Andrew Stubbs [Fri, 8 Jul 2022 10:58:46 +0000 (11:58 +0100)]
openmp: fix max_vf setting for amdgcn offloading
Ensure that the "max_vf" figure used for the "safelen" attribute is large
enough for the largest configured offload device.
This change gives ~10x speed improvement on the Bablestream "dot" benchmark for
AMD GCN.
gcc/ChangeLog:
* gimple-loop-versioning.cc (loop_versioning::loop_versioning): Add
comment.
* omp-general.cc (omp_max_simd_vf): New function.
* omp-general.h (omp_max_simd_vf): New prototype.
* omp-low.cc (lower_rec_simd_input_clauses): Select largest from
omp_max_vf, omp_max_simt_vf, and omp_max_simd_vf.
gcc/testsuite/ChangeLog:
* lib/target-supports.exp
(check_effective_target_amdgcn_offloading_enabled): New.
(check_effective_target_nvptx_offloading_enabled): New.
* gcc.dg/gomp/target-vf.c: New test.
This also means finally switching Intel MIC 'mkoffload' to
'GOMP_offload_register_ver', 'GOMP_offload_unregister_ver',
making 'GOMP_offload_register', 'GOMP_offload_unregister'
legacy entry points.
gcc/
* config/i386/intelmic-mkoffload.cc (generate_host_descr_file)
(prepare_target_image, main): Handle OpenMP 'requires'.
(generate_host_descr_file): Switch to 'GOMP_offload_register_ver',
'GOMP_offload_unregister_ver'.
libgomp/
* target.c (GOMP_offload_register, GOMP_offload_unregister):
Denote as legacy entry points.
* testsuite/lib/libgomp.exp
(check_effective_target_offload_target_any): New proc.
* testsuite/libgomp.c-c++-common/requires-1.c: Enable for
'offload_target_any'.
* testsuite/libgomp.c-c++-common/requires-3.c: Likewise.
* testsuite/libgomp.c-c++-common/requires-7.c: Likewise.
* testsuite/libgomp.fortran/requires-1.f90: Likewise.
Thomas Schwinge [Thu, 7 Jul 2022 12:05:17 +0000 (14:05 +0200)]
Adjust 'libgomp.c-c++-common/requires-3.c'
As documented, this one does "Check diagnostic by device-compiler's lto1".
Indeed there are none when compiling with '-foffload=disable' with an
offloading-enabled compiler, so we should use 'offload_target_[...]', as
used in other similar test cases.
The recent commit 683f11843974f0bdf42f79cdcbb0c2b43c7b81b0
"OpenMP: Move omp requires checks to libgomp" changed the
'GOMP_offload_register_ver' interface but didn't change
'GOMP_offload_unregister_ver' accordingly, so we're no longer
actually unregistering.
Thomas Schwinge [Tue, 5 Jul 2022 14:41:03 +0000 (16:41 +0200)]
libgomp nvptx plugin: Remove '--with-cuda-driver=[...]' etc. configuration option
That means, exposing to the user only the '--without-cuda-driver' behavior:
including the GCC-shipped 'include/cuda/cuda.h' (not system <cuda.h>), and
'dlopen'ing the CUDA Driver library (not linking it).
For development purposes, the libgomp nvptx plugin developer may still manually
override that, to get the previous '--with-cuda-driver' behavior.
It also updates
* libgomp/plugin/plugin-gcn.c (GOMP_OFFLOAD_get_num_devices)
to permit GOMP_REQUIRES_UNIFIED_ADDRESS | GOMP_REQUIRES_UNIFIED_SHARED_MEMORY,
moved from the now no-longer existing function GOMP_OFFLOAD_supported_features;
the flags were added in OG12 commit 12d14a9a255c1cc10e4506935327aabd9766967d,
'amdgcn: libgomp plugin USM implementation'.
And, likewise, it updates
* libgomp/plugin/plugin-nvptx.c (GOMP_OFFLOAD_get_num_devices)
from OG12 commit f3fd38e31fafdc7f6b6a721b53bdc595c4ec1e09,
'libgomp, nvptx: report USM'.
* * *
Handle reverse_offload, unified_address, and unified_shared_memory
requirements in libgomp by saving them alongside the offload table.
When the device lto1 runs, it extracts the data for mkoffload. The
latter than passes the value on to GOMP_offload_register_ver.
lto1 (either the host one, with -flto [+ ENABLE_OFFLOADING], or in the
offload-device lto1) also does the the consistency check is done,
erroring out when the 'omp requires' clause use is inconsistent.
For all in-principle supported devices, if a requirement cannot be fulfilled,
the device is excluded from the (supported) devices list. Currently, none of
those requirements are marked as supported for any of the non-host devices.
* config/gcn/mkoffload.cc (process_asm): Write '#include <stdint.h>'.
(process_obj): Pass omp_requires_mask to GOMP_offload_register_ver.
(main): Ask lto1 to obtain omp_requires_mask and pass it on.
* config/nvptx/mkoffload.cc (process, main): Likewise.
* lto-cgraph.cc (omp_requires_to_name): New.
(input_offload_tables): Save omp_requires_mask.
(output_offload_tables): Read it, check for consistency,
save value for mkoffload.
* omp-low.cc (lower_omp_target): Force output_offloadtables
call for OMP_REQUIRES_TARGET_USED.
* gfortran.h (gfc_omp_namelist): Update by creating 'linear' struct,
move 'linear_op' as 'op' to id and add 'old_modifier' to it.
* dump-parse-tree.cc (show_omp_namelist): Update accordingly.
* module.cc (mio_omp_declare_simd): Likewise.
* trans-openmp.cc (gfc_trans_omp_clauses): Likewise.
* openmp.cc (resolve_omp_clauses): Likewise; accept new-style
'val' modifier with do/simd.
(gfc_match_omp_clauses): Handle OpenMP 5.2 linear clause syntax.
libgomp/ChangeLog:
* libgomp.texi (OpenMP 5.2): Mark linear-clause change as 'Y'.
gcc/testsuite/ChangeLog:
* c-c++-common/gomp/linear-4.c: New test.
* gfortran.dg/gomp/linear-2.f90: New test.
* gfortran.dg/gomp/linear-3.f90: New test.
* gfortran.dg/gomp/linear-4.f90: New test.
* gfortran.dg/gomp/linear-5.f90: New test.
* gfortran.dg/gomp/linear-6.f90: New test.
* gfortran.dg/gomp/linear-7.f90: New test.
* gfortran.dg/gomp/linear-8.f90: New test.
In 5.2, a map clause can be map-entering or map-exiting,
either containing 'tofrom'. The main reason for this is
permit 'map(x)' with 'omp target enter/exit data',
avoiding to specify 'to:/from:' explicitly. (OpenMP
defaults to 'tofrom'.)
gcc/c/ChangeLog:
* c-parser.cc (c_parser_omp_target_enter_data,
c_parser_omp_target_exit_data): Accept tofrom
map-type modifier but use 'to' / 'from' internally.
gcc/cp/ChangeLog:
* parser.cc (cp_parser_omp_target_enter_data,
cp_parser_omp_target_exit_data): Accept tofrom
map-type modifier but use 'to' / 'from' internally.
gcc/fortran/ChangeLog:
* dump-parse-tree.cc (show_omp_namelist): For the map-type,
also handle the always modifer and release/delete.
* openmp.cc (resolve_omp_clauses): Accept tofrom
map-type modifier for target enter/exit data,
but use 'to' / 'from' internally.
libgomp/ChangeLog:
* libgomp.texi (OpenMP 5.2): Mark target enter/exit data
with fromto as implemented.
gcc/testsuite/ChangeLog:
* c-c++-common/gomp/target-data-2.c: New test.
* c-c++-common/gomp/target-data-3.c: New test.
* gfortran.dg/gomp/target-data-1.f90: New test.
* gfortran.dg/gomp/target-data-2.f90: New test.
