gcc.git
4 years agoRISC-V: Make unique SECCAT_SRODATA names start with .srodata (not .sdata2)
Keith Packard [Wed, 29 Apr 2020 16:49:56 +0000 (09:49 -0700)]
RISC-V: Make unique SECCAT_SRODATA names start with .srodata (not .sdata2)

default_unique_section uses ".sdata2" as a prefix for SECCAT_SRODATA
unique sections, but RISC-V uses ".srodata" instead. Override the
TARGET_ASM_UNIQUE_SECTION function to catch this case, allowing the
default to be used for all other sections.

Signed-off-by: Keith Packard <keithp@keithp.com>
gcc/
* config/riscv/riscv.c (riscv_unique_section): New.
(TARGET_ASM_UNIQUE_SECTION): New.

4 years agoRISC-V: Add shorten_memrefs pass.
Craig Blackmore [Tue, 12 May 2020 21:41:08 +0000 (14:41 -0700)]
RISC-V: Add shorten_memrefs pass.

gcc/
* config.gcc:  Add riscv-shorten-memrefs.o to extra_objs for riscv.
* config/riscv/riscv-passes.def: New file.
* config/riscv/riscv-protos.h (make_pass_shorten_memrefs): Declare.
* config/riscv/riscv-shorten-memrefs.c: New file.
* config/riscv/riscv.c (tree-pass.h): New include.
(riscv_compressed_reg_p): New Function
(riscv_compressed_lw_offset_p): Likewise.
(riscv_compressed_lw_address_p): Likewise.
(riscv_shorten_lw_offset): Likewise.
(riscv_legitimize_address): Attempt to convert base + large_offset
to compressible new_base + small_offset.
(riscv_address_cost): Make anticipated compressed load/stores
cheaper for code size than uncompressed load/stores.
(riscv_register_priority): Move compressed register check to
riscv_compressed_reg_p.
* config/riscv/riscv.h (C_S_BITS): Define.
(CSW_MAX_OFFSET): Define.
* config/riscv/riscv.opt (mshorten-memefs): New option.
* config/riscv/t-riscv (riscv-shorten-memrefs.o): New rule.
(PASSES_EXTRA): Add riscv-passes.def.
* doc/invoke.texi: Document -mshorten-memrefs.

* config/riscv/riscv.c (riscv_new_address_profitable_p): New function.
(TARGET_NEW_ADDRESS_PROFITABLE_P): Define.
* doc/tm.texi: Regenerate.
* doc/tm.texi.in (TARGET_NEW_ADDRESS_PROFITABLE_P):  New hook.
* sched-deps.c (attempt_change): Use old address if it is cheaper than
new address.
* target.def (new_address_profitable_p): New hook.
* targhooks.c (default_new_address_profitable_p): New function.
* targhooks.h (default_new_address_profitable_p): Declare.

gcc/testsuite/
* gcc.target/riscv/shorten-memrefs-1.c: New test.
* gcc.target/riscv/shorten-memrefs-2.c: New test.
* gcc.target/riscv/shorten-memrefs-3.c: New test.
* gcc.target/riscv/shorten-memrefs-4.c: New test.
* gcc.target/riscv/shorten-memrefs-5.c: New test.
* gcc.target/riscv/shorten-memrefs-6.c: New test.
* gcc.target/riscv/shorten-memrefs-7.c: New test.

4 years agoSuppress warning for Interfaces.C with -fdump-ada-spec
Eric Botcazou [Tue, 12 May 2020 20:41:09 +0000 (22:41 +0200)]
Suppress warning for Interfaces.C with -fdump-ada-spec

The C/C++ bindings generated by means of -fdump-ada-spec always contain
with and use clauses for Interfaces.C, but they can be unused in some
cases so make sure to avoid warning about that.

* c-ada-spec.c (dump_ads): Output pragma Warnings ("U"); on entry.

4 years agopreprocessor: EOF location is at end of file [PR95013]
Nathan Sidwell [Tue, 12 May 2020 20:33:11 +0000 (13:33 -0700)]
preprocessor: EOF location is at end of file [PR95013]

My recent C++ parser change to pay attention to EOF location uncovered
a separate bug.  The preprocesor's EOF logic would set the EOF
location to be the beginning of the last line of text in the file --
not the 'line' after that, which contains no characters.  Mostly.
This fixes things so that when we attempt to read the last line of the
main file, we don't pop the buffer until the tokenizer has a chance to
create an EOF token with the correct location information.  It is then
responsible for popping the buffer.  As it happens, raw string literal
tokenizing contained a bug -- it would increment the line number
prematurely, because it cached buffer->cur in a local variable, but
checked buffer->cur before updating it to figure out if it was at end
of file.   We fix up that too.

The EOF token intentionally doesn't have a column number -- it's not a
position on a line, it's a non-existant line.

The testsuite churn is just correcting the EOF location diagnostics.

libcpp/
PR preprocessor/95013
* lex.c (lex_raw_string): Process line notes before incrementing.
Correct incrementing condition.  Adjust for new
_cpp_get_fresh_line EOF behaviour.
(_cpp_get_fresh_line): Do not pop buffer at EOF, increment line
instead.
(_cpp_lex_direct): Adjust for new _cpp_get_fresh_line behaviour.
(cpp_directive_only_process): Assert we got a fresh line.
* traditional.c (_cpp_read_logical_line_trad): Adjust for new
_cpp_get_fresh_line behaviour.

gcc/testsuite/
* c-c++-common/goacc/pr79428-1.c: Adjust EOF diagnostic location.
* c-c++-common/gomp/pr79428-2.c: Likewise.
* g++.dg/cpp0x/decltype63.C: Likewise.
* g++.dg/cpp0x/gen-attrs-64.C: Likewise.
* g++.dg/cpp0x/pr68726.C: Likewise.
* g++.dg/cpp0x/pr78341.C: Likewise.
* g++.dg/cpp1y/pr65202.C: Likewise.
* g++.dg/cpp1y/pr65340.C: Likewise.
* g++.dg/cpp1y/pr68578.C: Likewise.
* g++.dg/cpp1z/class-deduction44.C: Likewise.
* g++.dg/diagnostic/unclosed-extern-c.C: Likewise.
* g++.dg/diagnostic/unclosed-function.C: Likewise.
* g++.dg/diagnostic/unclosed-namespace.C: Likewise.
* g++.dg/diagnostic/unclosed-struct.C: Likewise.
* g++.dg/ext/pr84598.C: Likewise.
* g++.dg/other/switch4.C: Likewise.
* g++.dg/parse/attr4.C: Likewise.
* g++.dg/parse/cond4.C: Likewise.
* g++.dg/parse/crash10.C: Likewise.
* g++.dg/parse/crash18.C: Likewise.
* g++.dg/parse/crash27.C: Likewise.
* g++.dg/parse/crash34.C: Likewise.
* g++.dg/parse/crash35.C: Likewise.
* g++.dg/parse/crash52.C: Likewise.
* g++.dg/parse/crash59.C: Likewise.
* g++.dg/parse/crash61.C: Likewise.
* g++.dg/parse/crash67.C: Likewise.
* g++.dg/parse/error14.C: Likewise.
* g++.dg/parse/error56.C: Likewise.
* g++.dg/parse/invalid1.C: Likewise.
* g++.dg/parse/parameter-declaration-1.C: Likewise.
* g++.dg/parse/parser-pr28152-2.C: Likewise.
* g++.dg/parse/parser-pr28152.C: Likewise.
* g++.dg/parse/pr68722.C: Likewise.
* g++.dg/pr46852.C: Likewise.
* g++.dg/pr46868.C: Likewise.
* g++.dg/template/crash115.C: Likewise.
* g++.dg/template/crash43.C: Likewise.
* g++.dg/template/crash90.C: Likewise.
* g++.dg/template/error-recovery1.C: Likewise.
* g++.dg/template/error57.C: Likewise.
* g++.old-deja/g++.other/crash31.C: Likewise.
* gcc.dg/empty-source-2.c: Likewise.
* gcc.dg/empty-source-3.c: Likewise.
* gcc.dg/noncompile/pr30552-3.c: Likewise.
* gcc.dg/noncompile/pr35447-1.c: Likewise.
* gcc.dg/pr20245-1.c: Likewise.
* gcc.dg/pr28419.c: Likewise.
* gcc.dg/rtl/truncated-rtl-file.c: Likewise.
* gcc.dg/unclosed-init.c: Likewise.
* obj-c++.dg/property/property-neg-6.mm: Likewise.
* obj-c++.dg/syntax-error-10.mm: Likewise.
* obj-c++.dg/syntax-error-8.mm: Likewise.
* obj-c++.dg/syntax-error-9.mm: Likewise.

4 years agoBe prepared for more aggregates in gigi
Eric Botcazou [Tue, 12 May 2020 20:34:50 +0000 (22:34 +0200)]
Be prepared for more aggregates in gigi

This makes sure that gigi is prepared to handle more aggregates in the
special memset code path.

* sem_aggr.ads (Is_Single_Aggregate): New function.
* sem_aggr.adb (Is_Others_Aggregate): Use local variable.
(Is_Single_Aggregate): New function to recognize an aggregate with
a single association containing a single choice.
* fe.h (Is_Others_Aggregate): Delete.
(Is_Single_Aggregate): New declaration.
* gcc-interface/trans.c (gnat_to_gnu) <N_Assignment_Statement>: Call
Is_Single_Aggregate instead of Is_Others_Aggregate.

4 years agoc++: Function found via ADL when it should not [PR95074]
Marek Polacek [Tue, 12 May 2020 18:56:13 +0000 (14:56 -0400)]
c++: Function found via ADL when it should not [PR95074]

I noticed that we don't implement [basic.lookup.argdep]/3: quite correctly;
it says "If X (the lookup set produced by unqualified lookup) contains
-- a block-scope function declaration that is not a using-declaration
[...]
then Y (the lookup set produced by ADL) is empty."
but we were still performing ADL in fn1 in the attached test.  The
problem was that we were only looking at the first function in the
overload set which in this case happened to be a using-declaration, and
those don't suppress ADL.  We have to look through the whole set to find
out if unqualified lookup found a block-scope function declaration, or
a member function declaration.

PR c++/95074
* parser.c (cp_parser_postfix_expression) <case CPP_OPEN_PAREN>: When
looking for a block-scope function declaration, look through the whole
set, not just the first function in the overload set.

* g++.dg/lookup/koenig15.C: New test.

4 years agotestsuite: Fix up gcc.dg/asan/pr95051.c testcase [PR95051]
Jakub Jelinek [Tue, 12 May 2020 19:21:30 +0000 (21:21 +0200)]
testsuite: Fix up gcc.dg/asan/pr95051.c testcase [PR95051]

asan.exp adds -fsanitize=address which is incompatible with -fsanitize=kernel-address,
so we need to disable it first.

2020-05-12  Jakub Jelinek  <jakub@redhat.com>

PR sanitizer/95051
* gcc.dg/asan/pr95051.c: Add -fno-sanitize=all to dg-options.

4 years agoFix throw specifiers on interface.
Nathan Sidwell [Tue, 12 May 2020 17:54:53 +0000 (10:54 -0700)]
Fix throw specifiers on interface.

