rsandifo [Tue, 18 Jun 2019 09:18:17 +0000 (09:18 +0000)]
Restore correct iv step for fully-masked loops
r272233 introduced a large number of execution failures on SVE.
The patch hard-coded an IV step of VF, but for SLP groups it needs
to be VF * group size.
Also, iv_precision had type widest_int but only needs to be unsigned int.
2019-06-18 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-loop-manip.c (vect_set_loop_masks_directly): Remove
vf parameter. Restore the previous iv step of nscalars_step,
but give it iv_type rather than compare_type. Tweak code order
to match the comments.
(vect_set_loop_condition_masked): Update accordingly.
* tree-vect-loop.c (vect_verify_full_masking): Use "unsigned int"
for iv_precision. Tweak comment formatting.
alejandro [Tue, 18 Jun 2019 08:09:00 +0000 (08:09 +0000)]
[Vectorizer] Support masking fold left reductions
This patch adds support in the vectorizer for masking fold left reductions.
This avoids the need to insert a conditional assignement with some identity
value.
iains [Tue, 18 Jun 2019 08:01:00 +0000 (08:01 +0000)]
libsanitizer - remove reference to __cxa_rethrow_primary_exception.
For some Darwin versions the absence of the rethrow_primary_exception
symbol causes almost all sanitizer tests to fail.
The symbol is not present in libstdc++ and, therefore is is correct to
remove the reference to it for all platforms. We do this by adding a
new guard "ASAN_HAS_CXA_RETHROW_PRIMARY_EXCEPTION".
linkw [Tue, 18 Jun 2019 05:08:02 +0000 (05:08 +0000)]
Add one target hook predict_doloop_p, it return true if we can predict it
is possible to use a low-overhead loop, it can help ivopts to make some
better decisions.
PR middle-end/80791
* target.def (predict_doloop_p): New hook.
* targhooks.h (default_predict_doloop_p): New declaration.
* targhooks.c (default_predict_doloop_p): New function.
* doc/tm.texi.in (TARGET_PREDICT_DOLOOP_P): New hook.
* doc/tm.texi: Regenerate.
* config/rs6000/rs6000.c (rs6000_predict_doloop_p): New function.
(TARGET_PREDICT_DOLOOP_P): New macro.
* tree-ssa-loop-ivopts.c (generic_predict_doloop_p): New function.
jakub [Mon, 17 Jun 2019 21:20:00 +0000 (21:20 +0000)]
* omp-low.c (struct omp_context): Add scan_inclusive field.
(scan_omp_1_stmt) <case GIMPLE_OMP_SCAN>: Set ctx->scan_inclusive
if inclusive scan.
(struct omplow_simd_context): Add lastlane member.
(lower_rec_simd_input_clauses): Add rvar argument, handle inscan
reductions. Build 2 or 3 argument .GOMP_SIMD_LANE calls rather than
1 or 2 argument.
(lower_rec_input_clauses): Handle inscan reductions in simd contexts.
(lower_lastprivate_clauses): Set TREE_THIS_NOTRAP on the ARRAY_REF.
(lower_omp_scan): New function.
(lower_omp_1) <case GIMPLE_OMP_SCAN>: Use lower_omp_scan.
* tree-ssa-dce.c (eliminate_unnecessary_stmts): For IFN_GOMP_SIMD_LANE
check 3rd argument if present rather than 2nd.
* tree-vectorizer.h (struct _loop_vec_info): Add scan_map member.
(struct _stmt_vec_info): Change simd_lane_access_p from bool into
2-bit bitfield.
* tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
scan_map. For IFN_GOMP_SIMD_LANE check 3rd argument if present rather
than 2nd.
(_loop_vec_info::~_loop_vec_info): Delete scan_map.
* tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Allow two
different STMT_VINFO_SIMD_LANE_ACCESS_P refs if they have the same
init.
(vect_find_stmt_data_reference): Encode in ->aux the 2nd
IFN_GOMP_SIMD_LANE argument.
(vect_analyze_data_refs): Set STMT_VINFO_SIMD_LANE_ACCESS_P from the
encoded ->aux value.
* tree-vect-stmts.c: Include attribs.h.
(vectorizable_call): Adjust comment about IFN_GOMP_SIMD_LANE.
(scan_operand_equal_p, check_scan_store, vectorizable_scan_store): New
functions.