Jakub Jelinek [Tue, 5 Jul 2022 09:27:03 +0000 (11:27 +0200)]
openmp: Fix up get-mapped-ptr-1.{c,f90} tests
On Tue, Jun 14, 2022 at 06:41:37PM +0200, Thomas Schwinge wrote:
> In an offloading configuration, I'm seeing:
>
> PASS: libgomp.fortran/get-mapped-ptr-1.f90 -O (test for excess errors)
> [-PASS:-]{+FAIL:+} libgomp.fortran/get-mapped-ptr-1.f90 -O execution test
>
> Does that one need similar treatment?
I assume not just that but libgomp.c-c++-common/get-mapped-ptr-1.c too?
It both needs the same treatment, and in the get-mapped-ptr-1.c
case there is even UB, while the Fortran version was using c_loc (q)
as the host pointer, in C/C++ it was using q which was value of
uninitialized pointer.
2022-06-15 Jakub Jelinek <jakub@redhat.com>
* testsuite/libgomp.c-c++-common/get-mapped-ptr-1.c (main): Initialize
q to ddress of an automatic variable. Use -5 instead of -1 in
omp_get_mapped_ptr call. Add test with omp_initial_device.
* testsuite/libgomp.fortran/get-mapped-ptr-1.f90 (main): Use -5 instead
of -1 in omp_get_mapped_ptr call. Add test with omp_initial_device.
Renumber stop arguments afterwards.
Jakub Jelinek [Tue, 5 Jul 2022 09:22:20 +0000 (11:22 +0200)]
openmp: Conforming device numbers and omp_{initial,invalid}_device
OpenMP 5.2 changed once more what device numbers are allowed.
In 5.1, valid device numbers were [0, omp_get_num_devices()].
5.2 makes also -1 valid (calls it omp_initial_device), which is equivalent
in behavior to omp_get_num_devices() number but has the advantage that it
is a constant. And it also introduces omp_invalid_device which is
also a constant with implementation defined value < -1. That value should
act like sNaN, any time any device construct (GOMP_target*) or OpenMP runtime
API routine is asked for such a device, the program is terminated.
And if OMP_TARGET_OFFLOAD=mandatory, all non-conforming device numbers (which
is all but [-1, omp_get_num_devices()] other than omp_invalid_device)
must be treated like omp_invalid_device.
For device constructs, we have a compatibility problem, we've historically
used 2 magic negative values to mean something special.
GOMP_DEVICE_ICV (-1) means device clause wasn't present, pick the
omp_get_default_device () number
GOMP_DEVICE_FALLBACK (-2) means the host device (this is used e.g. for
#pragma omp target if (cond)
where if cond is false, we pass -2
But 5.2 requires that omp_initial_device is -1 (there were discussions
about it, advantage of -1 is that one can say iterate over the
[-1, omp_get_num_devices()-1] range to get all devices starting with
the host/initial one.
And also, if user passes -2, unless it is omp_invalid_device, we need to
treat it like non-conforming with OMP_TARGET_OFFLOAD=mandatory.
So, the patch does on the compiler side some number remapping,
user_device_num >= -2U ? user_device_num - 1 : user_device_num.
This remapping is done at compile time if device clause has constant
argument, otherwise at runtime, and means that for user -1 (omp_initial_device)
we pass -2 to GOMP_* in the runtime library where it treats it like host
fallback, while -2 is remapped to -3 (one of the non-conforming device numbers,
for those it doesn't matter which one is which).
omp_invalid_device is then -4.
For the OpenMP device runtime APIs, no remapping is done.
This patch doesn't deal with the initial default-device-var for
OMP_TARGET_OFFLOAD=mandatory , the spec says that the inital ICV value
for that should in that case depend on whether there are any offloading
devices or not (if not, should be omp_invalid_device), but that means
we can't determine the number of devices lazily (and let libraries have the
possibility to register their offloading data etc.).
2022-06-13 Jakub Jelinek <jakub@redhat.com>
gcc/
* omp-expand.cc (expand_omp_target): Remap user provided
device clause arguments, -1 to -2 and -2 to -3, either
at compile time if constant, or at runtime.
include/
* gomp-constants.h (GOMP_DEVICE_INVALID): Define.
libgomp/
* omp.h.in (omp_initial_device, omp_invalid_device): New enumerators.
* omp_lib.f90.in (omp_initial_device, omp_invalid_device): New
parameters.
* omp_lib.h.in (omp_initial_device, omp_invalid_device): Likewise.
* target.c (resolve_device): Add remapped argument, handle
GOMP_DEVICE_ICV only if remapped is true (and clear remapped),
for negative values, treat GOMP_DEVICE_FALLBACK as fallback only
if remapped, otherwise treat omp_initial_device that way. For
omp_invalid_device, always emit gomp_fatal, even when
OMP_TARGET_OFFLOAD isn't mandatory.
(GOMP_target, GOMP_target_ext, GOMP_target_data, GOMP_target_data_ext,
GOMP_target_update, GOMP_target_update_ext,
GOMP_target_enter_exit_data): Pass true as remapped argument to
resolve_device.
(omp_target_alloc, omp_target_free, omp_target_is_present,
omp_target_memcpy_check, omp_target_associate_ptr,
omp_target_disassociate_ptr, omp_get_mapped_ptr,
omp_target_is_accessible): Pass false as remapped argument to
resolve_device. Treat omp_initial_device the same as
gomp_get_num_devices (). Don't bypass resolve_device calls if
device_num is negative.
(omp_pause_resource): Treat omp_initial_device the same as
gomp_get_num_devices (). Call resolve_device.
* icv-device.c (omp_set_default_device): Always set to device_num
even when it is negative.
* libgomp.texi: Document that Conforming device numbers,
omp_initial_device and omp_invalid_device is implemented.
* testsuite/libgomp.c/target-41.c (main): Add test with
omp_initial_device.
* testsuite/libgomp.c/target-45.c: New test.
* testsuite/libgomp.c/target-46.c: New test.
* testsuite/libgomp.c/target-47.c: New test.
* testsuite/libgomp.c-c++-common/target-is-accessible-1.c (main): Add
test with omp_initial_device. Use -5 instead of -1 for negative value
test.
* testsuite/libgomp.fortran/target-is-accessible-1.f90 (main):
Likewise. Reorder stop numbers.
Jakub Jelinek [Tue, 5 Jul 2022 09:15:15 +0000 (11:15 +0200)]
openmp: Add support for OpenMP 5.2 linear clause syntax for C/C++
The syntax for linear clause changed in 5.2, the original syntax
which is still valid is:
linear (var1, var2)
linear (var3, var4 : step1)
The 4.5 syntax with modifiers like:
linear (val (var5, var6))
linear (val (var7, var8) : step2)
is still supported in 5.2, but is deprecated there.
Instead, one can use a new syntax:
linear (var9, var10 : val)
linear (var11, var12 : step (step3), val)
As val, ref, uval or step (someexpr) can be valid expressions (and especially
in C++ can be const / constexpr / consteval), the spec says that
when the whole step expression is val (or ref or uval) or step ( ... )
then it is the new modifier syntax, one can use + 0 or 0 + or 1 * or * 1
or ()s to say it is the old step expression.
Also, 5.2 now allows val modifier to be specified even outside of declare simd
(but not the other modifiers). I've implemented this for the new modifier
syntax only, the old one keeps the old restriction (which is why
OMP_CLAUSE_LINEAR_OLD_LINEAR_MODIFIER flag has been introduced).
2022-06-07 Jakub Jelinek <jakub@redhat.com>
gcc/
* tree.h (OMP_CLAUSE_LINEAR_OLD_LINEAR_MODIFIER): Define.
* tree-pretty-print.cc (dump_omp_clause) <case OMP_CLAUSE_LINEAR>:
Adjust clause printing style depending on
OMP_CLAUSE_LINEAR_OLD_LINEAR_MODIFIER.
gcc/c/
* c-parser.cc (c_parser_omp_clause_linear): Parse OpenMP 5.2
style linear clause modifiers. Set
OMP_CLAUSE_LINEAR_OLD_LINEAR_MODIFIER flag on the clauses when
old style modifiers are used.