I discovered that libitm:
(a) declares __cxa_allocate_exception and friends directly,
(b) doesn't mark them as 'throw()'
(c) doesn't mark the replacment fns _ITM_$foo as nothrow either

We happen to get away with it because of code in the compiler that,
although it checks the parameter types, doesn't check the exception
specification.  (One reason being they used to not be part of the
language's type system, but now they are.)  I suspect this can lead us
to generate pessimal code later, if we've seen one of these decls
earlier.  Anyway, with modules it becomes trickier[*], so I'm trying
to clean it up and not be a problem.  I see Jakub fixed part of the
problem
(https://gcc.gnu.org/pipermail/gcc-patches/2018-December/513302.html)
AFAICT, he did fix libitm's decls, but left the lax parm-type checking
in the compiler.

libitm.h is not very informative about specification:
  in version 1 of http://www.intel.com/some/path/here.pdf.  */

Anyway, it was too fiddly to have libitm pick up the declarations from
libsupc++.  Besides it makes them weak declarations, and then provides
definitions for non-elf systems.  So this patch adds the expected
'throw()'

* libitm/libitm.h (_ITM_NOTHROW): Define.
(_ITM_cxa_allocate_exception, _ITM_cxa_free_exception)
(_ITM_cxa_begin_catch): Use it.
* eh_cpp.cc: Add throw() to __cxa_allocate_exception,
__cxa_free_exception, __cxa_begin_catch, __cxa_tm_cleanup,
__cxa_get_globals.
(_ITM_cxa_allocate_exception, _ITM_cxa_free_exception)
(_ITM_cxa_begin_catch): Likewise.

4 years agoEnable CET in cross compiler if possible
H.J. Lu [Tue, 12 May 2020 17:39:42 +0000 (10:39 -0700)]
Enable CET in cross compiler if possible

Don't perform CET run-time check for host when cross compiling.  Instead,
enable CET in cross compiler if possible so that it will run on both CET
and non-CET hosts.

config/

PR bootstrap/94998
* cet.m4 (GCC_CET_HOST_FLAGS): Enable CET in cross compiler if
possible.

libiberty/

PR bootstrap/94998
* configure: Regenerated.

lto-plugin/

PR bootstrap/94998
* configure: Regenerated.

4 years agoi386: Add V2SFmode copysign, xorsign and signbit expanders [PR95046]
Uros Bizjak [Tue, 12 May 2020 17:24:53 +0000 (19:24 +0200)]
i386: Add V2SFmode copysign, xorsign and signbit expanders [PR95046]

gcc/ChangeLog:

PR target/95046
* config/i386/mmx.md (copysignv2sf3): New expander.
(xorsignv2sf3): Ditto.
(signbitv2sf3): Ditto.

testsuite/ChangeLog:

PR target/95046
* gcc.target/i386/pr95046-4.c: New test.

4 years agoi386: Add V2SFmode FMA insn patterns [PR95046]
Uros Bizjak [Tue, 12 May 2020 16:36:27 +0000 (18:36 +0200)]
i386: Add V2SFmode FMA insn patterns [PR95046]

gcc/ChangeLog:

PR target/95046
* config/i386/mmx.md (fmav2sf4): New insn pattern.
(fmsv2sf4): Ditto.
(fnmav2sf4): Ditto.
(fnmsv2sf4): Ditto.

testsuite/ChangeLog:

PR target/95046
* gcc.target/i386/pr95046-3.c: New test.

4 years agoc++: Add abbreviated fn template test for [PR78752]
Patrick Palka [Tue, 12 May 2020 16:20:55 +0000 (12:20 -0400)]
c++: Add abbreviated fn template test for [PR78752]

This adds an abbreviated function template version of the testcase in
PR78752, which seems to already be fixed.

gcc/testsuite/ChangeLog:

PR c++/78752
* g++.dg/cpp2a/concepts-pr78752-2.C: New test.

4 years agozlib: Add --enable-cet to configure
H.J. Lu [Tue, 12 May 2020 16:27:24 +0000 (09:27 -0700)]
zlib: Add --enable-cet to configure

When --enable-cet is used to configure GCC, enable Intel CET in target
zlib.

* Makefile.am (AM_CFLAGS): New.
* configure.ac (CET_FLAGS): Add GCC_CET_FLAGS and AC_SUBST.
* Makefile.in: Regenerated.
* aclocal.m4: Likewise.
* configure.ac: Likewise.

4 years agolibbacktrace: Enable Intel CET on Intel CET enabled host for jit
H.J. Lu [Tue, 12 May 2020 16:23:56 +0000 (09:23 -0700)]
libbacktrace: Enable Intel CET on Intel CET enabled host for jit

Since on Intel CET enabled host, dlopen in Intel CET enabled applications
fails on shared libraries which aren't Intel CET enabled, compile with
-fcf-protection on Intel CET enabled host when jit is enabled to enable
Intel CET on libgccjit.

* Makefile.am (AM_CFLAGS): Add $(CET_HOST_FLAGS).
* configure.ac: Add GCC_CET_HOST_FLAGS(CET_HOST_FLAGS) and
AC_SUBST(CET_HOST_FLAGS).  Clear CET_HOST_FLAGS if jit isn't
enabled.
* Makefile.in: Regenerated.
* configure: Likewise.

4 years agolibdecnumber: Enable Intel CET on Intel CET enabled host for jit
H.J. Lu [Tue, 12 May 2020 16:19:14 +0000 (09:19 -0700)]
libdecnumber: Enable Intel CET on Intel CET enabled host for jit

Since on Intel CET enabled host, dlopen in Intel CET enabled applications
fails on shared libraries which aren't Intel CET enabled, compile with
-fcf-protection on Intel CET enabled host when jit is enabled to enable
Intel CET on libgccjit.

* Makefile.in (CET_HOST_FLAGS): New.
(COMPILER): Add $(CET_HOST_FLAGS).
* configure.ac: Add GCC_CET_HOST_FLAGS(CET_HOST_FLAGS) and
AC_SUBST(CET_HOST_FLAGS).  Clear CET_HOST_FLAGS if jit isn't
enabled.
* aclocal.m4: Regenerated.
* configure: Likewise.

4 years agolibcpp: Enable Intel CET on Intel CET enabled host for jit
H.J. Lu [Tue, 12 May 2020 16:17:34 +0000 (09:17 -0700)]
libcpp: Enable Intel CET on Intel CET enabled host for jit

Since on Intel CET enabled host, dlopen in Intel CET enabled applications
fails on shared libraries which aren't Intel CET enabled, compile with
-fcf-protection on Intel CET enabled host when jit is enabled to enable
Intel CET on libgccjit.

* Makefile.in (CET_HOST_FLAGS): New.
(COMPILER): Add $(CET_HOST_FLAGS).
* configure.ac: Add GCC_CET_HOST_FLAGS(CET_HOST_FLAGS) and
AC_SUBST(CET_HOST_FLAGS).  Clear CET_HOST_FLAGS if jit isn't
enabled.
* aclocal.m4: Regenerated.
* configure: Likewise.

4 years agogcc: Enable Intel CET on Intel CET enabled host for jit
H.J. Lu [Tue, 12 May 2020 16:14:52 +0000 (09:14 -0700)]
gcc: Enable Intel CET on Intel CET enabled host for jit

Since on Intel CET enabled host, dlopen in Intel CET enabled applications
fails on shared libraries which aren't Intel CET enabled, compile with
-fcf-protection on Intel CET enabled host when jit is enabled to enable
Intel CET on libgccjit.

* Makefile.in (CET_HOST_FLAGS): New.
(COMPILER): Add $(CET_HOST_FLAGS).
* configure.ac: Add GCC_CET_HOST_FLAGS(CET_HOST_FLAGS) and
AC_SUBST(CET_HOST_FLAGS).  Clear CET_HOST_FLAGS if jit isn't
enabled.
* aclocal.m4: Regenerated.
* configure: Likewise.

4 years agolibcc1: Enable Intel CET on Intel CET enabled host
H.J. Lu [Tue, 12 May 2020 16:12:26 +0000 (09:12 -0700)]
libcc1: Enable Intel CET on Intel CET enabled host

Since on Intel CET enabled host, dlopen in Intel CET enabled applications
fails on shared libraries which aren't Intel CET enabled, enable Intel
CET in libcc1 on Intel CET enabled host.

* Makefile.am (AM_CXXFLAGS): Add $(CET_HOST_FLAGS).
* configure.ac: Add GCC_CET_HOST_FLAGS(CET_HOST_FLAGS) and
AC_SUBST(CET_HOST_FLAGS).
* Makefile.in: Regenerated.
* aclocal.m4: Likewise.
* configure: Likewise.

4 years agoi386: Add V2SFmode NEG, ABS and logic insn patterns [PR95046]
Uros Bizjak [Tue, 12 May 2020 15:08:03 +0000 (17:08 +0200)]
i386: Add V2SFmode NEG, ABS and logic insn patterns [PR95046]

gcc/ChangeLog:

PR target/95046
* config/i386/mmx.md (<code>v2sf2): New insn pattern.
(*mmx_<code>v2sf2): New insn_and_split pattern.
(*mmx_nabsv2sf2): Ditto.
(*mmx_andnotv2sf3): New insn pattern.
(*mmx_<code>v2sf3): Ditto.
* config/i386/i386.md (absneg_op): New code attribute.
* config/i386/i386.c (ix86_build_const_vector): Handle V2SFmode.
(ix86_build_signbit_mask): Ditto.

testsuite/ChangeLog:

PR target/95046
* gcc.target/i386/pr95046-2.c: New test.

4 years agoi386: Add V2SFmode NEG, ABS and logic insn patterns [PR95046]
Uros Bizjak [Tue, 12 May 2020 15:06:13 +0000 (17:06 +0200)]
i386: Add V2SFmode NEG, ABS and logic insn patterns [PR95046]

gcc/ChangeLog:

PR target/95046
* config/i386/mmx.md (<code>v2sf2): New insn pattern.
(*mmx_<code>v2sf2): New insn_and_split pattern.
(*mmx_nabsv2sf2): Ditto.
(*mmx_andnotv2sf3): New insn pattern.
(*mmx_<code>v2sf3): Ditto.
* config/i386/i386.md (absneg_op): New code attribute.
* config/i386/i386.c (ix86_build_const_vector): Handle V2SFmode.
(ix86_build_signbit_mask): Ditto.

testsuite/ChangeLog:

PR target/95046
* gcc.target/i386/pr95046-2.c: New test.

4 years agoremove dead debug-bind resets
Richard Biener [Tue, 14 Jan 2020 14:31:18 +0000 (15:31 +0100)]
remove dead debug-bind resets

This removes debug-bind resets aka

 # DEBUG b = NULL

when the reset variable is otherwise unused.  I've gathered statistics
for a single TU, fold-const.ii which at -O2 -g shows

28 ssa "dead debug bind reset" 1
34 einline "dead debug bind reset" 340
54 release_ssa "dead debug bind reset" 176
54 release_ssa "live debug bind reset of dead var" 4
86 inline "dead debug bind reset" 5131
86 inline "live debug bind reset of dead var" 61
241 optimized "dead debug bind reset" 970
241 optimized "live debug bind reset of dead var" 287

where "live debug bind reset of dead var" means the variable is unused
but there were debug binds with a value for them and
"dead debug bind reset" means the variable is unused and there were
only debug bind resets (each reset of the same variable is counted
for both counters).  This shows A considerable amount of dead stmts
removed esp. after IPA inlining.

2020-05-12  Richard Biener  <rguenther@suse.de>

* tree-ssa-live.c (remove_unused_locals): Remove dead debug
bind resets.

4 years agomiddle-end/94988 fix testcase for big-endian
Richard Biener [Tue, 12 May 2020 12:13:32 +0000 (14:13 +0200)]
middle-end/94988 fix testcase for big-endian

The testcase only works for little-endian, mark it so.

2020-05-12  Richard Biener  <rguenther@suse.de>

PR middle-end/94988
* gcc.dg/torture/pr94988.c: Disable runtime test for
* non-little-endian.

4 years agoMSP430: Define ASM_OUTPUT_ALIGNED_DECL_LOCAL
Jozef Lawrynowicz [Fri, 8 May 2020 13:45:20 +0000 (14:45 +0100)]
MSP430: Define ASM_OUTPUT_ALIGNED_DECL_LOCAL

gcc/ChangeLog:

2020-05-12  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

* config/msp430/msp430-protos.h (msp430_output_aligned_decl_common):
Update prototype to include "local" argument.
* config/msp430/msp430.c (msp430_output_aligned_decl_common): Add
"local" argument.  Handle local common decls.
* config/msp430/msp430.h (ASM_OUTPUT_ALIGNED_DECL_COMMON): Adjust
msp430_output_aligned_decl_common call with 0 for "local" argument.
(ASM_OUTPUT_ALIGNED_DECL_LOCAL): Define.

gcc/testsuite/ChangeLog:

2020-05-12  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

* gcc.c-torture/execute/noinit-attribute.c: Skip for msp430
in the large memory model.

