This is libgomp.info, produced by makeinfo version 6.8 from
libgomp.texi.

Copyright (C) 2006-2024 Free Software Foundation, Inc.

   Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software Foundation; with the
Invariant Sections being "Funding Free Software", the Front-Cover texts
being (a) (see below), and with the Back-Cover Texts being (b) (see
below).  A copy of the license is included in the section entitled "GNU
Free Documentation License".

   (a) The FSF's Front-Cover Text is:

   A GNU Manual

   (b) The FSF's Back-Cover Text is:

   You have freedom to copy and modify this GNU Manual, like GNU
software.  Copies published by the Free Software Foundation raise funds
for GNU development.
INFO-DIR-SECTION GNU Libraries
START-INFO-DIR-ENTRY
* libgomp: (libgomp).          GNU Offloading and Multi Processing Runtime Library.
END-INFO-DIR-ENTRY

   This manual documents libgomp, the GNU Offloading and Multi
Processing Runtime library.  This is the GNU implementation of the
OpenMP and OpenACC APIs for parallel and accelerator programming in
C/C++ and Fortran.

   Published by the Free Software Foundation 51 Franklin Street, Fifth
Floor Boston, MA 02110-1301 USA

   Copyright (C) 2006-2024 Free Software Foundation, Inc.

   Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software Foundation; with the
Invariant Sections being "Funding Free Software", the Front-Cover texts
being (a) (see below), and with the Back-Cover Texts being (b) (see
below).  A copy of the license is included in the section entitled "GNU
Free Documentation License".

   (a) The FSF's Front-Cover Text is:

   A GNU Manual

   (b) The FSF's Back-Cover Text is:

   You have freedom to copy and modify this GNU Manual, like GNU
software.  Copies published by the Free Software Foundation raise funds
for GNU development.


File: libgomp.info,  Node: Top,  Next: Enabling OpenMP,  Up: (dir)

Introduction
************

This manual documents the usage of libgomp, the GNU Offloading and Multi
Processing Runtime Library.  This includes the GNU implementation of the
OpenMP (https://www.openmp.org) Application Programming Interface (API)
for multi-platform shared-memory parallel programming in C/C++ and
Fortran, and the GNU implementation of the OpenACC
(https://www.openacc.org) Application Programming Interface (API) for
offloading of code to accelerator devices in C/C++ and Fortran.

   Originally, libgomp implemented the GNU OpenMP Runtime Library.
Based on this, support for OpenACC and offloading (both OpenACC and
OpenMP 4's target construct) has been added later on, and the library's
name changed to GNU Offloading and Multi Processing Runtime Library.

* Menu:

* Enabling OpenMP::            How to enable OpenMP for your applications.
* OpenMP Implementation Status:: List of implemented features by OpenMP version
* OpenMP Runtime Library Routines: Runtime Library Routines.
                               The OpenMP runtime application programming
                               interface.
* OpenMP Environment Variables: Environment Variables.
                               Influencing OpenMP runtime behavior with
                               environment variables.
* Enabling OpenACC::           How to enable OpenACC for your
                               applications.
* OpenACC Runtime Library Routines:: The OpenACC runtime application
                               programming interface.
* OpenACC Environment Variables:: Influencing OpenACC runtime behavior with
                               environment variables.
* CUDA Streams Usage::         Notes on the implementation of
                               asynchronous operations.
* OpenACC Library Interoperability:: OpenACC library interoperability with the
                               NVIDIA CUBLAS library.
* OpenACC Profiling Interface::
* OpenMP-Implementation Specifics:: Notes specifics of this OpenMP
                               implementation
* Offload-Target Specifics::   Notes on offload-target specific internals
* The libgomp ABI::            Notes on the external ABI presented by libgomp.
* Reporting Bugs::             How to report bugs in the GNU Offloading and
                               Multi Processing Runtime Library.
* Copying::                    GNU general public license says
                               how you can copy and share libgomp.
* GNU Free Documentation License::
                               How you can copy and share this manual.
* Funding::                    How to help assure continued work for free
                               software.
* Library Index::              Index of this documentation.


File: libgomp.info,  Node: Enabling OpenMP,  Next: OpenMP Implementation Status,  Up: Top

1 Enabling OpenMP
*****************

To activate the OpenMP extensions for C/C++ and Fortran, the
compile-time flag '-fopenmp' must be specified.  For C and C++, this
enables the handling of the OpenMP directives using '#pragma omp' and
the '[[omp::directive(...)]]', '[[omp::sequence(...)]]' and
'[[omp::decl(...)]]' attributes.  For Fortran, it enables for free
source form the '!$omp' sentinel for directives and the '!$' conditional
compilation sentinel and for fixed source form the 'c$omp', '*$omp' and
'!$omp' sentinels for directives and the 'c$', '*$' and '!$' conditional
compilation sentinels.  The flag also arranges for automatic linking of
the OpenMP runtime library (*note Runtime Library Routines::).

   The '-fopenmp-simd' flag can be used to enable a subset of OpenMP
directives that do not require the linking of either the OpenMP runtime
library or the POSIX threads library.

   A complete description of all OpenMP directives may be found in the
OpenMP Application Program Interface (https://www.openmp.org) manuals.
See also *note OpenMP Implementation Status::.


File: libgomp.info,  Node: OpenMP Implementation Status,  Next: Runtime Library Routines,  Prev: Enabling OpenMP,  Up: Top

2 OpenMP Implementation Status
******************************

* Menu:

* OpenMP 4.5:: Feature completion status to 4.5 specification
* OpenMP 5.0:: Feature completion status to 5.0 specification
* OpenMP 5.1:: Feature completion status to 5.1 specification
* OpenMP 5.2:: Feature completion status to 5.2 specification
* OpenMP Technical Report 12:: Feature completion status to second 6.0 preview

The '_OPENMP' preprocessor macro and Fortran's 'openmp_version'
parameter, provided by 'omp_lib.h' and the 'omp_lib' module, have the
value '201511' (i.e.  OpenMP 4.5).


File: libgomp.info,  Node: OpenMP 4.5,  Next: OpenMP 5.0,  Up: OpenMP Implementation Status

2.1 OpenMP 4.5
==============

The OpenMP 4.5 specification is fully supported.


File: libgomp.info,  Node: OpenMP 5.0,  Next: OpenMP 5.1,  Prev: OpenMP 4.5,  Up: OpenMP Implementation Status

2.2 OpenMP 5.0
==============

New features listed in Appendix B of the OpenMP specification
-------------------------------------------------------------

Description                                 Status  Comments
-----------------------------------------------------------------------
Array shaping                               N
Array sections with non-unit strides in C   N
and C++
Iterators                                   Y
'metadirective' directive                   N
'declare variant' directive                 P       _simd_ traits
                                                    not handled
                                                    correctly
TARGET-OFFLOAD-VAR ICV and                  Y
'OMP_TARGET_OFFLOAD' env variable
Nested-parallel changes to                  Y
MAX-ACTIVE-LEVELS-VAR ICV
'requires' directive                        P       complete but no
                                                    non-host device
                                                    provides
                                                    'unified_shared_memory'
'teams' construct outside an enclosing      Y
target region
Non-rectangular loop nests                  P       Full support for
                                                    C/C++, partial
                                                    for Fortran
                                                    (PR110735
                                                    (https://gcc.gnu.org/PR110735))
'!=' as relational-op in canonical loop     Y
form for C/C++
'nonmonotonic' as default loop schedule     Y
modifier for worksharing-loop constructs
Collapse of associated loops that are       Y
imperfectly nested loops
Clauses 'if', 'nontemporal' and             Y
'order(concurrent)' in 'simd' construct
'atomic' constructs in 'simd'               Y
'loop' construct                            Y
'order(concurrent)' clause                  Y
'scan' directive and 'in_scan' modifier     Y
for the 'reduction' clause
'in_reduction' clause on 'task'             Y
constructs
'in_reduction' clause on 'target'           P       'nowait' only
constructs                                          stub
'task_reduction' clause with 'taskgroup'    Y
'task' modifier to 'reduction' clause       Y
'affinity' clause to 'task' construct       Y       Stub only
'detach' clause to 'task' construct         Y
'omp_fulfill_event' runtime routine         Y
'reduction' and 'in_reduction' clauses on   Y
'taskloop' and 'taskloop simd' constructs
'taskloop' construct cancelable by          Y
'cancel' construct
'mutexinoutset' _dependence-type_ for       Y
'depend' clause
Predefined memory spaces, memory            Y       See also
allocators, allocator traits                        *note Memory allocation::
Memory management routines                  Y
'allocate' directive                        P       Only C for
                                                    stack/automatic
                                                    and Fortran for
                                                    stack/automatic
                                                    and
                                                    allocatable/pointer
                                                    variables
'allocate' clause                           P       Initial support
'use_device_addr' clause on 'target data'   Y
'ancestor' modifier on 'device' clause      Y
Implicit declare target directive           Y
Discontiguous array section with 'target    N
update' construct
C/C++'s lvalue expressions in 'to',         Y
'from' and 'map' clauses
C/C++'s lvalue expressions in 'depend'      Y
clauses
Nested 'declare target' directive           Y
Combined 'master' constructs                Y
'depend' clause on 'taskwait'               Y
Weak memory ordering clauses on 'atomic'    Y
and 'flush' construct
'hint' clause on the 'atomic' construct     Y       Stub only
'depobj' construct and depend objects       Y
Lock hints were renamed to                  Y
synchronization hints
'conditional' modifier to 'lastprivate'     Y
clause
Map-order clarifications                    P
'close' _map-type-modifier_                 Y
Mapping C/C++ pointer variables and to      P
assign the address of device memory
mapped by an array section
Mapping of Fortran pointer and              P       Mapping of vars
allocatable variables, including pointer            with allocatable
and allocatable components of variables             components
                                                    unsupported
'defaultmap' extensions                     Y
'declare mapper' directive                  N
'omp_get_supported_active_levels' routine   Y
Runtime routines and environment            Y
variables to display runtime thread
affinity information
'omp_pause_resource' and                    Y
'omp_pause_resource_all' runtime routines
'omp_get_device_num' runtime routine        Y
OMPT interface                              N
OMPD interface                              N

Other new OpenMP 5.0 features
-----------------------------

Description                                 Status  Comments
-----------------------------------------------------------------------
Supporting C++'s range-based for loop       Y


File: libgomp.info,  Node: OpenMP 5.1,  Next: OpenMP 5.2,  Prev: OpenMP 5.0,  Up: OpenMP Implementation Status

2.3 OpenMP 5.1
==============

New features listed in Appendix B of the OpenMP specification
-------------------------------------------------------------

Description                                 Status  Comments
-----------------------------------------------------------------------
OpenMP directive as C++ attribute           Y
specifiers
'omp_all_memory' reserved locator           Y
_target_device trait_ in OpenMP Context     N
'target_device' selector set in context     N
selectors
C/C++'s 'declare variant' directive:        N
elision support of preprocessed code
'declare variant': new clauses              N
'adjust_args' and 'append_args'
'dispatch' construct                        N
device-specific ICV settings with           Y
environment variables
'assume' and 'assumes' directives           Y
'nothing' directive                         Y
'error' directive                           Y
'masked' construct                          Y
'scope' directive                           Y
Loop transformation constructs              N
'strict' modifier in the 'grainsize' and    Y
'num_tasks' clauses of the 'taskloop'
construct
'align' clause in 'allocate' directive      P       Only C and
                                                    Fortran (and not
                                                    for static
                                                    variables)
'align' modifier in 'allocate' clause       Y
'thread_limit' clause to 'target'           Y
construct
'has_device_addr' clause to 'target'        Y
construct
Iterators in 'target update' motion         N
clauses and 'map' clauses
Indirect calls to the device version of a   Y
procedure or function in 'target' regions
'interop' directive                         N
'omp_interop_t' object support in runtime   N
routines
'nowait' clause in 'taskwait' directive     Y
Extensions to the 'atomic' directive        Y
'seq_cst' clause on a 'flush' construct     Y
'inoutset' argument to the 'depend'         Y
clause
'private' and 'firstprivate' argument to    Y
'default' clause in C and C++
'present' argument to 'defaultmap' clause   Y
'omp_set_num_teams',                        Y
'omp_set_teams_thread_limit',
'omp_get_max_teams',
'omp_get_teams_thread_limit' runtime
routines
'omp_target_is_accessible' runtime          Y
routine
'omp_target_memcpy_async' and               Y
'omp_target_memcpy_rect_async' runtime
routines
'omp_get_mapped_ptr' runtime routine        Y
'omp_calloc', 'omp_realloc',                Y
'omp_aligned_alloc' and
'omp_aligned_calloc' runtime routines
'omp_alloctrait_key_t' enum:                Y
'omp_atv_serialized' added,
'omp_atv_default' changed
'omp_display_env' runtime routine           Y
'ompt_scope_endpoint_t' enum:               N
'ompt_scope_beginend'
'ompt_sync_region_t' enum additions         N
'ompt_state_t' enum:                        N
'ompt_state_wait_barrier_implementation'
and 'ompt_state_wait_barrier_teams'
'ompt_callback_target_data_op_emi_t',       N
'ompt_callback_target_emi_t',
'ompt_callback_target_map_emi_t' and
'ompt_callback_target_submit_emi_t'
'ompt_callback_error_t' type                N
'OMP_PLACES' syntax extensions              Y
'OMP_NUM_TEAMS' and                         Y
'OMP_TEAMS_THREAD_LIMIT' environment
variables

Other new OpenMP 5.1 features
-----------------------------

Description                                 Status  Comments
-----------------------------------------------------------------------
Support of strictly structured blocks in    Y
Fortran
Support of structured block sequences in    Y
C/C++
'unconstrained' and 'reproducible'          Y
modifiers on 'order' clause
Support 'begin/end declare target' syntax   Y
in C/C++
Pointer predetermined firstprivate          N
getting initialized to address of
matching mapped list item per 5.1, Sect.
2.21.7.2
For Fortran, diagnose placing declarative   N
before/between 'USE', 'IMPORT', and
'IMPLICIT' as invalid
Optional comma between directive and        Y
clause in the '#pragma' form
'indirect' clause in 'declare target'       Y
'device_type(nohost)'/'device_type(host)'   N
for variables
'present' modifier to the 'map', 'to' and   Y
'from' clauses


File: libgomp.info,  Node: OpenMP 5.2,  Next: OpenMP Technical Report 12,  Prev: OpenMP 5.1,  Up: OpenMP Implementation Status