(vectorizable_load): For STMT_VINFO_SIMD_LANE_ACCESS_P tests use != 0.
(vectorizable_store): Handle STMT_VINFO_SIMD_LANE_ACCESS_P > 1.
cp/
* semantics.c (finish_omp_clauses): For OMP_CLAUSE_REDUCTION_INSCAN
set need_copy_assignment.
testsuite/
* gcc.dg/vect/vect-simd-8.c: New test.
* gcc.dg/vect/vect-simd-9.c: New test.
* g++.dg/vect/simd-2.cc: New test.
* g++.dg/gomp/scan-1.C: New test.
redi [Mon, 17 Jun 2019 15:51:31 +0000 (15:51 +0000)]
Fix AIX test failure due to replacement operator delete
On AIX the sized delete defined in the library will call the non-sized
delete defined in the library, not the replacement version defined in
the test file. By also replacing sized delete we make the test pass
everywhere.
* testsuite/20_util/allocator/1.cc: Add sized delete, which fixes a
failure on AIX.
Build and reg-tested on x86_64 with nvptx accelerator.
2019-06-17 Tom de Vries <tdevries@suse.de>
* config/nvptx/nvptx-protos.h (gen_set_softstack_insn): Declare.
* config/nvptx/nvptx.c (gen_set_softstack_insn): New function.
* config/nvptx/nvptx.md (define_insn "set_softstack_insn"): Rename to
...
(define_insn "set_softstack_<mode>"): ... this. Use P iterator on
match_operand 0.
(define_insn "omp_simt_enter_insn"): Rename to ...
(define_insn "omp_simt_enter_<mode>"): ... this. Use P iterator on
match_operand 0, 1 and 2, as well as the unspec_volatile result.
(define_expand "omp_simt_enter): Use gen_omp_simt_enter_di and
gen_omp_simt_enter_si.
(define_expand "omp_simt_exit"): New.
(define_insn "omp_simt_exit"): Rename to ...
(define_insn "omp_simt_exit_<mode>"): ... this. Use P iterator on
match_operand 0.
redi [Mon, 17 Jun 2019 14:19:04 +0000 (14:19 +0000)]
PR libstdc++/90281 Fix string conversions for filesystem::path
Fix several bugs in the encoding conversions for filesystem::path that
prevent conversion of Unicode characters outside the Basic Multilingual
Plane, and prevent returning basic_string specializations with
alternative allocator types.
The std::codecvt_utf8 class template is not suitable for UTF-16
conversions because it uses UCS-2 instead. For conversions between UTF-8
and UTF-16 either std::codecvt<C, char, mbstate> or
codecvt_utf8_utf16<C> must be used.
The __str_codecvt_in and __str_codecvt_out utilities do not
return false on a partial conversion (e.g. for invalid or incomplete
Unicode input). Add new helpers that treat partial conversions as
errors, and use them for all filesystem::path conversions.
PR libstdc++/90281 Fix string conversions for filesystem::path
* include/bits/fs_path.h (u8path) [_GLIBCXX_FILESYSTEM_IS_WINDOWS]:
Use codecvt_utf8_utf16 instead of codecvt_utf8. Use
__str_codecvt_in_all to fail for partial conversions and throw on
error.
[!_GLIBCXX_FILESYSTEM_IS_WINDOWS && _GLIBCXX_USE_CHAR8_T]
(path::_Cvt<char8_t>): Add explicit specialization.
[_GLIBCXX_FILESYSTEM_IS_WINDOWS] (path::_Cvt::_S_wconvert): Remove
overloads.
[_GLIBCXX_FILESYSTEM_IS_WINDOWS] (path::_Cvt::_S_convert): Use
if-constexpr instead of dispatching to _S_wconvert. Use codecvt
instead of codecvt_utf8. Use __str_codecvt_in_all and
__str_codecvt_out_all.
[!_GLIBCXX_FILESYSTEM_IS_WINDOWS] (path::_Cvt::_S_convert): Use
codecvt instead of codecvt_utf8. Use __str_codecvt_out_all.
(path::_S_str_convert) [_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Use
codecvt_utf8_utf16 instead of codecvt_utf8. Construct return values
with allocator. Use __str_codecvt_out_all. Fallthrough to POSIX code
after converting to UTF-8.
(path::_S_str_convert): Use codecvt instead of codecvt_utf8. Use
__str_codecvt_in_all.
(path::string): Fix initialization of string types with different
allocators.
