the constraint strings are ignored at expand
time and are therefore redundant/useless. We now avoid specifying
constraints in new define_expands, but we should clean up the existing
define_expand definitions.

For example, the constraint "=w" is removed in the following case:
(define_expand "sqrt<mode>2"
    [(set (match_operand:GPF_F16 0 "register_operand" "=w")
The "" marks with an empty constraint in define_expand are removed as well.

2019-07-03  Dennis Zhang  <dennis.zhang@arm.com>

gcc/
* config/aarch64/aarch64.md: Remove redundant constraints from
define_expand but keep some patterns untouched if they are
specially selected by TARGET_SECONDARY_RELOAD hook.
* config/aarch64/aarch64-sve.md: Likewise.
* config/aarch64/atomics.md: Remove redundant constraints from
define_expand.
* config/aarch64/aarch64-simd.md: Likewise.

From-SVN: r273021

5 years ago[Darwin] Revise pie,no-pie and rdynamic driver specs.
Iain Sandoe [Wed, 3 Jul 2019 18:43:55 +0000 (18:43 +0000)]
[Darwin] Revise pie,no-pie and rdynamic driver specs.

Processing these early and pushing the corresponding Xlinker lines
has the effect that the driver then supposes that there are linker
inputs, and causes a link line to be created when it is not needed

The solution is to place these drive specs into the link spec and
claim them at the end of that.

2019-07-03  Iain Sandoe  <iain@sandoe.co.uk>

* config/darwin.h (DRIVER_SELF_SPECS): Remove the linker cases.
(RDYNAMIC): Rename to, DARWIN_RDYNAMIC.
(DARWIN_PIE_SPEC, DARWIN_NOPIE_SPEC): Adjust to remove the Xlinker
clauses.
(LINK_COMMAND_SPEC_A): Add DARWIN_RDYNAMIC, DARWIN_PIE_SPEC and
DARWIN_NOPIE_SPEC.

From-SVN: r273017

5 years ago[Darwin] Some TLC for older Darwin versions.
Iain Sandoe [Wed, 3 Jul 2019 18:36:28 +0000 (18:36 +0000)]
[Darwin] Some TLC for older Darwin versions.

The library handling and some of the options for creating the crts for
the older PPC Darwin versions had bit-rotted somewhat.  This adjusts the
build criteria for the crts to avoid newer ld64 versions warnings about
mismatches in build and object versions.

Added to some of the comments that it's documented why the specs are as
they are.

2019-07-03  Iain Sandoe  <iain@sandoe.co.uk>

gcc/

* config/darwin.h (REAL_LIBGCC_SPEC): Adjust for earlier Darwin.
(STARTFILE_SPEC): Split crt3 into a separate spec.
(DARWIN_EXTRA_SPECS): Add crt2 and crt3 spec.
(DARWIN_CRT2_SPEC): New.
(DARWIN_CRT3_SPEC): New.
(MIN_LD64_OMIT_STUBS): Revise to 62.1.
* config/rs6000/darwin.h (DARWIN_CRT2_SPEC): Revise conditions.
(DARWIN_CRT3_SPEC): New.

libgcc/

2019-07-03  Iain Sandoe  <iain@sandoe.co.uk>

* config.host (powerpc-*-darwin*,powerpc64-*-darwin*): Revise crt
list.
* config/rs6000/t-darwin: Build crt3_2 for older systems.  Revise
mmacosx-version-min for crts to run across all system versions.
* config/rs6000/t-darwin64 (LIB2ADD): Remove.
* config/t-darwin: Revise mmacosx-version-min for crts to run across
system versions >= 10.4.

From-SVN: r273016

5 years agoaltivec.md (altivec_mov<mode>, [...]): Change the RTL attribute "length" from "4...
Michael Meissner [Wed, 3 Jul 2019 17:42:09 +0000 (17:42 +0000)]
altivec.md (altivec_mov<mode>, [...]): Change the RTL attribute "length" from "4" to "*" to allow the length attribute...

2019-07-03  Michael Meissner  <meissner@linux.ibm.com>

* config/rs6000/altivec.md (altivec_mov<mode>, VM2 iterator):
Change the RTL attribute "length" from "4" to "*" to allow the
length attribute to be adjusted automatically for prefixed load,
store, and add immediate instructions.
* config/rs6000/rs6000.md (extendhi<mode>2, EXTHI iterator):
Likewise.
(extendsi<mode>2, EXTSI iterator): Likewise.
(movsi_internal1): Likewise.
(movsi_from_sf): Likewise.
(movdi_from_sf_zero_ext): Likewise.
(mov<mode>_internal): Likewise.
(movcc_internal1, QHI iterator): Likewise.
(mov<mode>_softfloat, FMOVE32 iterator): Likewise.
(movsf_from_si): Likewise.
(mov<mode>_hardfloat32, FMOVE64 iterator): Likewise.
(mov<mode>_softfloat64, FMOVE64 iterator): Likewise.
(mov<mode>, FMOVE128 iterator): Likewise.
(movdi_internal64): Likewise.
* config/rs6000/vsx.md (vsx_le_permute_<mode>, VSX_TI iterator):
Likewise.
(vsx_le_undo_permute_<mode>, VSX_TI iterator): Likewise.
(vsx_mov<mode>_64bit, VSX_M iterator): Likewise.
(vsx_mov<mode>_32bit, VSX_M iterator): Likewise.
(vsx_splat_v4sf): Likewise.

From-SVN: r273013

5 years agoFix store merging tests on Arm
Wilco Dijkstra [Wed, 3 Jul 2019 16:40:17 +0000 (16:40 +0000)]
Fix store merging tests on Arm

Fix the failing store merging test on Arm.  Aligning variables fixes a
few cases, otherwise disable the test on Arm.  All store merging tests
now pass.  Committed as obvious.

    testsuite/
* gcc.dg/store_merging_27.c: Fix test for Arm.
* gcc.dg/store_merging_28.c: Likewise.
* gcc.dg/store_merging_29.c: Likewise.
* gcc.dg/tree-ssa/dump-6.c: Likewise.

From-SVN: r273011

5 years agocompiler: include transitive imports in the type descriptor list
Ian Lance Taylor [Wed, 3 Jul 2019 13:23:39 +0000 (13:23 +0000)]
compiler: include transitive imports in the type descriptor list

    In CL 179598, we were using Gogo::packages_, when compiling the
    main package, as the list of packages of which we need to
    register the type descriptors. This is not complete. It only
    includes main's direct import and one-level indirect imports. It
    does not include all the packages transitively imported.

    To fix that, we need to track all the transitive imports. We
    have almost already done that, for init functions. However, there
    may be packages that don't need init functions but do need to
    register type descriptors. For them, we add a dummy init function
    to its export data. So when we compile the main package we will
    see all the transitive imports. The dummy init functions are not
    real functions and are not called.

    Fixes golang/go#32901.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184717

From-SVN: r273009

5 years agoPR debug/90981 Empty .debug_addr crashes -gdwarf-5 -gsplit-dwarf
Mark Wielaard [Wed, 3 Jul 2019 13:08:01 +0000 (13:08 +0000)]
PR debug/90981 Empty .debug_addr crashes -gdwarf-5 -gsplit-dwarf

Even if there was no, or an empty address list we would try to generate
a header for the .debug_addr section with -gdwarf-5 and -gsplit-dwarf.
The skeleton DIE would also get a (dangling) DW_AT_addr_base in that case.

PR debug/90981
* dwarf2out.c (add_top_level_skeleton_die_attrs): Only add
DW_AT_addr_base if there is actually a .debug_addr section with
addresses.
(output_addr_table): Add DWARF5 table header generation here after
checking there are actually any addresses from...
(dwarf2out_finish): ...here.
* testsuite/g++.dg/pr90981.C: New test.

From-SVN: r273008

5 years agore PR tree-optimization/91069 (Miscompare of 453.povray since r272843)
Richard Biener [Wed, 3 Jul 2019 12:47:07 +0000 (12:47 +0000)]
re PR tree-optimization/91069 (Miscompare of 453.povray since r272843)

2019-07-03  Richard Biener  <rguenther@suse.de>

PR middle-end/91069
* match.pd (vec_perm -> bit_insert): Fix element read from
first vector.

* gcc.dg/pr91069.c: New testcase.

From-SVN: r273007

5 years agoAdd dbgcnt for gimple_match and generic_match.
Martin Liska [Wed, 3 Jul 2019 12:43:44 +0000 (14:43 +0200)]
Add dbgcnt for gimple_match and generic_match.

2019-07-03  Martin Liska  <mliska@suse.cz>

* dbgcnt.def (DEBUG_COUNTER): Add match debug counter.
* genmatch.c (dt_simplify::gen_1): Generate dbgcnt
condition.
* generic-match-head.c: Include dbgcnt.h.
* gimple-match-head.c: Likewise.

From-SVN: r273006

5 years agoRename SINGE_VALUE to TOPN_VALUES counters.
Martin Liska [Wed, 3 Jul 2019 12:42:23 +0000 (14:42 +0200)]
Rename SINGE_VALUE to TOPN_VALUES counters.

2019-07-03  Martin Liska  <mliska@suse.cz>

* gcov-counter.def (GCOV_COUNTER_V_SINGLE): Remove.
(GCOV_COUNTER_V_TOPN): New.
(GCOV_COUNTER_V_INDIR): Use _topn.
* gcov-io.h (GCOV_DISK_SINGLE_VALUES): Remove.
(GCOV_TOPN_VALUES): New.
(GCOV_SINGLE_VALUE_COUNTERS): Remove.
(GCOV_TOPN_VALUES_COUNTERS): New.
* profile.c (instrument_values): Use HIST_TYPE_TOPN_VALUES.
* tree-profile.c:
(gimple_init_gcov_profiler): Rename variables from one_value
to topn_values.
(gimple_gen_one_value_profiler): Remove.
(gimple_gen_topn_values_profiler): New function.
* value-prof.c (dump_histogram_value): Use TOPN_VALUES
names instead of SINGLE_VALUE.
(stream_out_histogram_value): Likewise.
(stream_in_histogram_value): Likewise.
(get_most_common_single_value): Likewise.
(gimple_divmod_fixed_value_transform): Likewise.
(gimple_stringops_transform): Likewise.
(gimple_divmod_values_to_profile): Likewise.
(gimple_stringops_values_to_profile): Likewise.
(gimple_find_values_to_profile): Likewise.
* value-prof.h (enum hist_type): Rename to TOPN.
(gimple_gen_one_value_profiler): Remove.
(gimple_gen_topn_values_profiler): New.
2019-07-03  Martin Liska  <mliska@suse.cz>

* Makefile.in: Use topn_values instead of one_value names.
* libgcov-merge.c (__gcov_merge_single): Move to ...
(__gcov_merge_topn): ... this.
(merge_single_value_set): Move to ...
(merge_topn_values_set): ... this.
* libgcov-profiler.c (__gcov_one_value_profiler_body): Move to
...
(__gcov_topn_values_profiler_body): ... this.
(__gcov_one_value_profiler_v2): Move to ...
(__gcov_topn_values_profiler): ... this.
(__gcov_one_value_profiler_v2_atomic): Move to ...
(__gcov_topn_values_profiler_atomic): ... this.
(__gcov_indirect_call_profiler_v4): Remove.
* libgcov-util.c (__gcov_single_counter_op): Move to ...
(__gcov_topn_counter_op): ... this.
* libgcov.h (L_gcov_merge_single): Remove.
(L_gcov_merge_topn): New.
(__gcov_merge_single): Remove.
(__gcov_merge_topn): New.
(__gcov_one_value_profiler_v2): Move to ..
(__gcov_topn_values_profiler): ... this.
(__gcov_one_value_profiler_v2_atomic): Move to ...
(__gcov_topn_values_profiler_atomic): ... this.