2.4 OpenMP 5.2
==============

New features listed in Appendix B of the OpenMP specification
-------------------------------------------------------------

Description                                 Status  Comments
-----------------------------------------------------------------------
'omp_in_explicit_task' routine and          Y
EXPLICIT-TASK-VAR ICV
'omp'/'ompx'/'omx' sentinels and            N/A     warning for
'omp_'/'ompx_' namespaces                           'ompx/omx'
                                                    sentinels(1)
Clauses on 'end' directive can be on        Y
directive
'destroy' clause with destroy-var           Y
argument on 'depobj'
Deprecation of no-argument 'destroy'        N
clause on 'depobj'
'linear' clause syntax changes and 'step'   Y
modifier
Deprecation of minus operator for           N
reductions
Deprecation of separating 'map' modifiers   N
without comma
'declare mapper' with iterator and          N
'present' modifiers
If a matching mapped list item is not       Y
found in the data environment, the
pointer retains its original value
New 'enter' clause as alias for 'to' on     Y
declare target directive
Deprecation of 'to' clause on declare       N
target directive
Extended list of directives permitted in    Y
Fortran pure procedures
New 'allocators' directive for Fortran      Y
Deprecation of 'allocate' directive for     N
Fortran allocatables/pointers
Optional paired 'end' directive with        N
'dispatch'
New 'memspace' and 'traits' modifiers for   N
'uses_allocators'
Deprecation of traits array following the   N
allocator_handle expression in
'uses_allocators'
New 'otherwise' clause as alias for         N
'default' on metadirectives
Deprecation of 'default' clause on          N
metadirectives
Deprecation of delimited form of 'declare   N
target'
Reproducible semantics changed for          N
'order(concurrent)'
'allocate' and 'firstprivate' clauses on    Y
'scope'
'ompt_callback_work'                        N
Default map-type for the 'map' clause in    Y
'target enter/exit data'
New 'doacross' clause as alias for          Y
'depend' with 'source'/'sink' modifier
Deprecation of 'depend' with                N
'source'/'sink' modifier
'omp_cur_iteration' keyword                 Y

Other new OpenMP 5.2 features
-----------------------------

Description                                 Status  Comments
-----------------------------------------------------------------------
For Fortran, optional comma between         N
directive and clause
Conforming device numbers and               Y
'omp_initial_device' and
'omp_invalid_device' enum/PARAMETER
Initial value of DEFAULT-DEVICE-VAR ICV     Y
with 'OMP_TARGET_OFFLOAD=mandatory'
'all' as _implicit-behavior_ for            Y
'defaultmap'
_interop_types_ in any position of the      N
modifier list for the 'init' clause of
the 'interop' construct
Invoke virtual member functions of C++      N
objects created on the host device on
other devices

   ---------- Footnotes ----------

   (1) The 'ompx' sentinel as C/C++ pragma and C++ attributes are warned
for with '-Wunknown-pragmas' (implied by '-Wall') and '-Wattributes'
(enabled by default), respectively; for Fortran free-source code, there
is a warning enabled by default and, for fixed-source code, the 'omx'
sentinel is warned for with '-Wsurprising' (enabled by '-Wall').
Unknown clauses are always rejected with an error.


File: libgomp.info,  Node: OpenMP Technical Report 12,  Prev: OpenMP 5.2,  Up: OpenMP Implementation Status

2.5 OpenMP Technical Report 12
==============================

Technical Report (TR) 12 is the second preview for OpenMP 6.0.

New features listed in Appendix B of the OpenMP specification
-------------------------------------------------------------

Features deprecated in versions 5.2, 5.1    N/A     Backward
and 5.0 were removed                                compatibility
Full support for C23 was added              P
Full support for C++23 was added            P
'_ALL' suffix to the device-scope           P       Host device
environment variables                               number wrongly
                                                    accepted
'num_threads' now accepts a list            N
Supporting increments with abstract names   N
in 'OMP_PLACES'
Extension of 'OMP_DEFAULT_DEVICE' and new   N
'OMP_AVAILABLE_DEVICES' environment vars
New 'OMP_THREADS_RESERVE' environment       N
variable
The 'decl' attribute was added to the C++   Y
attribute syntax
The OpenMP directive syntax was extended    Y
to include C 23 attribute specifiers
All inarguable clauses take now an          N
optional Boolean argument
For Fortran, _locator list_ can be also     N
function reference with data pointer
result
Concept of _assumed-size arrays_ in C and   N
C++
_directive-name-modifier_ accepted in all   N
clauses
For Fortran, atomic with BLOCK construct    N
and, for C/C++, with unlimited curly
braces supported
For Fortran, atomic compare with storing    N
the comparison result
New 'looprange' clause                      N
Ref-count change for                        N
'use_device_ptr'/'use_device_addr'
Support for inductions                      N
Implicit reduction identifiers of C++       N
classes
Change of the _map-type_ property from      N
_ultimate_ to _default_
'self' modifier to 'map' and 'self' as      N
'defaultmap' argument
Mapping of _assumed-size arrays_ in C,      N
C++ and Fortran
'groupprivate' directive                    N
'local' clause to 'declare target'          N
directive
'part_size' allocator trait                 N
'pin_device', 'preferred_device' and        N
'target_access' allocator traits
'access' allocator trait changes            N
Extension of 'interop' operation of         N
'append_args', allowing all modifiers of
the 'init' clause
'interop' clause to 'dispatch'              N
'message' and 'severity' clauses to         N
'parallel' directive
'self' clause to 'requires' directive       N
'no_openmp_constructs' assumptions clause   N
'reverse' loop-transformation construct     N
'interchange' loop-transformation           N
construct
'fuse' loop-transformation construct        N
'apply' code to loop-transforming           N
constructs
'omp_curr_progress_width' identifier        N
'safesync' clause to the 'parallel'         N
construct
'omp_get_max_progress_width' runtime        N
routine
'strict' modifier keyword to                N
'num_threads'
'atomic' permitted in a construct with      N
'order(concurrent)'
'workdistribute' directive for Fortran      N       Renamed just
                                                    after TR12;
                                                    added in TR12 as
                                                    'coexecute'
Fortran DO CONCURRENT as associated loop    N
in a 'loop' construct
'threadset' clause in task-generating       N
constructs
'nowait' clause with reverse-offload        N
'target' directives
Boolean argument to 'nowait' and            N
'nogroup' may be non constant
'memscope' clause to 'atomic' and 'flush'   N
'omp_is_free_agent' and                     N
'omp_ancestor_is_free_agent' routines
'omp_target_memset' and                     N
'omp_target_memset_rect_async' routines
Routines for obtaining memory               N
spaces/allocators for shared/device
memory
'omp_get_memspace_num_resources' routine    N
'omp_get_submemspace' routine               N
'ompt_target_data_transfer' and             N
'ompt_target_data_transfer_async' values
in 'ompt_target_data_op_t' enum
'ompt_get_buffer_limits' OMPT routine       N

Other new TR 12 features
------------------------

Canonical loop nest enclosed in             N
(multiple) curly braces (C/C++) or BLOCK
constructs (Fortran)
Relaxed Fortran restrictions to the         N
'aligned' clause
Mapping lambda captures                     N
New 'omp_pause_stop_tool' constant for      N
omp_pause_resource


File: libgomp.info,  Node: Runtime Library Routines,  Next: Environment Variables,  Prev: OpenMP Implementation Status,  Up: Top

3 OpenMP Runtime Library Routines
*********************************

The runtime routines described here are defined by Section 18 of the
OpenMP specification in version 5.2.

* Menu:

* Thread Team Routines::
* Thread Affinity Routines::
* Teams Region Routines::
* Tasking Routines::
* Resource Relinquishing Routines::
* Device Information Routines::
* Device Memory Routines::
* Lock Routines::
* Timing Routines::
* Event Routine::
* Memory Management Routines::
* Environment Display Routine::


File: libgomp.info,  Node: Thread Team Routines,  Next: Thread Affinity Routines,  Up: Runtime Library Routines

3.1 Thread Team Routines
========================

Routines controlling threads in the current contention group.  They have
C linkage and do not throw exceptions.

* Menu:

* omp_set_num_threads::         Set upper team size limit
* omp_get_num_threads::         Size of the active team
* omp_get_max_threads::         Maximum number of threads of parallel region
* omp_get_thread_num::          Current thread ID
* omp_in_parallel::             Whether a parallel region is active
* omp_set_dynamic::             Enable/disable dynamic teams
* omp_get_dynamic::             Dynamic teams setting
* omp_get_cancellation::        Whether cancellation support is enabled
* omp_set_nested::              Enable/disable nested parallel regions
* omp_get_nested::              Nested parallel regions
* omp_set_schedule::            Set the runtime scheduling method
* omp_get_schedule::            Obtain the runtime scheduling method
* omp_get_teams_thread_limit::  Maximum number of threads imposed by teams
* omp_get_supported_active_levels:: Maximum number of active regions supported
* omp_set_max_active_levels::   Limits the number of active parallel regions
* omp_get_max_active_levels::   Current maximum number of active regions
* omp_get_level::               Number of parallel regions
* omp_get_ancestor_thread_num:: Ancestor thread ID
* omp_get_team_size::           Number of threads in a team
* omp_get_active_level::        Number of active parallel regions


File: libgomp.info,  Node: omp_set_num_threads,  Next: omp_get_num_threads,  Up: Thread Team Routines

3.1.1 'omp_set_num_threads' - Set upper team size limit
-------------------------------------------------------

_Description_:
     Specifies the number of threads used by default in subsequent
     parallel sections, if those do not specify a 'num_threads' clause.
     The argument of 'omp_set_num_threads' shall be a positive integer.

_C/C++_:
     _Prototype_:   'void omp_set_num_threads(int num_threads);'

_Fortran_:
     _Interface_:   'subroutine omp_set_num_threads(num_threads)'
                    'integer, intent(in) :: num_threads'

_See also_:
     *note OMP_NUM_THREADS::, *note omp_get_num_threads::, *note
     omp_get_max_threads::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.1.


File: libgomp.info,  Node: omp_get_num_threads,  Next: omp_get_max_threads,  Prev: omp_set_num_threads,  Up: Thread Team Routines

3.1.2 'omp_get_num_threads' - Size of the active team
-----------------------------------------------------

_Description_:
     Returns the number of threads in the current team.  In a sequential
     section of the program 'omp_get_num_threads' returns 1.

     The default team size may be initialized at startup by the
     'OMP_NUM_THREADS' environment variable.  At runtime, the size of
     the current team may be set either by the 'NUM_THREADS' clause or
     by 'omp_set_num_threads'.  If none of the above were used to define
     a specific value and 'OMP_DYNAMIC' is disabled, one thread per CPU
     online is used.

_C/C++_:
     _Prototype_:   'int omp_get_num_threads(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_num_threads()'

_See also_:
     *note omp_get_max_threads::, *note omp_set_num_threads::, *note
     OMP_NUM_THREADS::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.2.


File: libgomp.info,  Node: omp_get_max_threads,  Next: omp_get_thread_num,  Prev: omp_get_num_threads,  Up: Thread Team Routines

3.1.3 'omp_get_max_threads' - Maximum number of threads of parallel region
--------------------------------------------------------------------------

_Description_:
     Return the maximum number of threads used for the current parallel
     region that does not use the clause 'num_threads'.

_C/C++_:
     _Prototype_:   'int omp_get_max_threads(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_max_threads()'

_See also_:
     *note omp_set_num_threads::, *note omp_set_dynamic::, *note
     omp_get_thread_limit::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.3.


File: libgomp.info,  Node: omp_get_thread_num,  Next: omp_in_parallel,  Prev: omp_get_max_threads,  Up: Thread Team Routines

3.1.4 'omp_get_thread_num' - Current thread ID
----------------------------------------------

_Description_:
     Returns a unique thread identification number within the current
     team.  In a sequential parts of the program, 'omp_get_thread_num'
     always returns 0.  In parallel regions the return value varies from
     0 to 'omp_get_num_threads'-1 inclusive.  The return value of the
     primary thread of a team is always 0.

_C/C++_:
     _Prototype_:   'int omp_get_thread_num(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_thread_num()'

_See also_:
     *note omp_get_num_threads::, *note omp_get_ancestor_thread_num::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.4.


File: libgomp.info,  Node: omp_in_parallel,  Next: omp_set_dynamic,  Prev: omp_get_thread_num,  Up: Thread Team Routines

3.1.5 'omp_in_parallel' - Whether a parallel region is active
-------------------------------------------------------------

_Description_:
     This function returns 'true' if currently running in parallel,
     'false' otherwise.  Here, 'true' and 'false' represent their
     language-specific counterparts.

_C/C++_:
     _Prototype_:   'int omp_in_parallel(void);'

_Fortran_:
     _Interface_:   'logical function omp_in_parallel()'

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.6.


File: libgomp.info,  Node: omp_set_dynamic,  Next: omp_get_dynamic,  Prev: omp_in_parallel,  Up: Thread Team Routines

3.1.6 'omp_set_dynamic' - Enable/disable dynamic teams
------------------------------------------------------

_Description_:
     Enable or disable the dynamic adjustment of the number of threads
     within a team.  The function takes the language-specific equivalent
     of 'true' and 'false', where 'true' enables dynamic adjustment of
     team sizes and 'false' disables it.

_C/C++_:
     _Prototype_:   'void omp_set_dynamic(int dynamic_threads);'

_Fortran_:
     _Interface_:   'subroutine omp_set_dynamic(dynamic_threads)'
                    'logical, intent(in) :: dynamic_threads'

_See also_:
     *note OMP_DYNAMIC::, *note omp_get_dynamic::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.7.


File: libgomp.info,  Node: omp_get_dynamic,  Next: omp_get_cancellation,  Prev: omp_set_dynamic,  Up: Thread Team Routines

3.1.7 'omp_get_dynamic' - Dynamic teams setting
-----------------------------------------------

_Description_:
     This function returns 'true' if enabled, 'false' otherwise.  Here,
     'true' and 'false' represent their language-specific counterparts.

     The dynamic team setting may be initialized at startup by the
     'OMP_DYNAMIC' environment variable or at runtime using
     'omp_set_dynamic'.  If undefined, dynamic adjustment is disabled by
     default.