* c-typeck.cc (c_finish_omp_clauses): Only reject linear clause
with val modifier on simd or for if the old style modifiers are
used.
gcc/cp/
* parser.cc (cp_parser_omp_clause_linear): Parse OpenMP 5.2
style linear clause modifiers. Set
OMP_CLAUSE_LINEAR_OLD_LINEAR_MODIFIER flag on the clauses when
old style modifiers are used.
* semantics.cc (finish_omp_clauses): Only reject linear clause
with val modifier on simd or for if the old style modifiers are
used.
gcc/fortran/
* trans-openmp.cc (gfc_trans_omp_clauses): Set
OMP_CLAUSE_LINEAR_OLD_LINEAR_MODIFIER on OMP_CLAUSE_LINEAR
clauses unconditionally for now.
gcc/testsuite/
* c-c++-common/gomp/linear-2.c: New test.
* c-c++-common/gomp/linear-3.c: New test.
* g++.dg/gomp/linear-3.C: New test.
* g++.dg/gomp/linear-4.C: New test.
* g++.dg/gomp/linear-5.C: New test.
Jakub Jelinek [Tue, 5 Jul 2022 08:01:09 +0000 (10:01 +0200)]
openmp: Add support for firstprivate and allocate clauses on scope construct
OpenMP 5.2 adds support for firstprivate and allocate clauses on the scope
construct and this patch adds that support to GCC.
5.2 unfortunately (IMNSHO mistakenly) marked scope construct as worksharing,
which implies that it isn't possible to nest inside of it other scope,
worksharing loop, sections, explicit barriers, single etc. which would
make scope far less useful. I'm not implementing that part, keeping the
5.1 behavior here, and will file an issue to revert that for OpenMP 6.0.
But, for firstprivate it keeps the restriction that is now implied from
worksharing construct that listed var can't be private in outer context,
where for reduction 5.1 had similar restriction explicit even for scope
and 5.2 has it implicitly through worksharing construct.
2022-05-31 Jakub Jelinek <jakub@redhat.com>
gcc/
* omp-low.cc (build_outer_var_ref): For code == OMP_CLAUSE_ALLOCATE
allow var to be private in the outer context.
(lower_private_allocate): Pass OMP_CLAUSE_ALLOCATE as last argument
to build_outer_var_ref.
gcc/c/
* c-parser.cc (OMP_SCOPE_CLAUSE_MASK): Add firstprivate and allocate
clauses.
gcc/cp/
* parser.cc (OMP_SCOPE_CLAUSE_MASK): Add firstprivate and allocate
clauses.
gcc/testsuite/
* c-c++-common/gomp/scope-5.c: New test.
* c-c++-common/gomp/scope-6.c: New test.
* g++.dg/gomp/attrs-1.C (bar): Add firstprivate and allocate clauses
to scope construct.
* g++.dg/gomp/attrs-2.C (bar): Likewise.
libgomp/
* testsuite/libgomp.c-c++-common/allocate-1.c (foo): Add testcase for
scope construct with allocate clause.
* testsuite/libgomp.c-c++-common/allocate-3.c (foo): Likewise.
* testsuite/libgomp.c-c++-common/scope-2.c: New test.
* libgomp.texi (OpenMP 5.2): Mark 'enter' clause as supported.
* testsuite/libgomp.fortran/declare-target-1.f90: Extend to test
explicit 'to' and 'enter' clause.
* testsuite/libgomp.fortran/declare-target-2.f90: Update accordingly.
Jakub Jelinek [Tue, 5 Jul 2022 07:29:23 +0000 (09:29 +0200)]
openmp: Adjust diagnostics about same variable in link and to/enter clauses
On Fri, May 27, 2022 at 04:52:17PM +0200, Tobias Burnus wrote:
> The 'TO'/'ENTER' usage is first stored in a linked list – and
> then as attribute to the symbol. I am not sure how to handle it best.
This reminds me I've left the C/C++ FE diagnostics about mixing link and
to/enter on the same variable in separate directives as is, so it always
talked about mixing link and to clauses.
This patch adjusts it, so that if link is first, it talks about the
clause actually used and if link is later, uses to or enter together
in the wording.
2022-05-28 Jakub Jelinek <jakub@redhat.com>
gcc/c/
* c-parser.cc (c_parser_omp_declare_target): If OMP_CLAUSE_LINK was
seen first, use "%<to%>" or "%<enter%>" depending on
OMP_CLAUSE_ENTER_TO of the current clause, otherwise use
"%<to%> or %<enter%>" wording.
gcc/cp/
* parser.cc (handle_omp_declare_target_clause): If OMP_CLAUSE_LINK was
seen first, use "%<to%>" or "%<enter%>" depending on
OMP_CLAUSE_ENTER_TO of the current clause, otherwise use
"%<to%> or %<enter%>" wording.
gcc/testsuite/
* c-c++-common/gomp/declare-target-2.c: Add further tests for mixing of
link and to/enter clauses on separate directives.
Jakub Jelinek [Tue, 5 Jul 2022 07:27:36 +0000 (09:27 +0200)]
openmp: Add support for enter clause on declare target
OpenMP 5.1 and earlier had 2 different uses of to clause, one for target
update construct with one semantics, and one for declare target directive
with a different semantics.
Under the hood we were using OMP_CLAUSE_TO_DECLARE to represent the latter.
OpenMP 5.2 renamed the declare target clause to to enter, the old one is
kept as a deprecated alias.
As we are far from having full OpenMP 5.2 support, this patch adds support
for the enter clause (and renames OMP_CLAUSE_TO_DECLARE to OMP_CLAUSE_ENTER
with a flag to tell the spelling of the clause for better diagnostics),
but doesn't deprecate the to clause on declare target just yet (that
should be done as one of the last steps in 5.2 support).
2022-05-27 Jakub Jelinek <jakub@redhat.com>
gcc/
* tree-core.h (enum omp_clause_code): Rename OMP_CLAUSE_TO_DECLARE
to OMP_CLAUSE_ENTER.
* tree.h (OMP_CLAUSE_ENTER_TO): Define.
* tree.cc (omp_clause_num_ops, omp_clause_code_name): Rename
OMP_CLAUSE_TO_DECLARE to OMP_CLAUSE_ENTER.
* tree-pretty-print.cc (dump_omp_clause): Handle OMP_CLAUSE_ENTER
instead of OMP_CLAUSE_TO_DECLARE, if OMP_CLAUSE_ENTER_TO, print
"to" instead of "enter".
* tree-nested.cc (convert_nonlocal_omp_clauses,
convert_local_omp_clauses): Handle OMP_CLAUSE_ENTER instead of
OMP_CLAUSE_TO_DECLARE.
gcc/c-family/
* c-pragma.h (enum pragma_omp_clause): Add PRAGMA_OMP_CLAUSE_ENTER.
gcc/c/
* c-parser.cc (c_parser_omp_clause_name): Parse enter clause.
(c_parser_omp_all_clauses): For to clause on declare target, use
OMP_CLAUSE_ENTER clause with OMP_CLAUSE_ENTER_TO instead of
OMP_CLAUSE_TO_DECLARE clause. Handle PRAGMA_OMP_CLAUSE_ENTER.
(OMP_DECLARE_TARGET_CLAUSE_MASK): Add enter clause.
(c_parser_omp_declare_target): Use OMP_CLAUSE_ENTER instead of
OMP_CLAUSE_TO_DECLARE.
* c-typeck.cc (c_finish_omp_clauses): Handle OMP_CLAUSE_ENTER instead
of OMP_CLAUSE_TO_DECLARE, to OMP_CLAUSE_ENTER_TO use "to" as clause
name in diagnostics instead of
omp_clause_code_name[OMP_CLAUSE_CODE (c)].
gcc/cp/
* parser.cc (cp_parser_omp_clause_name): Parse enter clause.