From-SVN: r273005

5 years agoSupport N values in libgcov for single value counter type.
Martin Liska [Wed, 3 Jul 2019 12:42:02 +0000 (14:42 +0200)]
Support N values in libgcov for single value counter type.

2019-07-03  Martin Liska  <mliska@suse.cz>

* gcc.dg/tree-prof/val-prof-2.c: Update scanned pattern
as we do now better.
2019-07-03  Martin Liska  <mliska@suse.cz>

* libgcov-merge.c (merge_single_value_set): Support N values.
* libgcov-profiler.c (__gcov_one_value_profiler_body): Likewise.

From-SVN: r273004

5 years agodwarf2out.c (add_scalar_info): Add back refererence to existing DIE if it has the...
Eric Botcazou [Wed, 3 Jul 2019 11:58:49 +0000 (11:58 +0000)]
dwarf2out.c (add_scalar_info): Add back refererence to existing DIE if it has the DW_AT_data_member_location...

* dwarf2out.c (add_scalar_info): Add back refererence to existing DIE
if it has the DW_AT_data_member_location attribute.

From-SVN: r273002

5 years agogimple-pretty-print.c (dump_ternary_rhs): Fix BIT_INSERT_EXPR dumping.
Richard Biener [Wed, 3 Jul 2019 11:01:01 +0000 (11:01 +0000)]
gimple-pretty-print.c (dump_ternary_rhs): Fix BIT_INSERT_EXPR dumping.

2019-07-03  Richard Biener  <rguenther@suse.de>

* gimple-pretty-print.c (dump_ternary_rhs): Fix BIT_INSERT_EXPR
dumping.

From-SVN: r273001

5 years ago[aarch64]: force frame pointer setup before tlsdesc call
Sylvia Taylor [Wed, 3 Jul 2019 10:57:28 +0000 (10:57 +0000)]
[aarch64]: force frame pointer setup before tlsdesc call

The operation that sets up the tlsdesc register is really a function
call; as such, the frame pointer should really be correctly set up
before this happens.  However, because it does not appear as a
call_insn in the RTL the normal dependencies are not added
automatically.  We fix this by adding a USE of FP_REGNUM to prevent
migration of the frame-pointer setup past the call by the scheduler.
If a frame pointer is not being used, this is harmless since it only
represents a scheduling constraint.

2019-07-03  Sylvia Taylor  <sylvia.taylor@arm.com>

* config/aarch64/aarch64.md (FP_REGNUM): New constant.
(tlsdesc_small_advsimd_<mode>): Add use of FP_REGNUM.
(tlsdesc_small_sve_<mode>): Likewise.

From-SVN: r273000

5 years agoFix ChangeLog date for last commit
Andrea Corallo [Wed, 3 Jul 2019 09:42:05 +0000 (09:42 +0000)]
Fix ChangeLog date for last commit

From-SVN: r272999

5 years agoAdd myself to MAINTAINERS
Andrea Corallo [Wed, 3 Jul 2019 09:09:26 +0000 (09:09 +0000)]
Add myself to MAINTAINERS

2019-07-02  Andrea Corallo  <andrea.corallo@arm.com>

* MAINTAINERS (Write After Approval): Add myself.

From-SVN: r272997

5 years agoAdd zstd support for LTO bytecode compression.
Martin Liska [Wed, 3 Jul 2019 08:36:54 +0000 (10:36 +0200)]
Add zstd support for LTO bytecode compression.

2019-07-03  Martin Liska  <mliska@suse.cz>

* Makefile.in: Define ZSTD_LIB.
* common.opt: Adjust compression level
to support also zstd levels.
* config.in: Regenerate.
* configure: Likewise.
* configure.ac: Add --with-zstd and --with-zstd-include options
and detect ZSTD.
* doc/install.texi: Mention zstd dependency.
* gcc.c: Print supported LTO compression algorithms.
* lto-compress.c (lto_normalized_zstd_level): Likewise.
(lto_compression_zstd): Likewise.
(lto_uncompression_zstd): Likewise.
(lto_end_compression): Dispatch in between zlib and zstd.
(lto_compression_zlib): Mark with ATTRIBUTE_UNUSED.
(lto_uncompression_zlib): Make it static.
* lto-compress.h (lto_end_uncompression): Fix GNU coding style.
* lto-section-in.c (lto_get_section_data): Pass info
about used compression.
* lto-streamer-out.c: By default use zstd when possible.
* timevar.def (TV_IPA_LTO_DECOMPRESS): Rename to decompression
(TV_IPA_LTO_COMPRESS): Likewise for compression.

From-SVN: r272996

5 years agoAdd .gnu.lto_.lto section.
Martin Liska [Wed, 3 Jul 2019 08:36:35 +0000 (10:36 +0200)]
Add .gnu.lto_.lto section.

2019-07-03  Martin Liska  <mliska@suse.cz>

* lto-section-in.c (lto_get_section_data): Add "lto" section.
* lto-section-out.c (lto_destroy_simple_output_block): Never
compress LTO_section_lto section.
* lto-streamer-out.c (produce_asm): Do not set major_version
and minor_version.
(lto_output_toplevel_asms): Likewise.
(produce_lto_section): New function.
(lto_output): Call produce_lto_section.
(lto_write_mode_table): Do not set major_version and
minor_version.
(produce_asm_for_decls): Likewise.
* lto-streamer.h (enum lto_section_type): Add LTO_section_lto
type.
(struct lto_header): Remove.
(struct lto_section): New struct.
(struct lto_simple_header): Do not inherit from lto_header.
(struct lto_file_decl_data): Add lto_section_header field.
2019-07-03  Martin Liska  <mliska@suse.cz>

* lto-common.c: Read LTO section and verify header.

From-SVN: r272995

5 years agoRemove another bunch of dead assignment.
Martin Liska [Wed, 3 Jul 2019 08:34:20 +0000 (10:34 +0200)]
Remove another bunch of dead assignment.

2019-07-03  Martin Liska  <mliska@suse.cz>

* lra-eliminations.c (eliminate_regs_in_insn): Remove
dead assignemts.
* reg-stack.c (check_asm_stack_operands): Likewise.
* tree-ssa-structalias.c (create_function_info_for): Likewise.
* tree-vect-generic.c (expand_vector_operations_1): Likewise.
* config/i386/i386-expand.c (ix86_expand_sse2_mulvxdi3): Use
force_expand_binop.
2019-07-03  Martin Liska  <mliska@suse.cz>

* c-common.c (try_to_locate_new_include_insertion_point): Remove
dead assignemts.
2019-07-03  Martin Liska  <mliska@suse.cz>

* call.c (build_new_op_1): Remove
dead assignemts.
* typeck.c (cp_build_binary_op): Likewise.
2019-07-03  Martin Liska  <mliska@suse.cz>

* check.c (gfc_check_c_funloc): Remove
dead assignemts.
* decl.c (variable_decl): Likewise.
* resolve.c (resolve_typebound_function): Likewise.
* simplify.c (gfc_simplify_matmul): Likewise.
(gfc_simplify_scan): Likewise.
* trans-array.c (gfc_could_be_alias): Likewise.
* trans-common.c (add_equivalences): Likewise.
* trans-expr.c (trans_class_vptr_len_assignment): Likewise.
(gfc_trans_array_constructor_copy): Likewise.
(gfc_trans_assignment_1): Likewise.
* trans-intrinsic.c (conv_intrinsic_atomic_op): Likewise.
* trans-openmp.c (gfc_omp_finish_clause): Likewise.
* trans-types.c (gfc_get_array_descriptor_base): Likewise.
* trans.c (gfc_build_final_call): Likewise.
2019-07-03  Martin Liska  <mliska@suse.cz>

* line-map.c (linemap_get_expansion_filename): Remove
dead assignemts.
* mkdeps.c (make_write): Likewise.

From-SVN: r272994

5 years agoHandle '\0' in strcmp in RTL expansion (PR tree-optimization/90892).
Martin Liska [Wed, 3 Jul 2019 08:32:25 +0000 (10:32 +0200)]
Handle '\0' in strcmp in RTL expansion (PR tree-optimization/90892).

2019-07-03  Martin Liska  <mliska@suse.cz>

PR tree-optimization/90892
* builtins.c (inline_expand_builtin_string_cmp): Handle '\0'
in string constants.
2019-07-03  Martin Liska  <mliska@suse.cz>

PR tree-optimization/90892
* gcc.dg/pr90892.c: New test.

From-SVN: r272993

5 years agoAdd to same comdate group only if set (PR middle-end/90899)
Martin Liska [Wed, 3 Jul 2019 08:31:35 +0000 (10:31 +0200)]
Add to same comdate group only if set (PR middle-end/90899)

2019-07-03  Martin Liska  <mliska@suse.cz>

PR middle-end/90899
* multiple_target.c (create_dispatcher_calls): Add to comdat
group only if set for ifunc.
2019-07-03  Martin Liska  <mliska@suse.cz>

PR middle-end/90899
* gcc.target/i386/pr90899.c: New test.

From-SVN: r272992

5 years agoFix use-after-scope in host-mingw32.c (PR target/88056).
Martin Liska [Wed, 3 Jul 2019 08:29:48 +0000 (10:29 +0200)]
Fix use-after-scope in host-mingw32.c (PR target/88056).

2019-07-03  Martin Liska  <mliska@suse.cz>

PR target/88056
* config/i386/host-mingw32.c (mingw32_gt_pch_use_address):
Define local_object_name in outer scope in order to handle
use-after-scope issue.

From-SVN: r272991

5 years agoAdd -fprofile-note option.
Martin Liska [Wed, 3 Jul 2019 08:29:08 +0000 (10:29 +0200)]
Add -fprofile-note option.

2019-07-03  Martin Liska  <mliska@suse.cz>

* common.opt: Add fprofile-note.
* coverage.c (coverage_init): Append the option
to bbg_file_name.
* doc/invoke.texi: Document -fprofile-note.

From-SVN: r272990

5 years agore PR tree-optimization/91033 (ICE in vect_analyze_loop, at tree-vect-loop.c:2416)
Jakub Jelinek [Wed, 3 Jul 2019 08:25:22 +0000 (10:25 +0200)]
re PR tree-optimization/91033 (ICE in vect_analyze_loop, at tree-vect-loop.c:2416)

PR tree-optimization/91033
* tree-vectorizer.h (vect_mark_stmts_to_be_vectorized,
vect_analyze_data_refs): Add bool * arguments.
* tree-vect-data-refs.c (vect_analyze_data_refs): Add fatal argument,
if failure is due to scatter/gather, set *fatal to false if non-NULL.
* tree-vect-stmts.c (vect_mark_stmts_to_be_vectorized): Likewise.
* tree-vect-loop.c (vect_analyze_loop_2): Adjust
vect_mark_stmts_to_be_vectorized and vect_analyze_data_refs callers.
* tree-vect-slp.c (vect_slp_analyze_bb_1): Adjust
vect_analyze_data_refs caller.