(path::u8string) [_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Use
codecvt_utf8_utf16 instead of codecvt_utf8. Use __str_codecvt_out_all.
* include/bits/locale_conv.h (__do_str_codecvt): Reorder static and
runtime conditions.
(__str_codecvt_out_all, __str_codecvt_in_all): New functions that
return false for partial conversions.
* include/experimental/bits/fs_path.h (u8path):
[_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Implement correctly for mingw.
[_GLIBCXX_FILESYSTEM_IS_WINDOWS] (path::_Cvt::_S_wconvert): Add
missing handling for char8_t. Use codecvt and codecvt_utf8_utf16
instead of codecvt_utf8. Use __str_codecvt_in_all and
__str_codecvt_out_all.
[!_GLIBCXX_FILESYSTEM_IS_WINDOWS] (path::_Cvt::_S_convert): Use
codecvt instead of codecvt_utf8. Use __str_codecvt_out_all.
(path::string) [_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Use
codecvt_utf8_utf16 instead of codecvt_utf8. Construct return values
with allocator. Use __str_codecvt_out_all and __str_codecvt_in_all.
(path::string) [!_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Use
__str_codecvt_in_all.
(path::u8string) [_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Use
codecvt_utf8_utf16 instead of codecvt_utf8. Use __str_codecvt_out_all.
* src/c++17/fs_path.cc (path::_S_convert_loc): Use
__str_codecvt_in_all.
* src/filesystem/path.cc (path::_S_convert_loc): Likewise.
* testsuite/27_io/filesystem/path/construct/90281.cc: New test.
* testsuite/27_io/filesystem/path/factory/u8path.cc: New test.
* testsuite/27_io/filesystem/path/native/string.cc: Test with empty
strings and with Unicode characters outside the basic multilingual
plane.
* testsuite/27_io/filesystem/path/native/alloc.cc: New test.
* testsuite/experimental/filesystem/path/construct/90281.cc: New test.
* testsuite/experimental/filesystem/path/factory/u8path.cc: New test.
* testsuite/experimental/filesystem/path/native/alloc.cc: New test.
* testsuite/experimental/filesystem/path/native/string.cc: Test with
empty strings and with Unicode characters outside the basic
multilingual plane.
redi [Mon, 17 Jun 2019 10:25:04 +0000 (10:25 +0000)]
Simplify node ownership in _Hashtable members
Introduce an RAII type to manage nodes in unordered containers while
they are being inserted. If the caller always owns a node until it is
inserted, then the insertion functions don't need to deallocate on
failure. This allows a FIXME in the node re-insertion API to be removed.
Also change extract(const key_type&) to not call extract(const_iterator)
anymore. This avoids looping through the bucket nodes again to find the
node before the one being extracted.
2019-06-17 François Dumont <fdumont@gcc.gnu.org>
Jonathan Wakely <jwakely@redhat.com>
* include/bits/hashtable.h (struct _Hashtable::_Scoped_node): New type.
(_Hashtable::_M_insert_unique_node): Add key_type parameter. Don't
deallocate node if insertion fails.
(_Hashtable::_M_insert_multi_node): Likewise.
(_Hashtable::_M_reinsert_node): Pass additional key argument.
(_Hashtable::_M_reinsert_node_multi): Likewise. Remove FIXME.
(_Hashtable::_M_extract_node(size_t, __node_base*)): New function.
(_Hashtable::extract(const_iterator)): Use _M_extract_node.
(_Hashtable::extract(const _Key&)): Likewise.
(_Hashtable::_M_merge_unique): Pass additional key argument.
(_Hashtable::_M_emplace<Args>(true_type, Args&&...)): Likewise. Use
_Scoped_node.
(_Hashtable::_M_insert): Likewise.
* include/bits/hashtable_policy.h (_Map_base::operator[]): Likewise.
(_Hashtable_alloc): Add comments to functions with misleading names.
redi [Mon, 17 Jun 2019 08:18:17 +0000 (08:18 +0000)]
Fix tests that fail without PCH
The recent change to stop transitively including <string> broke some
tests, but only when the library is configured without PCH, because
otherwise the <string> header still gets included via the precompiled
<bits/stdc++.h> header.
* testsuite/20_util/bad_function_call/what.cc: Include <string> header
for std::string.