_C/C++_:
     _Prototype_:   'int omp_get_dynamic(void);'

_Fortran_:
     _Interface_:   'logical function omp_get_dynamic()'

_See also_:
     *note omp_set_dynamic::, *note OMP_DYNAMIC::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.8.


File: libgomp.info,  Node: omp_get_cancellation,  Next: omp_set_nested,  Prev: omp_get_dynamic,  Up: Thread Team Routines

3.1.8 'omp_get_cancellation' - Whether cancellation support is enabled
----------------------------------------------------------------------

_Description_:
     This function returns 'true' if cancellation is activated, 'false'
     otherwise.  Here, 'true' and 'false' represent their
     language-specific counterparts.  Unless 'OMP_CANCELLATION' is set
     true, cancellations are deactivated.

_C/C++_:
     _Prototype_:   'int omp_get_cancellation(void);'

_Fortran_:
     _Interface_:   'logical function omp_get_cancellation()'

_See also_:
     *note OMP_CANCELLATION::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.9.


File: libgomp.info,  Node: omp_set_nested,  Next: omp_get_nested,  Prev: omp_get_cancellation,  Up: Thread Team Routines

3.1.9 'omp_set_nested' - Enable/disable nested parallel regions
---------------------------------------------------------------

_Description_:
     Enable or disable nested parallel regions, i.e., whether team
     members are allowed to create new teams.  The function takes the
     language-specific equivalent of 'true' and 'false', where 'true'
     enables dynamic adjustment of team sizes and 'false' disables it.

     Enabling nested parallel regions also sets the maximum number of
     active nested regions to the maximum supported.  Disabling nested
     parallel regions sets the maximum number of active nested regions
     to one.

     Note that the 'omp_set_nested' API routine was deprecated in the
     OpenMP specification 5.2 in favor of 'omp_set_max_active_levels'.

_C/C++_:
     _Prototype_:   'void omp_set_nested(int nested);'

_Fortran_:
     _Interface_:   'subroutine omp_set_nested(nested)'
                    'logical, intent(in) :: nested'

_See also_:
     *note omp_get_nested::, *note omp_set_max_active_levels::, *note
     OMP_MAX_ACTIVE_LEVELS::, *note OMP_NESTED::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.10.


File: libgomp.info,  Node: omp_get_nested,  Next: omp_set_schedule,  Prev: omp_set_nested,  Up: Thread Team Routines

3.1.10 'omp_get_nested' - Nested parallel regions
-------------------------------------------------

_Description_:
     This function returns 'true' if nested parallel regions are
     enabled, 'false' otherwise.  Here, 'true' and 'false' represent
     their language-specific counterparts.

     The state of nested parallel regions at startup depends on several
     environment variables.  If 'OMP_MAX_ACTIVE_LEVELS' is defined and
     is set to greater than one, then nested parallel regions will be
     enabled.  If not defined, then the value of the 'OMP_NESTED'
     environment variable will be followed if defined.  If neither are
     defined, then if either 'OMP_NUM_THREADS' or 'OMP_PROC_BIND' are
     defined with a list of more than one value, then nested parallel
     regions are enabled.  If none of these are defined, then nested
     parallel regions are disabled by default.

     Nested parallel regions can be enabled or disabled at runtime using
     'omp_set_nested', or by setting the maximum number of nested
     regions with 'omp_set_max_active_levels' to one to disable, or
     above one to enable.

     Note that the 'omp_get_nested' API routine was deprecated in the
     OpenMP specification 5.2 in favor of 'omp_get_max_active_levels'.

_C/C++_:
     _Prototype_:   'int omp_get_nested(void);'

_Fortran_:
     _Interface_:   'logical function omp_get_nested()'

_See also_:
     *note omp_get_max_active_levels::, *note omp_set_nested::, *note
     OMP_MAX_ACTIVE_LEVELS::, *note OMP_NESTED::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.11.


File: libgomp.info,  Node: omp_set_schedule,  Next: omp_get_schedule,  Prev: omp_get_nested,  Up: Thread Team Routines

3.1.11 'omp_set_schedule' - Set the runtime scheduling method
-------------------------------------------------------------

_Description_:
     Sets the runtime scheduling method.  The KIND argument can have the
     value 'omp_sched_static', 'omp_sched_dynamic', 'omp_sched_guided'
     or 'omp_sched_auto'.  Except for 'omp_sched_auto', the chunk size
     is set to the value of CHUNK_SIZE if positive, or to the default
     value if zero or negative.  For 'omp_sched_auto' the CHUNK_SIZE
     argument is ignored.

_C/C++_
     _Prototype_:   'void omp_set_schedule(omp_sched_t kind, int
                    chunk_size);'

_Fortran_:
     _Interface_:   'subroutine omp_set_schedule(kind, chunk_size)'
                    'integer(kind=omp_sched_kind) kind'
                    'integer chunk_size'

_See also_:
     *note omp_get_schedule:: *note OMP_SCHEDULE::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.12.


File: libgomp.info,  Node: omp_get_schedule,  Next: omp_get_teams_thread_limit,  Prev: omp_set_schedule,  Up: Thread Team Routines

3.1.12 'omp_get_schedule' - Obtain the runtime scheduling method
----------------------------------------------------------------

_Description_:
     Obtain the runtime scheduling method.  The KIND argument is set to
     'omp_sched_static', 'omp_sched_dynamic', 'omp_sched_guided' or
     'omp_sched_auto'.  The second argument, CHUNK_SIZE, is set to the
     chunk size.

_C/C++_
     _Prototype_:   'void omp_get_schedule(omp_sched_t *kind, int
                    *chunk_size);'

_Fortran_:
     _Interface_:   'subroutine omp_get_schedule(kind, chunk_size)'
                    'integer(kind=omp_sched_kind) kind'
                    'integer chunk_size'

_See also_:
     *note omp_set_schedule::, *note OMP_SCHEDULE::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.13.


File: libgomp.info,  Node: omp_get_teams_thread_limit,  Next: omp_get_supported_active_levels,  Prev: omp_get_schedule,  Up: Thread Team Routines

3.1.13 'omp_get_teams_thread_limit' - Maximum number of threads imposed by teams
--------------------------------------------------------------------------------

_Description_:
     Return the maximum number of threads that are able to participate
     in each team created by a teams construct.

_C/C++_:
     _Prototype_:   'int omp_get_teams_thread_limit(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_teams_thread_limit()'

_See also_:
     *note omp_set_teams_thread_limit::, *note OMP_TEAMS_THREAD_LIMIT::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.4.6.


File: libgomp.info,  Node: omp_get_supported_active_levels,  Next: omp_set_max_active_levels,  Prev: omp_get_teams_thread_limit,  Up: Thread Team Routines

3.1.14 'omp_get_supported_active_levels' - Maximum number of active regions supported
-------------------------------------------------------------------------------------

_Description_:
     This function returns the maximum number of nested, active parallel
     regions supported by this implementation.

_C/C++_
     _Prototype_:   'int omp_get_supported_active_levels(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_supported_active_levels()'

_See also_:
     *note omp_get_max_active_levels::, *note
     omp_set_max_active_levels::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.2.15.


File: libgomp.info,  Node: omp_set_max_active_levels,  Next: omp_get_max_active_levels,  Prev: omp_get_supported_active_levels,  Up: Thread Team Routines

3.1.15 'omp_set_max_active_levels' - Limits the number of active parallel regions
---------------------------------------------------------------------------------

_Description_:
     This function limits the maximum allowed number of nested, active
     parallel regions.  MAX_LEVELS must be less or equal to the value
     returned by 'omp_get_supported_active_levels'.

_C/C++_
     _Prototype_:   'void omp_set_max_active_levels(int max_levels);'

_Fortran_:
     _Interface_:   'subroutine omp_set_max_active_levels(max_levels)'
                    'integer max_levels'

_See also_:
     *note omp_get_max_active_levels::, *note omp_get_active_level::,
     *note omp_get_supported_active_levels::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.15.


File: libgomp.info,  Node: omp_get_max_active_levels,  Next: omp_get_level,  Prev: omp_set_max_active_levels,  Up: Thread Team Routines

3.1.16 'omp_get_max_active_levels' - Current maximum number of active regions
-----------------------------------------------------------------------------

_Description_:
     This function obtains the maximum allowed number of nested, active
     parallel regions.

_C/C++_
     _Prototype_:   'int omp_get_max_active_levels(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_max_active_levels()'

_See also_:
     *note omp_set_max_active_levels::, *note omp_get_active_level::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.16.


File: libgomp.info,  Node: omp_get_level,  Next: omp_get_ancestor_thread_num,  Prev: omp_get_max_active_levels,  Up: Thread Team Routines

3.1.17 'omp_get_level' - Obtain the current nesting level
---------------------------------------------------------

_Description_:
     This function returns the nesting level for the parallel blocks,
     which enclose the calling call.

_C/C++_
     _Prototype_:   'int omp_get_level(void);'

_Fortran_:
     _Interface_:   'integer function omp_level()'

_See also_:
     *note omp_get_active_level::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.17.


File: libgomp.info,  Node: omp_get_ancestor_thread_num,  Next: omp_get_team_size,  Prev: omp_get_level,  Up: Thread Team Routines

3.1.18 'omp_get_ancestor_thread_num' - Ancestor thread ID
---------------------------------------------------------

_Description_:
     This function returns the thread identification number for the
     given nesting level of the current thread.  For values of LEVEL
     outside zero to 'omp_get_level' -1 is returned; if LEVEL is
     'omp_get_level' the result is identical to 'omp_get_thread_num'.

_C/C++_
     _Prototype_:   'int omp_get_ancestor_thread_num(int level);'

_Fortran_:
     _Interface_:   'integer function omp_get_ancestor_thread_num(level)'
                    'integer level'

_See also_:
     *note omp_get_level::, *note omp_get_thread_num::, *note
     omp_get_team_size::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.18.


File: libgomp.info,  Node: omp_get_team_size,  Next: omp_get_active_level,  Prev: omp_get_ancestor_thread_num,  Up: Thread Team Routines

3.1.19 'omp_get_team_size' - Number of threads in a team
--------------------------------------------------------

_Description_:
     This function returns the number of threads in a thread team to
     which either the current thread or its ancestor belongs.  For
     values of LEVEL outside zero to 'omp_get_level', -1 is returned; if
     LEVEL is zero, 1 is returned, and for 'omp_get_level', the result
     is identical to 'omp_get_num_threads'.

_C/C++_:
     _Prototype_:   'int omp_get_team_size(int level);'

_Fortran_:
     _Interface_:   'integer function omp_get_team_size(level)'
                    'integer level'

_See also_:
     *note omp_get_num_threads::, *note omp_get_level::, *note
     omp_get_ancestor_thread_num::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.19.


File: libgomp.info,  Node: omp_get_active_level,  Prev: omp_get_team_size,  Up: Thread Team Routines

3.1.20 'omp_get_active_level' - Number of parallel regions
----------------------------------------------------------

_Description_:
     This function returns the nesting level for the active parallel
     blocks, which enclose the calling call.

_C/C++_
     _Prototype_:   'int omp_get_active_level(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_active_level()'

_See also_:
     *note omp_get_level::, *note omp_get_max_active_levels::, *note
     omp_set_max_active_levels::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.20.


File: libgomp.info,  Node: Thread Affinity Routines,  Next: Teams Region Routines,  Prev: Thread Team Routines,  Up: Runtime Library Routines

3.2 Thread Affinity Routines
============================

Routines controlling and accessing thread-affinity policies.  They have
C linkage and do not throw exceptions.

* Menu:

* omp_get_proc_bind::           Whether threads may be moved between CPUs


File: libgomp.info,  Node: omp_get_proc_bind,  Up: Thread Affinity Routines

3.2.1 'omp_get_proc_bind' - Whether threads may be moved between CPUs
---------------------------------------------------------------------

_Description_:
     This functions returns the currently active thread affinity policy,
     which is set via 'OMP_PROC_BIND'.  Possible values are
     'omp_proc_bind_false', 'omp_proc_bind_true',
     'omp_proc_bind_primary', 'omp_proc_bind_master',
     'omp_proc_bind_close' and 'omp_proc_bind_spread', where
     'omp_proc_bind_master' is an alias for 'omp_proc_bind_primary'.

_C/C++_:
     _Prototype_:   'omp_proc_bind_t omp_get_proc_bind(void);'

_Fortran_:
     _Interface_:   'integer(kind=omp_proc_bind_kind) function
                    omp_get_proc_bind()'

_See also_:
     *note OMP_PROC_BIND::, *note OMP_PLACES::, *note
     GOMP_CPU_AFFINITY::,

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.22.


File: libgomp.info,  Node: Teams Region Routines,  Next: Tasking Routines,  Prev: Thread Affinity Routines,  Up: Runtime Library Routines

3.3 Teams Region Routines
=========================

Routines controlling the league of teams that are executed in a 'teams'
region.  They have C linkage and do not throw exceptions.

* Menu:

* omp_get_num_teams::           Number of teams
* omp_get_team_num::            Get team number
* omp_set_num_teams::           Set upper teams limit for teams region
* omp_get_max_teams::           Maximum number of teams for teams region
* omp_set_teams_thread_limit::  Set upper thread limit for teams construct
* omp_get_thread_limit::        Maximum number of threads


File: libgomp.info,  Node: omp_get_num_teams,  Next: omp_get_team_num,  Up: Teams Region Routines

3.3.1 'omp_get_num_teams' - Number of teams
-------------------------------------------

_Description_:
     Returns the number of teams in the current team region.

_C/C++_:
     _Prototype_:   'int omp_get_num_teams(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_num_teams()'

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.32.


File: libgomp.info,  Node: omp_get_team_num,  Next: omp_set_num_teams,  Prev: omp_get_num_teams,  Up: Teams Region Routines

3.3.2 'omp_get_team_num' - Get team number
------------------------------------------

_Description_:
     Returns the team number of the calling thread.

_C/C++_:
     _Prototype_:   'int omp_get_team_num(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_team_num()'

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.33.


File: libgomp.info,  Node: omp_set_num_teams,  Next: omp_get_max_teams,  Prev: omp_get_team_num,  Up: Teams Region Routines

3.3.3 'omp_set_num_teams' - Set upper teams limit for teams construct
---------------------------------------------------------------------

_Description_:
     Specifies the upper bound for number of teams created by the teams
     construct which does not specify a 'num_teams' clause.  The
     argument of 'omp_set_num_teams' shall be a positive integer.