4 years agoMSP430: Allow .bss section to be created in region-attribute-misuse test
Jozef Lawrynowicz [Mon, 11 May 2020 15:58:11 +0000 (16:58 +0100)]
MSP430: Allow .bss section to be created in region-attribute-misuse test

2020-05-12  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

* gcc.target/msp430/region-attribute-misuse.c: Allow a .bss section to
be created.

4 years agoFix incorrect scalar storage order handling
Eric Botcazou [Tue, 12 May 2020 11:14:20 +0000 (13:14 +0200)]
Fix incorrect scalar storage order handling

This fixes an oversight in the new canonicalization code for packable
types: it does not take into account the scalar storage order.

PR ada/95035
* gcc-interface/utils.c (packable_type_hasher::equal): Also compare
the scalar storage order.
(hash_packable_type): Also hash the scalar storage order.
(hash_pad_type): Likewise.

4 years agopreserve EDGE_DFS_BACK across split_edge
Richard Biener [Tue, 14 Jan 2020 13:14:54 +0000 (14:14 +0100)]
preserve EDGE_DFS_BACK across split_edge

This moves EDGE_DFS_BACK to the appropriate edge when the split
edge had it set.

2020-05-12  Richard Biener  <rguenther@suse.de>

* cfghooks.c (split_edge): Preserve EDGE_DFS_BACK if set.

4 years agoASAN: clear DECL_NOT_GIMPLE_REG_P.
Martin Liska [Tue, 12 May 2020 08:23:09 +0000 (10:23 +0200)]
ASAN: clear DECL_NOT_GIMPLE_REG_P.

PR sanitizer/95033
PR sanitizer/95051
* sanopt.c (sanitize_rewrite_addressable_params):
Clear DECL_NOT_GIMPLE_REG_P for argument.
PR sanitizer/95033
PR sanitizer/95051
* g++.dg/asan/function-argument-4.C: New test.
* gcc.dg/asan/pr95033.c: New test.
* gcc.dg/asan/pr95051.c: New test.

4 years agoopenmp: Fix up handling of DECL_OMP_PRIVATIZED_MEMBER for bit-fields [PR95063]
Jakub Jelinek [Tue, 12 May 2020 08:00:32 +0000 (10:00 +0200)]
openmp: Fix up handling of DECL_OMP_PRIVATIZED_MEMBER for bit-fields [PR95063]

The r11-15 change broke this testcase, as it now asserts type is equal to
the type of the DECL_VALUE_EXPR, but for DECL_OMP_PRIVATIZED_MEMBER artificial
vars mapping to bitfields it wasn't.  Fixed by changing the
DECL_OMP_PRIVATIZED_MEMBER var type in that case.

2020-05-12  Jakub Jelinek  <jakub@redhat.com>

PR c++/95063
* pt.c (tsubst_decl): Deal with DECL_OMP_PRIVATIZED_MEMBER for
a bit-field.

* g++.dg/gomp/pr95063.C: New test.

4 years agotree-vect-generic: Fix bitfield widths [PR94980 3/3]
Richard Sandiford [Tue, 12 May 2020 08:01:13 +0000 (09:01 +0100)]
tree-vect-generic: Fix bitfield widths [PR94980 3/3]

This third patch of three actually fixes the PR.  We were using
8-bit BIT_FIELD_REFs to access single-bit elements, and multiplying
the vector index by 8 bits rather than 1 bit.

2020-05-12  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
PR tree-optimization/94980
* tree-vect-generic.c (expand_vector_comparison): Use
vector_element_bits_tree to get the element size in bits,
rather than using TYPE_SIZE.
(expand_vector_condition, vector_element): Likewise.

gcc/testsuite/
PR tree-optimization/94980
* gcc.target/i386/pr94980.c: New test.

4 years agotree-vect-generic: Tweak build_replicated_const [PR94980 2/3]
Richard Sandiford [Tue, 12 May 2020 08:01:12 +0000 (09:01 +0100)]
tree-vect-generic: Tweak build_replicated_const [PR94980 2/3]

This patch makes build_replicated_const take the number of bits
in VALUE rather than calculating the width from the element type.
The callers can then use vector_element_bits to calculate the
correct element size from the vector type.

2020-05-12  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
PR tree-optimization/94980
* tree-vect-generic.c (build_replicated_const): Take the number
of bits as a parameter, instead of the type of the elements.
(do_plus_minus): Update accordingly, using vector_element_bits
to calculate the correct number of bits.
(do_negate): Likewise.

4 years agotree: Add vector_element_bits(_tree) [PR94980 1/3]
Richard Sandiford [Tue, 12 May 2020 08:01:10 +0000 (09:01 +0100)]
tree: Add vector_element_bits(_tree) [PR94980 1/3]

A lot of code that wants to know the number of bits in a vector
element gets that information from the element's TYPE_SIZE,
which is always equal to TYPE_SIZE_UNIT * BITS_PER_UNIT.
This doesn't work for SVE and AVX512-style packed boolean vectors,
where several elements can occupy a single byte.

This patch introduces a new pair of helpers for getting the true
(possibly sub-byte) size.  I made a token attempt to convert obvious
element size calculations, but I'm sure I missed some.

2020-05-12  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
PR tree-optimization/94980
* tree.h (vector_element_bits, vector_element_bits_tree): Declare.
* tree.c (vector_element_bits, vector_element_bits_tree): New.
* match.pd: Use the new functions instead of determining the
vector element size directly from TYPE_SIZE(_UNIT).
* tree-vect-data-refs.c (vect_gather_scatter_fn_p): Likewise.
* tree-vect-patterns.c (vect_recog_mask_conversion_pattern): Likewise.
* tree-vect-stmts.c (vect_is_simple_cond): Likewise.
* tree-vect-generic.c (expand_vector_piecewise): Likewise.
(expand_vector_conversion): Likewise.
(expand_vector_addition): Likewise for a TYPE_SIZE_UNIT used as
a divisor.  Convert the dividend to bits to compensate.
* tree-vect-loop.c (vectorizable_live_operation): Call
vector_element_bits instead of open-coding it.

4 years agoopenmp: Implement discovery of implicit declare target to clauses
Jakub Jelinek [Tue, 12 May 2020 07:17:09 +0000 (09:17 +0200)]
openmp: Implement discovery of implicit declare target to clauses

This attempts to implement what the OpenMP 5.0 spec in declare target section
says as ammended by the 5.1 changes so far (related to device_type(host)), except
that it doesn't have the device(ancestor: ...) handling yet because we do not
support it yet, and I've left so far out the except lambda note, because I need
that clarified.

2020-05-12  Jakub Jelinek  <jakub@redhat.com>

* omp-offload.h (omp_discover_implicit_declare_target): Declare.
* omp-offload.c: Include context.h.
(omp_declare_target_fn_p, omp_declare_target_var_p,
omp_discover_declare_target_fn_r, omp_discover_declare_target_var_r,
omp_discover_implicit_declare_target): New functions.
* cgraphunit.c (analyze_functions): Call
omp_discover_implicit_declare_target.

* testsuite/libgomp.c/target-39.c: New test.

4 years agoFold &MEM[0 + CST]->a.b.c to a constant
Richard Biener [Tue, 14 Jan 2020 14:25:26 +0000 (15:25 +0100)]
Fold &MEM[0 + CST]->a.b.c to a constant

This canonicalizes those to a constant literal.

2020-05-12  Richard Biener  <rguenther@suse.de>

* gimple-fold.c (maybe_canonicalize_mem_ref_addr): Canonicalize
literal constant &MEM[..] to a constant literal.

4 years agotree-optimization/95045 - fix SM with exit exiting multiple loops
Richard Biener [Mon, 11 May 2020 13:26:09 +0000 (15:26 +0200)]
tree-optimization/95045 - fix SM with exit exiting multiple loops

Since we apply SM to an edge which exits multiple loops we have
to make sure to commit insertions on it immediately since otherwise
store order is not preserved.

2020-05-12  Richard Biener  <rguenther@suse.de>

PR tree-optimization/95045
* dbgcnt.def (lim): Add debug-counter.
* tree-ssa-loop-im.c: Include dbgcnt.h.
(find_refs_for_sm): Use lim debug counter for store motion
candidates.
(do_store_motion): Rename form store_motion.  Commit edge
insertions...
(store_motion_loop): ... here.
(tree_ssa_lim): Adjust.

4 years agoActually comment the new tests
Ulrich Drepper [Tue, 12 May 2020 05:38:28 +0000 (07:38 +0200)]
Actually comment the new tests

4 years agoImplent C++20 std::atomic_flag::test
Ulrich Drepper [Tue, 12 May 2020 05:37:09 +0000 (07:37 +0200)]
Implent C++20 std::atomic_flag::test

        * include/bits/atomic_base.h (atomic_flag): Implement test member
        function.
        * include/std/version: Define __cpp_lib_atomic_flag_test.
        * testsuite/29_atomics/atomic_flag/test/explicit.cc: New file.
        * testsuite/29_atomics/atomic_flag/test/implicit.cc: New file.

4 years agors6000: Built-in cleanups for vec_clzm, vec_ctzm, and vec_gnb
Kelvin Nilsen [Tue, 12 May 2020 02:37:41 +0000 (21:37 -0500)]
rs6000: Built-in cleanups for vec_clzm, vec_ctzm, and vec_gnb

Changes to the built-in specification occurred after early patches
added support for these.  The name of vec_clzm became vec_cntlzm,
and vec_ctzm became vec_cnttzm.  Four of the overloaded forms of
vec_gnb were removed, and the fourth argument redefined as an
unsigned int, not an unsigned char.  This patch reflects those
changes in the code and test cases.  Eight of the vec_gnb test
cases are removed as a result.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/altivec.h (vec_clzm): Rename to vec_cntlzm.
(vec_ctzm): Rename to vec_cnttzm.
* config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
Change fourth operand for vec_ternarylogic to require
compatibility with unsigned SImode rather than unsigned QImode.
* config/rs6000/rs6000-call.c (altivec_overloaded_builtins):
Remove overloaded forms of vec_gnb that are no longer needed.
* doc/extend.texi (PowerPC AltiVec Built-in Functions Available
for a Future Architecture): Replace vec_clzm with vec_cntlzm;
replace vec_ctzm with vec_cntlzm; remove four unwanted forms of
vec_gnb; move vec_ternarylogic documentation into this section
and replace const unsigned char with const unsigned int as its
fourth argument.

[gcc/testsuite]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/vec-clzm-0.c: Rename to...
* gcc.target/powerpc/vec-cntlzm-0.c: ...this.
* gcc.target/powerpc/vec-clzm-1.c: Rename to...
* gcc.target/powerpc/vec-cntlzm-1.c: ...this.
* gcc.target/powerpc/vec-ctzm-0.c: Rename to...
* gcc.target/powerpc/vec-cnttzm-0.c: ...this.
* gcc.target/powerpc/vec-ctzm-1.c: Rename to...
* gcc.target/powerpc/vec-cnttzm-1.c: ...this.
* gcc.target/powerpc/vec-gnb-8.c: Rename to...
* gcc.target/powerpc/vec-gnb-0.c: ...this, deleting the old file.
* gcc.target/powerpc/vec-gnb-9.c: Rename to...
* gcc.target/powerpc/vec-gnb-1.c: ...this, deleting the old file.
* gcc.target/powerpc/vec-gnb-10.c: Rename to...
* gcc.target/powerpc/vec-gnb-2.c: ...this, deleting the old file.
* gcc.target/powerpc/vec-gnb-3.c: Delete.
* gcc.target/powerpc/vec-gnb-4.c: Delete.
* gcc.target/powerpc/vec-gnb-5.c: Delete.
* gcc.target/powerpc/vec-gnb-6.c: Delete.
* gcc.target/powerpc/vec-gnb-7.c: Delete.