* testsuite/20_util/shared_ptr/cons/weak_ptr_expired.cc: Likewise.
* testsuite/20_util/tuple/cons/allocator_with_any.cc: Include <memory>
header for std::allocator.
* testsuite/23_containers/array/tuple_interface/tuple_element.cc: Add
using-declaration for std::size_t.
* testsuite/23_containers/array/tuple_interface/tuple_size.cc:
Likewise.
* testsuite/23_containers/deque/cons/55977.cc: Include <istream> for
std::istream.
* testsuite/23_containers/vector/cons/55977.cc: Likewise.
* testsuite/experimental/map/erasure.cc: Include <string> for
std::string.
* testsuite/experimental/unordered_map/erasure.cc: Likewise.
Build and reg-tested on x86_64 with nvptx accelerator.
2019-06-17 Tom de Vries <tdevries@suse.de>
* config/nvptx/nvptx.md (define_insn "call_insn"): Rename to ...
(define_insn "call_insn_<mode>"): ... this. Use P iterator on
match_operand 0.
(define_insn "call_value_insn"): Rename to ...
(define_insn "call_value_insn_<mode>"): this. Use P iterator on
match_operand 0.
(define_insn "nvptx_red_partition"): Set unspec_volatile result mode to
DI.
jozefl [Sun, 16 Jun 2019 21:24:56 +0000 (21:24 +0000)]
MSP430: Implement 64-bit shifts in assembly code
gcc/ChangeLog:
2019-06-16 Jozef Lawrynowicz <jozef.l@mittosystems.com>
* config/msp430/msp430.c (msp430_expand_helper): Setup arguments which
describe how to perform MSPABI compliant 64-bit shift.
* config/msp430/msp430.md (ashldi3): New define_expand.
(ashrdi3): New define_expand.
(lshrdi3): New define_expand.
libgcc/ChangeLog:
2019-06-16 Jozef Lawrynowicz <jozef.l@mittosystems.com>
* config/msp430/slli.S (__mspabi_sllll): New library function for
performing a logical left shift of a 64-bit value.
* config/msp430/srai.S (__mspabi_srall): New library function for
performing a arithmetic right shift of a 64-bit value.
* config/msp430/srll.S (__mspabi_srlll): New library function for
performing a logical right shift of a 64-bit value.
gcc/testsuite/ChangeLog:
2019-06-16 Jozef Lawrynowicz <jozef.l@mittosystems.com>
* gcc.target/msp430/mspabi_sllll.c: New test.
* gcc.target/msp430/mspabi_srall.c: New test.
* gcc.target/msp430/mspabi_srlll.c: New test.
* gcc.c-torture/execute/shiftdi-2.c: New test.
iains [Sun, 16 Jun 2019 16:21:28 +0000 (16:21 +0000)]
[Darwin] The need for picsym stubs is dependent on linker used, not OS rev.
For very old toolchains, the compiler generated pic symbol stubs that
provide the necessary indirections. We no longer need this (since xcode
3 era) and it's more efficient for the linker to make one stub when it
knows a symbol is needed that for us to emit them speculatively in every
object.
Our current codegen is making the assumption that a specific OS version
uses a specific linker version - and therefore the presence of support
could be based on the target OS rev. Of course, that's way too simplistic
(most likely bogus for cross-toolchains) and we want to make things explict.
Since we have detection of the linker version, we can use that directly
(A config test for support for stub-less linking might also be feasible
but much more involved).
Finally, should the user wish to generate code that caters for export to
use in an environment with an older toolchain, the generation of stubs
can be forced from the command line.
In addition to the points above, branch islanding and a long branch opt
for PowerPC Darwin has become conflated with the emission of these stubs
(which is confusing, and will be rectified in a following patch).
2019-06-16 Iain Sandoe <iain@sandoe.co.uk>
* config/darwin.c (machopic_indirect_call_target): Use renamed
darwin_picsymbol_stubs to decide on output.
(darwin_override_options): Handle darwin_picsymbol_stubs.
* config/darwin.h (MIN_LD64_OMIT_STUBS): New.
(LD64_VERSION): Revise default.
* config/darwin.opt: (mpic-symbol-stubs): New option.
(darwin_picsymbol_stubs): New variable.
* config/i386/darwin.h (TARGET_MACHO_BRANCH_ISLANDS):
rename to TARGET_MACHO_PICSYM_STUBS.