_C/C++_:
     _Prototype_:   'void omp_set_num_teams(int num_teams);'

_Fortran_:
     _Interface_:   'subroutine omp_set_num_teams(num_teams)'
                    'integer, intent(in) :: num_teams'

_See also_:
     *note OMP_NUM_TEAMS::, *note omp_get_num_teams::, *note
     omp_get_max_teams::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.4.3.


File: libgomp.info,  Node: omp_get_max_teams,  Next: omp_set_teams_thread_limit,  Prev: omp_set_num_teams,  Up: Teams Region Routines

3.3.4 'omp_get_max_teams' - Maximum number of teams of teams region
-------------------------------------------------------------------

_Description_:
     Return the maximum number of teams used for the teams region that
     does not use the clause 'num_teams'.

_C/C++_:
     _Prototype_:   'int omp_get_max_teams(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_max_teams()'

_See also_:
     *note omp_set_num_teams::, *note omp_get_num_teams::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.4.4.


File: libgomp.info,  Node: omp_set_teams_thread_limit,  Next: omp_get_thread_limit,  Prev: omp_get_max_teams,  Up: Teams Region Routines

3.3.5 'omp_set_teams_thread_limit' - Set upper thread limit for teams construct
-------------------------------------------------------------------------------

_Description_:
     Specifies the upper bound for number of threads that are available
     for each team created by the teams construct which does not specify
     a 'thread_limit' clause.  The argument of
     'omp_set_teams_thread_limit' shall be a positive integer.

_C/C++_:
     _Prototype_:   'void omp_set_teams_thread_limit(int thread_limit);'

_Fortran_:
     _Interface_:   'subroutine omp_set_teams_thread_limit(thread_limit)'
                    'integer, intent(in) :: thread_limit'

_See also_:
     *note OMP_TEAMS_THREAD_LIMIT::, *note omp_get_teams_thread_limit::,
     *note omp_get_thread_limit::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.4.5.


File: libgomp.info,  Node: omp_get_thread_limit,  Prev: omp_set_teams_thread_limit,  Up: Teams Region Routines

3.3.6 'omp_get_thread_limit' - Maximum number of threads
--------------------------------------------------------

_Description_:
     Return the maximum number of threads of the program.

_C/C++_:
     _Prototype_:   'int omp_get_thread_limit(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_thread_limit()'

_See also_:
     *note omp_get_max_threads::, *note OMP_THREAD_LIMIT::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.14.


File: libgomp.info,  Node: Tasking Routines,  Next: Resource Relinquishing Routines,  Prev: Teams Region Routines,  Up: Runtime Library Routines

3.4 Tasking Routines
====================

Routines relating to explicit tasks.  They have C linkage and do not
throw exceptions.

* Menu:

* omp_get_max_task_priority::   Maximum task priority value that can be set
* omp_in_explicit_task::        Whether a given task is an explicit task
* omp_in_final::                Whether in final or included task region


File: libgomp.info,  Node: omp_get_max_task_priority,  Next: omp_in_explicit_task,  Up: Tasking Routines

3.4.1 'omp_get_max_task_priority' - Maximum priority value
----------------------------------------------------------

that can be set for tasks.
_Description_:
     This function obtains the maximum allowed priority number for
     tasks.

_C/C++_
     _Prototype_:   'int omp_get_max_task_priority(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_max_task_priority()'

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.29.


File: libgomp.info,  Node: omp_in_explicit_task,  Next: omp_in_final,  Prev: omp_get_max_task_priority,  Up: Tasking Routines

3.4.2 'omp_in_explicit_task' - Whether a given task is an explicit task
-----------------------------------------------------------------------

_Description_:
     The function returns the EXPLICIT-TASK-VAR ICV; it returns true
     when the encountering task was generated by a task-generating
     construct such as 'target', 'task' or 'taskloop'.  Otherwise, the
     encountering task is in an implicit task region such as generated
     by the implicit or explicit 'parallel' region and
     'omp_in_explicit_task' returns false.

_C/C++_
     _Prototype_:   'int omp_in_explicit_task(void);'

_Fortran_:
     _Interface_:   'logical function omp_in_explicit_task()'

_Reference_:
     OpenMP specification v5.2 (https://www.openmp.org), Section 18.5.2.


File: libgomp.info,  Node: omp_in_final,  Prev: omp_in_explicit_task,  Up: Tasking Routines

3.4.3 'omp_in_final' - Whether in final or included task region
---------------------------------------------------------------

_Description_:
     This function returns 'true' if currently running in a final or
     included task region, 'false' otherwise.  Here, 'true' and 'false'
     represent their language-specific counterparts.

_C/C++_:
     _Prototype_:   'int omp_in_final(void);'

_Fortran_:
     _Interface_:   'logical function omp_in_final()'

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.21.


File: libgomp.info,  Node: Resource Relinquishing Routines,  Next: Device Information Routines,  Prev: Tasking Routines,  Up: Runtime Library Routines

3.5 Resource Relinquishing Routines
===================================

Routines releasing resources used by the OpenMP runtime.  They have C
linkage and do not throw exceptions.

* Menu:

* omp_pause_resource:: Release OpenMP resources on a device
* omp_pause_resource_all:: Release OpenMP resources on all devices


File: libgomp.info,  Node: omp_pause_resource,  Next: omp_pause_resource_all,  Up: Resource Relinquishing Routines

3.5.1 'omp_pause_resource' - Release OpenMP resources on a device
-----------------------------------------------------------------

_Description_:
     Free resources used by the OpenMP program and the runtime library
     on and for the device specified by DEVICE_NUM; on success, zero is
     returned and non-zero otherwise.

     The value of DEVICE_NUM must be a conforming device number.  The
     routine may not be called from within any explicit region and all
     explicit threads that do not bind to the implicit parallel region
     have finalized execution.

_C/C++_:
     _Prototype_:   'int omp_pause_resource(omp_pause_resource_t kind, int
                    device_num);'

_Fortran_:
     _Interface_:   'integer function omp_pause_resource(kind, device_num)'
                    'integer (kind=omp_pause_resource_kind) kind'
                    'integer device_num'

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.2.43.


File: libgomp.info,  Node: omp_pause_resource_all,  Prev: omp_pause_resource,  Up: Resource Relinquishing Routines

3.5.2 'omp_pause_resource_all' - Release OpenMP resources on all devices
------------------------------------------------------------------------

_Description_:
     Free resources used by the OpenMP program and the runtime library
     on all devices, including the host.  On success, zero is returned
     and non-zero otherwise.

     The routine may not be called from within any explicit region and
     all explicit threads that do not bind to the implicit parallel
     region have finalized execution.

_C/C++_:
     _Prototype_:   'int omp_pause_resource(omp_pause_resource_t kind);'

_Fortran_:
     _Interface_:   'integer function omp_pause_resource(kind)'
                    'integer (kind=omp_pause_resource_kind) kind'

_See also_:
     *note omp_pause_resource::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.2.44.


File: libgomp.info,  Node: Device Information Routines,  Next: Device Memory Routines,  Prev: Resource Relinquishing Routines,  Up: Runtime Library Routines

3.6 Device Information Routines
===============================

Routines related to devices available to an OpenMP program.  They have C
linkage and do not throw exceptions.

* Menu:

* omp_get_num_procs::           Number of processors online
* omp_set_default_device::      Set the default device for target regions
* omp_get_default_device::      Get the default device for target regions
* omp_get_num_devices::         Number of target devices
* omp_get_device_num::          Get device that current thread is running on
* omp_is_initial_device::       Whether executing on the host device
* omp_get_initial_device::      Device number of host device


File: libgomp.info,  Node: omp_get_num_procs,  Next: omp_set_default_device,  Up: Device Information Routines

3.6.1 'omp_get_num_procs' - Number of processors online
-------------------------------------------------------

_Description_:
     Returns the number of processors online on that device.

_C/C++_:
     _Prototype_:   'int omp_get_num_procs(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_num_procs()'

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.5.


File: libgomp.info,  Node: omp_set_default_device,  Next: omp_get_default_device,  Prev: omp_get_num_procs,  Up: Device Information Routines

3.6.2 'omp_set_default_device' - Set the default device for target regions
--------------------------------------------------------------------------

_Description_:
     Set the default device for target regions without device clause.
     The argument shall be a nonnegative device number.

_C/C++_:
     _Prototype_:   'void omp_set_default_device(int device_num);'

_Fortran_:
     _Interface_:   'subroutine omp_set_default_device(device_num)'
                    'integer device_num'

_See also_:
     *note OMP_DEFAULT_DEVICE::, *note omp_get_default_device::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.29.


File: libgomp.info,  Node: omp_get_default_device,  Next: omp_get_num_devices,  Prev: omp_set_default_device,  Up: Device Information Routines

3.6.3 'omp_get_default_device' - Get the default device for target regions
--------------------------------------------------------------------------

_Description_:
     Get the default device for target regions without device clause.

_C/C++_:
     _Prototype_:   'int omp_get_default_device(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_default_device()'

_See also_:
     *note OMP_DEFAULT_DEVICE::, *note omp_set_default_device::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.30.


File: libgomp.info,  Node: omp_get_num_devices,  Next: omp_get_device_num,  Prev: omp_get_default_device,  Up: Device Information Routines

3.6.4 'omp_get_num_devices' - Number of target devices
------------------------------------------------------

_Description_:
     Returns the number of target devices.

_C/C++_:
     _Prototype_:   'int omp_get_num_devices(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_num_devices()'

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.31.


File: libgomp.info,  Node: omp_get_device_num,  Next: omp_is_initial_device,  Prev: omp_get_num_devices,  Up: Device Information Routines

3.6.5 'omp_get_device_num' - Return device number of current device
-------------------------------------------------------------------

_Description_:
     This function returns a device number that represents the device
     that the current thread is executing on.  For OpenMP 5.0, this must
     be equal to the value returned by the 'omp_get_initial_device'
     function when called from the host.

_C/C++_
     _Prototype_:   'int omp_get_device_num(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_device_num()'

_See also_:
     *note omp_get_initial_device::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.2.37.


File: libgomp.info,  Node: omp_is_initial_device,  Next: omp_get_initial_device,  Prev: omp_get_device_num,  Up: Device Information Routines

3.6.6 'omp_is_initial_device' - Whether executing on the host device
--------------------------------------------------------------------

_Description_:
     This function returns 'true' if currently running on the host
     device, 'false' otherwise.  Here, 'true' and 'false' represent
     their language-specific counterparts.

_C/C++_:
     _Prototype_:   'int omp_is_initial_device(void);'

_Fortran_:
     _Interface_:   'logical function omp_is_initial_device()'

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.34.


File: libgomp.info,  Node: omp_get_initial_device,  Prev: omp_is_initial_device,  Up: Device Information Routines

3.6.7 'omp_get_initial_device' - Return device number of initial device
-----------------------------------------------------------------------

_Description_:
     This function returns a device number that represents the host
     device.  For OpenMP 5.1, this must be equal to the value returned
     by the 'omp_get_num_devices' function.

_C/C++_
     _Prototype_:   'int omp_get_initial_device(void);'

_Fortran_:
     _Interface_:   'integer function omp_get_initial_device()'

_See also_:
     *note omp_get_num_devices::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.2.35.


File: libgomp.info,  Node: Device Memory Routines,  Next: Lock Routines,  Prev: Device Information Routines,  Up: Runtime Library Routines

3.7 Device Memory Routines
==========================

Routines related to memory allocation and managing corresponding
pointers on devices.  They have C linkage and do not throw exceptions.

* Menu:

* omp_target_alloc:: Allocate device memory
* omp_target_free:: Free device memory
* omp_target_is_present:: Check whether storage is mapped
* omp_target_is_accessible:: Check whether memory is device accessible
* omp_target_memcpy:: Copy data between devices
* omp_target_memcpy_async:: Copy data between devices asynchronously
* omp_target_memcpy_rect:: Copy a subvolume of data between devices
* omp_target_memcpy_rect_async:: Copy a subvolume of data between devices asynchronously
* omp_target_associate_ptr:: Associate a device pointer with a host pointer
* omp_target_disassociate_ptr:: Remove device-host pointer association
* omp_get_mapped_ptr:: Return device pointer to a host pointer


File: libgomp.info,  Node: omp_target_alloc,  Next: omp_target_free,  Up: Device Memory Routines

3.7.1 'omp_target_alloc' - Allocate device memory
-------------------------------------------------

_Description_:
     This routine allocates SIZE bytes of memory in the device
     environment associated with the device number DEVICE_NUM.  If
     successful, a device pointer is returned, otherwise a null pointer.

     In GCC, when the device is the host or the device shares memory
     with the host, the memory is allocated on the host; in that case,
     when SIZE is zero, either NULL or a unique pointer value that can
     later be successfully passed to 'omp_target_free' is returned.
     When the allocation is not performed on the host, a null pointer is
     returned when SIZE is zero; in that case, additionally a diagnostic
     might be printed to standard error (stderr).

     Running this routine in a 'target' region except on the initial
     device is not supported.

_C/C++_
     _Prototype_:   'void *omp_target_alloc(size_t size, int device_num)'

_Fortran_:
     _Interface_:   'type(c_ptr) function omp_target_alloc(size, device_num)
                    bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_int,
                    c_size_t'
                    'integer(c_size_t), value :: size'
                    'integer(c_int), value :: device_num'

_See also_:
     *note omp_target_free::, *note omp_target_associate_ptr::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.1


File: libgomp.info,  Node: omp_target_free,  Next: omp_target_is_present,  Prev: omp_target_alloc,  Up: Device Memory Routines

3.7.2 'omp_target_free' - Free device memory
--------------------------------------------

_Description_:
     This routine frees memory allocated by the 'omp_target_alloc'
     routine.  The DEVICE_PTR argument must be either a null pointer or
     a device pointer returned by 'omp_target_alloc' for the specified
     'device_num'.  The device number DEVICE_NUM must be a conforming
     device number.

     Running this routine in a 'target' region except on the initial
     device is not supported.

_C/C++_
     _Prototype_:   'void omp_target_free(void *device_ptr, int device_num)'

_Fortran_:
     _Interface_:   'subroutine omp_target_free(device_ptr, device_num)
                    bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_int'
                    'type(c_ptr), value :: device_ptr'
                    'integer(c_int), value :: device_num'

_See also_:
     *note omp_target_alloc::, *note omp_target_disassociate_ptr::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.2


File: libgomp.info,  Node: omp_target_is_present,  Next: omp_target_is_accessible,  Prev: omp_target_free,  Up: Device Memory Routines

3.7.3 'omp_target_is_present' - Check whether storage is mapped
---------------------------------------------------------------

_Description_:
     This routine tests whether storage, identified by the host pointer
     PTR is mapped to the device specified by DEVICE_NUM.  If so, it
     returns a nonzero value and otherwise zero.