(cp_parser_omp_all_clauses): For to clause on declare target, use
OMP_CLAUSE_ENTER clause with OMP_CLAUSE_ENTER_TO instead of
OMP_CLAUSE_TO_DECLARE clause. Handle PRAGMA_OMP_CLAUSE_ENTER.
(OMP_DECLARE_TARGET_CLAUSE_MASK): Add enter clause.
(cp_parser_omp_declare_target): Use OMP_CLAUSE_ENTER instead of
OMP_CLAUSE_TO_DECLARE.
* semantics.cc (finish_omp_clauses): Handle OMP_CLAUSE_ENTER instead
of OMP_CLAUSE_TO_DECLARE, to OMP_CLAUSE_ENTER_TO use "to" as clause
name in diagnostics instead of
omp_clause_code_name[OMP_CLAUSE_CODE (c)].
gcc/testsuite/
* c-c++-common/gomp/clauses-3.c: Add tests with enter clause instead
of to or modify some existing to clauses to enter.
* c-c++-common/gomp/declare-target-1.c: Likewise.
* c-c++-common/gomp/declare-target-2.c: Likewise.
* c-c++-common/gomp/declare-target-3.c: Likewise.
* g++.dg/gomp/attrs-9.C: Likewise.
* g++.dg/gomp/declare-target-1.C: Likewise.
libgomp/
* testsuite/libgomp.c-c++-common/target-40.c: Modify some existing to
clauses to enter.
* testsuite/libgomp.c/target-41.c: Likewise.
Jakub Jelinek [Tue, 5 Jul 2022 07:19:47 +0000 (09:19 +0200)]
libgomp: Fix occassional hangs with taskwait nowait depend
Richi reported occassional hangs with taskwait-depend-nowait-1.*
tests and I've finally manged to reproduce. The problem is if
taskwait depend without nowait is encountered soon after
taskwait depend nowait and the former depends on the latter and there
is no other work to do, the taskwait depend without nowait is put
to sleep, but the empty_task optimization in
gomp_task_run_post_handle_dependers wouldn't wake it up in that
case. gomp_task_run_post_handle_dependers normally does some wakeups
because it schedules more work (another task), which is not the
case of empty_task, but we need to do the wakeups that would be done
upon task completion so that we awake sleeping threads when the
last child is done.
So, the taskwait-depend-nowait-1.* testcase is fixed with the
else if (__builtin_expect (task->parent_depends_on, 0) part of
the patch.
The new testcase can hang on another problem, if the empty task
is the last task of a taskgroup, we need to use atomic store
like elsewhere to decrease the counter to 0, and wake up taskgroup
end if needed.
Yet another spot which can sleep is normal taskwait (without depend),
but I believe nothing needs to be done for that - in that case we
await solely until the children's queue has no tasks, tasks still
waiting for dependencies aren't accounted in that, but the reason
is that if taskwait should wait for something, there needs to be at least
one active child doing something (in the children queue), which then
possibly awakes some of its siblings when the dependencies are met,
or in the empty task case awakes further dependencies, but in any
case the child that finished is still handled as active child and
will awake taskwait at the end if there is nothing further to
do.
Last sleeping case are barriers, but that is handled by ++ret and
awaking the barrier.
2022-05-25 Jakub Jelinek <jakub@redhat.com>
* task.c (gomp_task_run_post_handle_dependers): If empty_task
is the last task taskwait depend depends on, wake it up.
Similarly if it is the last child of a taskgroup, use atomic
store instead of decrement and awak taskgroup wait if any.
* testsuite/libgomp.c-c++-common/taskwait-depend-nowait-2.c: New test.
Jakub Jelinek [Tue, 5 Jul 2022 07:11:46 +0000 (09:11 +0200)]
openmp: Add taskwait nowait depend support [PR105378]
This patch adds support for (so far C/C++)
#pragma omp taskwait nowait depend(...)
directive, which is like
#pragma omp task depend(...)
;
but slightly optimized on the library side, so that it creates
the task only for the purpose of dependency tracking and doesn't actually
schedule it and wait for it when the dependencies are satisfied, instead
makes its dependencies satisfied right away.
2022-05-24 Jakub Jelinek <jakub@redhat.com>
PR c/105378
gcc/
* omp-builtins.def (BUILT_IN_GOMP_TASKWAIT_DEPEND_NOWAIT): New
builtin.
* gimplify.cc (gimplify_omp_task): Diagnose taskwait with nowait
clause but no depend clauses.
* omp-expand.cc (expand_taskwait_call): Use
BUILT_IN_GOMP_TASKWAIT_DEPEND_NOWAIT rather than
BUILT_IN_GOMP_TASKWAIT_DEPEND if nowait clause is present.
gcc/c/
* c-parser.cc (OMP_TASKWAIT_CLAUSE_MASK): Add nowait clause.
gcc/cp/
* parser.cc (OMP_TASKWAIT_CLAUSE_MASK): Add nowait clause.
gcc/testsuite/
* c-c++-common/gomp/taskwait-depend-nowait-1.c: New test.
libgomp/
* libgomp_g.h (GOMP_taskwait_depend_nowait): Declare.
* libgomp.map (GOMP_taskwait_depend_nowait): Export at GOMP_5.1.1.
* task.c (empty_task): New function.
(gomp_task_run_post_handle_depend_hash): Declare earlier.
(gomp_task_run_post_handle_depend): Declare.
(GOMP_task): Optimize fn == empty_task if there is nothing to wait
for.
(gomp_task_run_post_handle_dependers): Optimize task->fn == empty_task.
(GOMP_taskwait_depend_nowait): New function.
* testsuite/libgomp.c-c++-common/taskwait-depend-nowait-1.c: New test.
libgomp: Add new runtime routines omp_target_memcpy_async and omp_target_memcpy_rect_async
This patch adds two new OpenMP runtime routines: omp_target_memcpy_async and
omp_target_memcpy_rect_async. Both functions are introduced in OpenMP 5.1 as
asynchronous variants of omp_target_memcpy and omp_target_memcpy_rect.
In contrast to the synchronous variants, the asynchronous functions have two
additional function parameters to allow the specification of task dependences:
int depobj_count
omp_depend_t *depobj_list
integer(c_int), value :: depobj_count
integer(omp_depend_kind), optional :: depobj_list(*)
The implementation splits the synchronous functions into two parts: (a) check
and (b) copy. Then (a) is used in the asynchronous functions for the sequential
part, and the actual copy process (b) is executed in a new created task. The
sequential part (a) takes into account the requirements for the return values:
"The routine returns zero if successful. Otherwise, it returns a non-zero
value." (omp_target_memcpy_async, OpenMP 5.1 spec, section 3.8.7)
"An application can determine the number of inclusive dimensions supported by an
implementation by passing NULL pointers (or C_NULL_PTR, for Fortran) for both
dst and src. The routine returns the number of dimensions supported by the
implementation for the specified device numbers. No copy operation is
performed." (omp_target_memcpy_rect_async, OpenMP 5.1 spec, section 3.8.8)
Due to asynchronicity an error is thrown if the asynchronous memcpy is not
successful (in contrast to the synchronous functions which use a return
value unequal to zero).
gcc/ChangeLog:
* omp-low.cc (omp_runtime_api_call): Added target_memcpy_async and
target_memcpy_rect_async to omp_runtime_apis array.
libgomp/ChangeLog:
* libgomp.map: Added omp_target_memcpy_async and
omp_target_memcpy_rect_async.
* libgomp.texi: Both functions are now supported.
* omp.h.in: Added omp_target_memcpy_async and
omp_target_memcpy_rect_async.
* omp_lib.f90.in: Added interfaces for both new functions.
* omp_lib.h.in: Likewise.
* target.c (ialias_redirect): Added for GOMP_task.