* gcc.target/i386/pr91033.c: New test.

From-SVN: r272989

5 years ago[Ada] Document new flags in GNATpp
Bob Duff [Wed, 3 Jul 2019 08:16:34 +0000 (08:16 +0000)]
[Ada] Document new flags in GNATpp

2019-07-03  Bob Duff  <duff@adacore.com>

gcc/ada/

* doc/gnat_ugn/gnat_utility_programs.rst: Document new flags in
GNATpp.

From-SVN: r272988

5 years ago[Ada] Forced elaboration order in Elaboration order v4.0
Hristian Kirtchev [Wed, 3 Jul 2019 08:16:29 +0000 (08:16 +0000)]
[Ada] Forced elaboration order in Elaboration order v4.0

This patch refactors the forced elaboration order functionality,
reintegrates it in Binde, and impelements it in Bindo.

------------
-- Source --
------------

--  server.ads

package Server is
end Server;

--  client.ads

with Server;

package Client is
end Client;

--  main.adb

with Client;

procedure Main is begin null; end Main;

--  duplicate_1.txt

server (spec)
client (spec)
server (spec)

--  error_unit_1.txt

no such unit
client (spec)

--  error_unit_2.txt

no such unit
client (spec)

--  error_unit_3.txt

no such unit     --  comment
client (spec)

--  error_unit_4.txt

         no such unit     --  comment

client (spec)

--  error_unit_5.txt

no such unit (body)
client (spec)

--  error_unit_6.txt

    no such unit (body)
client (spec)

--  error_unit_7.txt

    no such unit (body)    --  comment
client (spec)

--  error_unit_8.txt

    no such unit (body)--  comment
client (spec)

--  error_unit_9.txt

    no such unit--  comment
client (spec)

--  no_unit_1.txt

--  no_unit_2.txt

--  no_unit_3.txt

      --  comment

--  no_unit_4.txt

--  no_unit_5.txt

--  no_unit_6.txt

       --  comment

--  no_unit_7.txt

--  no_unit_8.txt

    --  comment
--  comment

--  ok_unit_1.txt

server (spec)
client (spec)

--  ok_unit_2.txt

    server (spec)
client (spec)

--  ok_unit_3.txt

    server (spec)
client (spec)

--  ok_unit_4.txt

    server (spec)      --  comment
client (spec)

--  ok_unit_5.txt

server (spec)
client (spec)

--  ok_unit_6.txt

server (spec)
client (spec)    --  comment

--  ok_unit_7.txt

server (spec)
client (spec)    --  comment

--  ok_unit_8.txt

    --  comment
--  comment
    server (spec)

   --  comment
--  comment

client (spec)    --  comment

--  ok_unit_9.txt

server (spec)--  comment
client (spec)

----------------------------
-- Compilation and output --
----------------------------
$ gnatmake -q main.adb
$ gnatbind -fno_unit_1.txt main.ali
$ gnatbind -fno_unit_2.txt main.ali
$ gnatbind -fno_unit_3.txt main.ali
$ gnatbind -fno_unit_4.txt main.ali
$ gnatbind -fno_unit_5.txt main.ali
$ gnatbind -fno_unit_6.txt main.ali
$ gnatbind -fno_unit_7.txt main.ali
$ gnatbind -fno_unit_8.txt main.ali
$ gnatbind -ferror_unit_1.txt main.ali
$ gnatbind -ferror_unit_2.txt main.ali
$ gnatbind -ferror_unit_3.txt main.ali
$ gnatbind -ferror_unit_4.txt main.ali
$ gnatbind -ferror_unit_5.txt main.ali
$ gnatbind -ferror_unit_6.txt main.ali
$ gnatbind -ferror_unit_7.txt main.ali
$ gnatbind -ferror_unit_8.txt main.ali
$ gnatbind -ferror_unit_9.txt main.ali
$ gnatbind -fduplicate_1.txt main.ali
$ gnatbind -fok_unit_1.txt main.ali
$ gnatbind -fok_unit_2.txt main.ali
$ gnatbind -fok_unit_3.txt main.ali
$ gnatbind -fok_unit_4.txt main.ali
$ gnatbind -fok_unit_5.txt main.ali
$ gnatbind -fok_unit_6.txt main.ali
$ gnatbind -fok_unit_7.txt main.ali
$ gnatbind -fok_unit_8.txt main.ali
$ gnatbind -fok_unit_9.txt main.ali
"no such unit": not present; ignored
"no such unit": not present; ignored
"no such unit": not present; ignored
"no such unit": not present; ignored
"no such unit%b": not present; ignored
"no such unit%b": not present; ignored
"no such unit%b": not present; ignored
"no such unit%b": not present; ignored
"no such unit": not present; ignored
server (spec) <-- client (spec)
error: duplicate_1.txt:3: duplicate unit name "server (spec)" from line 1
server (spec) <-- client (spec)
server (spec) <-- client (spec)
server (spec) <-- client (spec)
server (spec) <-- client (spec)
server (spec) <-- client (spec)
server (spec) <-- client (spec)
server (spec) <-- client (spec)
server (spec) <-- client (spec)
server (spec) <-- client (spec)

2019-07-03  Hristian Kirtchev  <kirtchev@adacore.com>

gcc/ada/

* binde.adb: Remove with clause for System.OS_Lib.
(Force_Elab_Order): Refactor the majority of the code in Butil.
Use the new forced units iterator to obtain unit names.
* bindo-builders.adb: Add with and use clauses for Binderr,
Butil, Opt, Output, Types, GNAT, and GNAT.Dynamic_HTables.  Add
a hash table which maps units to line number in the forced
elaboration order file.
(Add_Unit): New routine.
(Build_Library_Graph): Create forced edges between pairs of
units listed in the forced elaboration order file.
(Create_Forced_Edge, Create_Forced_Edges, Destroy_Line_Number,
Duplicate_Unit_Error, Hash_Unit, Internal_Unit_Info,
Is_Duplicate_Unit, Missing_Unit_Info): New routines.
* bindo-graphs.adb (Is_Internal_Unit, Is_Predefined_Unit):
Refactor some of the behavior to Bindo-Units.
* bindo-graphs.ads: Enable the enumeration literal for forced
edges.
* bindo-units.adb, bindo-units.ads (Is_Internal_Unit,
Is_Predefined_Unit): New routines.
* butil.adb: Add with and use clauses for Opt, GNAT, and
System.OS_Lib.  Add with clause for Unchecked_Deallocation.
(Has_Next, Iterate_Forced_Units, Next, Parse_Next_Unit_Name,
Read_Forced_Elab_Order_File): New routines.
* butil.ads: Add with and use clauses for Types.  Add new
iterator over the units listed in the forced elaboration order
file.
(Has_Next, Iterate_Forced_Units, Next): New routine.
* namet.adb, namet.ads (Present): New routine.

From-SVN: r272987

5 years ago[Ada] Crash on anonymous access-to-class-wide with tasks
Bob Duff [Wed, 3 Jul 2019 08:16:24 +0000 (08:16 +0000)]
[Ada] Crash on anonymous access-to-class-wide with tasks

This patch fixes a bug in which if an object declaration is of an
anonymous access type whose designated type is a limited class-wide type
(but not an interface), and the object is initialized with an allocator,
and the designated type of the allocator contains tasks, the compiler
would crash.

2019-07-03  Bob Duff  <duff@adacore.com>

gcc/ada/

* sem_ch3.adb (Access_Definition): The code was creating a
master in the case where the designated type is a class-wide
interface type. Create a master in the noninterface case as
well. That is, create a master for all limited class-wide types.

gcc/testsuite/

* gnat.dg/task2.adb, gnat.dg/task2_pkg.adb,
gnat.dg/task2_pkg.ads: New testcase.

From-SVN: r272986

5 years ago[Ada] Suppress warnings in generic instantiations with pragma Warnings
Yannick Moy [Wed, 3 Jul 2019 08:16:20 +0000 (08:16 +0000)]
[Ada] Suppress warnings in generic instantiations with pragma Warnings

Warnings issued by GNAT or GNATprove inside generic instantiations can
now be suppressed by using pragma Warnings Off/On around the instance.
This has mostly an effect on GNATprove, since GNAT typically does not
issue warnings on instantiations, only on the generic code itself.

2019-07-03  Yannick Moy  <moy@adacore.com>

gcc/ada/

* erroutc.adb (Sloc_In_Range): New function to determine whether
the range of a pragma Warnings covers a location, taking
instantiations into account.

From-SVN: r272985

5 years ago[Ada] New routine to access file on command line
Johannes Kanig [Wed, 3 Jul 2019 08:16:15 +0000 (08:16 +0000)]
[Ada] New routine to access file on command line

This patch adds a new routine to query the first file argument of the
commandline without moving to the next file. This is needed in SPARK.

There is no impact on compilation.

2019-07-03  Johannes Kanig  <kanig@adacore.com>

gcc/ada/

* osint.ads, osint.adb (Get_First_Main_File_Name): New routine
to access the first file provided on the command line.

From-SVN: r272984

5 years ago[Ada] Crash on front-end inlining of subp. with aspect specifications
Ed Schonberg [Wed, 3 Jul 2019 08:16:11 +0000 (08:16 +0000)]
[Ada] Crash on front-end inlining of subp. with aspect specifications

This patch fixes a gap in the handling of formals when inlining a call
to a subprogram marked Inline_Always. For the inlining, the formals are
replaced by the actuals in the block constructed for inlining, The
traversal that performs this replacement does not apply to aspect
specifications that may appear in the original body, because these
aspects are only indirectly reachable from the nodes to which they
apply: a separate traversal is required to perform the replacement in
the expressions for any aspect specification present in the source.

2019-07-03  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* inline.adb (Process_Formals_In_Aspects): New procedure within
Expand_Inlined_Call, to perform a replacement of references to
formals that appear in aspect specifications within the body
being inlined.

gcc/testsuite/

* gnat.dg/inline16.adb, gnat.dg/inline16_gen.adb,
gnat.dg/inline16_gen.ads, gnat.dg/inline16_types.ads: New
testcase.

From-SVN: r272983

5 years ago[Ada] Incorrect expansion on renamings of formal parameters
Justin Squirek [Wed, 3 Jul 2019 08:16:06 +0000 (08:16 +0000)]
[Ada] Incorrect expansion on renamings of formal parameters

This patch fixes an issue whereby a renaming of an unconstrained formal
parameter leads to spurious runtime errors; manifesting either as a
storage or constraint error due to incorrect bounds being assumed.

This issue also occurs when the renamings are implicit such as through
generic instantiations.

2019-07-03  Justin Squirek  <squirek@adacore.com>

gcc/ada/

* sem_ch8.adb (Analyze_Object_Renaming): Add call to search for
the appropriate actual subtype of the object renaming being
analyzed.
(Check_Constrained_Object): Minor cleanup.

gcc/testsuite/

* gnat.dg/renaming13.adb, gnat.dg/renaming14.adb: New testcases.