* config/i386/i386.c (output_pic_addr_const): Likewise.
* config/i386/i386.h Likewise.
* config/rs6000/darwin.h: Likewise.
* config/rs6000/rs6000.c (rs6000_call_darwin_1): Use renamed
darwin_picsymbol_stubs.
vries [Sun, 16 Jun 2019 07:47:15 +0000 (07:47 +0000)]
[openacc, parloops] Fix SIGSEGV in oacc_entry_exit_ok_1
When compiling the test-case with r268755, we run into a SIGSEGV in
oacc_entry_exit_ok_1 when trying to dereference a NULL red:
...
struct reduction_info *red;
red = reduction_phi (reduction_list, use_stmt);
tree val = PHI_RESULT (red->keep_res);
...
Fix this by handling ref == NULL.
Bootstrapped and reg-tested on x86_64.
Build and reg-tested on x86_64 with nvptx accelerator.
2019-06-16 Tom de Vries <tdevries@suse.de>
PR tree-optimization/89376
* tree-parloops.c (oacc_entry_exit_ok_1): Handle red == NULL.
* testsuite/libgomp.oacc-c-c++-common/pr89376.c: New test.
iains [Sat, 15 Jun 2019 14:03:44 +0000 (14:03 +0000)]
[objective-c/c++, testsuite, 3/3] Workaround for PR90709.
Use the CF-CFString.h, F-NS*.h proxy headers where needed in the objective-c++
testsuite. Make minor adjustments to tests as required, providing that those
do not alter the test intent.
2019-06-15 Iain Sandoe <iain@sandoe.co.uk>
PR objc/90709
* obj-c++.dg/proto-lossage-7.mm: Use proxy headers.
* obj-c++.dg/strings/const-cfstring-2.mm: Likewise.
* obj-c++.dg/strings/const-cfstring-5.mm: Likewise
* obj-c++.dg/strings/const-str-12.mm: Likewise.
* obj-c++.dg/syntax-error-1.mm: Likewise.
* obj-c++.dg/torture/strings/const-cfstring-1.mm: Likewise.
* obj-c++.dg/torture/strings/const-str-10.mm: Likewise.
* obj-c++.dg/torture/strings/const-str-11.mm: Likewise.
* obj-c++.dg/torture/strings/const-str-9.mm: Likewise.
* obj-c++.dg/cxx-ivars-3.mm: Skip on later Darwin, where the 10.4 API
in no longer supported, also on m64 where there's no meaning to it.
* obj-c++.dg/isa-field-1.mm: Suppress unwanted warning, add comment why.
* obj-c++.dg/objc-gc-3.mm: Skip for Darwin > 16, the API use is an error
there.
* obj-c++.dg/qual-types-1.mm: Prune a spurious l64 warning.
* obj-c++.dg/stubify-1.mm: Tidy up after better compiler warnings.
* obj-c++.dg/stubify-2.mm: Likewise.
* obj-c++.dg/try-catch-1.mm: Likewise.
* obj-c++.dg/try-catch-3.mm: Likewise.
iains [Sat, 15 Jun 2019 14:00:24 +0000 (14:00 +0000)]
[objective-c/c++, testsuite, 2/3] Workaround for PR90709.
Use the CF-CFString.h, F-NS*.h proxy headers where needed in the objective-c
testsuite. Make minor adjustments to tests as required, providing that those
do not alter the test intent.
2019-06-15 Iain Sandoe <iain@sandoe.co.uk>
PR objc/90709
* objc.dg/encode-7-next-64bit.m: Use proxy headers.
* objc.dg/image-info.m: Likewise.
* objc.dg/method-6.m: Likewise.
* objc.dg/no-extra-load.m: Likewise.
* objc.dg/objc-foreach-4.m: Likewise.
* objc.dg/objc-foreach-5.m: Likewise.
* objc.dg/proto-lossage-7.m: Likewise.
* objc.dg/strings/const-cfstring-2.m: Likewise.
* objc.dg/strings/const-cfstring-5.m: Likewise.
* objc.dg/strings/const-str-12b.m: Likewise.
* objc.dg/symtab-1.m: Likewise.
* objc.dg/torture/strings/const-cfstring-1.m: Likewise.
* objc.dg/torture/strings/const-str-10.m: Likewise.
* objc.dg/torture/strings/const-str-11.m: Likewise.