4 years agors6000: Add xxgenpcvwm and xxgenpcvdm
Carl Love [Tue, 12 May 2020 02:22:07 +0000 (21:22 -0500)]
rs6000: Add xxgenpcvwm and xxgenpcvdm

Add support for xxgenpcv[dw]m, along with individual and overloaded
built-in functions for access.

[gcc]

2020-05-11  Carl Love  <cel@us.ibm.com>

* config/rs6000/altivec.h (vec_genpcvm): New #define.
* config/rs6000/rs6000-builtin.def (XXGENPCVM_V16QI): New built-in
instantiation.
(XXGENPCVM_V8HI): Likewise.
(XXGENPCVM_V4SI): Likewise.
(XXGENPCVM_V2DI): Likewise.
(XXGENPCVM): New overloaded built-in instantiation.
* config/rs6000/rs6000-call.c (altivec_overloaded_builtins): Add
entries for FUTURE_BUILTIN_VEC_XXGENPCVM.
(altivec_expand_builtin): Add special handling for
FUTURE_BUILTIN_VEC_XXGENPCVM.
(builtin_function_type): Add handling for
FUTURE_BUILTIN_XXGENPCVM_{V16QI,V8HI,V4SI,V2DI}.
* config/rs6000/vsx.md (VSX_EXTRACT_I4): New mode iterator.
(UNSPEC_XXGENPCV): New constant.
(xxgenpcvm_<mode>_internal): New insn.
(xxgenpcvm_<mode>): New expansion.
* doc/extend.texi: Add documentation for vec_genpcvm built-ins.

[gcc/testsuite]

2020-05-11  Carl Love  <cel@us.ibm.com>

* gcc.target/powerpc/xxgenpc-runnable.c: New.

4 years agoruntime: fix TestCallersNilPointerPanic
eric fang [Mon, 20 Apr 2020 08:42:01 +0000 (08:42 +0000)]
runtime: fix TestCallersNilPointerPanic

The expected result of TestCallersNilPointerPanic has changed in
GoLLVM.  This CL makes some elements of the expected result optional
so that this test passes in both gccgo and GoLLVM.

Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/230138

4 years agosyscall: append to environment in tests, don't clobber it
Ian Lance Taylor [Mon, 11 May 2020 23:23:44 +0000 (16:23 -0700)]
syscall: append to environment in tests, don't clobber it

This is a partial backport of https://golang.org/cl/233318.
It's only a partial backport because part of the change was
already applied to libgo in CL 193497 as part of the update
to the Go 1.13beta1 release.

Fixes PR go/95061

Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/233359

4 years agoDaily bump.
GCC Administrator [Tue, 12 May 2020 00:16:19 +0000 (00:16 +0000)]
Daily bump.

4 years agocompiler: use const std::string& in a couple of places
Ian Lance Taylor [Mon, 11 May 2020 22:28:55 +0000 (15:28 -0700)]
compiler: use const std::string& in a couple of places

Use a reference to avoid copying a std::string.

Fixes go/94766

Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/233320

4 years agors6000: Vector string isolate instructions
Kelvin Nilsen [Mon, 11 May 2020 21:33:19 +0000 (16:33 -0500)]
rs6000: Vector string isolate instructions

Adds new instructions vstribr, vstrihr, vstribl, and vstrihl, with
overloaded built-in support.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/altivec.h (vec_strir): New #define.
(vec_stril): Likewise.
(vec_strir_p): Likewise.
(vec_stril_p): Likewise.
* config/rs6000/altivec.md (UNSPEC_VSTRIR): New constant.
(UNSPEC_VSTRIL): Likewise.
(vstrir_<mode>): New expansion.
(vstrir_code_<mode>): New insn.
(vstrir_p_<mode>): New expansion.
(vstrir_p_code_<mode>): New insn.
(vstril_<mode>): New expansion.
(vstril_code_<mode>): New insn.
(vstril_p_<mode>): New expansion.
(vstril_p_code_<mode>): New insn.
* config/rs6000/rs6000-builtin.def (__builtin_altivec_vstribr):
New built-in function.
(__builtin_altivec_vstrihr): Likewise.
(__builtin_altivec_vstribl): Likewise.
(__builtin_altivec_vstrihl): Likewise.
(__builtin_altivec_vstribr_p): Likewise.
(__builtin_altivec_vstrihr_p): Likewise.
(__builtin_altivec_vstribl_p): Likewise.
(__builtin_altivec_vstrihl_p): Likewise.
(__builtin_vec_strir): New overloaded built-in function.
(__builtin_vec_stril): Likewise.
(__builtin_vec_strir_p): Likewise.
(__builtin_vec_stril_p): Likewise.
* config/rs6000/rs6000-call.c (altivec_overloaded_builtins):
Define overloaded forms of __builtin_vec_strir,
__builtin_vec_stril, __builtin_vec_strir_p, and
__builtin_vec_stril_p.
* doc/extend.texi (PowerPC AltiVec Built-in Functions Available
for a Future Architecture): Add description of vec_stril,
vec_stril_p, vec_strir, and vec_strir_p built-in functions.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/vec-stril-0.c: New.
* gcc.target/powerpc/vec-stril-1.c: New.
* gcc.target/powerpc/vec-stril-10.c: New.
* gcc.target/powerpc/vec-stril-11.c: New.
* gcc.target/powerpc/vec-stril-12.c: New.
* gcc.target/powerpc/vec-stril-13.c: New.
* gcc.target/powerpc/vec-stril-14.c: New.
* gcc.target/powerpc/vec-stril-15.c: New.
* gcc.target/powerpc/vec-stril-16.c: New.
* gcc.target/powerpc/vec-stril-17.c: New.
* gcc.target/powerpc/vec-stril-18.c: New.
* gcc.target/powerpc/vec-stril-19.c: New.
* gcc.target/powerpc/vec-stril-2.c: New.
* gcc.target/powerpc/vec-stril-20.c: New.
* gcc.target/powerpc/vec-stril-21.c: New.
* gcc.target/powerpc/vec-stril-22.c: New.
* gcc.target/powerpc/vec-stril-23.c: New.
* gcc.target/powerpc/vec-stril-3.c: New.
* gcc.target/powerpc/vec-stril-4.c: New.
* gcc.target/powerpc/vec-stril-5.c: New.
* gcc.target/powerpc/vec-stril-6.c: New.
* gcc.target/powerpc/vec-stril-7.c: New.
* gcc.target/powerpc/vec-stril-8.c: New.
* gcc.target/powerpc/vec-stril-9.c: New.
* gcc.target/powerpc/vec-stril_p-0.c: New.
* gcc.target/powerpc/vec-stril_p-1.c: New.
* gcc.target/powerpc/vec-stril_p-10.c: New.
* gcc.target/powerpc/vec-stril_p-11.c: New.
* gcc.target/powerpc/vec-stril_p-2.c: New.
* gcc.target/powerpc/vec-stril_p-3.c: New.
* gcc.target/powerpc/vec-stril_p-4.c: New.
* gcc.target/powerpc/vec-stril_p-5.c: New.
* gcc.target/powerpc/vec-stril_p-6.c: New.
* gcc.target/powerpc/vec-stril_p-7.c: New.
* gcc.target/powerpc/vec-stril_p-8.c: New.
* gcc.target/powerpc/vec-stril_p-9.c: New.
* gcc.target/powerpc/vec-strir-0.c: New.
* gcc.target/powerpc/vec-strir-1.c: New.
* gcc.target/powerpc/vec-strir-10.c: New.
* gcc.target/powerpc/vec-strir-11.c: New.
* gcc.target/powerpc/vec-strir-12.c: New.
* gcc.target/powerpc/vec-strir-13.c: New.
* gcc.target/powerpc/vec-strir-14.c: New.
* gcc.target/powerpc/vec-strir-15.c: New.
* gcc.target/powerpc/vec-strir-16.c: New.
* gcc.target/powerpc/vec-strir-17.c: New.
* gcc.target/powerpc/vec-strir-18.c: New.
* gcc.target/powerpc/vec-strir-19.c: New.
* gcc.target/powerpc/vec-strir-2.c: New.
* gcc.target/powerpc/vec-strir-20.c: New.
* gcc.target/powerpc/vec-strir-21.c: New.
* gcc.target/powerpc/vec-strir-22.c: New.
* gcc.target/powerpc/vec-strir-23.c: New.
* gcc.target/powerpc/vec-strir-3.c: New.
* gcc.target/powerpc/vec-strir-4.c: New.
* gcc.target/powerpc/vec-strir-5.c: New.
* gcc.target/powerpc/vec-strir-6.c: New.
* gcc.target/powerpc/vec-strir-7.c: New.
* gcc.target/powerpc/vec-strir-8.c: New.
* gcc.target/powerpc/vec-strir-9.c: New.
* gcc.target/powerpc/vec-strir_p-0.c: New.
* gcc.target/powerpc/vec-strir_p-1.c: New.
* gcc.target/powerpc/vec-strir_p-10.c: New.
* gcc.target/powerpc/vec-strir_p-11.c: New.
* gcc.target/powerpc/vec-strir_p-2.c: New.
* gcc.target/powerpc/vec-strir_p-3.c: New.
* gcc.target/powerpc/vec-strir_p-4.c: New.
* gcc.target/powerpc/vec-strir_p-5.c: New.
* gcc.target/powerpc/vec-strir_p-6.c: New.
* gcc.target/powerpc/vec-strir_p-7.c: New.
* gcc.target/powerpc/vec-strir_p-8.c: New.
* gcc.target/powerpc/vec-strir_p-9.c: New.

4 years agors6000: Add xxeval and vec_ternarylogic
Kelvin Nilsen [Mon, 11 May 2020 21:25:03 +0000 (16:25 -0500)]
rs6000: Add xxeval and vec_ternarylogic

Add the xxeval insn and access it via the vec_ternarylogic built-in
function.  As part of this, add support to the built-in function
infrastructure for functions that take four arguments.

[gcc]

2020-05-11  Kelvin Nilsen  <wschmidt@linux.ibm.com>

* config/rs6000/altivec.h (vec_ternarylogic): New #define.
* config/rs6000/altivec.md (UNSPEC_XXEVAL): New constant.
(xxeval): New insn.
* config/rs6000/predicates.md (u8bit_cint_operand): New predicate.
* config/rs6000/rs6000-builtin.def: Add handling of new macro
RS6000_BUILTIN_4.
(BU_FUTURE_V_4): New macro. Use it.
(BU_FUTURE_OVERLOAD_4): Likewise.
* config/rs6000/rs6000-c.c (altivec_build_resolved_builtin): Add
handling for quaternary built-in functions.
(altivec_resolve_overloaded_builtin): Add special-case handling
for __builtin_vec_xxeval.
* config/rs6000/rs6000-call.c: Add handling of new macro
RS6000_BUILTIN_4 in initialization of rs6000_builtin_info,
bdesc0_arg, bdesc1_arg, bdesc2_arg, bdesc_3arg,
bdesc_altivec_preds, bdesc_abs, and bdesc_htm arrays.
(altivec_overloaded_builtins): Add definitions for
FUTURE_BUILTIN_VEC_XXEVAL.
(bdesc_4arg): New array.
(htm_expand_builtin): Add handling for quaternary built-in
functions.
(rs6000_expand_quaternop_builtin): New function.
(rs6000_expand_builtin): Add handling for quaternary built-in
functions.
(rs6000_init_builtins): Initialize builtin_mode_to_type entries
for unsigned QImode and unsigned HImode.
(builtin_quaternary_function_type): New function.
(rs6000_common_init_builtins): Add handling of quaternary
operations.
* config/rs6000/rs6000.h (RS6000_BTC_QUATERNARY): New defined
constant.
(RS6000_BTC_PREDICATE): Change value of constant.
(RS6000_BTC_ABS): Likewise.
(rs6000_builtins): Add support for new macro RS6000_BUILTIN_4.
* doc/extend.texi (PowerPC AltiVec Built-In Functions Available
for a Future Architecture): Add description of vec_ternarylogic
built-in function.