(omp_target_memcpy): Restructured into check and copy part.
(omp_target_memcpy_check): New helper function for omp_target_memcpy and
omp_target_memcpy_async that checks requirements.
(omp_target_memcpy_copy): New helper function for omp_target_memcpy and
omp_target_memcpy_async that performs the memcpy.
(omp_target_memcpy_async_helper): New helper function that is used in
omp_target_memcpy_async for the asynchronous task.
(omp_target_memcpy_async): Added.
(omp_target_memcpy_rect): Restructured into check and copy part.
(omp_target_memcpy_rect_check): New helper function for
omp_target_memcpy_rect and omp_target_memcpy_rect_async that checks
requirements.
(omp_target_memcpy_rect_copy): New helper function for
omp_target_memcpy_rect and omp_target_memcpy_rect_async that performs
the memcpy.
(omp_target_memcpy_rect_async_helper): New helper function that is used
in omp_target_memcpy_rect_async for the asynchronous task.
(omp_target_memcpy_rect_async): Added.
* task.c (ialias): Added for GOMP_task.
* testsuite/libgomp.c-c++-common/target-memcpy-async-1.c: New test.
* testsuite/libgomp.c-c++-common/target-memcpy-async-2.c: New test.
* testsuite/libgomp.c-c++-common/target-memcpy-rect-async-1.c: New test.
* testsuite/libgomp.c-c++-common/target-memcpy-rect-async-2.c: New test.
* testsuite/libgomp.fortran/target-memcpy-async-1.f90: New test.
* testsuite/libgomp.fortran/target-memcpy-async-2.f90: New test.
* testsuite/libgomp.fortran/target-memcpy-rect-async-1.f90: New test.
* testsuite/libgomp.fortran/target-memcpy-rect-async-2.f90: New test.
Jakub Jelinek [Mon, 4 Jul 2022 19:32:48 +0000 (21:32 +0200)]
openmp: Add support for inoutset depend-kind
This patch adds support for inoutset depend-kind in depend
clauses. It is very similar to the in depend-kind in that
a task with a dependency with that depend-kind is dependent
on all previously created sibling tasks with matching address
unless they have the same depend-kind.
In the in depend-kind case everything is dependent except
for in -> in dependency, for inoutset everything is
dependent except for inoutset -> inoutset dependency.
mutexinoutset is also similar (everything is dependent except
for mutexinoutset -> mutexinoutset dependency), but there is
also the additional restriction that only one task with
mutexinoutset for each address can be scheduled at once (i.e.
mutual exclusitivty). For now we support mutexinoutset
the same as inout/out, but the inoutset support is full.
In order not to bump the ABI for dependencies each time
(we've bumped it already once, the old ABI supports only
inout/out and in depend-kind, the new ABI supports
inout/out, mutexinoutset, in and depobj), this patch arranges
for inoutset to be at least for the time being always handled
as if it was specified through depobj even when it is not.
So it uses the new ABI for that and inoutset are represented
like depobj - pointer to a pair of pointers where the first one
will be the actual address of the object mentioned in depend
clause and second pointer will be (void *) GOMP_DEPEND_INOUTSET.
2022-05-17 Jakub Jelinek <jakub@redhat.com>
gcc/
* tree-core.h (enum omp_clause_depend_kind): Add
OMP_CLAUSE_DEPEND_INOUTSET.
* tree-pretty-print.cc (dump_omp_clause): Handle
OMP_CLAUSE_DEPEND_INOUTSET.
* gimplify.cc (gimplify_omp_depend): Likewise.
* omp-low.cc (lower_depend_clauses): Likewise.
gcc/c-family/
* c-omp.cc (c_finish_omp_depobj): Handle
OMP_CLAUSE_DEPEND_INOUTSET.
gcc/c/
* c-parser.cc (c_parser_omp_clause_depend): Parse
inoutset depend-kind.
(c_parser_omp_depobj): Likewise.
gcc/cp/
* parser.cc (cp_parser_omp_clause_depend): Parse
inoutset depend-kind.
(cp_parser_omp_depobj): Likewise.
* cxx-pretty-print.cc (cxx_pretty_printer::statement): Handle
OMP_CLAUSE_DEPEND_INOUTSET.
gcc/testsuite/
* c-c++-common/gomp/all-memory-1.c (boo): Add test with
inoutset depend-kind.
* c-c++-common/gomp/all-memory-2.c (boo): Likewise.
* c-c++-common/gomp/depobj-1.c (f1): Likewise.
(f2): Adjusted expected diagnostics.
* g++.dg/gomp/depobj-1.C (f4): Adjust expected diagnostics.
include/
* gomp-constants.h (GOMP_DEPEND_INOUTSET): Define.
libgomp/
* libgomp.h (struct gomp_task_depend_entry): Change is_in type
from bool to unsigned char.
* task.c (gomp_task_handle_depend): Handle GOMP_DEPEND_INOUTSET.
Ignore dependencies where
task->depend[i].is_in && task->depend[i].is_in == ent->is_in
rather than just task->depend[i].is_in && ent->is_in. Remember
whether GOMP_DEPEND_IN loop is needed and guard the loop with that
conditional.
(gomp_task_maybe_wait_for_dependencies): Handle GOMP_DEPEND_INOUTSET.
Ignore dependencies where elem.is_in && elem.is_in == ent->is_in
rather than just elem.is_in && ent->is_in.
* testsuite/libgomp.c-c++-common/depend-1.c (test): Add task with
inoutset depend-kind.
* testsuite/libgomp.c-c++-common/depend-2.c (test): Likewise.
* testsuite/libgomp.c-c++-common/depend-3.c (test): Likewise.
* testsuite/libgomp.c-c++-common/depend-inoutset-1.c: New test.
* libgomp.texi (OpenMP 5.1): Set omp_all_memory to 'Y'.
* testsuite/libgomp.fortran/depend-5.f90: New test.
* testsuite/libgomp.fortran/depend-6.f90: New test.
* testsuite/libgomp.fortran/depend-7.f90: New test.
gcc/testsuite/ChangeLog:
* gfortran.dg/gomp/all-memory-1.f90: New test.
* gfortran.dg/gomp/all-memory-2.f90: New test.
* gfortran.dg/gomp/all-memory-3.f90: New test.
* testsuite/libgomp.c++/target-has-device-addr-7.C: New test.
* testsuite/libgomp.c++/target-has-device-addr-8.C: New test.
* testsuite/libgomp.c++/target-has-device-addr-9.C: New test.
Tobias Burnus [Fri, 13 May 2022 18:00:34 +0000 (20:00 +0200)]
OpenMP/Fortran: Use firstprivat not alloc for ptr attach for arrays
For a non-descriptor array, map(A(n:m)) was mapped as
map(tofrom:A[n-1] [len: ...]) map(alloc:A [pointer assign, bias: ...])
with this patch, it is changed to
map(tofrom:A[n-1] [len: ...]) map(firstprivate:A [pointer assign, bias: ...])
The latter avoids an alloc - and also avoids the race condition with
nowait in the enclosed testcase. (Note: predantically, the testcase is
invalid since OpenMP 5.1, violating the map clause restriction at [354:10-13].
gcc/fortran/ChangeLog:
* trans-openmp.cc (gfc_trans_omp_clauses): When mapping nondescriptor
array sections, use GOMP_MAP_FIRSTPRIVATE_POINTER instead of
GOMP_MAP_POINTER for the pointer attachment.
libgomp/ChangeLog:
* testsuite/libgomp.fortran/target-nowait-array-section.f90: New test.
Andrew MacLeod [Mon, 4 Jul 2022 15:21:34 +0000 (11:21 -0400)]
Don't use gori depedencies to optimize.
The routine fold_using_range::relation_fold_and_or needs to verify that both
operands of 2 stmts are the same, and uses GORIs dependency cache for this.
This cache cannot be counted on to reflect the current contents of a
stmt, expecially in the presence of an IL changing pass. Instead, look at the
statement operands.