     In GCC, this includes self mapping such that
     'omp_target_is_present' returns _true_ when DEVICE_NUM specifies
     the host or when the host and the device share memory.  If PTR is a
     null pointer, TRUE is returned and if DEVICE_NUM is an invalid
     device number, FALSE is returned.

     If those conditions do not apply, _true_ is returned if the
     association has been established by an explicit or implicit 'map'
     clause, the 'declare target' directive or a call to the
     'omp_target_associate_ptr' routine.

     Running this routine in a 'target' region except on the initial
     device is not supported.

_C/C++_
     _Prototype_:   'int omp_target_is_present(const void *ptr,'
                    ' int device_num)'

_Fortran_:
     _Interface_:   'integer(c_int) function omp_target_is_present(ptr, &'
                    ' device_num) bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_int'
                    'type(c_ptr), value :: ptr'
                    'integer(c_int), value :: device_num'

_See also_:
     *note omp_target_associate_ptr::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.3


File: libgomp.info,  Node: omp_target_is_accessible,  Next: omp_target_memcpy,  Prev: omp_target_is_present,  Up: Device Memory Routines

3.7.4 'omp_target_is_accessible' - Check whether memory is device accessible
----------------------------------------------------------------------------

_Description_:
     This routine tests whether memory, starting at the address given by
     PTR and extending SIZE bytes, is accessibly on the device specified
     by DEVICE_NUM.  If so, it returns a nonzero value and otherwise
     zero.

     The address given by PTR is interpreted to be in the address space
     of the device and SIZE must be positive.

     Note that GCC's current implementation assumes that PTR is a valid
     host pointer.  Therefore, all addresses given by PTR are assumed to
     be accessible on the initial device.  And, to err on the safe side,
     this memory is only available on a non-host device that can access
     all host memory ([uniform] shared memory access).

     Running this routine in a 'target' region except on the initial
     device is not supported.

_C/C++_
     _Prototype_:   'int omp_target_is_accessible(const void *ptr,'
                    ' size_t size,'
                    ' int device_num)'

_Fortran_:
     _Interface_:   'integer(c_int) function omp_target_is_accessible(ptr,
                    &'
                    ' size, device_num) bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_size_t,
                    c_int'
                    'type(c_ptr), value :: ptr'
                    'integer(c_size_t), value :: size'
                    'integer(c_int), value :: device_num'

_See also_:
     *note omp_target_associate_ptr::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.4


File: libgomp.info,  Node: omp_target_memcpy,  Next: omp_target_memcpy_async,  Prev: omp_target_is_accessible,  Up: Device Memory Routines

3.7.5 'omp_target_memcpy' - Copy data between devices
-----------------------------------------------------

_Description_:
     This routine copies LENGTH of bytes of data from the device
     identified by device number SRC_DEVICE_NUM to device
     DST_DEVICE_NUM.  The data is copied from the source device from the
     address provided by SRC, shifted by the offset of SRC_OFFSET bytes,
     to the destination device's DST address shifted by DST_OFFSET.  The
     routine returns zero on success and non-zero otherwise.

     Running this routine in a 'target' region except on the initial
     device is not supported.

_C/C++_
     _Prototype_:   'int omp_target_memcpy(void *dst,'
                    ' const void *src,'
                    ' size_t length,'
                    ' size_t dst_offset,'
                    ' size_t src_offset,'
                    ' int dst_device_num,'
                    ' int src_device_num)'

_Fortran_:
     _Interface_:   'integer(c_int) function omp_target_memcpy( &'
                    ' dst, src, length, dst_offset, src_offset, &'
                    ' dst_device_num, src_device_num) bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_size_t,
                    c_int'
                    'type(c_ptr), value :: dst, src'
                    'integer(c_size_t), value :: length, dst_offset,
                    src_offset'
                    'integer(c_int), value :: dst_device_num,
                    src_device_num'

_See also_:
     *note omp_target_memcpy_async::, *note omp_target_memcpy_rect::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.5


File: libgomp.info,  Node: omp_target_memcpy_async,  Next: omp_target_memcpy_rect,  Prev: omp_target_memcpy,  Up: Device Memory Routines

3.7.6 'omp_target_memcpy_async' - Copy data between devices asynchronously
--------------------------------------------------------------------------

_Description_:
     This routine copies asynchronously LENGTH of bytes of data from the
     device identified by device number SRC_DEVICE_NUM to device
     DST_DEVICE_NUM.  The data is copied from the source device from the
     address provided by SRC, shifted by the offset of SRC_OFFSET bytes,
     to the destination device's DST address shifted by DST_OFFSET.
     Task dependence is expressed by passing an array of depend objects
     to DEPOBJ_LIST, where the number of array elements is passed as
     DEPOBJ_COUNT; if the count is zero, the DEPOBJ_LIST argument is
     ignored.  The routine returns zero if the copying process has
     successfully been started and non-zero otherwise.

     Running this routine in a 'target' region except on the initial
     device is not supported.

_C/C++_
     _Prototype_:   'int omp_target_memcpy_async(void *dst,'
                    ' const void *src,'
                    ' size_t length,'
                    ' size_t dst_offset,'
                    ' size_t src_offset,'
                    ' int dst_device_num,'
                    ' int src_device_num,'
                    ' int depobj_count,'
                    ' omp_depend_t *depobj_list)'

_Fortran_:
     _Interface_:   'integer(c_int) function omp_target_memcpy_async( &'
                    ' dst, src, length, dst_offset, src_offset, &'
                    ' dst_device_num, src_device_num, &'
                    ' depobj_count, depobj_list) bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_size_t,
                    c_int'
                    'type(c_ptr), value :: dst, src'
                    'integer(c_size_t), value :: length, dst_offset,
                    src_offset'
                    'integer(c_int), value :: dst_device_num,
                    src_device_num, depobj_count'
                    'integer(omp_depend_kind), optional :: depobj_list(*)'

_See also_:
     *note omp_target_memcpy::, *note omp_target_memcpy_rect_async::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.7


File: libgomp.info,  Node: omp_target_memcpy_rect,  Next: omp_target_memcpy_rect_async,  Prev: omp_target_memcpy_async,  Up: Device Memory Routines

3.7.7 'omp_target_memcpy_rect' - Copy a subvolume of data between devices
-------------------------------------------------------------------------

_Description_:
     This routine copies a subvolume of data from the device identified
     by device number SRC_DEVICE_NUM to device DST_DEVICE_NUM.  The
     array has NUM_DIMS dimensions and each array element has a size of
     ELEMENT_SIZE bytes.  The VOLUME array specifies how many elements
     per dimension are copied.  The full sizes of the destination and
     source arrays are given by the DST_DIMENSIONS and SRC_DIMENSIONS
     arguments, respectively.  The offset per dimension to the first
     element to be copied is given by the DST_OFFSET and SRC_OFFSET
     arguments.  The routine returns zero on success and non-zero
     otherwise.

     The OpenMP specification only requires that NUM_DIMS up to three is
     supported.  In order to find implementation-specific maximally
     supported number of dimensions, the routine returns this value when
     invoked with a null pointer to both the DST and SRC arguments.  As
     GCC supports arbitrary dimensions, it returns 'INT_MAX'.

     The device-number arguments must be conforming device numbers, the
     SRC and DST must be either both null pointers or all of the
     following must be fulfilled: ELEMENT_SIZE and NUM_DIMS must be
     positive and the VOLUME, offset and dimension arrays must have at
     least NUM_DIMS dimensions.

     Running this routine in a 'target' region is not supported except
     on the initial device.

_C/C++_
     _Prototype_:   'int omp_target_memcpy_rect(void *dst,'
                    ' const void *src,'
                    ' size_t element_size,'
                    ' int num_dims,'
                    ' const size_t *volume,'
                    ' const size_t *dst_offset,'
                    ' const size_t *src_offset,'
                    ' const size_t *dst_dimensions,'
                    ' const size_t *src_dimensions,'
                    ' int dst_device_num,'
                    ' int src_device_num)'

_Fortran_:
     _Interface_:   'integer(c_int) function omp_target_memcpy_rect( &'
                    ' dst, src, element_size, num_dims, volume, &'
                    ' dst_offset, src_offset, dst_dimensions, &'
                    ' src_dimensions, dst_device_num, src_device_num)
                    bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_size_t,
                    c_int'
                    'type(c_ptr), value :: dst, src'
                    'integer(c_size_t), value :: element_size, dst_offset,
                    src_offset'
                    'integer(c_size_t), value :: volume, dst_dimensions,
                    src_dimensions'
                    'integer(c_int), value :: num_dims, dst_device_num,
                    src_device_num'

_See also_:
     *note omp_target_memcpy_rect_async::, *note omp_target_memcpy::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.6


File: libgomp.info,  Node: omp_target_memcpy_rect_async,  Next: omp_target_associate_ptr,  Prev: omp_target_memcpy_rect,  Up: Device Memory Routines

3.7.8 'omp_target_memcpy_rect_async' - Copy a subvolume of data between devices asynchronously
----------------------------------------------------------------------------------------------

_Description_:
     This routine copies asynchronously a subvolume of data from the
     device identified by device number SRC_DEVICE_NUM to device
     DST_DEVICE_NUM.  The array has NUM_DIMS dimensions and each array
     element has a size of ELEMENT_SIZE bytes.  The VOLUME array
     specifies how many elements per dimension are copied.  The full
     sizes of the destination and source arrays are given by the
     DST_DIMENSIONS and SRC_DIMENSIONS arguments, respectively.  The
     offset per dimension to the first element to be copied is given by
     the DST_OFFSET and SRC_OFFSET arguments.  Task dependence is
     expressed by passing an array of depend objects to DEPOBJ_LIST,
     where the number of array elements is passed as DEPOBJ_COUNT; if
     the count is zero, the DEPOBJ_LIST argument is ignored.  The
     routine returns zero on success and non-zero otherwise.

     The OpenMP specification only requires that NUM_DIMS up to three is
     supported.  In order to find implementation-specific maximally
     supported number of dimensions, the routine returns this value when
     invoked with a null pointer to both the DST and SRC arguments.  As
     GCC supports arbitrary dimensions, it returns 'INT_MAX'.

     The device-number arguments must be conforming device numbers, the
     SRC and DST must be either both null pointers or all of the
     following must be fulfilled: ELEMENT_SIZE and NUM_DIMS must be
     positive and the VOLUME, offset and dimension arrays must have at
     least NUM_DIMS dimensions.

     Running this routine in a 'target' region is not supported except
     on the initial device.

_C/C++_
     _Prototype_:   'int omp_target_memcpy_rect_async(void *dst,'
                    ' const void *src,'
                    ' size_t element_size,'
                    ' int num_dims,'
                    ' const size_t *volume,'
                    ' const size_t *dst_offset,'
                    ' const size_t *src_offset,'
                    ' const size_t *dst_dimensions,'
                    ' const size_t *src_dimensions,'
                    ' int dst_device_num,'
                    ' int src_device_num,'
                    ' int depobj_count,'
                    ' omp_depend_t *depobj_list)'

_Fortran_:
     _Interface_:   'integer(c_int) function omp_target_memcpy_rect_async(
                    &'
                    ' dst, src, element_size, num_dims, volume, &'
                    ' dst_offset, src_offset, dst_dimensions, &'
                    ' src_dimensions, dst_device_num, src_device_num, &'
                    ' depobj_count, depobj_list) bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_size_t,
                    c_int'
                    'type(c_ptr), value :: dst, src'
                    'integer(c_size_t), value :: element_size, dst_offset,
                    src_offset'
                    'integer(c_size_t), value :: volume, dst_dimensions,
                    src_dimensions'
                    'integer(c_int), value :: num_dims, dst_device_num,
                    src_device_num'
                    'integer(c_int), value :: depobj_count'
                    'integer(omp_depend_kind), optional :: depobj_list(*)'

_See also_:
     *note omp_target_memcpy_rect::, *note omp_target_memcpy_async::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.8


File: libgomp.info,  Node: omp_target_associate_ptr,  Next: omp_target_disassociate_ptr,  Prev: omp_target_memcpy_rect_async,  Up: Device Memory Routines

3.7.9 'omp_target_associate_ptr' - Associate a device pointer with a host pointer
---------------------------------------------------------------------------------

_Description_:
     This routine associates storage on the host with storage on a
     device identified by DEVICE_NUM.  The device pointer is usually
     obtained by calling 'omp_target_alloc' or by other means (but not
     by using the 'map' clauses or the 'declare target' directive).  The
     host pointer should point to memory that has a storage size of at
     least SIZE.

     The DEVICE_OFFSET parameter specifies the offset into DEVICE_PTR
     that is used as the base address for the device side of the
     mapping; the storage size should be at least DEVICE_OFFSET plus
     SIZE.

     After the association, the host pointer can be used in a 'map'
     clause and in the 'to' and 'from' clauses of the 'target update'
     directive to transfer data between the associated pointers.  The
     reference count of such associated storage is infinite.  The
     association can be removed by calling 'omp_target_disassociate_ptr'
     which should be done before the lifetime of either storage ends.

     The routine returns nonzero ('EINVAL') when the DEVICE_NUM invalid,
     for when the initial device or the associated device shares memory
     with the host.  'omp_target_associate_ptr' returns zero if HOST_PTR
     points into already associated storage that is fully inside of a
     previously associated memory.  Otherwise, if the association was
     successful zero is returned; if none of the cases above apply,
     nonzero ('EINVAL') is returned.

     The 'omp_target_is_present' routine can be used to test whether
     associated storage for a device pointer exists.

     Running this routine in a 'target' region except on the initial
     device is not supported.