[gcc/testsuite]

2020-05-11  Kelvin Nilsen  <wschmidt@linux.ibm.com>

* gcc.target/powerpc/vec-ternarylogic-0.c: New.
* gcc.target/powerpc/vec-ternarylogic-1.c: New.
* gcc.target/powerpc/vec-ternarylogic-10.c: New.
* gcc.target/powerpc/vec-ternarylogic-2.c: New.
* gcc.target/powerpc/vec-ternarylogic-3.c: New.
* gcc.target/powerpc/vec-ternarylogic-4.c: New.
* gcc.target/powerpc/vec-ternarylogic-5.c: New.
* gcc.target/powerpc/vec-ternarylogic-6.c: New.
* gcc.target/powerpc/vec-ternarylogic-7.c: New.
* gcc.target/powerpc/vec-ternarylogic-8.c: New.
* gcc.target/powerpc/vec-ternarylogic-9.c: New.

4 years agors6000: Add pdepd and pextd
Kelvin Nilsen [Mon, 11 May 2020 21:16:15 +0000 (16:16 -0500)]
rs6000: Add pdepd and pextd

Add scalar instructions for parallel bit deposit and extract, with
built-in function support.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/rs6000-builtin.def (__builtin_pdepd): New built-in
function.
(__builtin_pextd): Likewise.
* config/rs6000/rs6000.md (UNSPEC_PDEPD): New constant.
(UNSPEC_PEXTD): Likewise.
(pdepd): New insn.
(pextd): Likewise.
* doc/extend.texi (Basic PowerPC Built-in Functions Available for
a Future Architecture): Add descriptions of __builtin_pdepd and
__builtin_pextd functions.

[gcc/testsuite]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/pdep-0.c: New.
* gcc.target/powerpc/pdep-1.c: New.
* gcc.target/powerpc/pextd-0.c: New.
* gcc.target/powerpc/pextd-1.c: New.

4 years agors6000: Add vclrlb and vclrrb
Kelvin Nilsen [Mon, 11 May 2020 21:09:53 +0000 (16:09 -0500)]
rs6000: Add vclrlb and vclrrb

Add new vector instructions to clear leftmost and rightmost bytes.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/altivec.h (vec_clrl): New #define.
(vec_clrr): Likewise.
* config/rs6000/altivec.md (UNSPEC_VCLRLB): New constant.
(UNSPEC_VCLRRB): Likewise.
(vclrlb): New insn.
(vclrrb): Likewise.
* config/rs6000/rs6000-builtin.def (__builtin_altivec_vclrlb): New
built-in function.
(__builtin_altivec_vclrrb): Likewise.
(__builtin_vec_clrl): New overloaded built-in function.
(__builtin_vec_clrr): Likewise.
* config/rs6000/rs6000-call.c (altivec_overloaded_builtins):
Define overloaded forms of __builtin_vec_clrl and
__builtin_vec_clrr.
* doc/extend.texi (PowerPC AltiVec Built-in Functions Available
for a Future Architecture): Add descriptions of vec_clrl and
vec_clrr.

[gcc/testsuite]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/vec-clrl-0.c: New.
* gcc.target/powerpc/vec-clrl-1.c: New.
* gcc.target/powerpc/vec-clrr-0.c: New.
* gcc.target/powerpc/vec-clrr-1.c: New.

4 years agoFix change log ordering from previous commit
Bill Schmidt [Mon, 11 May 2020 21:04:55 +0000 (16:04 -0500)]
Fix change log ordering from previous commit

4 years agoUpdate gcc .po files.
Joseph Myers [Mon, 11 May 2020 20:42:46 +0000 (20:42 +0000)]
Update gcc .po files.

* be.po, da.po, de.po, el.po, es.po, fi.po, fr.po, hr.po, id.po,
ja.po, nl.po, ru.po, sr.po, sv.po, tr.po, uk.po, vi.po, zh_CN.po,
zh_TW.po: Update.

4 years agors6000: Add cntlzdm and cnttzdm
Kelvin Nilsen [Mon, 11 May 2020 20:10:24 +0000 (15:10 -0500)]
rs6000: Add cntlzdm and cnttzdm

Add support for new scalar instructions for counting leading or
trailing zeros under control of a bitmask.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/rs6000-builtin.def (__builtin_cntlzdm): New
built-in function definition.
(__builtin_cnttzdm): Likewise.
* config/rs6000/rs6000.md (UNSPEC_CNTLZDM): New constant.
(UNSPEC_CNTTZDM): Likewise.
(cntlzdm): New insn.
(cnttzdm): Likewise.
* doc/extend.texi (Basic PowerPC Built-in Functions available for
a Future Architecture): Add descriptions of __builtin_cntlzdm and
__builtin_cnttzdm functions.

[gcc/testsuite]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/cntlzdm-0.c: New test.
* gcc.target/powerpc/cntlzdm-1.c: New test.
* gcc.target/powerpc/cnttzdm-0.c: New test.
* gcc.target/powerpc/cnttzdm-1.c: New test.

4 years agoc++: Fix specialization of constrained member template.
Jason Merrill [Mon, 11 May 2020 19:46:59 +0000 (15:46 -0400)]
c++: Fix specialization of constrained member template.

The resolution of comment CA104 clarifies that we need to do direct
substitution of constraints in order to determine which member template
corresponds to an explicit specialization.

gcc/cp/ChangeLog
2020-05-11  Jason Merrill  <jason@redhat.com>

Resolve C++20 NB comment CA104
* pt.c (determine_specialization): Compare constraints for
specialization of member template of class instantiation.

4 years agoc++: tree walk into TYPENAME_TYPE.
Jason Merrill [Mon, 11 May 2020 19:46:59 +0000 (15:46 -0400)]
c++: tree walk into TYPENAME_TYPE.

While looking at 92583/92654 it occurred to me that typename types needed
the same fix.  So extract_locals_r also needs to see the TYPE_CONTEXT of a
TYPENAME_TYPE.  But it must not look through a typedef.

Most tree walking in the front end wants to walk through the syntactic form
of a type of expression, and doesn't care about the type referred to by a
typedef.  But min_vis_r does care.

gcc/cp/ChangeLog
2020-05-11  Jason Merrill  <jason@redhat.com>

PR c++/92583
PR c++/92654
* tree.c (cp_walk_subtrees): Stop at typedefs.
Handle TYPENAME_TYPE here.
* pt.c (find_parameter_packs_r): Not here.
(for_each_template_parm_r): Clear *walk_subtrees.
* decl2.c (min_vis_r): Look through typedefs.

4 years agoc++: Better diagnostic in converted const expr.
Jason Merrill [Mon, 11 May 2020 19:39:44 +0000 (15:39 -0400)]
c++: Better diagnostic in converted const expr.

This improves the diagnostic from

error: could not convert ‘((A<>*)(void)0)->A<>::e’ from
       ‘<unresolved overloaded function type>’ to ‘bool’

to

error: cannot convert ‘A<>::e’ from type ‘void (A<>::)()’ to type ‘bool’

gcc/cp/ChangeLog
2020-05-11  Jason Merrill  <jason@redhat.com>

* call.c (implicit_conversion_error): Split out from...
(perform_implicit_conversion_flags): ...here.
(build_converted_constant_expr_internal): Use it.

4 years agoc++: Use of 'this' in parameter declaration [PR90748]
Jason Merrill [Mon, 11 May 2020 19:39:44 +0000 (15:39 -0400)]
c++: Use of 'this' in parameter declaration [PR90748]

We were incorrectly accepting the use of 'this' at parse time and then
crashing when we tried to instantiate it.  It is invalid because 'this' is
not in scope until after the function-cv-quals.  So let's hoist setting
current_class_ptr up from cp_parser_late_return_type_opt into
cp_parser_direct_declarator where it can work for noexcept as well.

gcc/cp/ChangeLog
2020-05-11  Jason Merrill  <jason@redhat.com>

PR c++/90748
* parser.c (inject_parm_decls): Set current_class_ptr here.
(cp_parser_direct_declarator): And here.
(cp_parser_late_return_type_opt): Not here.
(cp_parser_noexcept_specification_opt): Nor here.
(cp_parser_exception_specification_opt)
(cp_parser_late_noexcept_specifier): Remove unneeded parameters.

4 years agoPR fortran/95053 - ICE in gfc_divide(): Bad basic type
Harald Anlauf [Mon, 11 May 2020 19:27:11 +0000 (21:27 +0200)]
PR fortran/95053 - ICE in gfc_divide(): Bad basic type

The fix for PR 93499 introduced a too strict check in gfc_divide
that could trigger errors in the early parsing phase.  Relax the
check and defer to a later stage.

gcc/fortran/

2020-05-11  Harald Anlauf  <anlauf@gmx.de>

PR fortran/95053
* arith.c (gfc_divide): Do not error out if operand 2 is
non-numeric.  Defer checks to later stage.

gcc/testsuite/

2020-05-11  Harald Anlauf  <anlauf@gmx.de>

PR fortran/95053
* gfortran.dg/pr95053.f: New test.

4 years agoc++: Make references to __cxa_pure_virtual weak.
Jason Merrill [Mon, 11 May 2020 18:05:46 +0000 (14:05 -0400)]
c++: Make references to __cxa_pure_virtual weak.

If a program has no other dependencies on libstdc++, we shouldn't require it
just for __cxa_pure_virtual, which is only there to give a prettier
diagnostic before crashing the program; resolving the reference to NULL will
also crash, just without the diagnostic.

gcc/cp/ChangeLog
2020-05-11  Jason Merrill  <jason@redhat.com>

* decl.c (cxx_init_decl_processing): Call declare_weak for
__cxa_pure_virtual.

4 years agoc++: Improve print_tree of static_assert.
Jason Merrill [Mon, 11 May 2020 18:05:46 +0000 (14:05 -0400)]
c++: Improve print_tree of static_assert.

We weren't printing the condition and message of a STATIC_ASSERT.

It's also unnecessary to duplicate the code for instantiating a
STATIC_ASSERT between tsubst_expr and instantiate_class_template_1.

gcc/cp/ChangeLog
2020-05-11  Jason Merrill  <jason@redhat.com>

* pt.c (instantiate_class_template_1): Call tsubst_expr for
STATIC_ASSERT member.
* ptree.c (cxx_print_xnode): Handle STATIC_ASSERT.

4 years agoc++: Remove redundant code.
Jason Merrill [Mon, 11 May 2020 18:05:46 +0000 (14:05 -0400)]
c++: Remove redundant code.

We walk the lambda captures in cp_walk_subtrees, so we don't also need to
walk them here.

gcc/cp/ChangeLog
2020-05-11  Jason Merrill  <jason@redhat.com>

* pt.c (find_parameter_packs_r) [LAMBDA_EXPR]: Remove redundant
walking of capture list.

4 years agoc++: Remove LOOKUP_EXPLICIT_TMPL_ARGS.
Jason Merrill [Mon, 11 May 2020 18:05:46 +0000 (14:05 -0400)]
c++: Remove LOOKUP_EXPLICIT_TMPL_ARGS.

This flag is redundant with the explicit_targs field in the overload
candidate information.

gcc/cp/ChangeLog
2020-05-11  Jason Merrill  <jason@redhat.com>

* cp-tree.h (LOOKUP_EXPLICIT_TMPL_ARGS): Remove.
* call.c (build_new_function_call): Don't set it.
(build_new_method_call_1): Likewise.
(build_over_call): Check cand->explicit_targs instead.