Iain Buclaw [Wed, 29 Jun 2022 19:52:39 +0000 (21:52 +0200)]
d: Fix error: aggregate value used where floating point was expected
Casting from vector to static array is permitted, and the frontend
generates a reinterpret cast, but casting back the other way resulted in
an error. This has been fixed to be properly handled in the code
generation pass of VectorExp, and the conversion for lvalue and rvalue
handling done in convert_expr and convert_for_rvalue respectively.
PR d/106139
gcc/d/ChangeLog:
* d-convert.cc (convert_expr): Handle casting from array to vector.
(convert_for_rvalue): Rewrite vector to array casts of the same
element type into a constructor.
(convert_for_assignment): Return calling convert_for_rvalue.
* expr.cc (ExprVisitor::visit (VectorExp *)): Handle generating a
vector expression from a static array.
* toir.cc (IRVisitor::visit (ReturnStatement *)): Call
convert_for_rvalue on return value.
gcc/testsuite/ChangeLog:
* gdc.dg/pr106139a.d: New test.
* gdc.dg/pr106139b.d: New test.
* gdc.dg/pr106139c.d: New test.
* gdc.dg/pr106139d.d: New test.
Roger Sayle [Mon, 4 Jul 2022 12:58:37 +0000 (13:58 +0100)]
PR target/105991: Recognize PLUS and XOR forms of rldimi in rs6000.md.
This patch addresses PR target/105991 where a change to prefer representing
shifts and adds at the tree-level as multiplications, causes problems for
the rldimi patterns in the powerpc backend. The issue is that rs6000.md
models this pattern using IOR, and some variants that have the equivalent
PLUS or XOR in the RTL fail to match some *rotl<mode>4_insert patterns.
This is fixed in this patch by adding a define_insn_and_split to locally
canonicalize the PLUS and XOR forms to the backend's preferred IOR form.
Backported from master.
2022-07-04 Roger Sayle <roger@nextmovesoftware.com>
Marek Polacek <polacek@redhat.com>
Segher Boessenkool <segher@kernel.crashing.org>
Kewen Lin <linkw@linux.ibm.com>
gcc/ChangeLog
PR target/105991
* config/rs6000/rs6000.md (rotl<mode>3_insert_3): Check that
exact_log2 doesn't return -1 (or zero).
(plus_xor): New code iterator.
(*rotl<mode>3_insert_3_<code>): New define_insn_and_split.
gcc/testsuite/ChangeLog
PR target/105991
* gcc.target/powerpc/pr105991.c: New test case.
Xi Ruoyao [Sat, 2 Jul 2022 08:16:44 +0000 (16:16 +0800)]
loongarch: use -mno-check-zero-division as the default for optimized code
Integer division by zero is undefined behavior anyway, and there are
already many platforms where neither the GCC port and the hardware do
anything to trap on division by zero. So any portable program shall not
rely on SIGFPE on division by zero, in both theory and practice. As the
result, there is no real reason to cost two additional instructions just
for the trap on division by zero with a new ISA.
One remaining reason to trap on division by zero may be debugging,
especially while -fsanitize=integer-divide-by-zero is not implemented
for LoongArch yet. To make debugging easier, keep -mcheck-zero-division
as the default for -O0 and -Og, but use -mno-check-zero-division as the
default for all other optimization levels.
Backport this behavior change for 12.2, so 12.1 will be the only release
with a different default.
* config/loongarch/loongarch.cc (loongarch_check_zero_div_p):
New static function.
(loongarch_idiv_insns): Use loongarch_check_zero_div_p instead
of TARGET_CHECK_ZERO_DIV.
(loongarch_output_division): Likewise.
* common/config/loongarch/loongarch-common.cc
(TARGET_DEFAULT_TARGET_FLAGS): Remove unneeded hook.
* doc/invoke.texi: Update to match the new behavior.
gcc/testsuite/ChangeLog:
* gcc.c-torture/execute/20101011-1.c (dg-additional-options):
add -mcheck-zero-division for LoongArch targets.
Harald Anlauf [Wed, 29 Jun 2022 19:36:17 +0000 (21:36 +0200)]
Fortran: error recovery on invalid CLASS(), PARAMETER declarations [PR105243]
gcc/fortran/ChangeLog:
PR fortran/103137
PR fortran/103138
PR fortran/103693
PR fortran/105243
* decl.cc (gfc_match_data_decl): Reject CLASS entity declaration
when it is given the PARAMETER attribute.
Harald Anlauf [Tue, 28 Jun 2022 20:29:28 +0000 (22:29 +0200)]
Fortran: improve error recovery for EXTENDS_TYPE_OF() [PR106121]
gcc/fortran/ChangeLog:
PR fortran/106121
* simplify.cc (gfc_simplify_extends_type_of): Do not attempt to
simplify when one of the arguments is a CLASS variable that was
not properly declared.
gcc/testsuite/ChangeLog:
PR fortran/106121
* gfortran.dg/extends_type_of_4.f90: New test.
Roger Sayle [Sun, 3 Jul 2022 13:01:17 +0000 (14:01 +0100)]
PR target/106122: Don't update %esp via the stack with -Oz on x86.
When optimizing for size with -Oz, setting a register can be minimized by
pushing an immediate value to the stack and popping it to the destination.
Alas the one general register that shouldn't be updated via the stack is
the stack pointer itself, where "pop %esp" can't be represented in GCC's
RTL ("use of a register mentioned in pre_inc, pre_dec, post_inc or
post_dec is not permitted within the same instruction"). This patch
fixes PR target/106122 by explicitly checking for SP_REG in the
problematic peephole2.
2022-07-O3 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog
PR target/106122
* config/i386/i386.md (peephole2): Avoid generating pop %esp
when optimizing for size.
gcc/testsuite/ChangeLog
PR target/106122
* gcc.target/i386/pr106122.c: New test case.
On musl <pthread.h> uses calloc() (via <sched.h>). <memory> includes
it indirectly and exposes use of poisoned calloc() when module code
is built:
/build/build/./prev-gcc/xg++ ... ../../gcc-13-20220626/gcc/cp/mapper-resolver.cc
In file included from /<<NIX>>/musl-1.2.3-dev/include/pthread.h:30,
from /build/build/prev-x86_64-unknown-linux-musl/libstdc++-v3/include/x86_64-unknown-linux-musl/bits/gthr-default.h:35,
....
from /build/build/prev-x86_64-unknown-linux-musl/libstdc++-v3/include/memory:77,
from ../../gcc-13-20220626/gcc/../libcody/cody.hh:24,
from ../../gcc-13-20220626/gcc/cp/../../c++tools/resolver.h:25,
from ../../gcc-13-20220626/gcc/cp/../../c++tools/resolver.cc:23,
from ../../gcc-13-20220626/gcc/cp/mapper-resolver.cc:32:
/<<NIX>>/musl-1.2.3-dev/include/sched.h:84:7: error: attempt to use poisoned "calloc"
84 | void *calloc(size_t, size_t);
| ^
/<<NIX>>/musl-1.2.3-dev/include/sched.h:124:36: error: attempt to use poisoned "calloc"
124 | #define CPU_ALLOC(n) ((cpu_set_t *)calloc(1,CPU_ALLOC_SIZE(n)))
| ^
gcc/cp/
PR c++/106102
* mapper-client.cc: Include <memory> via "system.h".
* mapper-resolver.cc: Ditto.
* module.cc: Ditto.
libcc1/
PR c++/106102
* libcc1plugin.cc: Include <memory> via "system.h".
* libcp1plugin.cc: Ditto.
Andrew Stubbs [Mon, 20 Jun 2022 14:51:15 +0000 (15:51 +0100)]
amdgcn: libgomp plugin USM implementation
Implement the Unified Shared Memory API calls in the GCN plugin.