_C/C++_
     _Prototype_:   'int omp_target_associate_ptr(const void *host_ptr,'
                    ' const void *device_ptr,'
                    ' size_t size,'
                    ' size_t device_offset,'
                    ' int device_num)'

_Fortran_:
     _Interface_:   'integer(c_int) function
                    omp_target_associate_ptr(host_ptr, &'
                    ' device_ptr, size, device_offset, device_num) bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_int,
                    c_size_t'
                    'type(c_ptr), value :: host_ptr, device_ptr'
                    'integer(c_size_t), value :: size, device_offset'
                    'integer(c_int), value :: device_num'

_See also_:
     *note omp_target_disassociate_ptr::, *note omp_target_is_present::,
     *note omp_target_alloc::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.9


File: libgomp.info,  Node: omp_target_disassociate_ptr,  Next: omp_get_mapped_ptr,  Prev: omp_target_associate_ptr,  Up: Device Memory Routines

3.7.10 'omp_target_disassociate_ptr' - Remove device-host pointer association
-----------------------------------------------------------------------------

_Description_:
     This routine removes the storage association established by calling
     'omp_target_associate_ptr' and sets the reference count to zero,
     even if 'omp_target_associate_ptr' was invoked multiple times for
     for host pointer 'ptr'.  If applicable, the device memory needs to
     be freed by the user.

     If an associated device storage location for the DEVICE_NUM was
     found and has infinite reference count, the association is removed
     and zero is returned.  In all other cases, nonzero ('EINVAL') is
     returned and no other action is taken.

     Note that passing a host pointer where the association to the
     device pointer was established with the 'declare target' directive
     yields undefined behavior.

     Running this routine in a 'target' region except on the initial
     device is not supported.

_C/C++_
     _Prototype_:   'int omp_target_disassociate_ptr(const void *ptr,'
                    ' int device_num)'

_Fortran_:
     _Interface_:   'integer(c_int) function
                    omp_target_disassociate_ptr(ptr, &'
                    ' device_num) bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_int'
                    'type(c_ptr), value :: ptr'
                    'integer(c_int), value :: device_num'

_See also_:
     *note omp_target_associate_ptr::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.10


File: libgomp.info,  Node: omp_get_mapped_ptr,  Prev: omp_target_disassociate_ptr,  Up: Device Memory Routines

3.7.11 'omp_get_mapped_ptr' - Return device pointer to a host pointer
---------------------------------------------------------------------

_Description_:
     If the device number is refers to the initial device or to a device
     with memory accessible from the host (shared memory), the
     'omp_get_mapped_ptr' routines returns the value of the passed PTR.
     Otherwise, if associated storage to the passed host pointer PTR
     exists on device associated with DEVICE_NUM, it returns that
     pointer.  In all other cases and in cases of an error, a null
     pointer is returned.

     The association of storage location is established either via an
     explicit or implicit 'map' clause, the 'declare target' directive
     or the 'omp_target_associate_ptr' routine.

     Running this routine in a 'target' region except on the initial
     device is not supported.

_C/C++_
     _Prototype_:   'void *omp_get_mapped_ptr(const void *ptr, int
                    device_num);'

_Fortran_:
     _Interface_:   'type(c_ptr) function omp_get_mapped_ptr(ptr,
                    device_num) bind(C)'
                    'use, intrinsic :: iso_c_binding, only: c_ptr, c_int'
                    'type(c_ptr), value :: ptr'
                    'integer(c_int), value :: device_num'

_See also_:
     *note omp_target_associate_ptr::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.8.11


File: libgomp.info,  Node: Lock Routines,  Next: Timing Routines,  Prev: Device Memory Routines,  Up: Runtime Library Routines

3.8 Lock Routines
=================

Initialize, set, test, unset and destroy simple and nested locks.  The
routines have C linkage and do not throw exceptions.

* Menu:

* omp_init_lock::            Initialize simple lock
* omp_init_nest_lock::       Initialize nested lock
* omp_destroy_lock::         Destroy simple lock
* omp_destroy_nest_lock::    Destroy nested lock
* omp_set_lock::             Wait for and set simple lock
* omp_set_nest_lock::        Wait for and set simple lock
* omp_unset_lock::           Unset simple lock
* omp_unset_nest_lock::      Unset nested lock
* omp_test_lock::            Test and set simple lock if available
* omp_test_nest_lock::       Test and set nested lock if available


File: libgomp.info,  Node: omp_init_lock,  Next: omp_init_nest_lock,  Up: Lock Routines

3.8.1 'omp_init_lock' - Initialize simple lock
----------------------------------------------

_Description_:
     Initialize a simple lock.  After initialization, the lock is in an
     unlocked state.

_C/C++_:
     _Prototype_:   'void omp_init_lock(omp_lock_t *lock);'

_Fortran_:
     _Interface_:   'subroutine omp_init_lock(svar)'
                    'integer(omp_lock_kind), intent(out) :: svar'

_See also_:
     *note omp_destroy_lock::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.3.1.


File: libgomp.info,  Node: omp_init_nest_lock,  Next: omp_destroy_lock,  Prev: omp_init_lock,  Up: Lock Routines

3.8.2 'omp_init_nest_lock' - Initialize nested lock
---------------------------------------------------

_Description_:
     Initialize a nested lock.  After initialization, the lock is in an
     unlocked state and the nesting count is set to zero.

_C/C++_:
     _Prototype_:   'void omp_init_nest_lock(omp_nest_lock_t *lock);'

_Fortran_:
     _Interface_:   'subroutine omp_init_nest_lock(nvar)'
                    'integer(omp_nest_lock_kind), intent(out) :: nvar'

_See also_:
     *note omp_destroy_nest_lock::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.3.1.


File: libgomp.info,  Node: omp_destroy_lock,  Next: omp_destroy_nest_lock,  Prev: omp_init_nest_lock,  Up: Lock Routines

3.8.3 'omp_destroy_lock' - Destroy simple lock
----------------------------------------------

_Description_:
     Destroy a simple lock.  In order to be destroyed, a simple lock
     must be in the unlocked state.

_C/C++_:
     _Prototype_:   'void omp_destroy_lock(omp_lock_t *lock);'

_Fortran_:
     _Interface_:   'subroutine omp_destroy_lock(svar)'
                    'integer(omp_lock_kind), intent(inout) :: svar'

_See also_:
     *note omp_init_lock::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.3.3.


File: libgomp.info,  Node: omp_destroy_nest_lock,  Next: omp_set_lock,  Prev: omp_destroy_lock,  Up: Lock Routines

3.8.4 'omp_destroy_nest_lock' - Destroy nested lock
---------------------------------------------------

_Description_:
     Destroy a nested lock.  In order to be destroyed, a nested lock
     must be in the unlocked state and its nesting count must equal
     zero.

_C/C++_:
     _Prototype_:   'void omp_destroy_nest_lock(omp_nest_lock_t *);'

_Fortran_:
     _Interface_:   'subroutine omp_destroy_nest_lock(nvar)'
                    'integer(omp_nest_lock_kind), intent(inout) :: nvar'

_See also_:
     *note omp_init_lock::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.3.3.


File: libgomp.info,  Node: omp_set_lock,  Next: omp_set_nest_lock,  Prev: omp_destroy_nest_lock,  Up: Lock Routines

3.8.5 'omp_set_lock' - Wait for and set simple lock
---------------------------------------------------

_Description_:
     Before setting a simple lock, the lock variable must be initialized
     by 'omp_init_lock'.  The calling thread is blocked until the lock
     is available.  If the lock is already held by the current thread, a
     deadlock occurs.

_C/C++_:
     _Prototype_:   'void omp_set_lock(omp_lock_t *lock);'

_Fortran_:
     _Interface_:   'subroutine omp_set_lock(svar)'
                    'integer(omp_lock_kind), intent(inout) :: svar'

_See also_:
     *note omp_init_lock::, *note omp_test_lock::, *note
     omp_unset_lock::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.3.4.


File: libgomp.info,  Node: omp_set_nest_lock,  Next: omp_unset_lock,  Prev: omp_set_lock,  Up: Lock Routines

3.8.6 'omp_set_nest_lock' - Wait for and set nested lock
--------------------------------------------------------

_Description_:
     Before setting a nested lock, the lock variable must be initialized
     by 'omp_init_nest_lock'.  The calling thread is blocked until the
     lock is available.  If the lock is already held by the current
     thread, the nesting count for the lock is incremented.

_C/C++_:
     _Prototype_:   'void omp_set_nest_lock(omp_nest_lock_t *lock);'

_Fortran_:
     _Interface_:   'subroutine omp_set_nest_lock(nvar)'
                    'integer(omp_nest_lock_kind), intent(inout) :: nvar'

_See also_:
     *note omp_init_nest_lock::, *note omp_unset_nest_lock::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.3.4.


File: libgomp.info,  Node: omp_unset_lock,  Next: omp_unset_nest_lock,  Prev: omp_set_nest_lock,  Up: Lock Routines

3.8.7 'omp_unset_lock' - Unset simple lock
------------------------------------------

_Description_:
     A simple lock about to be unset must have been locked by
     'omp_set_lock' or 'omp_test_lock' before.  In addition, the lock
     must be held by the thread calling 'omp_unset_lock'.  Then, the
     lock becomes unlocked.  If one or more threads attempted to set the
     lock before, one of them is chosen to, again, set the lock to
     itself.

_C/C++_:
     _Prototype_:   'void omp_unset_lock(omp_lock_t *lock);'

_Fortran_:
     _Interface_:   'subroutine omp_unset_lock(svar)'
                    'integer(omp_lock_kind), intent(inout) :: svar'

_See also_:
     *note omp_set_lock::, *note omp_test_lock::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.3.5.


File: libgomp.info,  Node: omp_unset_nest_lock,  Next: omp_test_lock,  Prev: omp_unset_lock,  Up: Lock Routines

3.8.8 'omp_unset_nest_lock' - Unset nested lock
-----------------------------------------------

_Description_:
     A nested lock about to be unset must have been locked by
     'omp_set_nested_lock' or 'omp_test_nested_lock' before.  In
     addition, the lock must be held by the thread calling
     'omp_unset_nested_lock'.  If the nesting count drops to zero, the
     lock becomes unlocked.  If one ore more threads attempted to set
     the lock before, one of them is chosen to, again, set the lock to
     itself.

_C/C++_:
     _Prototype_:   'void omp_unset_nest_lock(omp_nest_lock_t *lock);'

_Fortran_:
     _Interface_:   'subroutine omp_unset_nest_lock(nvar)'
                    'integer(omp_nest_lock_kind), intent(inout) :: nvar'

_See also_:
     *note omp_set_nest_lock::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.3.5.


File: libgomp.info,  Node: omp_test_lock,  Next: omp_test_nest_lock,  Prev: omp_unset_nest_lock,  Up: Lock Routines

3.8.9 'omp_test_lock' - Test and set simple lock if available
-------------------------------------------------------------

_Description_:
     Before setting a simple lock, the lock variable must be initialized
     by 'omp_init_lock'.  Contrary to 'omp_set_lock', 'omp_test_lock'
     does not block if the lock is not available.  This function returns
     'true' upon success, 'false' otherwise.  Here, 'true' and 'false'
     represent their language-specific counterparts.

_C/C++_:
     _Prototype_:   'int omp_test_lock(omp_lock_t *lock);'

_Fortran_:
     _Interface_:   'logical function omp_test_lock(svar)'
                    'integer(omp_lock_kind), intent(inout) :: svar'

_See also_:
     *note omp_init_lock::, *note omp_set_lock::, *note omp_set_lock::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.3.6.


File: libgomp.info,  Node: omp_test_nest_lock,  Prev: omp_test_lock,  Up: Lock Routines

3.8.10 'omp_test_nest_lock' - Test and set nested lock if available
-------------------------------------------------------------------

_Description_:
     Before setting a nested lock, the lock variable must be initialized
     by 'omp_init_nest_lock'.  Contrary to 'omp_set_nest_lock',
     'omp_test_nest_lock' does not block if the lock is not available.
     If the lock is already held by the current thread, the new nesting
     count is returned.  Otherwise, the return value equals zero.

_C/C++_:
     _Prototype_:   'int omp_test_nest_lock(omp_nest_lock_t *lock);'

_Fortran_:
     _Interface_:   'logical function omp_test_nest_lock(nvar)'
                    'integer(omp_nest_lock_kind), intent(inout) :: nvar'

_See also_:
     *note omp_init_lock::, *note omp_set_lock::, *note omp_set_lock::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.3.6.


File: libgomp.info,  Node: Timing Routines,  Next: Event Routine,  Prev: Lock Routines,  Up: Runtime Library Routines

3.9 Timing Routines
===================

Portable, thread-based, wall clock timer.  The routines have C linkage
and do not throw exceptions.

* Menu:

* omp_get_wtick::            Get timer precision.
* omp_get_wtime::            Elapsed wall clock time.


File: libgomp.info,  Node: omp_get_wtick,  Next: omp_get_wtime,  Up: Timing Routines

3.9.1 'omp_get_wtick' - Get timer precision
-------------------------------------------

_Description_:
     Gets the timer precision, i.e., the number of seconds between two
     successive clock ticks.

_C/C++_:
     _Prototype_:   'double omp_get_wtick(void);'

_Fortran_:
     _Interface_:   'double precision function omp_get_wtick()'

_See also_:
     *note omp_get_wtime::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.4.2.


File: libgomp.info,  Node: omp_get_wtime,  Prev: omp_get_wtick,  Up: Timing Routines

3.9.2 'omp_get_wtime' - Elapsed wall clock time
-----------------------------------------------

_Description_:
     Elapsed wall clock time in seconds.  The time is measured per
     thread, no guarantee can be made that two distinct threads measure
     the same time.  Time is measured from some "time in the past",
     which is an arbitrary time guaranteed not to change during the
     execution of the program.

_C/C++_:
     _Prototype_:   'double omp_get_wtime(void);'

_Fortran_:
     _Interface_:   'double precision function omp_get_wtime()'

_See also_:
     *note omp_get_wtick::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 3.4.1.


File: libgomp.info,  Node: Event Routine,  Next: Memory Management Routines,  Prev: Timing Routines,  Up: Runtime Library Routines

3.10 Event Routine
==================

Support for event objects.  The routine has C linkage and do not throw
exceptions.

* Menu:

* omp_fulfill_event::        Fulfill and destroy an OpenMP event.


File: libgomp.info,  Node: omp_fulfill_event,  Up: Event Routine

3.10.1 'omp_fulfill_event' - Fulfill and destroy an OpenMP event
----------------------------------------------------------------

_Description_:
     Fulfill the event associated with the event handle argument.
     Currently, it is only used to fulfill events generated by detach
     clauses on task constructs - the effect of fulfilling the event is
     to allow the task to complete.