4 years agoc++: Tweak VLA representation.
Jason Merrill [Mon, 11 May 2020 18:05:46 +0000 (14:05 -0400)]
c++: Tweak VLA representation.

If we put the SAVE_EXPR for a VLA size inside the MINUS_EXPR rather than
outside, it will work better with constant folding.

The equivalent change was made in the C front-end in 2004, in commit
r0-64535-g8b0b9aefd29dfe6398857bcf5628662e2f0e21f6

gcc/cp/ChangeLog
2020-05-11  Jason Merrill  <jason@redhat.com>

* decl.c (compute_array_index_type_loc): Stabilize before building
the MINUS_EXPR.

4 years agoc++: Avoid unnecessary deprecated warnings.
Jason Merrill [Mon, 11 May 2020 18:05:46 +0000 (14:05 -0400)]
c++: Avoid unnecessary deprecated warnings.

There's no need to warn that a deprecated function uses a deprecated type,
that just adds noise.  We were preventing that in start_decl, but that
didn't help member declarations that go through grokfield.  So handle it in
grokdeclarator instead, which is shared between them.

gcc/cp/ChangeLog
2020-05-11  Jason Merrill  <jason@redhat.com>

* decl.c (grokdeclarator): Adjust deprecated_state here.
(start_decl): Not here.

4 years agoi386: Add V2SFmode sqrt insn pattern [PR95046]
Uros Bizjak [Mon, 11 May 2020 18:12:14 +0000 (20:12 +0200)]
i386: Add V2SFmode sqrt insn pattern [PR95046]

gcc/ChangeLog:

2020-05-11  Uroš Bizjak  <ubizjak@gmail.com>

PR target/95046
* config/i386/mmx.md (sqrtv2sf2): New insn pattern.

testsuite/ChangeLog:

2020-05-11  Uroš Bizjak  <ubizjak@gmail.com>

PR target/95046
* gcc.target/i386/pr95046-1.c (test_sqrt): Add.

4 years agolibbacktrace: declare getpagesize if necessary
Ian Lance Taylor [Mon, 11 May 2020 17:51:21 +0000 (10:51 -0700)]
libbacktrace: declare getpagesize if necessary

libbacktrace/
PR libbacktrace/95012
* configure.ac: Check for getpagesize declaration.
* mmap.c: Declare getpagesize if necessary.
* mmapio.c: Likewise.

4 years agors6000: Add vcfuged instruction
Kelvin Nilsen [Mon, 11 May 2020 16:41:23 +0000 (11:41 -0500)]
rs6000: Add vcfuged instruction

Add the new vector centrifuge-doubleword instruction and built-in
function access.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/altivec.h (vec_cfuge): New #define.
* config/rs6000/altivec.md (UNSPEC_VCFUGED): New constant.
(vcfuged): New insn.
* config/rs6000/rs6000-builtin.def (__builtin_altivec_vcfuged):
New built-in function.
* config/rs6000/rs6000-call.c (builtin_function_type): Add
handling for FUTURE_BUILTIN_VCFUGED case.
* doc/extend.texi (PowerPC AltiVec Built-in Functions Available
for a Future Architecture): Add description of vec_cfuge built-in
function.

[gcc/testsuite]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/vec-cfuged-0.c: New test.
* gcc.target/powerpc/vec-cfuged-1.c: New test.

4 years agors6000: Add scalar cfuged instruction
Kelvin Nilsen [Mon, 11 May 2020 16:01:32 +0000 (11:01 -0500)]
rs6000: Add scalar cfuged instruction

Add the centifuge-doubleword instruction and built-in access.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/rs6000-builtin.def (BU_FUTURE_MISC_0): New
#define.
(BU_FUTURE_MISC_1): Likewise.
(BU_FUTURE_MISC_2): Likewise.
(BU_FUTURE_MISC_3): Likewise.
(__builtin_cfuged): New built-in function definition.
* config/rs6000/rs6000.md (UNSPEC_CFUGED): New constant.
(cfuged): New insn.
* doc/extend.texi (Basic PowerPC Built-in Functions Available for
a Future Architecture): New subsubsection.

[gcc/testsuite]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target.powerpc/cfuged-0.c: New test.
* gcc.target.powerpc/cfuged-1.c: New test.

4 years agotree-optimization/95049 - fix not terminating RPO VN iteration
Richard Biener [Mon, 11 May 2020 11:40:37 +0000 (13:40 +0200)]
tree-optimization/95049 - fix not terminating RPO VN iteration

This rejects lattice changes from one constant to another.

2020-05-11  Richard Biener  <rguenther@suse.de>

PR tree-optimization/95049
* tree-ssa-sccvn.c (set_ssa_val_to): Reject lattice transition
between different constants.

* gcc.dg/torture/pr95049.c: New testcase.

4 years agotree-pretty-print: Handle boolean types
Richard Sandiford [Mon, 11 May 2020 15:51:48 +0000 (16:51 +0100)]
tree-pretty-print: Handle boolean types

AVX512-style masks and SVE-style predicates can be difficult
to debug in gimple dumps, since the types are printed like this:

  vector(4) <unnamed type> foo;

Some important details are hidden by that <unnamed type>,
such as the number of bits in an element and whether the type
is signed or unsigned.

This patch uses an ad-hoc syntax for printing unnamed
boolean types.  Normal frontend ones should be handled
by the earlier TYPE_NAME code.

2020-05-11  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* tree-pretty-print.c (dump_generic_node): Handle BOOLEAN_TYPEs.

4 years agors6000: Add vgnb
Kelvin Nilsen [Mon, 11 May 2020 15:13:14 +0000 (10:13 -0500)]
rs6000: Add vgnb

Add support for the vgnb instruction, which gathers every Nth bit
per vector element.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>
    Bill Schmidt  <wschmidt@linux.ibm.com>

* config/rs6000/altivec.h (vec_gnb): New #define.
* config/rs6000/altivec.md (UNSPEC_VGNB): New constant.
(vgnb): New insn.
* config/rs6000/rs6000-builtin.def (BU_FUTURE_OVERLOAD_1): New
#define.
(BU_FUTURE_OVERLOAD_2): Likewise.
(BU_FUTURE_OVERLOAD_3): Likewise.
(__builtin_altivec_gnb): New built-in function.
(__buiiltin_vec_gnb): New overloaded built-in function.
* config/rs6000/rs6000-call.c (altivec_overloaded_builtins):
Define overloaded forms of __builtin_vec_gnb.
(rs6000_expand_binop_builtin): Add error checking for 2nd argument
of __builtin_vec_gnb.
(builtin_function_type): Mark return value and arguments unsigned
for FUTURE_BUILTIN_VGNB.
* doc/extend.texi (PowerPC AltiVec Built-in Functions Available
for a Future Architecture): Add description of vec_gnb built-in
function.

[gcc/testsuite]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>
    Bill Schmidt  <wschmidt@linux.ibm.com>

* gcc.target/powerpc/vec-gnb-0.c: New test.
* gcc.target/powerpc/vec-gnb-1.c: New test.
* gcc.target/powerpc/vec-gnb-10.c: New test.
* gcc.target/powerpc/vec-gnb-2.c: New test.
* gcc.target/powerpc/vec-gnb-3.c: New test.
* gcc.target/powerpc/vec-gnb-4.c: New test.
* gcc.target/powerpc/vec-gnb-5.c: New test.
* gcc.target/powerpc/vec-gnb-6.c: New test.
* gcc.target/powerpc/vec-gnb-7.c: New test.
* gcc.target/powerpc/vec-gnb-8.c: New test.
* gcc.target/powerpc/vec-gnb-9.c: New test.

4 years agors6000: Add vector pdep/pext
Kelvin Nilsen [Mon, 11 May 2020 15:04:03 +0000 (10:04 -0500)]
rs6000: Add vector pdep/pext

Add support for the vpdepd and vpextd instructions that perform
vector parallel bit deposit and vector parallel bit extract.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>
    Bill Schmidt  <wschmidt@linux.ibm.com>

* config/rs6000/altivec.h (vec_pdep): New macro implementing new
built-in function.
(vec_pext): Likewise.
* config/rs6000/altivec.md (UNSPEC_VPDEPD): New constant.
(UNSPEC_VPEXTD): Likewise.
(vpdepd): New insn.
(vpextd): Likewise.
* config/rs6000/rs6000-builtin.def (__builtin_altivec_vpdepd): New
built-in function.
(__builtin_altivec_vpextd): Likewise.
* config/rs6000/rs6000-call.c (builtin_function_type): Add
handling for FUTURE_BUILTIN_VPDEPD and FUTURE_BUILTIN_VPEXTD
cases.
* doc/extend.texi (PowerPC Altivec Built-in Functions Available
for a Future Architecture): Add description of vec_pdep and
vec_pext built-in functions.

[gcc/testsuite]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/vec-pdep-0.c: New.
* gcc.target/powerpc/vec-pdep-1.c: New.
* gcc.target/powerpc/vec-pext-0.c: New.
* gcc.target/powerpc/vec-pext-1.c: New.

4 years agors6000: Add vector count under mask
Kelvin Nilsen [Mon, 11 May 2020 14:35:01 +0000 (09:35 -0500)]
rs6000: Add vector count under mask

Add support for new vclzdm and vctzdm vector instructions that
count leading and trailing zeros under control of a mask.

[gcc]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>
    Bill Schmidt  <wschmidt@linux.ibm.com>

* config/rs6000/altivec.h (vec_clzm): New macro.
(vec_ctzm): Likewise.
* config/rs6000/altivec.md (UNSPEC_VCLZDM): New constant.
(UNSPEC_VCTZDM): Likewise.
(vclzdm): New insn.
(vctzdm): Likewise.
* config/rs6000/rs6000-builtin.def (BU_FUTURE_V_0): New macro.
(BU_FUTURE_V_1): Likewise.
(BU_FUTURE_V_2): Likewise.
(BU_FUTURE_V_3): Likewise.
(__builtin_altivec_vclzdm): New builtin definition.
(__builtin_altivec_vctzdm): Likewise.
* config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Cause
_ARCH_PWR_FUTURE macro to be defined if OPTION_MASK_FUTURE flag is
set.
* config/rs6000/rs6000-call.c (builtin_function_type): Set return
value and parameter types to be unsigned for VCLZDM and VCTZDM.
* config/rs6000/rs6000.c (rs6000_builtin_mask_calculate): Add
support for TARGET_FUTURE flag.
* config/rs6000/rs6000.h (RS6000_BTM_FUTURE): New macro constant.
* doc/extend.texi (PowerPC Altivec Built-in Functions Available
for a Future Architecture): New subsubsection.

[gcc/testsuite]

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/vec-clzm-0.c: New test.
* gcc.target/powerpc/vec-clzm-1.c: New test.
* gcc.target/powerpc/vec-ctzm-0.c: New test.
* gcc.target/powerpc/vec-ctzm-1.c: New test.

4 years agotree-optimization/94988 - enhance SM some more
Richard Biener [Fri, 8 May 2020 10:03:30 +0000 (12:03 +0200)]
tree-optimization/94988 - enhance SM some more

This enhances store-order preserving store motion to handle the case
of non-invariant dependent stores in the sequence of unconditionally
executed stores on exit by re-issueing them as part of the sequence
of stores on the exit.  This fixes the observed regression of
gcc.target/i386/pr64110.c which relies on store-motion of 'b'
for a loop like

  for (int i = 0; i < j; ++i)
    *b++ = x;

where for correctness we now no longer apply store-motion.  With
the patch we emit the correct

  tem = b;
  for (int i = 0; i < j; ++i)
    {
      tem = tem + 1;
      *tem = x;
    }
  b = tem;
  *tem = x;

preserving the original order of stores.  A testcase reflecting
the miscompilation done by earlier GCC is added as well.

This also fixes the reported ICE in PR95025 and adds checking code
to catch it earlier - the issue was not-supported refs propagation
leaving stray refs in the sequence.