From-SVN: r272982

5 years ago[Ada] Refine pointer support in SPARK
Yannick Moy [Wed, 3 Jul 2019 08:16:01 +0000 (08:16 +0000)]
[Ada] Refine pointer support in SPARK

Refine the implementation of pointer support for SPARK analysis.

There is no impact on compilation.

2019-07-03  Yannick Moy  <moy@adacore.com>

gcc/ada/

* sem_spark.adb (Get_Observed_Or_Borrowed_Expr): New function to
return go through traversal function call.
(Check_Type): Consistently use underlying type.
(Get_Perm): Adapt for case of elaboration code where variables
are not declared in the environment. Remove incorrect handling
of borrow and observe.

From-SVN: r272981

5 years ago[Ada] Spurious visibility error in inlined function
Hristian Kirtchev [Wed, 3 Jul 2019 08:15:54 +0000 (08:15 +0000)]
[Ada] Spurious visibility error in inlined function

This patch corrects the use of tree replication when inlining a function
that returns an unconstrained result, and its sole statement is an
extended return statement. The use of New_Copy_Tree ensires that global
references saved in a generic template are properly carried over when
the function is instantiated and inlined.

2019-07-03  Hristian Kirtchev  <kirtchev@adacore.com>

gcc/ada/

* inline.adb (Build_Return_Object_Formal): New routine.
(Can_Split_Unconstrained_Function): Code clean up.
(Copy_Formals,Copy_Return_Object): New routines.
(Split_Unconstrained_Function): Code clean up and refactoring.

gcc/testsuite/

* gnat.dg/inline15.adb, gnat.dg/inline15_gen.adb,
gnat.dg/inline15_gen.ads, gnat.dg/inline15_types.ads: New
testcase.

From-SVN: r272980

5 years ago[Ada] Minor editorial corrections and reformatting
Gary Dismukes [Wed, 3 Jul 2019 08:15:39 +0000 (08:15 +0000)]
[Ada] Minor editorial corrections and reformatting

2019-07-03  Gary Dismukes  <dismukes@adacore.com>

gcc/ada/

* bindo-augmentors.adb, bindo-augmentors.ads,
bindo-builders.ads, bindo-elaborators.adb, sem_ch12.adb,
sem_ch13.adb, sem_spark.adb, sinfo.ads: Minor editorial
corrections and reformatting.

From-SVN: r272979

5 years ago[Ada] Improve warnings about infinite loops
Bob Duff [Wed, 3 Jul 2019 08:15:28 +0000 (08:15 +0000)]
[Ada] Improve warnings about infinite loops

The compiler now has fewer false alarms when warning about infinite
loops. For example, a loop of the form "for X of A ...", where A is an
array, cannot be infinite.  The compiler no longer warns in this case.

2019-07-03  Bob Duff  <duff@adacore.com>

gcc/ada/

* sem_warn.adb (Check_Infinite_Loop_Warning): Avoid the warning
if an Iterator_Specification is present.

gcc/testsuite/

* gnat.dg/warn20.adb, gnat.dg/warn20_pkg.adb,
gnat.dg/warn20_pkg.ads: New testcase.

From-SVN: r272978

5 years ago[Ada] Document default new-line behavior for GNATpp
Bob Duff [Wed, 3 Jul 2019 08:15:03 +0000 (08:15 +0000)]
[Ada] Document default new-line behavior for GNATpp

2019-07-03  Bob Duff  <duff@adacore.com>

gcc/ada/

* doc/gnat_ugn/gnat_utility_programs.rst: Document default
new-line behavior.

From-SVN: r272977

5 years ago[Ada] ABE checks v3.0, foundations of Elaboration order v4.0
Hristian Kirtchev [Wed, 3 Jul 2019 08:14:57 +0000 (08:14 +0000)]
[Ada] ABE checks v3.0, foundations of Elaboration order v4.0

------------------------
-- Elaboration checks --
------------------------

The dynamic ABE checks model now emits the same diagnostics as those of the
static ABE checks model.

The ABE checks mechanism has been redesigned and refactored in the face of
increasing requirements. Most of the functionality can now be toggled, thus
allowing for various combinations of behavior. The combinations are defined
as "initial states" and may be further altered.

Scenarios and targets have been distinctly separated at the higher level,
instead of directly working with nodes and entitites. Scenarios and targets
now carry a representation which removes the need to constantly recompute
relevant attributes, and offers a common interface for the various processors.

Most processing has now been refactored into "services" which perform a single
ABE-related function.

-----------------------
-- Elaboration order --
-----------------------

A new elaboration order mechanism based on the use of an invocation graph to
provide extra information about the flow of execution at elaboration time has
been introduced.

The ABE checks mechanism has been altered to encode pieces of the invocation
graph in the associated ALI files of units.

The new elaboration order mechanism reconstructs the full invocation graph at
bind time, and coupled with the library item graph, determines the elaboration
order of units.

The new elaboration order mechanism is currently inaccessible.

------------
-- Source --
------------

--  pack.ads

package Pack is
   procedure ABE_Proc;
   procedure Safe_Proc;
end Pack;

--  pack.adb

with Ada.Text_IO; use Ada.Text_IO;

package body Pack is
   function Call_Proc (ABE : Boolean) return Integer;

   procedure Safe_Proc is
   begin
      Put_Line ("safe");
   end Safe_Proc;

   function Call_Proc (ABE : Boolean) return Integer is
   begin
      if ABE then
         ABE_Proc;
      else
         Safe_Proc;
      end if;

      return 0;
   end Call_Proc;

   Elab_1 : constant Integer := Call_Proc (ABE => False);
   Elab_2 : constant Integer := Call_Proc (ABE => True);

   procedure ABE_Proc is
   begin
      Put_Line ("ABE");
   end ABE_Proc;
end Pack;

--  main.adb

with Pack;

procedure Main is begin null; end Main;

----------------------------
-- Compilation and output --
----------------------------

$ gnatmake -f -q -gnatE main.adb
$ ./main
$ gnatmake -f -q -gnatE main.adb -gnatDG -gnatwL
$ grep -c "safeE" pack.adb.dg
pack.adb:14:10: warning: cannot call "ABE_Proc" before body seen
pack.adb:14:10: warning: Program_Error may be raised at run time
pack.adb:14:10: warning:   body of unit "Pack" elaborated
pack.adb:14:10: warning:   function "Call_Proc" called at line 22
pack.adb:14:10: warning:   procedure "ABE_Proc" called at line 14
pack.adb:14:10: warning: cannot call "ABE_Proc" before body seen
pack.adb:14:10: warning: Program_Error may be raised at run time
pack.adb:14:10: warning:   body of unit "Pack" elaborated
pack.adb:14:10: warning:   function "Call_Proc" called at line 23
pack.adb:14:10: warning:   procedure "ABE_Proc" called at line 14
safe

raised PROGRAM_ERROR : pack.adb:14 access before elaboration
0

2019-07-03  Hristian Kirtchev  <kirtchev@adacore.com>

gcc/ada/

* ali.adb: Add with and use clauses for GNAT,
GNAT.Dynamic_HTables, and Snames.  Add a map from invocation
signature records to invocation signature ids.  Add various
encodings of invocation-related attributes.  Sort and update
table Known_ALI_Lines.
(Add_Invocation_Construct, Add_Invocation_Relation,
Body_Placement_Kind_To_Code, Code_To_Body_Placement_Kind,
Code_To_Invocation_Construct_Kind, Code_To_Invocation_Kind,
Code_To_Invocation_Graph_Line_Kind, Destroy, Hash): New
routines.
(Initialize_ALI): Sort the initialization sequence. Add
initialization for all invocation-related tables.
(Invocation_Construct_Kind_To_Code,
Invocation_Graph_Line_Kind_To_Code, Invocation_Kind_To_Code,
Invocation_Signature_Of, Present): New routines.
(Scan_ALI): Add the default values for invocation-related ids.
Scan invocation graph lines.
(Scan_Invocation_Graph_Line): New routine.
* ali.ads: Add with clause for GNAT.Dynamic_Tables.  Add types
for invocation constructs, relations, and signatures.  Add
tables for invocation constructs, relations, and signatures.
Update Unit_Record to capture invocation-related ids.  Relocate
table Unit_Id_Tables and subtypes Unit_Id_Table, Unit_Id_Array
from Binde.
(Add_Invocation_Construct, Add_Invocation_Relation,
Body_Placement_Kind_To_Code, Code_To_Body_Placement_Kind,
Code_To_Invocation_Construct_Kind, Code_To_Invocation_Kind,
Code_To_Invocation_Graph_Line_Kind,
Invocation_Construct_Kind_To_Code,
Invocation_Graph_Line_Kind_To_Code, Invocation_Kind_To_Code,
Invocation_Signature_Of, Present): New routines.
* binde.adb: Add with and use clause for Types.  Add use clause
for ALI.Unit_Id_Tables;
* binde.ads: Relocate table Unit_Id_Tables and subtypes
Unit_Id_Table, Unit_Id_Array to ALI.
* bindgen.adb: Remove with and use clause for ALI.
* bindgen.ads: Remove with and use clause for Binde.  Add with
and use clause for ALI.
* bindo.adb,  bindo.ads, bindo-augmentors.adb,
bindo-augmentors.ads, bindo-builders.adb, bindo-builders.ads,
bindo-diagnostics.adb, bindo-diagnostics.ads,
bindo-elaborators.adb, bindo-elaborators.ads, bindo-graphs.adb,
bindo-graphs.ads, bindo-units.adb, bindo-units.ads,
bindo-validators.adb, bindo-validators.ads, bindo-writers.adb,
bindo-writers.ads: New units.
* debug.adb: Use and describe GNAT debug switches -gnatd_F and
-gnatd_G.  Add GNATbind debug switches in the ranges dA .. dZ,
d.a .. d.z, d.A .. d.Z, d.1 .. d.9, d_a .. d_z, d_A .. d_Z, and
d_1 .. d_9.  Use and describe GNATbind debug switches -d_A,
-d_I, -d_L, -d_N, -d_O, -d_T, and -d_V.
* exp_util.adb, exp_util.ads (Exceptions_OK): Relocate to
Sem_Util.
* gnatbind.adb: Add with and use clause for Bindo.  Use the new
Bindo elaboration order only when -d_N is in effect.
* lib-writ.adb
(Column, Extra, Invoker, Kind, Line, Locations, Name, Placement,
Scope, Signature, Target): New routines.
(Write_ALI): Output all invocation-related data.
(Write_Invocation_Graph): New routine.
* lib-writ.ads: Document the invocation graph ALI line.
* namet.adb, namet.ads (Present): New routines.
* sem_ch8.adb (Find_Direct_Name): Capture the status of
elaboration checks and warnings of an identifier.
(Find_Expanded_Name): Capture the status of elaboration checks
and warnings of an expanded name.
* sem_ch12.adb (Analyze_Generic_Package_Declaration): Ensure
that invocation graph-related data within the body of the main
unit is encoded in the ALI file.
(Analyze_Generic_Subprogram_Declaration): Ensure that invocation
graph-related data within the body of the main unit is encoded
in the ALI file.
(Analyze_Package_Instantiation): Perform minimal decoration of
the instance entity.
(Analyze_Subprogram_Instantiation): Perform minimal decoration
of the instance entity.
* sem_elab.adb: Perform heavy refactoring of all code. The unit
is now split into "services" which specialize in one area of ABE
checks.  Add processing in order to capture invocation-graph
related attributes of the main unit, and encode them in the ALI
file.  The Processing phase can now operate in multiple modes,
all described by type Processing_Kind.  Scenarios and targets
are now distinct at the higher level, and carry their own
representations. This eliminates the need to constantly
recompute their attributes, and offers the various processors a
uniform interface.  The various initial states of the Processing
phase are now encoded using type Processing_In_State, and
xxx_State constants.
* sem_elab.ads: Update the literals of type
Enclosing_Level_Kind.  Add Inline pragmas on several routines.
* sem_prag.adb (Process_Inline): Ensure that invocation
graph-related data within the body of the main unit is encoded
in the ALI file.
* sem_util.adb (Enclosing_Generic_Body, Enclosing_Generic_Unit):
Code clean up.
(Exceptions_OK): Relocated from Sem_Util.
(Mark_Save_Invocation_Graph_Of_Body): New routine.
* sem_util.ads (Exceptions_OK): Relocated from Sem_Util.
(Mark_Save_Invocation_Graph_Of_Body): New routine.
* sinfo.adb (Is_Elaboration_Checks_OK_Node): Now applicable to
N_Variable_Reference_Marker.
(Is_Elaboration_Warnings_OK_Node): Now applicable to
N_Expanded_Name, N_Identifier, N_Variable_Reference_Marker.
(Is_Read): Use Flag4.
(Is_SPARK_Mode_On_Node): New applicable to
N_Variable_Reference_Marker.
(Is_Write): Use Flag5.
(Save_Invocation_Graph_Of_Body): New routine.
(Set_Is_Elaboration_Checks_OK_Node): Now applicable to
N_Variable_Reference_Marker.
(Set_Is_Elaboration_Warnings_OK_Node): Now applicable to
N_Expanded_Name, N_Identifier, N_Variable_Reference_Marker.
(Set_Is_SPARK_Mode_On_Node): New applicable to
N_Variable_Reference_Marker.
(Set_Save_Invocation_Graph_Of_Body): New routine.
* sinfo.ads: Update the documentation of attributes
Is_Elaboration_Checks_OK_Node, Is_Elaboration_Warnings_OK_Node,
Is_SPARK_Mode_On_Node.  Update the flag usage of attributes
Is_Read, Is_Write.  Add attribute Save_Invocation_Graph_Of_Body
and update its occurrence in nodes.
(Save_Invocation_Graph_Of_Body): New routine along with pragma
Inline.
(Set_Save_Invocation_Graph_Of_Body): New routine along with
pragma Inline.
* switch-b.adb (Scan_Binder_Switches): Refactor the scanning of
debug switches.
(Scan_Debug_Switches): New routine.
* libgnat/g-dynhta.adb, libgnat/g-dynhta.ads (Contains): New routine.
* libgnat/g-graphs.adb (Associate_Vertices): Update the use of
Component_Vertex_Iterator.
(Contains_Component, Contains_Edge, Contains_Vertex, Has_Next):
Reimplemented.
(Iterate_Component_Vertices): New routine.
(Iterate_Vertices): Removed.
(Next): Update the parameter profile.
(Number_Of_Component_Vertices, Number_Of_Outgoing_Edges): New
routines.
* libgnat/g-graphs.ads: Update the initialization of
No_Component.  Add type Component_Vertex_Iterator.  Remove type
Vertex_Iterator.
(Has_Next): Add new versions and remove old ones.
(Iterate_Component_Vertices): New routine.
(Iterate_Vertices): Removed.
(Next): Add new versions and remove old ones.
(Number_Of_Component_Vertices, Number_Of_Outgoing_Edges): New
routines.
* libgnat/g-sets.adb (Contains): Reimplemented.
* gcc-interface/Make-lang.in (GNATBIND_OBJS): Add
GNAT.Dynamic_HTables, GNAT.Graphs and Bindo units.
* rtsfind.ads: Remove extra space.