* objc.dg/torture/strings/const-str-9.m: Likewise.
* objc.dg/zero-link-1.m: Likewise.
* objc.dg/zero-link-2.m: Likewise.
* objc.dg/zero-link-3.m: Likewise.
* objc.dg/isa-field-1.m: Suppress unwanted warning, add comment why.
* objc.dg/headers.m: XFAIL for Darwin14-19.
* objc.dg/objc-gc-4.m: Skip for Darwin > 16, the API use is an error
there.
iains [Sat, 15 Jun 2019 13:54:36 +0000 (13:54 +0000)]
[objective-c/c++, testsuite, 1/3] Workaround for PR90709.
Since we cannot parse the current NeXT headers, because of PR90709 and its
dependents, we have a large amount of testsuite noise for Darwin platforms.
In order to restore the usefulness of the testsuite, we are going add headers
without the modern syntax elements that trigger the bug, and use these for
test runs on newer Darwin.
The headers are imported from GNUStep, with some local modifications to make
sure that __BLOCKS__ is honoured as a gate for Apple-style blocks closures.
CF-CFString.h, F-NS*.h are proxy headers that use the installed CoreFoundation
or Foundation headers on systems <= Darwin12 and the GNUStep headers for newer.
jakub [Sat, 15 Jun 2019 07:09:04 +0000 (07:09 +0000)]
PR middle-end/90779
* gimplify.c: Include omp-offload.h and context.h.
(gimplify_bind_expr): Add "omp declare target" attributes
to static block scope variables inside of target region or target
functions.
* c-c++-common/goacc/routine-5.c (func2): Don't expect error for
static block scope variable in #pragma acc routine.
* testsuite/libgomp.c/pr90779.c: New test.
* testsuite/libgomp.fortran/pr90779.f90: New test.
vries [Sat, 15 Jun 2019 07:06:19 +0000 (07:06 +0000)]
[openacc] Disable pass_thread_jumps for IFN_UNIQUE
If we compile the openacc testcase with -fopenacc -O2, we run into a SIGSEGV
or assert. The root cause for this is that pass_thread_jumps breaks the
invariant that OACC_FORK and OACC_JOIN mark the start and end of a
single-entry-single-exit region.
Fix this by bailing out when encountering an IFN_UNIQUE in
thread_jumps::profitable_jump_thread_path.
Bootstrapped and reg-tested on x86_64.
Build and reg-tested libgomp on x86_64 with nvptx accelerator.
segher [Sat, 15 Jun 2019 00:25:18 +0000 (00:25 +0000)]
rs6000: New iterator CCEITHER
This expands to both CC and CCUNS, it's useful for isel for example.
* config/rs6000/rs6000.md (CCEITHER): New define_mode_iterator.
(un): New define_mode_attr.
(isel_signed_<mode>, isel_unsigned_<mode>): Delete, merge into ...
(isel_<un>signed_<GPR:mode>): ... this. New define_insn.
(isel_reversed_signed_<mode>, isel_reversed_unsigned_<mode>): Delete,
merge into ...
(isel_reversed_<un>signed_<GPR:mode>): ... this. New define_insn.
iains [Fri, 14 Jun 2019 18:31:44 +0000 (18:31 +0000)]
Darwin, opts processing - Initial tidy.
This change adds RejectNegative in a number of places it was needed,
reorders the opts into two groups (one general and one driver-only).
We also add a minimal description to each opt, and note some that
are now obsolete. The only functional change is to permit
-mtarget-linker= as an alias of -mtarget-linker.
gcc/
2019-06-14 Iain Sandoe <iain@sandoe.co.uk>
* config/darwin.opt: Add RejectNegative where needed, reorder
and add minimal functional descriptions.
hjl [Fri, 14 Jun 2019 16:24:56 +0000 (16:24 +0000)]
Update preferred_stack_boundary only when expanding function call
locate_and_pad_parm is called when expanding function call from
initialize_argument_information and when generating function body
from assign_parm_find_entry_rtl:
/* Remember if the outgoing parameter requires extra alignment on the
calling function side. */
if (crtl->stack_alignment_needed < boundary)
crtl->stack_alignment_needed = boundary;
if (crtl->preferred_stack_boundary < boundary)
crtl->preferred_stack_boundary = boundary;
stack_alignment_needed and preferred_stack_boundary should be updated
only when expanding function call, not when generating function body.