2020-05-11  Richard Biener  <rguenther@suse.de>

PR tree-optimization/94988
PR tree-optimization/95025
* tree-ssa-loop-im.c (seq_entry): Make a struct, add from.
(sm_seq_push_down): Take extra parameter denoting where we
moved the ref to.
(execute_sm_exit): Re-issue sm_other stores in the correct
order.
(sm_seq_valid_bb): When always executed, allow sm_other to
prevail inbetween sm_ord and record their stored value.
(hoist_memory_references): Adjust refs_not_supported propagation
and prune sm_other from the end of the ordered sequences.

* gcc.dg/torture/pr94988.c: New testcase.
* gcc.dg/torture/pr95025.c: Likewise.
* gcc.dg/torture/pr95045.c: Likewise.
* g++.dg/asan/pr95025.C: New testcase.

4 years ago[Fortran] Fix/modify present() handling for assumed-shape optional (PR 94672)
Tobias Burnus [Mon, 11 May 2020 14:39:20 +0000 (16:39 +0200)]
[Fortran] Fix/modify present() handling for assumed-shape optional (PR 94672)

gcc/fortran/
2020-05-07  Tobias Burnus  <tobias@codesourcery.com>

PR fortran/94672
* trans.h (gfc_conv_expr_present): Add use_saved_decl=false argument.
* trans-expr.c (gfc_conv_expr_present): Likewise; use DECL directly
and only if use_saved_decl is true, use the actual PARAM_DECL arg (saved
descriptor).
* trans-array.c (gfc_trans_dummy_array_bias): Set local 'arg.0'
variable to NULL if 'arg' is not present.
* trans-openmp.c (gfc_omp_check_optional_argument): Simplify by checking
'arg.0' instead of the true PARM_DECL.
(gfc_omp_finish_clause): Remove setting 'arg.0' to NULL.

gcc/testsuite/
2020-05-07  Jakub Jelinek  <jakub@redhat.com>
    Tobias Burnus  <tobias@codesourcery.com>

PR fortran/94672
* gfortran.dg/gomp/pr94672.f90: New.
* gfortran.dg/missing_optional_dummy_6a.f90: Update scan-tree.

4 years agoi386: Improve basic vectorized V2SFmode operations [PR95046]
Uros Bizjak [Mon, 11 May 2020 14:37:19 +0000 (16:37 +0200)]
i386: Improve basic vectorized V2SFmode operations [PR95046]

Use plain "v" constraint for AVX alternatives and add "prefix" attribute.

gcc/ChangeLog:

PR target/95046
* config/i386/mmx.md (mmx_addv2sf3): Use "v" constraint
instead of "Yv" for AVX alternatives.  Add "prefix" attribute.
(*mmx_addv2sf3): Ditto.
(*mmx_subv2sf3): Ditto.
(*mmx_mulv2sf3): Ditto.
(*mmx_<code>v2sf3): Ditto.
(mmx_ieee_<ieee_maxmin>v2sf3): Ditto.

4 years agoaarch64: Fix ICE when expanding scalar floating move with -mgeneral-regs-only. [PR94991]
Fei Yang [Mon, 11 May 2020 14:18:47 +0000 (15:18 +0100)]
aarch64: Fix ICE when expanding scalar floating move with -mgeneral-regs-only. [PR94991]

In the testcase for PR94991, we are doing FAIL for scalar floating move expand
pattern since TARGET_FLOAT is false with option -mgeneral-regs-only. But move
expand pattern cannot fail. It would be better to replace the FAIL with code
that bitcasts to the equivalent integer mode using gen_lowpart.

2020-05-11  Felix Yang  <felix.yang@huawei.com>

gcc/
PR target/94991
* config/aarch64/aarch64.md (mov<mode>):
Bitcasts to the equivalent integer mode using gen_lowpart
instead of doing FAIL for scalar floating point move.

gcc/testsuite/
PR target/94991
* gcc.target/aarch64/mgeneral-regs_5.c: New test.

4 years ago[PATCH] aarch64: prefer using csinv, csneg in zero extend contexts
Alex Coplan [Mon, 11 May 2020 14:18:46 +0000 (15:18 +0100)]
[PATCH] aarch64: prefer using csinv, csneg in zero extend contexts

Given the C code:

unsigned long long inv(unsigned a, unsigned b, unsigned c)
{
  return a ? b : ~c;
}

Prior to this patch, AArch64 GCC at -O2 generates:

inv:
        cmp     w0, 0
        mvn     w2, w2
        csel    w0, w1, w2, ne
        ret

and after applying the patch, we get:

inv:
        cmp     w0, 0
        csinv   w0, w1, w2, ne
        ret

The new pattern also catches the optimization for the symmetric case where the
body of foo reads a ? ~b : c.

Similarly, with the following code:

unsigned long long neg(unsigned a, unsigned b, unsigned c)
{
  return a ? b : -c;
}

GCC at -O2 previously gave:

neg:
        cmp     w0, 0
        neg     w2, w2
        csel    w0, w1, w2, ne

but now gives:

neg:
        cmp     w0, 0
        csneg   w0, w1, w2, ne
        ret

with the corresponding code for the symmetric case as above.

2020-05-11  Alex Coplan  <alex.coplan@arm.com>

gcc/
* config/aarch64/aarch64.c (aarch64_if_then_else_costs): Add case
to correctly calculate cost for new pattern (*csinv3_uxtw_insn3).
* config/aarch64/aarch64.md (*csinv3_utxw_insn1): New.
(*csinv3_uxtw_insn2): New.
(*csinv3_uxtw_insn3): New.
* config/aarch64/iterators.md (neg_not_cs): New.

gcc/testsuite/
* gcc.target/aarch64/csinv-neg.c: New test.

4 years agoFix missing files from previous commit.
Kelvin Nilsen [Mon, 11 May 2020 13:49:06 +0000 (08:49 -0500)]
Fix missing files from previous commit.

4 years agors6000: powerpc_future_ok and powerpc_future_hw
Kelvin Nilsen [Mon, 11 May 2020 13:44:48 +0000 (08:44 -0500)]
rs6000: powerpc_future_ok and powerpc_future_hw

Dejagnu targets for these.

2020-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/dg-future-0.c: New.
* gcc.target/powerpc/dg-future-1.c: New.
* lib/target-supports.exp (check_powerpc_future_hw_available):
Replace -mfuture with -mcpu=future.
(check_effective_target_powerpc_future_ok): Likewise.
(is-effective-target): Add powerpc_future_hw.

4 years agoRevert "libstdc++ Enhance thread safety of debug mode iterators"
François Dumont [Mon, 11 May 2020 12:07:06 +0000 (14:07 +0200)]
Revert "libstdc++ Enhance thread safety of debug mode iterators"

This reverts commit 0b83c4fabb899fdbb3ae60ed75b7004b7859fae9.

4 years agotestsuite: Require gnu-tm support for pr94856.C
Kito Cheng [Mon, 11 May 2020 09:51:03 +0000 (17:51 +0800)]
testsuite: Require gnu-tm support for pr94856.C

 - The testcase uses the -fgnu-tm option but does not ensure that support
   is enabled. This patch adds the test to the testcase.

* gcc/testsuite/g++.dg/ipa/pr94856.C: Require fgnu-tm.

4 years agoi386: Vectorize basic V2SFmode operations [PR94913]
Uros Bizjak [Mon, 11 May 2020 09:16:31 +0000 (11:16 +0200)]
i386: Vectorize basic V2SFmode operations [PR94913]

Enable V2SFmode vectorization and vectorize V2SFmode PLUS,
MINUS, MULT, MIN and MAX operations using XMM registers.

To avoid unwanted secondary effects (e.g. exceptions), load values
to XMM registers using MOVQ that clears high bits of the XMM
register outside V2SFmode.

The compiler now vectorizes e.g.:

float r[2], a[2], b[2];

void
test_plus (void)
{
  for (int i = 0; i < 2; i++)
    r[i] = a[i] + b[i];
}

to:
        movq    a(%rip), %xmm0
        movq    b(%rip), %xmm1
        addps   %xmm1, %xmm0
        movlps  %xmm0, r(%rip)
        ret

gcc/ChangeLog:

PR target/95046
* config/i386/i386.c (ix86_vector_mode_supported_p):
Vectorize 3dNOW! vector modes for TARGET_MMX_WITH_SSE.
* config/i386/mmx.md (*mov<mode>_internal): Do not set
mode of alternative 13 to V2SF for TARGET_MMX_WITH_SSE.

(mmx_addv2sf3): Change operand predicates from
nonimmediate_operand to register_mmxmem_operand.
(addv2sf3): New expander.
(*mmx_addv2sf3): Add SSE/AVX alternatives.  Change operand
predicates from nonimmediate_operand to register_mmxmem_operand.
Enable instruction pattern for TARGET_MMX_WITH_SSE.

(mmx_subv2sf3): Change operand predicate from
nonimmediate_operand to register_mmxmem_operand.
(mmx_subrv2sf3): Ditto.
(subv2sf3): New expander.
(*mmx_subv2sf3): Add SSE/AVX alternatives.  Change operand
predicates from nonimmediate_operand to register_mmxmem_operand.
Enable instruction pattern for TARGET_MMX_WITH_SSE.

(mmx_mulv2sf3): Change operand predicates from
nonimmediate_operand to register_mmxmem_operand.
(mulv2sf3): New expander.
(*mmx_mulv2sf3): Add SSE/AVX alternatives.  Change operand
predicates from nonimmediate_operand to register_mmxmem_operand.
Enable instruction pattern for TARGET_MMX_WITH_SSE.

(mmx_<code>v2sf3): Change operand predicates from
nonimmediate_operand to register_mmxmem_operand.
(<code>v2sf3): New expander.
(*mmx_<code>v2sf3): Add SSE/AVX alternatives.  Change operand
predicates from nonimmediate_operand to register_mmxmem_operand.
Enable instruction pattern for TARGET_MMX_WITH_SSE.
(mmx_ieee_<ieee_maxmin>v2sf3): Ditto.

testsuite/ChangeLog:

PR target/95046
* gcc.target/i386/pr95046-1.c: New test.

4 years agoFortran : Spurious warning message with -Wsurprising PR59107
Mark Eggleston [Thu, 23 Apr 2020 09:33:14 +0000 (10:33 +0100)]
Fortran : Spurious warning message with -Wsurprising PR59107

This change is from a patch developed for gcc-5.  The code
has moved on since then requiring a change to interface.c

2020-05-11  Janus Weil  <janus@gcc.gnu.org>
    Dominique d'Humieres  <dominiq@lps.ens.fr>

gcc/fortran/

PR fortran/59107
* gfortran.h: Rename field resolved as resolve_symbol_called
and assign two 2 bits instead of 1.
* interface.c (check_dtio_interface1): Use new field name.
(gfc_find_typebound_dtio_proc): Use new field name.
* resolve.c (gfc_resolve_intrinsic): Replace check of the formal
field with resolve_symbol_called is at least 2, if it is not
set the field to 2.  (resolve_typebound_procedure): Use new field
name.  (resolve_symbol): Use new field name and check whether it
is at least 1, if it is not set the field to 1.

2020-05-11  Mark Eggleston  <markeggleston@gcc.gnu.org>

gcc/testsuite/

PR fortran/59107
* gfortran.dg/pr59107.f90: New test.

4 years agoFix typo in fprofile-prefix-path.
Martin Liska [Mon, 11 May 2020 07:34:22 +0000 (09:34 +0200)]
Fix typo in fprofile-prefix-path.

PR c/95040
* common.opt: Fix typo in option description.

4 years agoAdd caveat about parsing of .gcda and .gcno files.
Martin Liska [Mon, 11 May 2020 07:25:46 +0000 (09:25 +0200)]
Add caveat about parsing of .gcda and .gcno files.

PR gcov-profile/94928
* gcov-io.h: Add caveat about coverage format parsing and
possible outdated documentation.