From-SVN: r272976

5 years ago[Ada] SPARK pointer support extended to local borrowers and observers
Yannick Moy [Wed, 3 Jul 2019 08:14:52 +0000 (08:14 +0000)]
[Ada] SPARK pointer support extended to local borrowers and observers

SPARK rules allow local borrowers and observers to be declared. During
their lifetime, the access to the borrowed/observed object is
restricted.

There is no impact on compilation.

2019-07-03  Yannick Moy  <moy@adacore.com>

gcc/ada/

* sem_spark.adb: Add support for locally borrowing and observing
a path.
(Get_Root_Object): Add parameter Through_Traversal to denote
when we are interesting in getting to the traversed parameter.
(Is_Prefix_Or_Almost): New function to support detection of
illegal access to borrowed or observed paths.
(Check_Pragma): Add analysis of assertion pragmas.

From-SVN: r272975

5 years ago[Ada] Spurious error with static predicate in generic unit
Ed Schonberg [Wed, 3 Jul 2019 08:14:47 +0000 (08:14 +0000)]
[Ada] Spurious error with static predicate in generic unit

This patch fixes a spurious error in a generic unit that invludes a
subtype with a static predicate, when the type is used in a case
expression.

2019-07-03  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* sem_ch13.adb (Build_Predicate_Functions): In a generic context
we do not build the bodies of predicate fuctions, but the
expression in a static predicate must be elaborated to allow
case coverage checking within the generic unit.
(Build_Discrete_Static_Predicate): In a generic context, return
without building function body once the
Static_Discrete_Predicate expression for the type has been
constructed.

gcc/testsuite/

* gnat.dg/predicate6.adb, gnat.dg/predicate6.ads: New testcase.
* gnat.dg/static_pred1.adb: Remove expected error.

From-SVN: r272974

5 years ago[Ada] Minor reformatting
Hristian Kirtchev [Wed, 3 Jul 2019 08:14:43 +0000 (08:14 +0000)]
[Ada] Minor reformatting

2019-07-03  Hristian Kirtchev  <kirtchev@adacore.com>

gcc/ada/

* bindgen.adb, inline.adb, layout.adb, sem_ch12.adb,
sem_ch13.adb, sem_ch7.adb, styleg.adb: Minor reformatting.

From-SVN: r272973

5 years ago[Ada] Style check for mixed-case identifiers
Bob Duff [Wed, 3 Jul 2019 08:14:38 +0000 (08:14 +0000)]
[Ada] Style check for mixed-case identifiers

This patch implements a new switch, -gnatyD, enables a style check that
requires defining identifiers to be in mixed case.

2019-07-03  Bob Duff  <duff@adacore.com>

gcc/ada/

* par-ch3.adb (P_Defining_Identifier): Call
Check_Defining_Identifier_Casing.
* style.ads, styleg.ads, styleg.adb
(Check_Defining_Identifier_Casing): New procedure to check for
mixed-case defining identifiers.
* stylesw.ads, stylesw.adb (Style_Check_Mixed_Case_Decls): New
flag for checking for mixed-case defining identifiers.
* doc/gnat_ugn/building_executable_programs_with_gnat.rst:
Document new feature.
* gnat_ugn.texi: Regenerate.

From-SVN: r272972

5 years ago[Ada] Extend -gnatw.z warning to array types
Eric Botcazou [Wed, 3 Jul 2019 08:14:33 +0000 (08:14 +0000)]
[Ada] Extend -gnatw.z warning to array types

The -gnatw.z switch causes the compiler to issue a warning on record
types subject to both an alignment clause and a size clause, when the
specified size is not a multiple of the alignment in bits, because this
means that the Object_Size will be strictly larger than the specified
size.

It makes sense to extend this warning to array types, but not to the
cases of bit-packed arrays where the size is not a multiple of storage
unit and the specified alignment is the minimum one, because there would
be no way to get rid of it apart from explicitly silencing it.

The compiler must issue the warning:

p.ads:5:03: warning: size is not a multiple of alignment for "Triplet"
p.ads:5:03: warning: size of 24 specified at line 4
p.ads:5:03: warning: Object_Size will be increased to 32

on the following package:

package P is

  type Triplet is new String (1 .. 3);
  for Triplet'Size use 24;
  for Triplet'Alignment use 4;

  type Arr is array (1 .. 7) of Boolean;
  pragma Pack (Arr);
  for Arr'Size use 7;
  for Arr'Alignment use 1;

end P;

2019-07-03  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* doc/gnat_ugn/building_executable_programs_with_gnat.rst
(Warning message control): Document that -gnatw.z/Z apply to
array types.
* freeze.adb (Freeze_Entity): Give -gnatw.z warning for array
types as well, but not if the specified alignment is the minimum
one.
* gnat_ugn.texi: Regenerate.

From-SVN: r272971

5 years ago[Ada] Spell "laid" correctly
Bob Duff [Wed, 3 Jul 2019 08:14:29 +0000 (08:14 +0000)]
[Ada] Spell "laid" correctly

2019-07-03  Bob Duff  <duff@adacore.com>

gcc/ada/

* einfo.ads, exp_util.adb, layout.ads, sinfo.ads: Spell "laid"
correctly.

From-SVN: r272970

5 years ago[Ada] Spurious error on dynamic predicate in a generic context
Ed Schonberg [Wed, 3 Jul 2019 08:14:24 +0000 (08:14 +0000)]
[Ada] Spurious error on dynamic predicate in a generic context

This patch fixes a spurious error on the conformance checking between
the expression for an aspect analyzed at the freeze point of the type,
and the analysis of a copy of the expression performed at the end of the
enclosing list of declarationss. In a generic context the first may not
have been analyzed yet and this must be done before the conformance
check.

2019-07-03  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* sem_ch13.adb (Analyze_Attribute_Definition_Clause): No error
message on attribute applied to a renaming when the renamed
object is an aggregate (from code reading).
(Check_Aspect_At_End_Of_Declarations): In a generic context
where freeze nodes are not generated, the original expression
for an aspect may need to be analyzed to precent spurious
conformance errors when compared with the expression that is
anakyzed at the end of the current declarative list.

gcc/testsuite/

* gnat.dg/predicate5.adb, gnat.dg/predicate5.ads: New testcase.

From-SVN: r272969

5 years ago[Ada] Fix bogus error on array with overaligned scalar component
Eric Botcazou [Wed, 3 Jul 2019 08:14:15 +0000 (08:14 +0000)]
[Ada] Fix bogus error on array with overaligned scalar component

The compiler would wrongly reject an alignment clause larger than 8 on
the component type of an array of scalars, which is valid albeit
pathological.

2019-07-03  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* layout.adb (Layout_Type): Do not set the component size of an
array with a scalar component if the component type is
overaligned.

gcc/testsuite/

* gnat.dg/alignment14.adb: New testcase.

From-SVN: r272968

5 years ago[Ada] Make loop labels unique for front-end inlined calls
Ed Schonberg [Wed, 3 Jul 2019 08:14:10 +0000 (08:14 +0000)]
[Ada] Make loop labels unique for front-end inlined calls

This patch transforms loop labels in the body of subprograms that are to
be inlined by the front-end, to prevent accidental duplication of loop
labels, which might make the resulting source illegal.