     The result of calling 'omp_fulfill_event' with an event handle
     other than that generated by a detach clause is undefined.  Calling
     it with an event handle that has already been fulfilled is also
     undefined.

_C/C++_:
     _Prototype_:   'void omp_fulfill_event(omp_event_handle_t event);'

_Fortran_:
     _Interface_:   'subroutine omp_fulfill_event(event)'
                    'integer (kind=omp_event_handle_kind) :: event'

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.5.1.


File: libgomp.info,  Node: Memory Management Routines,  Next: Environment Display Routine,  Prev: Event Routine,  Up: Runtime Library Routines

3.11 Memory Management Routines
===============================

Routines to manage and allocate memory on the current device.  They have
C linkage and do not throw exceptions.

* Menu:

* omp_init_allocator:: Create an allocator
* omp_destroy_allocator:: Destroy an allocator
* omp_set_default_allocator:: Set the default allocator
* omp_get_default_allocator:: Get the default allocator
* omp_alloc:: Memory allocation with an allocator
* omp_aligned_alloc:: Memory allocation with an allocator and alignment
* omp_free:: Freeing memory allocated with OpenMP routines
* omp_calloc:: Allocate nullified memory with an allocator
* omp_aligned_calloc:: Allocate nullified aligned memory with an allocator
* omp_realloc:: Reallocate memory allocated with OpenMP routines


File: libgomp.info,  Node: omp_init_allocator,  Next: omp_destroy_allocator,  Up: Memory Management Routines

3.11.1 'omp_init_allocator' - Create an allocator
-------------------------------------------------

_Description_:
     Create an allocator that uses the specified memory space and has
     the specified traits; if an allocator that fulfills the
     requirements cannot be created, 'omp_null_allocator' is returned.

     The predefined memory spaces and available traits can be found at
     *note OMP_ALLOCATOR::, where the trait names have to be prefixed by
     'omp_atk_' (e.g.  'omp_atk_pinned') and the named trait values by
     'omp_atv_' (e.g.  'omp_atv_true'); additionally, 'omp_atv_default'
     may be used as trait value to specify that the default value should
     be used.

_C/C++_:
     _Prototype_:   'omp_allocator_handle_t omp_init_allocator('
                    ' omp_memspace_handle_t memspace,'
                    ' int ntraits,'
                    ' const omp_alloctrait_t traits[]);'

_Fortran_:
     _Interface_:   'function omp_init_allocator(memspace, ntraits, traits)'
                    'integer (omp_allocator_handle_kind) ::
                    omp_init_allocator'
                    'integer (omp_memspace_handle_kind), intent(in) ::
                    memspace'
                    'integer, intent(in) :: ntraits'
                    'type (omp_alloctrait), intent(in) :: traits(*)'

_See also_:
     *note OMP_ALLOCATOR::, *note Memory allocation::, *note
     omp_destroy_allocator::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.7.2


File: libgomp.info,  Node: omp_destroy_allocator,  Next: omp_set_default_allocator,  Prev: omp_init_allocator,  Up: Memory Management Routines

3.11.2 'omp_destroy_allocator' - Destroy an allocator
-----------------------------------------------------

_Description_:
     Releases all resources used by a memory allocator, which must not
     represent a predefined memory allocator.  Accessing memory after
     its allocator has been destroyed has unspecified behavior.  Passing
     'omp_null_allocator' to the routine is permitted but has no effect.

_C/C++_:
     _Prototype_:   'void omp_destroy_allocator (omp_allocator_handle_t
                    allocator);'

_Fortran_:
     _Interface_:   'subroutine omp_destroy_allocator(allocator)'
                    'integer (omp_allocator_handle_kind), intent(in) ::
                    allocator'

_See also_:
     *note omp_init_allocator::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.7.3


File: libgomp.info,  Node: omp_set_default_allocator,  Next: omp_get_default_allocator,  Prev: omp_destroy_allocator,  Up: Memory Management Routines

3.11.3 'omp_set_default_allocator' - Set the default allocator
--------------------------------------------------------------

_Description_:
     Sets the default allocator that is used when no allocator has been
     specified in the 'allocate' or 'allocator' clause or if an OpenMP
     memory routine is invoked with the 'omp_null_allocator' allocator.

_C/C++_:
     _Prototype_:   'void omp_set_default_allocator(omp_allocator_handle_t
                    allocator);'

_Fortran_:
     _Interface_:   'subroutine omp_set_default_allocator(allocator)'
                    'integer (omp_allocator_handle_kind), intent(in) ::
                    allocator'

_See also_:
     *note omp_get_default_allocator::, *note omp_init_allocator::,
     *note OMP_ALLOCATOR::, *note Memory allocation::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.7.4


File: libgomp.info,  Node: omp_get_default_allocator,  Next: omp_alloc,  Prev: omp_set_default_allocator,  Up: Memory Management Routines

3.11.4 'omp_get_default_allocator' - Get the default allocator
--------------------------------------------------------------

_Description_:
     The routine returns the default allocator that is used when no
     allocator has been specified in the 'allocate' or 'allocator'
     clause or if an OpenMP memory routine is invoked with the
     'omp_null_allocator' allocator.

_C/C++_:
     _Prototype_:   'omp_allocator_handle_t omp_get_default_allocator();'

_Fortran_:
     _Interface_:   'function omp_get_default_allocator()'
                    'integer (omp_allocator_handle_kind) ::
                    omp_get_default_allocator'

_See also_:
     *note omp_set_default_allocator::, *note OMP_ALLOCATOR::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.7.5


File: libgomp.info,  Node: omp_alloc,  Next: omp_aligned_alloc,  Prev: omp_get_default_allocator,  Up: Memory Management Routines

3.11.5 'omp_alloc' - Memory allocation with an allocator
--------------------------------------------------------

_Description_:
     Allocate memory with the specified allocator, which can either be a
     predefined allocator, an allocator handle or 'omp_null_allocator'.
     If the allocators is 'omp_null_allocator', the allocator specified
     by the DEF-ALLOCATOR-VAR ICV is used.  SIZE must be a nonnegative
     number denoting the number of bytes to be allocated; if SIZE is
     zero, 'omp_alloc' will return a null pointer.  If successful, a
     pointer to the allocated memory is returned, otherwise the
     'fallback' trait of the allocator determines the behavior.  The
     content of the allocated memory is unspecified.

     In 'target' regions, either the 'dynamic_allocators' clause must
     appear on a 'requires' directive in the same compilation unit - or
     the ALLOCATOR argument may only be a constant expression with the
     value of one of the predefined allocators and may not be
     'omp_null_allocator'.

     Memory allocated by 'omp_alloc' must be freed using 'omp_free'.

_C_:
     _Prototype_:   'void* omp_alloc(size_t size,'
                    ' omp_allocator_handle_t allocator)'

_C++_:
     _Prototype_:   'void* omp_alloc(size_t size,'
                    ' omp_allocator_handle_t allocator=omp_null_allocator)'

_Fortran_:
     _Interface_:   'type(c_ptr) function omp_alloc(size, allocator)
                    bind(C)'
                    'use, intrinsic :: iso_c_binding, only : c_ptr,
                    c_size_t'
                    'integer (c_size_t), value :: size'
                    'integer (omp_allocator_handle_kind), value ::
                    allocator'

_See also_:
     *note OMP_ALLOCATOR::, *note Memory allocation::, *note
     omp_set_default_allocator::, *note omp_free::, *note
     omp_init_allocator::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.7.6


File: libgomp.info,  Node: omp_aligned_alloc,  Next: omp_free,  Prev: omp_alloc,  Up: Memory Management Routines

3.11.6 'omp_aligned_alloc' - Memory allocation with an allocator and alignment
------------------------------------------------------------------------------

_Description_:
     Allocate memory with the specified allocator, which can either be a
     predefined allocator, an allocator handle or 'omp_null_allocator'.
     If the allocators is 'omp_null_allocator', the allocator specified
     by the DEF-ALLOCATOR-VAR ICV is used.  ALIGNMENT must be a positive
     power of two and SIZE must be a nonnegative number that is a
     multiple of the alignment and denotes the number of bytes to be
     allocated; if SIZE is zero, 'omp_aligned_alloc' will return a null
     pointer.  The alignment will be at least the maximal value required
     by 'alignment' trait of the allocator and the value of the passed
     ALIGNMENT argument.  If successful, a pointer to the allocated
     memory is returned, otherwise the 'fallback' trait of the allocator
     determines the behavior.  The content of the allocated memory is
     unspecified.

     In 'target' regions, either the 'dynamic_allocators' clause must
     appear on a 'requires' directive in the same compilation unit - or
     the ALLOCATOR argument may only be a constant expression with the
     value of one of the predefined allocators and may not be
     'omp_null_allocator'.

     Memory allocated by 'omp_aligned_alloc' must be freed using
     'omp_free'.

_C_:
     _Prototype_:   'void* omp_aligned_alloc(size_t alignment,'
                    ' size_t size,'
                    ' omp_allocator_handle_t allocator)'

_C++_:
     _Prototype_:   'void* omp_aligned_alloc(size_t alignment,'
                    ' size_t size,'
                    ' omp_allocator_handle_t allocator=omp_null_allocator)'

_Fortran_:
     _Interface_:   'type(c_ptr) function omp_aligned_alloc(alignment, size,
                    allocator) bind(C)'
                    'use, intrinsic :: iso_c_binding, only : c_ptr,
                    c_size_t'
                    'integer (c_size_t), value :: alignment, size'
                    'integer (omp_allocator_handle_kind), value ::
                    allocator'

_See also_:
     *note OMP_ALLOCATOR::, *note Memory allocation::, *note
     omp_set_default_allocator::, *note omp_free::, *note
     omp_init_allocator::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.13.6


File: libgomp.info,  Node: omp_free,  Next: omp_calloc,  Prev: omp_aligned_alloc,  Up: Memory Management Routines

3.11.7 'omp_free' - Freeing memory allocated with OpenMP routines
-----------------------------------------------------------------

_Description_:
     The 'omp_free' routine deallocates memory previously allocated by
     an OpenMP memory-management routine.  The PTR argument must point
     to such memory or be a null pointer; if it is a null pointer, no
     operation is performed.  If specified, the ALLOCATOR argument must
     be either the memory allocator that was used for the allocation or
     'omp_null_allocator'; if it is 'omp_null_allocator', the
     implementation will determine the value automatically.

     Calling 'omp_free' invokes undefined behavior if the memory was
     already deallocated or when the used allocator has already been
     destroyed.

_C_:
     _Prototype_:   'void omp_free(void *ptr,'
                    ' omp_allocator_handle_t allocator)'

_C++_:
     _Prototype_:   'void omp_free(void *ptr,'
                    ' omp_allocator_handle_t allocator=omp_null_allocator)'

_Fortran_:
     _Interface_:   'subroutine omp_free(ptr, allocator) bind(C)'
                    'use, intrinsic :: iso_c_binding, only : c_ptr'
                    'type (c_ptr), value :: ptr'
                    'integer (omp_allocator_handle_kind), value ::
                    allocator'

_See also_:
     *note omp_alloc::, *note omp_aligned_alloc::, *note omp_calloc::,
     *note omp_aligned_calloc::, *note omp_realloc::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.7.7


File: libgomp.info,  Node: omp_calloc,  Next: omp_aligned_calloc,  Prev: omp_free,  Up: Memory Management Routines

3.11.8 'omp_calloc' - Allocate nullified memory with an allocator
-----------------------------------------------------------------

_Description_:
     Allocate zero-initialized memory with the specified allocator,
     which can either be a predefined allocator, an allocator handle or
     'omp_null_allocator'.  If the allocators is 'omp_null_allocator',
     the allocator specified by the DEF-ALLOCATOR-VAR ICV is used.  The
     to-be allocated memory is for an array with NMEMB elements, each
     having a size of SIZE bytes.  Both NMEMB and SIZE must be
     nonnegative numbers; if either of them is zero, 'omp_calloc' will
     return a null pointer.  If successful, a pointer to the
     zero-initialized allocated memory is returned, otherwise the
     'fallback' trait of the allocator determines the behavior.

     In 'target' regions, either the 'dynamic_allocators' clause must
     appear on a 'requires' directive in the same compilation unit - or
     the ALLOCATOR argument may only be a constant expression with the
     value of one of the predefined allocators and may not be
     'omp_null_allocator'.

     Memory allocated by 'omp_calloc' must be freed using 'omp_free'.

_C_:
     _Prototype_:   'void* omp_calloc(size_t nmemb, size_t size,'
                    ' omp_allocator_handle_t allocator)'

_C++_:
     _Prototype_:   'void* omp_calloc(size_t nmemb, size_t size,'
                    ' omp_allocator_handle_t allocator=omp_null_allocator)'

_Fortran_:
     _Interface_:   'type(c_ptr) function omp_calloc(nmemb, size, allocator)
                    bind(C)'
                    'use, intrinsic :: iso_c_binding, only : c_ptr,
                    c_size_t'
                    'integer (c_size_t), value :: nmemb, size'
                    'integer (omp_allocator_handle_kind), value ::
                    allocator'

_See also_:
     *note OMP_ALLOCATOR::, *note Memory allocation::, *note
     omp_set_default_allocator::, *note omp_free::, *note
     omp_init_allocator::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.13.8


File: libgomp.info,  Node: omp_aligned_calloc,  Next: omp_realloc,  Prev: omp_calloc,  Up: Memory Management Routines

3.11.9 'omp_aligned_calloc' - Allocate aligned nullified memory with an allocator
---------------------------------------------------------------------------------

_Description_:
     Allocate zero-initialized memory with the specified allocator,
     which can either be a predefined allocator, an allocator handle or
     'omp_null_allocator'.  If the allocators is 'omp_null_allocator',
     the allocator specified by the DEF-ALLOCATOR-VAR ICV is used.  The
     to-be allocated memory is for an array with NMEMB elements, each
     having a size of SIZE bytes.  Both NMEMB and SIZE must be
     nonnegative numbers; if either of them is zero,
     'omp_aligned_calloc' will return a null pointer.  ALIGNMENT must be
     a positive power of two and SIZE must be a multiple of the
     alignment; the alignment will be at least the maximal value
     required by 'alignment' trait of the allocator and the value of the
     passed ALIGNMENT argument.  If successful, a pointer to the
     zero-initialized allocated memory is returned, otherwise the
     'fallback' trait of the allocator determines the behavior.