4 years agoAdd handling of MULT_EXPR/PLUS_EXPR for wrapping overflow in affine combination(PR83403)
Xionghu Luo [Mon, 11 May 2020 02:06:20 +0000 (21:06 -0500)]
Add handling of MULT_EXPR/PLUS_EXPR for wrapping overflow in affine combination(PR83403)

Use determine_value_range to get value range info for fold convert expressions
with internal operation PLUS_EXPR/MINUS_EXPR/MULT_EXPR when not overflow on
wrapping overflow inner type.  i.e.:

(long unsigned int)((unsigned int)n * 10 + 1)
=>
(long unsigned int)n * (long unsigned int)10 + (long unsigned int)1

With this patch for affine combination, load/store motion could detect
more address refs independency and promote some memory expressions to
registers within loop.

PS: Replace the previous "(T1)(X + CST) as (T1)X - (T1)(-CST))"
to "(T1)(X + CST) as (T1)X + (T1)(CST))" for wrapping overflow.

Bootstrap and regression tested pass on Power8-LE.

gcc/ChangeLog

2020-05-11  Xiong Hu Luo  <luoxhu@linux.ibm.com>

PR tree-optimization/83403
* tree-affine.c (expr_to_aff_combination): Replace SSA_NAME with
determine_value_range, Add fold conversion of MULT_EXPR, fix the
previous PLUS_EXPR.

gcc/testsuite/ChangeLog

2020-05-11  Xiong Hu Luo  <luoxhu@linux.ibm.com>

PR tree-optimization/83403
* gcc.dg/tree-ssa/pr83403-1.c: New test.
* gcc.dg/tree-ssa/pr83403-2.c: New test.
* gcc.dg/tree-ssa/pr83403.h: New header.

4 years agoDaily bump.
GCC Administrator [Mon, 11 May 2020 00:16:19 +0000 (00:16 +0000)]
Daily bump.

4 years agoi386: Define __ILP32__ and _ILP32 for all 32-bit targets
Gerald Pfeifer [Thu, 7 May 2020 22:48:52 +0000 (00:48 +0200)]
i386: Define __ILP32__ and _ILP32 for all 32-bit targets

* config/i386/i386-c.c (ix86_target_macros): Define _ILP32 and
__ILP32__ for 32-bit targets.

4 years agolibstdc++ Enhance thread safety of debug mode iterators
François Dumont [Sat, 29 Feb 2020 17:22:55 +0000 (18:22 +0100)]
libstdc++ Enhance thread safety of debug mode iterators

Avoids race condition when checking for an iterator to be singular or
to be comparable to another iterator.

* src/c++/debug.cc
(_Safe_sequence_base::_M_attach_single): Set attached iterator
sequence pointer and version.
(_Safe_sequence_base::_M_detach_single): Reset detached iterator.
(_Safe_iterator_base::_M_attach): Remove attached iterator sequence
pointer and version asignments.
(_Safe_iterator_base::_M_attach_single): Likewise.
(_Safe_iterator_base::_M_detach_single): Remove detached iterator
reset.
(_Safe_iterator_base::_M_singular): Use atomic load to access parent
sequence.
(_Safe_iterator_base::_M_can_compare): Likewise.
(_Safe_iterator_base::_M_get_mutex): Likewise.
(_Safe_local_iterator_base::_M_attach): Remove attached iterator container
pointer and version assignments.
(_Safe_local_iterator_base::_M_attach_single): Likewise.
(_Safe_unordered_container_base::_M_attach_local_single):
Set attached iterator container pointer and version.
(_Safe_unordered_container_base::_M_detach_local_single): Reset detached
iterator.

4 years agoPR fortran/93499 - ICE on division by zero in declaration statements
Harald Anlauf [Sun, 10 May 2020 17:46:06 +0000 (19:46 +0200)]
PR fortran/93499 - ICE on division by zero in declaration statements

Division by zero in declaration statements could sometimes
generate NULL pointers being passed around that lead to ICEs.

2020-05-10  Harald Anlauf  <anlauf@gmx.de>

gcc/fortran/
PR fortran/93499
* arith.c (gfc_divide): Catch division by zero.
(eval_intrinsic_f3): Safeguard for NULL operands.

gcc/testsuite/
PR fortran/93499
* gfortran.dg/pr93499.f90: New test.

4 years agolibbacktrace: don't crash if ELF file has no sections
Ian Lance Taylor [Sun, 10 May 2020 03:34:25 +0000 (20:34 -0700)]
libbacktrace: don't crash if ELF file has no sections

libbacktrace/
* elf.c (elf_add): Bail early if there are no section headers at all.

4 years agoDaily bump.
GCC Administrator [Sun, 10 May 2020 00:16:19 +0000 (00:16 +0000)]
Daily bump.

4 years agolibbacktrace: don't free ELF strtab if error occurs after saving syminfo
Ian Lance Taylor [Wed, 19 Feb 2020 18:30:51 +0000 (10:30 -0800)]
libbacktrace: don't free ELF strtab if error occurs after saving syminfo

* elf.c (elf_add): Don't free strtab if an error occurs after
recording symbol information.

4 years agolibbacktrace: add Mach-O support
Ian Lance Taylor [Sun, 16 Feb 2020 13:20:01 +0000 (05:20 -0800)]
libbacktrace: add Mach-O support

libbacktrace/
PR libbacktrace/88745
* macho.c: New file.
* filetype.awk: Recognize Mach-O files.
* Makefile.am (FORMAT_FILES): Add macho.c.
(check_DATA): New variable.  Set to .dSYM if HAVE_DSYMUTIL.
(%.dSYM): New pattern target.
(test_macho_SOURCES, test_macho_CFLAGS): New targets.
(test_macho_LDADD): New target.
(BUILDTESTS): Add test_macho.
(macho.lo): Add dependencies.
* configure.ac: Recognize macho file type.  Check for
mach-o/dyld.h.  Don't try to run objcopy if we don't find it.
Look for dsymutil and define a HAVE_DSYMUTIL conditional.
* Makefile.in: Regenerate.
* configure: Regenerate.
* config.h.in: Regenerate.

4 years agolibbacktrace: support short read
Ian Lance Taylor [Sun, 16 Feb 2020 02:13:28 +0000 (18:13 -0800)]
libbacktrace: support short read

* read.c (backtrace_get_view): Support short read.

4 years agolibbacktrace: sometimes read debug sections individually
Ian Lance Taylor [Sun, 16 Feb 2020 01:56:35 +0000 (17:56 -0800)]
libbacktrace: sometimes read debug sections individually

libbacktrace/
* elf.c (elf_add): If debug sections are very large or far apart,
read them individually rather than as a single view.

4 years agolibbacktrace: support fetching executable name using sysctl
Ian Lance Taylor [Sat, 15 Feb 2020 23:29:02 +0000 (15:29 -0800)]
libbacktrace: support fetching executable name using sysctl

This supports FreeBSD and NetBSD when /proc is not mounted.

libbacktrace/
* fileline.c (sysctl_exec_name): New static function.
(sysctl_exec_name1): New macro or static function.
(sysctl_exec_name2): Likewise.
(fileline_initialize): Try sysctl_exec_name[12].
* configure.ac: Check for sysctl args to fetch executable name.
* configure: Regenerate.
* config.h.in: Regenerate.

4 years agoUpdate copyright year
Eric Botcazou [Sat, 9 May 2020 21:17:39 +0000 (23:17 +0200)]
Update copyright year

4 years agoAdd assertion for access attributes
Eric Botcazou [Sat, 9 May 2020 21:08:18 +0000 (23:08 +0200)]
Add assertion for access attributes

* gcc-interface/trans.c (Attribute_to_gnu) <Attr_Access>: Assert
that the prefix is not a type.

4 years agoFix small issues with -fgnat-encodings=minimal
Eric Botcazou [Sat, 9 May 2020 21:04:38 +0000 (23:04 +0200)]
Fix small issues with -fgnat-encodings=minimal

This is the mode where the GNAT compiler does not use special encodings
in the debug info to describe some Ada constructs, for example packed
array types.

* gcc-interface/ada-tree.h (TYPE_PACKED_ARRAY_TYPE_P): Rename into...
(TYPE_BIT_PACKED_ARRAY_TYPE_P): ...this.
(TYPE_IS_PACKED_ARRAY_TYPE_P): Rename into...
(BIT_PACKED_ARRAY_TYPE_P): ...this.
(TYPE_IMPL_PACKED_ARRAY_P): Adjust to above renaming.
* gcc-interface/gigi.h (maybe_pad_type): Remove IS_USER_TYPE..
* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Variable>: Adjust
call to maybe_pad_type.
<E_Ordinary_Fixed_Point_Type>: Remove const qualifiers for tree.
<E_Signed_Integer_Subtype>: Remove redundant test and redundant call
to associate_original_type_to_packed_array.  Turn into assertion.
Call associate_original_type_to_packed_array and modify
gnu_entity_name accordingly.  Explicitly set the parallel type
for GNAT encodings.
Call create_type_decl in the misaligned case before maybe_pad_type.
<E_Array_Type>: Do not use the name of the implementation type for
a packed array when not using GNAT encodings.
<E_Array_Subtype>: Move around setting flags. Use the result of the
call to associate_original_type_to_packed_array for gnu_entity_name.
<E_Record_Subtype>: Create XVS type and XVZ variable only if debug
info is requested for the type.
Call create_type_decl if a padded type was created for a type entity
(gnat_to_gnu_component_type): Use local variable and adjust calls to
maybe_pad_type.
(gnat_to_gnu_subprog_type): Adjust call to maybe_pad_type.
(gnat_to_gnu_field): Likewise.
(validate_size): Adjust to renaming of macro.
(set_rm_size): Likewise.
(associate_original_type_to_packed_array): Adjust return type and
return the name of the original type if GNAT encodings are not used
* gcc-interface/misc.c (gnat_get_debug_typ): Remove obsolete stuff.
(gnat_get_fixed_point_type_info): Remove const qualifiers for tree.
(gnat_get_array_descr_info): Likewise and set variables lazily.
Remove call to maybe_debug_type.  Simplify a few computations.
(enumerate_modes): Remove const qualifier for tree.
* gcc-interface/utils.c (make_type_from_size): Adjust to renaming.
(maybe_pad_type): Remove IS_USER_TYPE parameter and adjust.  Remove
specific code for implementation types for packed arrays.
(compute_deferred_decl_context): Remove const qualifier for tree.
(convert): Adjust call to maybe_pad_type.
(unchecked_convert): Likewise.
* gcc-interface/utils2.c (is_simple_additive_expressio): Likewise.

4 years agoFix tree sharing issue with slices
Eric Botcazou [Sat, 9 May 2020 20:56:14 +0000 (22:56 +0200)]
Fix tree sharing issue with slices

This can happen because we build an array type on the fly in case there
is an apparent type inconsistency in the construct.

* gcc-interface/utils2.c (build_binary_op) <ARRAY_RANGE_REF>: Use
build_nonshared_array_type to build the common type and declare it.

4 years agoDo not override -fnon-call-exceptions in default mode
Eric Botcazou [Sat, 9 May 2020 20:52:21 +0000 (22:52 +0200)]
Do not override -fnon-call-exceptions in default mode

This was already the case in -gnatp mode.

* gcc-interface/misc.c (gnat_init_gcc_eh): Do not override the user
for -fnon-call-exceptions in default mode.

4 years agoDo not make a local copy of large aggregate
Eric Botcazou [Sat, 9 May 2020 20:44:39 +0000 (22:44 +0200)]
Do not make a local copy of large aggregate

This prevents gigi from making a local copy of large aggregates.

* gcc-interface/trans.c (lvalue_required_p) <N_Selected_Component>:
Merge with N_Slice.
<N_Allocator>: Move to...
(lvalue_for_aggregate_p): ...here.  New function.
(Identifier_to_gnu): For an identifier with aggregate type, also
call lvalue_for_aggregate_p if lvalue_required_p returned false
before substituting the identifier with the constant.