----
Source program:
----
package P is
   procedure Get_Rom_Addr_Offset
     with Inline_Always;
end P;
----
package body P is
   procedure Get_Rom_Addr_Offset is
      X : Integer;
   begin
      Main_Block :
      for I in 1 .. 10 loop
         X := 2;
         exit Main_Block when I > 4;
      other_loop:
         for J in character'('a') .. 'z' loop
            if I < 5 then
               exit Main_Block when J = 'k';
            else
               Exit Other_Loop;
            end if;
         end loop other_loop;
      end loop Main_Block;
   end Get_Rom_Addr_Offset;

   procedure P2 is
   begin
      Main_Block :
      for I in 1 .. 1 loop
         Get_Rom_Addr_Offset;
      end loop Main_Block;
   end P2;
end P;
----
Command:

   gcc -c -gnatN -gnatd.u -gnatDG p.adb

----
Output
----

package body p is

   procedure p__get_rom_addr_offset is
      x : integer;
      other_loop : label
      main_block : label
   begin
      main_block : for i in 1 .. 10 loop
         x := 2;
         exit main_block when i > 4;
         other_loop : for j in 'a' .. 'z' loop
            if i < 5 then
               exit main_block when j = 'k';
            else
               exit other_loop;
            end if;
         end loop other_loop;
      end loop main_block;
      return;
   end p__get_rom_addr_offset;

   procedure p__p2 is
      main_block : label
   begin
      main_block : for i in 1 .. 1 loop
         B6b : declare
            x : integer;
            other_loopL10b : label
            main_blockL9b : label
         begin
            main_blockL9b : for i in 1 .. 10 loop
               x := 2;
               exit main_blockL9b when i > 4;
               other_loopL10b : for j in 'a' .. 'z' loop
                  if i < 5 then
                     exit main_blockL9b when j = 'k';
                  else
                     exit other_loopL10b;
                  end if;
               end loop other_loopL10b;
            end loop main_blockL9b;
         end B6b;
      end loop main_block;
      return;
   end p__p2;
begin
   null;
end p;

2019-07-03  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* inline.adb (Make_Loop_Labels_Unique):  New procedure to modify
the source code of subprograms that are inlined by the
front-end, to prevent accidental duplication between loop labels
in the inlined code and the code surrounding the inlined call.

From-SVN: r272967

5 years ago[Ada] Update the section on resolving elaboration circularities
Hristian Kirtchev [Wed, 3 Jul 2019 08:14:05 +0000 (08:14 +0000)]
[Ada] Update the section on resolving elaboration circularities

2019-07-03  Hristian Kirtchev  <kirtchev@adacore.com>

gcc/ada/

* doc/gnat_ugn/elaboration_order_handling_in_gnat.rst: Update
the section on resolving elaboration circularities to eliminate
certain combinations of switches which together do not produce
the desired effect and confuse users.
* gnat_ugn.texi: Regenerate.

From-SVN: r272966

5 years ago[Ada] Add a gnatbind option to generate C code
Arnaud Charlet [Wed, 3 Jul 2019 08:14:00 +0000 (08:14 +0000)]
[Ada] Add a gnatbind option to generate C code

2019-07-03  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* bindgen.adb (Gen_Main): Disable generation of reference to
Ada_Main_Program_Name for CCG.
* bindusg.adb (Display): Add -G to the command-line usage for
gnatbind.
* opt.ads (Generate_C_Code): Update comment.
* switch-b.adb (Scan_Binder_Switches): Add handling for -G.

From-SVN: r272965

5 years ago[Ada] Do not consider inlined subprograms when generating C code
Arnaud Charlet [Wed, 3 Jul 2019 08:13:55 +0000 (08:13 +0000)]
[Ada] Do not consider inlined subprograms when generating C code

2019-07-03  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* sem_ch7.adb (Has_Referencer): Do not consider inlined
subprograms when generating C code, which allows us to generate
static inline subprograms.

From-SVN: r272964

5 years ago[Ada] Missing consistency check for constant modifier
Justin Squirek [Wed, 3 Jul 2019 08:13:51 +0000 (08:13 +0000)]
[Ada] Missing consistency check for constant modifier

This patch fixes an issue whereby instantiations of generic packages
were incorrectly allowed despite formal and actual subprograms not
having matching declarations with anonymous constant access type
parameters.

------------
-- Source --
------------

-- gen1.ads

package Gen1 is
   generic
      with procedure View (IA : not null access constant Integer);
   procedure Dispatch (IA : access Integer);
end;

-- gen2.adb

package body Gen1 is
   procedure Dispatch (IA : access Integer) is
   begin
      View (IA);
   end;
end;

-- bad1.ads

with Gen1;
package Bad1 is
   procedure Bad_View (IA : not null access Integer);
   procedure Bad_Dispatch is new Gen1.Dispatch (Bad_View);
end;

-- bad1.adb

package body Bad1 is
   procedure Bad_View (IA : not null access Integer) is
   begin
      IA.all := IA.all + 1;
   end;
end;

-- gen2.ads

package Gen2 is
   generic
      with procedure View (IA : access constant Integer);
   procedure Dispatch (IA : access Integer);
end;

-- gen2.adb

package body Gen2 is
   procedure Dispatch (IA : access Integer) is
   begin
      View (IA);
   end;
end;

-- bad2.ads

with Gen2;
package Bad2 is
   procedure Bad_View (IA : access Integer);
   procedure Bad_Dispatch is new Gen2.Dispatch (Bad_View);
end;

-- bad2.adb

package body Bad2 is
   procedure Bad_View (IA : access Integer) is
   begin
      IA.all := IA.all + 1;
   end;
end;

-----------------
-- Compilation --
-----------------

$ gnatmake -q bad1.adb
$ bad1.ads:4:04: instantiation error at gen1.ads:3
$ bad1.ads:4:04: not mode conformant with declaration at line 3
$ bad1.ads:4:04: constant modifier does not match
$ gnatmake: "bad1.adb" compilation error
$ gnatmake -q bad2.adb
$ bad2.ads:4:04: instantiation error at gen2.ads:3
$ bad2.ads:4:04: not mode conformant with declaration at line 3
$ bad2.ads:4:04: constant modifier does not match
$ gnatmake: "bad2.adb" compilation error

2019-07-03  Justin Squirek  <squirek@adacore.com>

gcc/ada/

* sem_ch6.adb (Check_Conformance): Add expression checking for
constant modifiers in anonymous access types (in addition to
"non-null" types) so that they are considered "matching" for
subsequent conformance tests.

From-SVN: r272963

5 years ago[Ada] Clarify wording on documentation for No_Multiple_Elaboration
Arnaud Charlet [Wed, 3 Jul 2019 08:13:46 +0000 (08:13 +0000)]
[Ada] Clarify wording on documentation for No_Multiple_Elaboration

2019-07-03  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* doc/gnat_rm/standard_and_implementation_defined_restrictions.rst:
Clarify wording on No_Multiple_Elaboration.
* gnat_rm.texi: Regenerate.

From-SVN: r272962

5 years ago[Ada] Spurious error on predicate of subtype in generic
Ed Schonberg [Wed, 3 Jul 2019 08:13:41 +0000 (08:13 +0000)]
[Ada] Spurious error on predicate of subtype in generic

This patch fixes a spurious error on a dynamic predicate of a record
subtype when the expression for the predicate includes a selected
component that denotes a component of the subtype.

2019-07-03  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* sem_ch8.adb (Find_Selected_Component): If the prefix is the
current instance of a type or subtype, complete the resolution
of the name by finding the component of the type denoted by the
selector name.

gcc/testsuite/

* gnat.dg/predicate4.adb, gnat.dg/predicate4_pkg.ads: New
testcase.

From-SVN: r272961

5 years ago[Ada] Document that boolean types with convention C now map to C99 bool
Eric Botcazou [Wed, 3 Jul 2019 08:13:34 +0000 (08:13 +0000)]
[Ada] Document that boolean types with convention C now map to C99 bool

2019-07-03  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* doc/gnat_rm/interfacing_to_other_languages.rst (Interfacing to C):
Document that boolean types with convention C now map to C99 bool.
* gnat_rm.texi: Regenerate.

From-SVN: r272960

5 years ago[Ada] Exp_Attr: remove dead code
Javier Miranda [Wed, 3 Jul 2019 08:13:29 +0000 (08:13 +0000)]
[Ada] Exp_Attr: remove dead code

2019-07-03  Javier Miranda  <miranda@adacore.com>

gcc/ada/

* exp_attr.adb (Expand_Min_Max_Attribute): Code cleanup:
removing code that it is now never executed in the CCG compiler
(dead code).

From-SVN: r272959

5 years agotree-core.h (enum omp_clause_code): Add OMP_CLAUSE__SCANTEMP_ clause.
Jakub Jelinek [Wed, 3 Jul 2019 05:03:58 +0000 (07:03 +0200)]
tree-core.h (enum omp_clause_code): Add OMP_CLAUSE__SCANTEMP_ clause.

* tree-core.h (enum omp_clause_code): Add OMP_CLAUSE__SCANTEMP_
clause.
* tree.h (OMP_CLAUSE_DECL): Use OMP_CLAUSE__SCANTEMP_ instead of
OMP_CLAUSE__CONDTEMP_ as range's upper bound.
(OMP_CLAUSE__SCANTEMP__ALLOC, OMP_CLAUSE__SCANTEMP__CONTROL): Define.
* tree.c (omp_clause_num_ops, omp_clause_code_name): Add
OMP_CLAUSE__SCANTEMP_ entry.
(walk_tree_1): Handle OMP_CLAUSE__SCANTEMP_.
* tree-pretty-print.c (dump_omp_clause): Likewise.
* tree-nested.c (convert_nonlocal_omp_clauses,
convert_local_omp_clauses): Likewise.
* omp-general.h (struct omp_for_data): Add have_scantemp and
have_nonctrl_scantemp members.
* omp-general.c (omp_extract_for_data): Initialize them.
* omp-low.c (struct omp_context): Add scan_exclusive member.
(scan_omp_1_stmt): Don't unnecessarily mask gimple_omp_for_kind
result again with GF_OMP_FOR_KIND_MASK.  Initialize also
ctx->scan_exclusive.
(lower_rec_simd_input_clauses): Use ctx->scan_exclusive instead
of !ctx->scan_inclusive.
(lower_rec_input_clauses): Simplify gimplification of dtors using
gimplify_and_add.  For non-is_simd test OMP_CLAUSE_REDUCTION_INSCAN
rather than rvarp.  Handle OMP_CLAUSE_REDUCTION_INSCAN in worksharing
loops.  Don't add barrier for reduction_omp_orig_ref if
ctx->scan_??xclusive.
(lower_reduction_clauses): Don't do anything for ctx->scan_??xclusive.
(lower_omp_scan): Use ctx->scan_exclusive instead
of !ctx->scan_inclusive.  Handle worksharing loops with inscan
reductions.  Use new_vard != new_var instead of repeated
omp_is_reference calls.
(omp_find_scan, lower_omp_for_scan): New functions.
(lower_omp_for): Call lower_omp_for_scan for worksharing loops with
inscan reductions.
* omp-expand.c (expand_omp_scantemp_alloc): New function.
(expand_omp_for_static_nochunk): Handle fd->have_nonctrl_scantemp
and fd->have_scantemp.

* c-c++-common/gomp/scan-3.c (f1): Don't expect a sorry message.
* c-c++-common/gomp/scan-5.c (foo): Likewise.