The allocate and free are pretty straight-forward because all "target" memory
allocations are compatible with USM, on the right hardware. However, there's
no known way to check what memory region was used, after the fact, so we use a
splay tree to record allocations so we can answer "is_usm_ptr" later.
Andrew Stubbs [Fri, 17 Jun 2022 12:07:11 +0000 (13:07 +0100)]
amdgcn, openmp: Auto-detect USM mode and set HSA_XNACK
The AMD GCN runtime must be set to the correct mode for Unified Shared Memory
to work, but this is not always clear at compile and link time due to the split
nature of the offload compilation pipeline.
This patch sets a new attribute on OpenMP offload functions to ensure that the
information is passed all the way to the backend. The backend then places a
marker in the assembler code for mkoffload to find. Finally mkoffload places a
constructor function into the final program to ensure that the HSA_XNACK
environment variable passes the correct mode to the GPU.
The HSA_XNACK variable must be set before the HSA runtime is even loaded, so
it makes more sense to have this set within the constructor than at some point
later within libgomp or the GCN plugin.
Andrew Stubbs [Fri, 10 Jun 2022 14:15:49 +0000 (15:15 +0100)]
amdgcn: Support XNACK mode
The XNACK feature allows memory load instructions to restart safely following
a page-miss interrupt. This is useful for shared-memory devices, like APUs,
and to implement OpenMP Unified Shared Memory.
To support the feature we must be able to set the appropriate meta-data and
set the load instructions to early-clobber. When the port supports scheduling
of s_waitcnt instructions there will be further requirements.
gcc/ChangeLog:
* config/gcn/gcn-hsa.h (XNACKOPT): New macro.
(ASM_SPEC): Use XNACKOPT.
* config/gcn/gcn-opts.h (enum sram_ecc_type): Rename to ...
(enum hsaco_attr_type): ... this, and generalize the names.
(TARGET_XNACK): New macro.
* config/gcn/gcn-valu.md (gather<mode>_insn_1offset<exec>):
Add xnack compatible alternatives.
(gather<mode>_insn_2offsets<exec>): Likewise.
* config/gcn/gcn.c (gcn_option_override): Permit -mxnack for devices
other than Fiji.
(gcn_expand_epilogue): Remove early-clobber problems.
(output_file_start): Emit xnack attributes.
(gcn_hsa_declare_function_name): Obey -mxnack setting.
* config/gcn/gcn.md (xnack): New attribute.
(enabled): Rework to include "xnack" attribute.
(*movbi): Add xnack compatible alternatives.
(*mov<mode>_insn): Likewise.
(*mov<mode>_insn): Likewise.
(*mov<mode>_insn): Likewise.
(*movti_insn): Likewise.
* config/gcn/gcn.opt (-mxnack): Add the "on/off/any" syntax.
(sram_ecc_type): Rename to ...
(hsaco_attr_type: ... this.)
* config/gcn/mkoffload.c (SET_XNACK_ANY): New macro.
(TEST_XNACK): Delete.
(TEST_XNACK_ANY): New macro.
(TEST_XNACK_ON): New macro.
(main): Support the new -mxnack=on/off/any syntax.
Jason Merrill [Wed, 11 May 2022 18:53:26 +0000 (14:53 -0400)]
c++: lambda template in requires [PR105541]
Since the patch for PR103408, the template parameters for the lambda in this
test have level 1 instead of 2, and we were treating null template args as 1
level of arguments, so tsubst_template_parms decided it had nothing to do.
Fixed by distinguishing between <> and no args at all, which is what we have
in our "substitution" in a requires-expression.
PR c++/105541
gcc/cp/ChangeLog:
* cp-tree.h (TMPL_ARGS_DEPTH): 0 for null args.
* parser.cc (cp_parser_enclosed_template_argument_list):
Use 0-length TREE_VEC for <>.
We were wrongly looking up the generic lambda op() in a dependent scope, and
then trying to look up its instantiation at substitution time, but lambdas
aren't instantiated, so we crashed. The fix is to not look into dependent
class scopes.
But this created trouble with wrongly trying to use a template from the
enclosing scope when we aren't actually looking at a template-argument-list,
in template/lookup18.C, so let's avoid that.
Harald Anlauf [Tue, 21 Jun 2022 21:20:18 +0000 (23:20 +0200)]
Fortran: fix simplification of INDEX(str1,str2) [PR105691]
gcc/fortran/ChangeLog:
PR fortran/105691
* simplify.cc (gfc_simplify_index): Replace old simplification
code by the equivalent of the runtime library implementation. Use
HOST_WIDE_INT instead of int for string index, length variables.
gcc/testsuite/ChangeLog:
PR fortran/105691
* gfortran.dg/index_6.f90: New test.
Harald Anlauf [Fri, 24 Jun 2022 20:21:39 +0000 (22:21 +0200)]
Fortran: fix checking of arguments to UNPACK when MASK is a variable [PR105813]
gcc/fortran/ChangeLog:
PR fortran/105813
* check.cc (gfc_check_unpack): Try to simplify MASK argument to
UNPACK so that checking of the VECTOR argument can work when MASK
is a variable.
gcc/testsuite/ChangeLog:
PR fortran/105813
* gfortran.dg/unpack_vector_1.f90: New test.
gcc/
* omp-low.cc (usm_transform): Remove unused function argument.
(This should be a fixup to 54c2d861ac62e30ebf34a4e62ee0d55478a742b9: 'Build
fix for 'openmp: Use libgomp memory allocation functions with unified shared
memory')
Tobias Burnus [Thu, 30 Jun 2022 06:30:48 +0000 (08:30 +0200)]
Build fix for 'openmp: allow requires unified_shared_memory'
OG12 commit fa65fc45972d27f2fd79a44eaba1978348177ee9 added an
error diagnostic (moved around in later commits); this diagnostic
caused bootstrap fails as %<...%> were missing. This commit adds
them.
gcc/c/
* c-parser.cc (c_parser_omp_requires): Add missing %<...%> in error.
gcc/cp/
* parser.cc (cp_parser_omp_requires): Add missing %<...%> in error.
Lulu Cheng [Mon, 27 Jun 2022 08:26:25 +0000 (16:26 +0800)]
LoongArch: Remove undefined behavior from code [PR 106097]
C++2017 and previous standard description:
The value of E1 << E2 is E1 left-shifted E2 bit positions;
vacated bits are zero-filled. If E1 has an unsigned type,
the value of the result is E1×2E2, reduced modulo one more
than the maximum value representable inthe result type.
Otherwise, if E1 has a signed type and non-negative value,
and E1×2E2 is representablein the corresponding unsigned
type of the result type, then that value, converted to the
result type, is the resulting value; otherwise, the behavior
is undefined.
The value of E1 >> E2 is E1 right-shifted E2 bit positions.
If E1 has an unsigned type or if E1 has a signed type and
a non-negative value, the value of the result is the integral
part of the quotient of E1/2E2. If E1 has a signed type and
a negative value, the resulting value is implementation-defined.
gcc/ChangeLog:
PR target/106097
* config/loongarch/loongarch.cc (loongarch_build_integer):
Remove undefined behavior from code.
Revert changes to acc_prof-init-1.c and acc_prof-parallel-1.c
There is an interaction between the commits "OpenACC profiling-interface
fixes for asynchronous operations" and "[OpenACC] Clarify sequencing of
'async' data copying vs. profiling events in
'libgomp.oacc-c-c++-common/acc_prof-{init,parallel}-1.c'", leading to an
execution test failure and hang in these two tests.
This reverts changes to the tests made by the second patch, allowing the
tests to pass.
The arguments to gfc_build_array_ref were recently updated in the commit
'fortran: Use pointer arithmetic to index arrays [PR102043]', but a call
from gfc_conv_array_ref used the old function signature. This went
unnoticed due to the use of default arguments.
This patch should be merged into 'Fortran: delinearize multi-dimensional
array accesses'.
The ICE occurs during Gimple verification after the ompexp stage because
one of the arguments to the generated builtin call is of a Gimple reg type,
but isn't a Gimple value (because it is marked addressable).