Add update_stack_alignment_for_call to update stack alignment when
outgoing parameter is passed in the stack.
gcc/
PR rtl-optimization/90765
* calls.c (update_stack_alignment_for_call): New function.
(expand_call): Call update_stack_alignment_for_call when
outgoing parameter is passed in the stack.
(emit_library_call_value_1): Likewise.
* function.c (locate_and_pad_parm): Don't update
stack_alignment_needed and preferred_stack_boundary.
gcc/testsuite/
PR rtl-optimization/90765
* gcc.target/i386/pr90765-1.c: New test.
* gcc.target/i386/pr90765-2.c: Likewise.
hjl [Fri, 14 Jun 2019 15:41:43 +0000 (15:41 +0000)]
i386: Update SSE <-> integer move costs
Since inline_secondary_memory_needed has
/* ??? This is a lie. We do have moves between mmx/general, and for
mmx/sse2. But by saying we need secondary memory we discourage the
register allocator from using the mmx registers unless needed. */
if (MMX_CLASS_P (class1) != MMX_CLASS_P (class2))
return true;
moves between MMX and non-MMX units require secondary memory. There
is no need to check moves between MMX and integer units.
struct processor_costs has:
const int mmxsse_to_integer; /* cost of moving mmxsse register to
integer. */
const int ssemmx_to_integer; /* cost of moving integer to mmxsse register. */
This patch also renames mmxsse_to_integer to sse_to_integer and
ssemmx_to_integer to integer_to_sse.
Tested on Linux/x86-64.
PR target/90877
* config/i386/i386-features.c
(dimode_scalar_chain::compute_convert_gain): Replace
mmxsse_to_integer with sse_to_integer.
* config/i386/i386.c (ix86_register_move_cost): Verify that
moves between MMX and non-MMX units require secondary memory.
Correct costs of moves between SSE and integer units.
* config/i386/i386.h (processor_costs): Rename cost of moving
SSE register to integer to sse_to_integer. Rename cost of
moving integer register to SSE to integer_to_sse.
rearnsha [Fri, 14 Jun 2019 14:04:20 +0000 (14:04 +0000)]
[NetBSD] Add support for the Arm EABI.
This is a roll-up of a set of changes needed to support the Arm EABI on NetBSD.
2019-06-14 Matt Thomas <matt@3am-software.com>
Matthew Green <mrg@eterna.com.au>
Nick Hudson <skrll@netbsd.org>
Maya Rashish <coypu@sdf.org>
Richard Earnshaw <rearnsha@arm.com>
gcc:
* config.gcc (arm*-*-netbsdelf*) Add support for EABI configuration.
* config.host (arm*-*-netbsd*): Use driver-arm.o on native NetBSD.
* config/arm/netbsd-eabi.h: New file.
* config/arm/netbsd-elf.h (TARGET_OS_CPP_BUILTINS): Undefine before
redefining.
(SUBTARGET_EXTRA_ASM_SPEC): Don't pass -matpcs to the assembler.
* config/netbsd-elf.h (NETBSD_LINK_LD_ELF_SO_SPEC): New define.
(NETBSD_SUBTARGET_EXTRA_SPECS): New define.
(SUBTARGET_EXTRA_SPECS): Define to NETBSD_SUBTARGET_EXTRA_SPECS.
libatomic:
* configure.tgt (arm*): Handle NetBSD in the same way as FreeBSD.
libgcc:
* config.host (arm*-*-netbsdelf*): Add support for EABI configurations.
* config/arm/t-netbsd (LIB1ASMFUNCS): Add some additional assembler
functions to build.
* config/arm/t-netbsd-eabi: New file.
redi [Fri, 14 Jun 2019 14:03:20 +0000 (14:03 +0000)]
Disable -Wctor-dtor-privacy warnings for some standard types
* include/experimental/type_traits (experimental::nonesuch): Use
pragma to disable -Wctor-dtor-privacy warnings.
* include/std/type_traits (__is_convertible_helper<From, To, false>)
(__is_nt_convertible_helper<From, To, false>, __nonesuch): Likewise.
jason [Fri, 14 Jun 2019 13:22:33 +0000 (13:22 +0000)]
PR c++/85552 - wrong instantiation of dtor for DMI.