* testsuite/libgomp.c++/scan-1.C: New test.
* testsuite/libgomp.c++/scan-2.C: New test.
* testsuite/libgomp.c++/scan-3.C: New test.
* testsuite/libgomp.c++/scan-4.C: New test.
* testsuite/libgomp.c++/scan-5.C: New test.
* testsuite/libgomp.c++/scan-6.C: New test.
* testsuite/libgomp.c++/scan-7.C: New test.
* testsuite/libgomp.c++/scan-8.C: New test.
* testsuite/libgomp.c/scan-1.c: New test.
* testsuite/libgomp.c/scan-2.c: New test.
* testsuite/libgomp.c/scan-3.c: New test.
* testsuite/libgomp.c/scan-4.c: New test.
* testsuite/libgomp.c/scan-5.c: New test.
* testsuite/libgomp.c/scan-6.c: New test.
* testsuite/libgomp.c/scan-7.c: New test.
* testsuite/libgomp.c/scan-8.c: New test.

From-SVN: r272958

5 years agogimplify.c (gimplify_scan_omp_clauses): For inscan reductions on worksharing loop...
Jakub Jelinek [Wed, 3 Jul 2019 04:56:25 +0000 (06:56 +0200)]
gimplify.c (gimplify_scan_omp_clauses): For inscan reductions on worksharing loop propagate it as shared clause to...

* gimplify.c (gimplify_scan_omp_clauses): For inscan reductions
on worksharing loop propagate it as shared clause to containing
combined parallel.

* c-omp.c (c_omp_split_clauses): Put OMP_CLAUSE_REDUCTION_INSCAN
clauses on OMP_FOR rather than OMP_PARALLEL when OMP_FOR is combined
with OMP_PARALLEL.

* c-c++-common/gomp/scan-5.c: New test.

From-SVN: r272957

5 years agoomp-expand.c (expand_omp_for_static_nochunk, [...]): For nowait worksharing loop...
Jakub Jelinek [Wed, 3 Jul 2019 04:51:45 +0000 (06:51 +0200)]
omp-expand.c (expand_omp_for_static_nochunk, [...]): For nowait worksharing loop with conditional lastprivate clause(s)...

* omp-expand.c (expand_omp_for_static_nochunk,
expand_omp_for_static_chunk): For nowait worksharing loop with
conditional lastprivate clause(s), emit GOMP_loop_end_nowait call
at the end.

* c-c++-common/gomp/lastprivate-conditional-5.c: New test.

From-SVN: r272956

5 years agocompiler: rework type and package tracking in exporter
Ian Lance Taylor [Wed, 3 Jul 2019 00:56:35 +0000 (00:56 +0000)]
compiler: rework type and package tracking in exporter

    Revamps the way the exporter tracks exported types and imported
    packages that need to be mentioned in the export data.

    The previous implementation wasn't properly handling the case where an
    exported non-inlinable function refers to an imported type whose
    method set includes an inlinable function whose body makes a call to a
    function in another package that's not directly used in the original
    package.

    This patch integrates together two existing traversal helper classes,
    "Collect_references_from_inline" and "Find_types_to_prepare" into a
    single helper "Collect_export_references", so as to have common/shared
    code that looks for indirectly imported packages.

    Fixes golang/go#32778

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/183850

From-SVN: r272955

5 years agore PR testsuite/91065 (gcc.dg/plugin/start_unit_plugin.c uses ggc memory without...
Joern Rennecke [Wed, 3 Jul 2019 00:22:53 +0000 (00:22 +0000)]
re PR testsuite/91065 (gcc.dg/plugin/start_unit_plugin.c uses ggc memory without registering a root_tab)

        PR testsuite/91065
        * testsuite/gcc.dg/plugin/start_unit_plugin.c: Register a root tab
        to reference fake_var.

From-SVN: r272954

5 years agoDaily bump.
GCC Administrator [Wed, 3 Jul 2019 00:16:15 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r272953

5 years agore PR tree-optimization/90883 (Generated code is worse if returned struct is unnamed)
Jeff Law [Tue, 2 Jul 2019 23:01:53 +0000 (17:01 -0600)]
re PR tree-optimization/90883 (Generated code is worse if returned struct is unnamed)

PR tree-optimization/90883
* g++.dg/tree-ssa/pr90883.c: Add -Os.  Check dse2 for the
deleted store on some targets.

From-SVN: r272949

5 years agore PR preprocessor/90581 (provide an option to adjust the maximum depth of nested...
Qing Zhao [Tue, 2 Jul 2019 20:23:30 +0000 (20:23 +0000)]
re PR preprocessor/90581 (provide an option to adjust the maximum depth of nested #include)

PR preprocessor/90581
Add a cpp option -fmax-include-depth to set the maximum depth of the nested #include.

From-SVN: r272948

5 years ago[PATCH, Ada, Darwin, PPC] PPC Darwin has stack check probes.
Iain Sandoe [Tue, 2 Jul 2019 19:03:48 +0000 (19:03 +0000)]
[PATCH, Ada, Darwin, PPC] PPC Darwin has stack check probes.

On PPC, Darwin uses the same code as other parts of the port.

2019-07-02  Iain Sandoe  <iain@sandoe.co.uk>

* libgnat/system-darwin-ppc.ads: Set Stack_Check_Probes True for
PPC Darwin.

From-SVN: r272947

5 years agooptabs.def (movmem_optab): Add movmem back for memmove().
Aaron Sawdey [Tue, 2 Jul 2019 18:51:23 +0000 (13:51 -0500)]
optabs.def (movmem_optab): Add movmem back for memmove().

2019-07-02  Aaron Sawdey  <acsawdey@linux.ibm.com>

* optabs.def (movmem_optab): Add movmem back for memmove().
* doc/md.texi: Add description of movmem pattern for overlapping move.

From-SVN: r272946

5 years agocompiler: use builtin memset for non-pointer memclr
Cherry Zhang [Tue, 2 Jul 2019 16:47:48 +0000 (16:47 +0000)]
compiler: use builtin memset for non-pointer memclr

    For zeroing a range of memory that doesn't contain pointer, we
    can use builtin memset directly.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184438

* go-gcc.cc (Gcc_backend::Gcc_backend): Define __builtin_memset.

From-SVN: r272944

5 years agommx.md (mmx_pack<s_trunsuffix>swb): Use TARGET_SSE2 && SSE_REGNO_P in split condition.
Uros Bizjak [Tue, 2 Jul 2019 15:48:36 +0000 (17:48 +0200)]
mmx.md (mmx_pack<s_trunsuffix>swb): Use TARGET_SSE2 && SSE_REGNO_P in split condition.

* config/i386/mmx.md (mmx_pack<s_trunsuffix>swb):
Use TARGET_SSE2 && SSE_REGNO_P in split condition.
(mmx_packssdw): Ditto.
(mmx_punpckhbw): Ditto.
(mmx_punpcklbw): Ditto.
(mmx_punpckhwd): Ditto.
(mmx_punpcklwd): Ditto.
(mmx_punpckhdq): Ditto.
(mmx_punpckldq): Ditto.
(*vec_dupv4hi): Ditto.
(*vec_dupv2si): Ditto.
(mmx_pmovmskb): Ditto.
* config/i386/sse.md (sse_cvtpi2ps): Use
TARGET_SSE2 && SSE_REG_P in split condition.
(ssse3_ph<plusminus_mnemonic>wv4hi3): Use
TARGET_SSSE3 && SSE_REGNO_P in split condition.
(ssse3_ph<plusminus_mnemonic>dv2si3): Ditto.
(ssse3_pshufbv8qi3): Ditto.
(ssse3_palignrdi): Ditto.

From-SVN: r272943

5 years agoFix amdgcn regrename ICE.
Andrew Stubbs [Tue, 2 Jul 2019 11:57:17 +0000 (11:57 +0000)]
Fix amdgcn regrename ICE.

2019-07-02  Andrew Stubbs  <ams@codesourcery.com>

gcc/
* config/gcn/gcn.md (movdi_symbol_save_scc): Convert to define_insn
with inlined save and restore.

From-SVN: r272932

5 years agoFix preprocessor checks for Clang builtins
Jonathan Wakely [Tue, 2 Jul 2019 11:50:27 +0000 (12:50 +0100)]
Fix preprocessor checks for Clang builtins

Clang seems to define built-ins that start with "__builtin_" as
non-keywords, which means that we need to use __has_builtin to detect
them, not __is_identifier. The built-ins that don't start with
"__builtin_" are keywords, and can only be detected using
__is_identifier and not by __has_builtin.

* include/bits/c++config (_GLIBCXX_HAVE_BUILTIN_LAUNDER)
(_GLIBCXX_HAVE_BUILTIN_IS_CONSTANT_EVALUATED): Use __has_builtin
instead of __is_identifier to detect Clang support.

From-SVN: r272931

5 years ago* cfgrtl.c (commit_edge_insertions): Rebuild jump labels chain.
Eric Botcazou [Tue, 2 Jul 2019 11:10:59 +0000 (11:10 +0000)]
* cfgrtl.c (commit_edge_insertions): Rebuild jump labels chain.

From-SVN: r272930

5 years agocfgexpand.c (pass_expand::execute): Deal specially with instructions to be inserted...
Eric Botcazou [Tue, 2 Jul 2019 09:44:47 +0000 (09:44 +0000)]
cfgexpand.c (pass_expand::execute): Deal specially with instructions to be inserted on single successor edge of the...

* cfgexpand.c (pass_expand::execute): Deal specially with instructions
to be inserted on single successor edge of the entry block.  Then call
commit_edge_insertions instead of inserting the instructions manually.
* cfgrtl.c (commit_edge_insertions): Do not verify flow info during
RTL expansion.

From-SVN: r272929

5 years agotree-core.h (enum tree_index): Add TI_CHREC_DONT_KNOW and TI_CHREC_KNOWN.
Richard Biener [Tue, 2 Jul 2019 09:35:12 +0000 (09:35 +0000)]
tree-core.h (enum tree_index): Add TI_CHREC_DONT_KNOW and TI_CHREC_KNOWN.

2019-07-02  Richard Biener  <rguenther@suse.de>

* tree-core.h (enum tree_index): Add TI_CHREC_DONT_KNOW and
TI_CHREC_KNOWN.
* tree.h (chrec_not_analyzed_yet, chrec_dont_know, chrec_known):
Define here.
* tree.c (build_common_tree_nodes): Initialize them.
* tree-chrec.h (chrec_not_analyzed_yet, chrec_dont_know, chrec_known):
Make declarations comments.
* tree-scalar-evolution.c (chrec_not_analyzed_yet, chrec_dont_know,
chrec_known): Remove definitions.
(initialize_scalar_evolutions_analyzer): Remove.
(scev_initialize): Do not call initialize_scalar_evolutions_analyzer.
* tree-streamer.c (preload_common_nodes): Do not preload
TI_CHREC_DONT_KNOW or TI_CHREC_KNOWN.

From-SVN: r272928

5 years agotree-ssa-alias.c (aliasing_component_refs_p): Remove forgotten sanity check.
Jan Hubicka [Tue, 2 Jul 2019 08:35:58 +0000 (10:35 +0200)]
tree-ssa-alias.c (aliasing_component_refs_p): Remove forgotten sanity check.