This appears to be fallout from the commit "OpenACC 'kernels' decomposition:
Mark variables used in synthesized data clauses as addressable [PR100280]".
The launch dimensions have been added to the arguments of a builtin call
by oacc_set_fn_attrib, but one of the dimensions has been marked addressable.
Fixed by forcing the added arguments to be re-gimplified.
The final tail mark has no LHS, causing code that assumes its presence to
segfault. The LHS and the assignment appear to have been removed as dead
code by the cddce1 stage.
Fixed by checking for the presence of the LHS before using it.
A change that was present in the OG11 version of
'openmp: in_reduction clause support on target construct' but
not in the mainline version resulted in non-contiguous
arrays being accepted in cache clauses, only to ICE later.
Thomas Schwinge [Thu, 5 May 2022 21:01:36 +0000 (23:01 +0200)]
Refactor '-ldl' handling for libgomp proper and plugins
Instead of implicit global 'LIBS="-ldl $LIBS"' via 'AC_CHECK_LIB', make
'-ldl' explicit for libgomp proper, and clean up 'PLUGIN_GCN_LIBS',
'PLUGIN_NVPTX_LIBS' accordingly.
Jakub Jelinek [Thu, 12 May 2022 06:31:20 +0000 (08:31 +0200)]
openmp: Add omp_all_memory support (C/C++ only so far)
The ugly part is that OpenMP 5.1 made omp_all_memory a reserved identifier
which isn't allowed to be used anywhere but in the depend clause, this is
against how everything else has been handled in OpenMP so far (where
some identifiers could have special meaning in some OpenMP clauses or
pragmas but not elsewhere).
The patch handles it by making it a conditional keyword (for -fopenmp
only) and emitting a better diagnostics when it is used in a primary
expression. Having a nicer diagnostics when e.g. trying to do
int omp_all_memory;
or
int *omp_all_memory[10];
etc. would mean changing too many spots and hooking into name lookups
to reject declaring any such symbols would be too ugly and I'm afraid
there are way too many spots where one can introduce a name
(variables, functions, namespaces, struct, enum, enumerators, template
arguments, ...).
Otherwise, the handling is quite simple, normal depend clauses lower
into addresses of variables being handed over to the library, for
omp_all_memory I'm using NULL pointers. omp_all_memory can only be
used with inout or out depend kinds and means that a task is dependent
on all previously created sibling tasks that have any dependency (of
any depend kind) and that any later created sibling tasks will be
dependent on it if they have any dependency.
2022-05-12 Jakub Jelinek <jakub@redhat.com>
gcc/
* gimplify.cc (gimplify_omp_depend): Don't build_fold_addr_expr
if null_pointer_node.
(gimplify_scan_omp_clauses): Likewise.
* tree-pretty-print.cc (dump_omp_clause): Print null_pointer_node
as omp_all_memory.
gcc/c-family/
* c-common.h (enum rid): Add RID_OMP_ALL_MEMORY.
* c-omp.cc (c_finish_omp_depobj): Don't build_fold_addr_expr
if null_pointer_node.
gcc/c/
* c-parser.cc (c_parse_init): Register omp_all_memory as keyword
if flag_openmp.
(c_parser_postfix_expression): Diagnose uses of omp_all_memory
in postfix expressions.
(c_parser_omp_variable_list): Handle omp_all_memory in depend
clause.
* c-typeck.cc (c_finish_omp_clauses): Handle omp_all_memory
keyword in depend clause as null_pointer_node, diagnose invalid
uses.
gcc/cp/
* lex.cc (init_reswords): Register omp_all_memory as keyword
if flag_openmp.
* parser.cc (cp_parser_primary_expression): Diagnose uses of
omp_all_memory in postfix expressions.
(cp_parser_omp_var_list_no_open): Handle omp_all_memory in depend
clause.
* semantics.cc (finish_omp_clauses): Handle omp_all_memory
keyword in depend clause as null_pointer_node, diagnose invalid
uses.
* pt.cc (tsubst_omp_clause_decl): Pass through omp_all_memory.
gcc/testsuite/
* c-c++-common/gomp/all-memory-1.c: New test.
* c-c++-common/gomp/all-memory-2.c: New test.
* c-c++-common/gomp/all-memory-3.c: New test.
* g++.dg/gomp/all-memory-1.C: New test.
* g++.dg/gomp/all-memory-2.C: New test.
libgomp/
* libgomp.h (struct gomp_task): Add depend_all_memory member.
* task.c (gomp_init_task): Initialize depend_all_memory.
(gomp_task_handle_depend): Handle omp_all_memory.
(gomp_task_run_post_handle_depend_hash): Clear
parent->depend_all_memory if equal to current task.
(gomp_task_maybe_wait_for_dependencies): Handle omp_all_memory.
* testsuite/libgomp.c-c++-common/depend-1.c: New test.
* testsuite/libgomp.c-c++-common/depend-2.c: New test.
* testsuite/libgomp.c-c++-common/depend-3.c: New test.
With recent commit 2e309a4eff80e55b53d32d26926a2a94eabfea21 "libgomp testsuite:
Don't amend 'LD_LIBRARY_PATH' for system-provided HSA Runtime library",
and commit d6adba307508c75f1ccb2121eb1a43c9ab1d4056 "libgomp GCN plugin:
Clean up unused references to system-provided HSA Runtime library", the last
uses of '--with-hsa-runtime' etc. are gone.
Thomas Schwinge [Wed, 6 Apr 2022 10:15:28 +0000 (12:15 +0200)]
libgomp GCN plugin: Clean up always-empty 'PLUGIN_GCN_CPPFLAGS', 'PLUGIN_GCN_LDFLAGS'
After recent commit d6adba307508c75f1ccb2121eb1a43c9ab1d4056
"libgomp GCN plugin: Clean up unused references to system-provided HSA Runtime
library", these aren't set anymore.
Thomas Schwinge [Wed, 6 Apr 2022 09:31:45 +0000 (11:31 +0200)]
libgomp GCN plugin: Clean up unused references to system-provided HSA Runtime library
This is only active if GCC is 'configure'd with '--with-hsa-runtime=[...]' or
'--with-hsa-runtime-include=[...]', '--with-hsa-runtime-lib=[...]' -- which
nobody really is doing, as far as I can tell.
Originally changed for the libgomp HSA plugin in
commit b8d89b03db5f212919e4571671ebb4f5f8b1e19d (r242749)
"Remove build dependence on HSA run-time", and later propagated into the GCN
plugin, these are no longer built against system-provided HSA Runtime library.
Instead, unconditionally built against the GCC-shipped 'include/hsa*.h' header
files, and at run time does 'dlopen("libhsa-runtime64.so.1")'. It thus doesn't
make sense to consider references to system-provided HSA Runtime library during
libgomp GCN plugin build.
Thomas Schwinge [Wed, 6 Apr 2022 08:39:56 +0000 (10:39 +0200)]
libgomp testsuite: Don't amend 'LD_LIBRARY_PATH' for system-provided HSA Runtime library
This is only active if GCC is 'configure'd with '--with-hsa-runtime=[...]' or
'--with-hsa-runtime-lib=[...]' -- which nobody really is doing, as far as I can
tell.
# For build-tree testing, also consider the library paths used for builing.
# For installed testing, we assume all that to be provided in the sysroot.
if { $blddir != "" } {
[...]
global hsa_runtime_lib
if { $hsa_runtime_lib != "" } {
append always_ld_library_path ":$hsa_runtime_lib"
}
}
However, the libgomp GCN plugin is unconditionally built against the
GCC-shipped 'include/hsa*.h' header files, and at run time does
'dlopen("libhsa-runtime64.so.1")', so there is no system-provided HSA Runtime
library "used for builing". It thus doesn't make sense to amend
'LD_LIBRARY_PATH' for system-provided HSA Runtime library.
projects / thirdparty / gcc.git / log