The problem here is that when processing direct-initialization of a data
member, we don't need to worry about destruction semantics; that will be
handled in the [cd]tor. Conveniently, we already have tf_no_cleanup from a
similar fix to new-expressions.
* typeck2.c (digest_nsdmi_init): Set tf_no_cleanup for direct-init.
rguenth [Fri, 14 Jun 2019 11:29:44 +0000 (11:29 +0000)]
2019-06-14 Richard Biener <rguenther@suse.de>
* tree-loop-distribution.c (classify_partition): Return
whether a reduction appeared in all partitions and do not
stop builtin detection because of this.
(distribute_loop): Sort a non-builtin partition last if
there's a reduction in all partitions and make sure the
partition prevailing as last is not a builtin.
uros [Thu, 13 Jun 2019 21:29:06 +0000 (21:29 +0000)]
* config/i386/i386.md (SWIM1248s): Rename from SWIM1248x.
Update all uses.
(and<mode>3): Use gen_extend_insn instead of indirect functions.
Do not generate DImode extends for 32bit targets.
(and->zext post-reload splitter): Use gen_extend_insn
instead of indirect functions.
(anddi->zext pre-reload splitter): New.
* config/i386/i386-expand.c (ix86_expand_int_sse_cmp):
Use gen_sub3_insn instead of indirect function.
(ix86_expand_ashl_const): Use gen_add2_insn instead of
indirect function.
(ix86_adjust_counter): Ditto.
iains [Thu, 13 Jun 2019 18:53:05 +0000 (18:53 +0000)]
Darwin, Driver - Improve processing of macosx-version-min=
For PR target/63810 some improvements were made in the parsing of
the version string at the point it's used to define the built-in
__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__. This is fine, but the
specs processing also uses the version, and specs version-compare
doesn't like leading zeros on components. This means that while we
succeed in processing -mmacosx-version-min=010.00002.000099 on compile
lines, it fails for any other line that uses the value as part of a spec
(in particular, link lines fail).
To fix this, we need to apply a bit of clean-up to the version that's
presented to the driver, and push that back into the command line opts.
The value can come from four places:
1. User-entered on the command line
2. User-entered as MACOSX_DEPLOYMENT_TARGET= environment var.
3. Absent those two
3a For self-hosting systems, look-up from the kernel
3b For cross-compilers, as a default supplied at configure time.
We apply the clean-up to all 4 (although it shouldn't really be needed
for the cases under 3).
We also supply a test-case that adapts to the target-version of the
system, so that the link requirements are met by the SDK in use (if you
try to link i686-darwin9 on an x86-64-darwin18 SDK, it will fail).
gcc/
2019-06-13 Iain Sandoe <iain@sandoe.co.uk>
* config/darwin-driver.c (validate_macosx_version_min): New.
(darwin_default_min_version): Cleanup and validate supplied version.
(darwin_driver_init): Likewise and push cleaned version into opts.
kargl [Thu, 13 Jun 2019 18:40:19 +0000 (18:40 +0000)]
2019-06-13 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/68544
* resolve.c (is_dt_name): New function to compare symbol name against
list of derived types.
(resolve_actual_arglist): Use it to find wrong code.
2019-06-13 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/68544
* gfortran.dg/pr68544.f90: New test.
* gfortran.dg/pr85687.f90: Modify test for new error message.
iains [Thu, 13 Jun 2019 18:32:32 +0000 (18:32 +0000)]
Darwin, testsuite - Fix pr71694 fail for m32.
This test fails for Darwin m32 because it's scanning for absence
of an instruction that's validly used in PIC code. Fixed, in this
case, by using non-PIC codegen.
2019-06-13 Iain Sandoe <iain@sandoe.co.uk>
* g++.dg/pr71694.C: Use non-PIC codegen for Darwin m32.
iains [Thu, 13 Jun 2019 17:15:37 +0000 (17:15 +0000)]
Darwin, testsuite - Adjust two tests that fail on newer OS versions.
Newer OS versions (10.14+) do not provide some of the CRTs that are used
for older ones (e.g. 10.5), and thus link tests that specify targeting
such a revision fail. We retain the testing of the correct defined OS
version number but switch to compile-only testing for these two tests.
2019-06-13 Iain Sandoe <iain@sandoe.co.uk>
* gcc.dg/darwin-minversion-1.c: Use compile rather than link/run.
* gcc.dg/darwin-minversion-2.c: Likewise.