* tree-ssa-alias.c (aliasing_component_refs_p): Remove forgotten
sanity check.

From-SVN: r272927

5 years agotree-ssa-alias.c (nonoverlapping_component_refs_for_decl_p): Rename to ..
Jan Hubicka [Tue, 2 Jul 2019 08:28:24 +0000 (10:28 +0200)]
tree-ssa-alias.c (nonoverlapping_component_refs_for_decl_p): Rename to ..

* tree-ssa-alias.c (nonoverlapping_component_refs_for_decl_p): Rename
to ..
(nonoverlapping_component_refs_since_match_p): ... this one;
handle also non-decl bases; return -1 if search gave up.
(alias_stats): Rename nonoverlapping_component_refs_of_decl_p_may_alias,
nonoverlapping_component_refs_of_decl_p_no_alias to
nonoverlapping_component_refs_since_match_p_may_alias,
nonoverlapping_component_refs_since_match_p_no_alias.
(dump_alias_stats): Update dumping.
(aliasing_matching_component_refs_p):  Break out from ...;
dispatch to nonoverlapping_component_refs_for_decl_p
and nonoverlapping_component_refs_since_match_p.
(aliasing_component_refs_p): ... here; call
nonoverlapping_component_refs_p in scenarios where we can not
precisely determine base match.
(decl_refs_may_alias_p): Use
nonoverlapping_component_refs_since_match_p.
(indirect_ref_may_alias_decl_p): Do not call
nonoverlapping_component_refs_p.
(indirect_refs_may_alias_p): Likewise.

* gcc.dg/tree-ssa/alias-access-path-7.c: New testcase.

From-SVN: r272926

5 years agotree-inline.c (remap_gimple_stmt): Do not subtitute handled components to clobber...
Jan Hubicka [Tue, 2 Jul 2019 08:26:16 +0000 (10:26 +0200)]
tree-inline.c (remap_gimple_stmt): Do not subtitute handled components to clobber of return value.

* tree-inline.c (remap_gimple_stmt): Do not subtitute handled components
to clobber of return value.
* g++.dg/lto/pr90990_0.C: New testcase.

From-SVN: r272925

5 years ago[arm/AArch64] Assume unhandled NEON types are neon_arith_basic types when scheduling...
Kyrylo Tkachov [Tue, 2 Jul 2019 08:24:54 +0000 (08:24 +0000)]
[arm/AArch64] Assume unhandled NEON types are neon_arith_basic types when scheduling for Cortex-A57

Some scheduling descriptions, like the Cortex-A57 one, are reused for multiple -mcpu options.
Sometimes those other -mcpu cores support more architecture features than the Armv8-A Cortex-A57.
For example, the Cortex-A75 and Cortex-A76 support Armv8.2-A as well as the Dot Product instructions.
These Dot Product instructions have the neon_dot and neon_dot_q scheduling type, but that type is not
handled in cortex-a57.md, since the Cortex-A57 itself doesn't need to care about these instructions.

But if we just ignore the neon_dot(_q) type at scheduling we get really terrible codegen when compiling
for -mcpu=cortex-a76, for example, because the scheduler just pools all the UDOT instructions at the end
of the basic block, since it doesn't assume anything about their behaviour.

This patch ameliorates the situation somewhat by telling the Cortex-A57 scheduling model to treat any
insn that doesn't get assigned a cortex_a57_neon_type but is actually a is_neon_type instruction as
a simple neon_arith_basic instruction. This allows us to treat post-Armv8-A SIMD instructions more sanely
without having to model each of them explicitly in cortex-a57.md.

* config/arm/cortex-a57.md (cortex_a57_neon_type): Use neon_arith_basic
for is_neon_type instructions that have not already been categorized.

From-SVN: r272924

5 years agolto-common.c (lto_register_canonical_types_for_odr_types): Copy CXX_ODR_P from the...
Jan Hubicka [Tue, 2 Jul 2019 08:23:02 +0000 (10:23 +0200)]
lto-common.c (lto_register_canonical_types_for_odr_types): Copy CXX_ODR_P from the main variant.

* lto-common.c (lto_register_canonical_types_for_odr_types):
Copy CXX_ODR_P from the main variant.

From-SVN: r272923

5 years agore PR tree-optimization/58483 (missing optimization opportunity for const std::vector...
Richard Biener [Tue, 2 Jul 2019 07:35:23 +0000 (07:35 +0000)]
re PR tree-optimization/58483 (missing optimization opportunity for const std::vector compared to std::array)

2019-07-02  Richard Biener  <rguenther@suse.de>

PR tree-optimization/58483
* tree-ssa-scopedtables.c (avail_expr_hash): Use OEP_ADDRESS_OF
for MEM_REF base hashing.
(equal_mem_array_ref_p): Likewise for base comparison.

* gcc.dg/tree-ssa/ssa-dom-cse-8.c: New testcase.

From-SVN: r272922

5 years agomklog/91048: Open ~/.mklog in string mode.
Janne Blomqvist [Tue, 2 Jul 2019 05:54:31 +0000 (08:54 +0300)]
mklog/91048: Open ~/.mklog in string mode.

2019-07-02  Janne Blomqvist  <jb@gcc.gnu.org>

PR other/91048
* mklog (read_user_info): Open ~/.mklog in string mode.

From-SVN: r272921

5 years agoFix libstdc++ install-pdf support.
Jim Wilson [Tue, 2 Jul 2019 02:30:52 +0000 (02:30 +0000)]
Fix libstdc++ install-pdf support.

Generating pdf files requires everything that is required for the xml files
except the style sheets.

libstdc++-v3/
* configure.ac (BUILD_PDF): Also test for doxygen, dot, xsltproc,
and xmllint.
* configure: Regenerate.

From-SVN: r272920

5 years agocompiler: refactoring in Export class to encapsulate type refs map
Ian Lance Taylor [Tue, 2 Jul 2019 01:39:19 +0000 (01:39 +0000)]
compiler: refactoring in Export class to encapsulate type refs map

    Convert the Export::type_refs map from a static object to a field
    contained (indirectly, via an impl class) in Export itself, for better
    encapsulation and to be able to reclaim its memory when exporting is
    done. No change in compiler functionality.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184170

From-SVN: r272919

5 years agoDaily bump.
GCC Administrator [Tue, 2 Jul 2019 00:16:22 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r272916

5 years agors6000.md (signbit<mode>2_dm): Make this a parameterized name.
Segher Boessenkool [Mon, 1 Jul 2019 21:58:47 +0000 (23:58 +0200)]
rs6000.md (signbit<mode>2_dm): Make this a parameterized name.

@signbit<mode>2_dm

* config/rs6000/rs6000.md (signbit<mode>2_dm): Make this a
parameterized name.
(signbit<mode>2): Use that name.  Simplify.

From-SVN: r272912

5 years agore PR tree-optimization/66726 (missed optimization, factor conversion out of COND_EXPR)
Joern Rennecke [Mon, 1 Jul 2019 21:48:55 +0000 (21:48 +0000)]
re PR tree-optimization/66726 (missed optimization, factor conversion out of COND_EXPR)

        PR middle-end/66726
        * tree-ssa-phiopt.c (factor_out_conditional_conversion):
        Tune heuristic from PR71016 to allow MIN / MAX.
        * testsuite/gcc.dg/tree-ssa/pr66726-4.c: New testcase.

From-SVN: r272911

5 years ago* config/i386/constraints.md: Remove stalled comment w.r.t. Yh constraint.
Uros Bizjak [Mon, 1 Jul 2019 19:04:05 +0000 (21:04 +0200)]
* config/i386/constraints.md: Remove stalled comment w.r.t. Yh constraint.

From-SVN: r272908

5 years agors6000.md (ieee_128bit_vsx_abs<mode>2): Make this a parameterized name.
Segher Boessenkool [Mon, 1 Jul 2019 18:47:56 +0000 (20:47 +0200)]
rs6000.md (ieee_128bit_vsx_abs<mode>2): Make this a parameterized name.

@ieee_128bit_vsx_abs<mode>2

* config/rs6000/rs6000.md (ieee_128bit_vsx_abs<mode>2): Make this a
parameterized name.
(abs<mode>2): Use that name.  Simplify.

From-SVN: r272907

5 years agors6000.md (ieee_128bit_vsx_neg<mode>2): Make this a parameterized name.
Segher Boessenkool [Mon, 1 Jul 2019 18:47:05 +0000 (20:47 +0200)]
rs6000.md (ieee_128bit_vsx_neg<mode>2): Make this a parameterized name.

@ieee_128bit_vsx_neg<mode>2

* config/rs6000/rs6000.md (ieee_128bit_vsx_neg<mode>2): Make this a
parameterized name.
(neg<mode>2): Use that name.  Simplify.

From-SVN: r272906

5 years agors6000.md (abs<mode>2_hw): Make this a parameterized name.
Segher Boessenkool [Mon, 1 Jul 2019 18:45:36 +0000 (20:45 +0200)]
rs6000.md (abs<mode>2_hw): Make this a parameterized name.

@abs<mode>2_hw

* config/rs6000/rs6000.md (abs<mode>2_hw): Make this a parameterized
name.
(abs<mode>2): Use that name.  Simplify.

From-SVN: r272905

5 years agors6000.md (neg<mode>2_hw): Make this a parameterized name.
Segher Boessenkool [Mon, 1 Jul 2019 18:44:18 +0000 (20:44 +0200)]
rs6000.md (neg<mode>2_hw): Make this a parameterized name.

@neg<mode>2_hw

* config/rs6000/rs6000.md (neg<mode>2_hw): Make this a parameterized
name.
(neg<mode>2): Use that name.  Simplify.

From-SVN: r272904

5 years agors6000.md (extenddf<mode>2): Make this a parameterized name.
Segher Boessenkool [Mon, 1 Jul 2019 18:43:10 +0000 (20:43 +0200)]
rs6000.md (extenddf<mode>2): Make this a parameterized name.

@extenddf<mode>2

* config/rs6000/rs6000.md (extenddf<mode>2): Make this a parameterized
name.
(floatsi<mode>2): Use that name.  Simplify.

From-SVN: r272903

5 years agoi386.md ("isa" attribute): Add sse_noavx.
Uros Bizjak [Mon, 1 Jul 2019 18:41:09 +0000 (20:41 +0200)]
i386.md ("isa" attribute): Add sse_noavx.

* config/i386/i386.md ("isa" attribute): Add sse_noavx.
("enabled" attribute): Handle sse_noavx isa attribute.
* config/i386/mmx.md (*vec_dupv2sf): Add "isa" attribute.
Use TARGET_SSE && SSE_REGNO_P in split condition.
(*vec_dupv2sf): Ditto.

From-SVN: r272902

5 years agors6000.md (extenddf<mode>2_fprs): Make this a parameterized name.
Segher Boessenkool [Mon, 1 Jul 2019 18:40:40 +0000 (20:40 +0200)]
rs6000.md (extenddf<mode>2_fprs): Make this a parameterized name.

@extenddf<mode>2_{fprs,vsx}

* config/rs6000/rs6000.md (extenddf<mode>2_fprs): Make this a
parameterized name.
(extenddf<mode>2_vsx): Make this a parameterized name.
(extenddf<mode>2): Use those names.  Simplify.

From-SVN: r272901