     In 'target' regions, either the 'dynamic_allocators' clause must
     appear on a 'requires' directive in the same compilation unit - or
     the ALLOCATOR argument may only be a constant expression with the
     value of one of the predefined allocators and may not be
     'omp_null_allocator'.

     Memory allocated by 'omp_aligned_calloc' must be freed using
     'omp_free'.

_C_:
     _Prototype_:   'void* omp_aligned_calloc(size_t nmemb, size_t size,'
                    ' omp_allocator_handle_t allocator)'

_C++_:
     _Prototype_:   'void* omp_aligned_calloc(size_t nmemb, size_t size,'
                    ' omp_allocator_handle_t allocator=omp_null_allocator)'

_Fortran_:
     _Interface_:   'type(c_ptr) function omp_aligned_calloc(nmemb, size,
                    allocator) bind(C)'
                    'use, intrinsic :: iso_c_binding, only : c_ptr,
                    c_size_t'
                    'integer (c_size_t), value :: nmemb, size'
                    'integer (omp_allocator_handle_kind), value ::
                    allocator'

_See also_:
     *note OMP_ALLOCATOR::, *note Memory allocation::, *note
     omp_set_default_allocator::, *note omp_free::, *note
     omp_init_allocator::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.13.8


File: libgomp.info,  Node: omp_realloc,  Prev: omp_aligned_calloc,  Up: Memory Management Routines

3.11.10 'omp_realloc' - Reallocate memory allocated with OpenMP routines
------------------------------------------------------------------------

_Description_:
     The 'omp_realloc' routine deallocates memory to which PTR points to
     and allocates new memory with the specified ALLOCATOR argument; the
     new memory will have the content of the old memory up to the
     minimum of the old size and the new SIZE, otherwise the content of
     the returned memory is unspecified.  If the new allocator is the
     same as the old one, the routine tries to resize the existing
     memory allocation, returning the same address as PTR if successful.
     PTR must point to memory allocated by an OpenMP memory-management
     routine.

     The ALLOCATOR and FREE_ALLOCATOR arguments must be a predefined
     allocator, an allocator handle or 'omp_null_allocator'.  If
     FREE_ALLOCATOR is 'omp_null_allocator', the implementation
     automatically determines the allocator used for the allocation of
     PTR.  If ALLOCATOR is 'omp_null_allocator' and PTR is not a null
     pointer, the same allocator as 'free_allocator' is used and when
     PTR is a null pointer the allocator specified by the
     DEF-ALLOCATOR-VAR ICV is used.

     The SIZE must be a nonnegative number denoting the number of bytes
     to be allocated; if SIZE is zero, 'omp_realloc' will return free
     the memory and return a null pointer.  When SIZE is nonzero: if
     successful, a pointer to the allocated memory is returned,
     otherwise the 'fallback' trait of the allocator determines the
     behavior.

     In 'target' regions, either the 'dynamic_allocators' clause must
     appear on a 'requires' directive in the same compilation unit - or
     the FREE_ALLOCATOR and ALLOCATOR arguments may only be a constant
     expression with the value of one of the predefined allocators and
     may not be 'omp_null_allocator'.

     Memory allocated by 'omp_realloc' must be freed using 'omp_free'.
     Calling 'omp_free' invokes undefined behavior if the memory was
     already deallocated or when the used allocator has already been
     destroyed.

_C_:
     _Prototype_:   'void* omp_realloc(void *ptr, size_t size,'
                    ' omp_allocator_handle_t allocator,'
                    ' omp_allocator_handle_t free_allocator)'

_C++_:
     _Prototype_:   'void* omp_realloc(void *ptr, size_t size,'
                    ' omp_allocator_handle_t allocator=omp_null_allocator,'
                    ' omp_allocator_handle_t
                    free_allocator=omp_null_allocator)'

_Fortran_:
     _Interface_:   'type(c_ptr) function omp_realloc(ptr, size, allocator,
                    free_allocator) bind(C)'
                    'use, intrinsic :: iso_c_binding, only : c_ptr,
                    c_size_t'
                    'type(C_ptr), value :: ptr'
                    'integer (c_size_t), value :: size'
                    'integer (omp_allocator_handle_kind), value ::
                    allocator, free_allocator'

_See also_:
     *note OMP_ALLOCATOR::, *note Memory allocation::, *note
     omp_set_default_allocator::, *note omp_free::, *note
     omp_init_allocator::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 3.7.9


File: libgomp.info,  Node: Environment Display Routine,  Prev: Memory Management Routines,  Up: Runtime Library Routines

3.12 Environment Display Routine
================================

Routine to display the OpenMP version number and the initial value of
ICVs.  It has C linkage and does not throw exceptions.

* Menu:

* omp_display_env:: print the initial ICV values


File: libgomp.info,  Node: omp_display_env,  Up: Environment Display Routine

3.12.1 'omp_display_env' - print the initial ICV values
-------------------------------------------------------

_Description_:
     Each time this routine is invoked, the OpenMP version number and
     initial value of internal control variables (ICVs) is printed on
     'stderr'.  The displayed values are those at startup after
     evaluating the environment variables; later calls to API routines
     or clauses used in enclosing constructs do not affect the output.

     If the VERBOSE argument is 'false', only the OpenMP version and
     standard OpenMP ICVs are shown; if it is 'true', additionally, the
     GCC-specific ICVs are shown.

     The output consists of multiple lines and starts with 'OPENMP
     DISPLAY ENVIRONMENT BEGIN' followed by the name-value lines and
     ends with 'OPENMP DISPLAY ENVIRONMENT END'.  The NAME is followed
     by an equal sign and the VALUE is enclosed in single quotes.

     The first line has as NAME either '_OPENMP' or 'openmp_version' and
     shows as value the supported OpenMP version number (4-digit year,
     2-digit month) of the implementation, matching the value of the
     '_OPENMP' macro and, in Fortran, the named constant
     'openmp_version'.

     In each of the succeeding lines, the NAME matches the
     environment-variable name of an ICV and shows its value.  Those
     line are might be prefixed by pair of brackets and a space, where
     the brackets enclose a comma-separated list of devices to which the
     ICV-value combination applies to; the value can either be a numeric
     device number or an abstract name denoting all devices ('all'), the
     initial host device ('host') or all devices but the host
     ('device').  Note that the same ICV might be printed multiple times
     for multiple devices, even if all have the same value.

     The effect when invoked from within a 'target' region is
     unspecified.

_C/C++_:
     _Prototype_:   'void omp_display_env(int verbose)'

_Fortran_:
     _Interface_:   'subroutine omp_display_env(vebose)'
                    'logical, intent(in) :: verbose'

_Example_:
     Note that the GCC-specific ICVs, such as the shown
     'GOMP_SPINCOUNT', are only printed when VARBOSE set to 'true'.

          OPENMP DISPLAY ENVIRONMENT BEGIN
            _OPENMP = '201511'
            [host] OMP_DYNAMIC = 'FALSE'
            [host] OMP_NESTED = 'FALSE'
            [all] OMP_CANCELLATION = 'FALSE'
            ...
            [host] GOMP_SPINCOUNT = '300000'
          OPENMP DISPLAY ENVIRONMENT END

_See also_:
     *note OMP_DISPLAY_ENV::, *note Environment Variables::, *note
     Implementation-defined ICV Initialization::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 3.15


File: libgomp.info,  Node: Environment Variables,  Next: Enabling OpenACC,  Prev: Runtime Library Routines,  Up: Top

4 OpenMP Environment Variables
******************************

The environment variables which beginning with 'OMP_' are defined by
section 4 of the OpenMP specification in version 4.5 or in a later
version of the specification, while those beginning with 'GOMP_' are GNU
extensions.  Most 'OMP_' environment variables have an associated
internal control variable (ICV).

   For any OpenMP environment variable that sets an ICV and is neither
'OMP_DEFAULT_DEVICE' nor has global ICV scope, associated
device-specific environment variables exist.  For them, the environment
variable without suffix affects the host.  The suffix '_DEV_' followed
by a non-negative device number less that the number of available
devices sets the ICV for the corresponding device.  The suffix '_DEV'
sets the ICV of all non-host devices for which a device-specific
corresponding environment variable has not been set while the '_ALL'
suffix sets the ICV of all host and non-host devices for which a more
specific corresponding environment variable is not set.

* Menu:

* OMP_ALLOCATOR::           Set the default allocator
* OMP_AFFINITY_FORMAT::     Set the format string used for affinity display
* OMP_CANCELLATION::        Set whether cancellation is activated
* OMP_DISPLAY_AFFINITY::    Display thread affinity information
* OMP_DISPLAY_ENV::         Show OpenMP version and environment variables
* OMP_DEFAULT_DEVICE::      Set the device used in target regions
* OMP_DYNAMIC::             Dynamic adjustment of threads
* OMP_MAX_ACTIVE_LEVELS::   Set the maximum number of nested parallel regions
* OMP_MAX_TASK_PRIORITY::   Set the maximum task priority value
* OMP_NESTED::              Nested parallel regions
* OMP_NUM_TEAMS::           Specifies the number of teams to use by teams region
* OMP_NUM_THREADS::         Specifies the number of threads to use
* OMP_PROC_BIND::           Whether threads may be moved between CPUs
* OMP_PLACES::              Specifies on which CPUs the threads should be placed
* OMP_STACKSIZE::           Set default thread stack size
* OMP_SCHEDULE::            How threads are scheduled
* OMP_TARGET_OFFLOAD::      Controls offloading behavior
* OMP_TEAMS_THREAD_LIMIT::  Set the maximum number of threads imposed by teams
* OMP_THREAD_LIMIT::        Set the maximum number of threads
* OMP_WAIT_POLICY::         How waiting threads are handled
* GOMP_CPU_AFFINITY::       Bind threads to specific CPUs
* GOMP_DEBUG::              Enable debugging output
* GOMP_STACKSIZE::          Set default thread stack size
* GOMP_SPINCOUNT::          Set the busy-wait spin count
* GOMP_RTEMS_THREAD_POOLS:: Set the RTEMS specific thread pools


File: libgomp.info,  Node: OMP_ALLOCATOR,  Next: OMP_AFFINITY_FORMAT,  Up: Environment Variables

4.1 'OMP_ALLOCATOR' - Set the default allocator
===============================================

_ICV:_ DEF-ALLOCATOR-VAR
_Scope:_ data environment
_Description_:
     Sets the default allocator that is used when no allocator has been
     specified in the 'allocate' or 'allocator' clause or if an OpenMP
     memory routine is invoked with the 'omp_null_allocator' allocator.
     If unset, 'omp_default_mem_alloc' is used.

     The value can either be a predefined allocator or a predefined
     memory space or a predefined memory space followed by a colon and a
     comma-separated list of memory trait and value pairs, separated by
     '='.

     Note: The corresponding device environment variables are currently
     not supported.  Therefore, the non-host DEF-ALLOCATOR-VAR ICVs are
     always initialized to 'omp_default_mem_alloc'.  However, on all
     devices, the 'omp_set_default_allocator' API routine can be used to
     change value.

     Predefined allocators            Associated predefined memory
                                      spaces
     ------------------------------------------------------------------
     omp_default_mem_alloc            omp_default_mem_space
     omp_large_cap_mem_alloc          omp_large_cap_mem_space
     omp_const_mem_alloc              omp_const_mem_space
     omp_high_bw_mem_alloc            omp_high_bw_mem_space
     omp_low_lat_mem_alloc            omp_low_lat_mem_space
     omp_cgroup_mem_alloc             omp_low_lat_mem_space
                                      (implementation defined)
     omp_pteam_mem_alloc              omp_low_lat_mem_space
                                      (implementation defined)
     omp_thread_mem_alloc             omp_low_lat_mem_space
                                      (implementation defined)

     The predefined allocators use the default values for the traits, as
     listed below.  Except that the last three allocators have the
     'access' trait set to 'cgroup', 'pteam', and 'thread',
     respectively.

     Trait              Allowed values                Default value
     --------------------------------------------------------------------
     'sync_hint'        'contended', 'uncontended',   'contended'
                        'serialized', 'private'
     'alignment'        Positive integer being a      1 byte
                        power of two
     'access'           'all', 'cgroup', 'pteam',     'all'
                        'thread'
     'pool_size'        Positive integer              See
                                                      *note Memory allocation::
     'fallback'         'default_mem_fb',             See below
                        'null_fb', 'abort_fb',
                        'allocator_fb'
     'fb_data'          _unsupported as it needs an   (none)
                        allocator handle_
     'pinned'           'true', 'false'               'false'
     'partition'        'environment', 'nearest',     'environment'
                        'blocked', 'interleaved'

     For the 'fallback' trait, the default value is 'null_fb' for the
     'omp_default_mem_alloc' allocator and any allocator that is
     associated with device memory; for all other allocators, it is
     'default_mem_fb' by default.

     Examples:
          OMP_ALLOCATOR=omp_high_bw_mem_alloc
          OMP_ALLOCATOR=omp_large_cap_mem_space
          OMP_ALLOCATOR=omp_low_lat_mem_space:pinned=true,partition=nearest

_See also_:
     *note Memory allocation::, *note omp_get_default_allocator::, *note
     omp_set_default_allocator::, *note Offload-Target Specifics::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 6.21


File: libgomp.info,  Node: OMP_AFFINITY_FORMAT,  Next: OMP_CANCELLATION,  Prev: OMP_ALLOCATOR,  Up: Environment Variables

4.2 'OMP_AFFINITY_FORMAT' - Set the format string used for affinity display
===========================================================================

_ICV:_ AFFINITY-FORMAT-VAR
_Scope:_ device
_Description_:
     Sets the format string used when displaying OpenMP thread affinity
     information.  Special values are output using '%' followed by an
     optional size specification and then either the single-character
     field type or its long name enclosed in curly braces; using '%%'
     displays a literal percent.  The size specification consists of an
     optional '0.' or '.' followed by a positive integer, specifying the
     minimal width of the output.  With '0.' and numerical values, the
     output is padded with zeros on the left; with '.', the output is
     padded by spaces on the left; otherwise, the output is padded by
     spaces on the right.  If unset, the value is "'level %L thread %i
     affinity %A'".

     Supported field types are:

     t       team_num           value returned by 'omp_get_team_num'
     T       num_teams          value returned by 'omp_get_num_teams'
     L       nesting_level      value returned by 'omp_get_level'
     n       thread_num         value returned by 'omp_get_thread_num'
     N       num_threads        value returned by 'omp_get_num_threads'
     a       ancestor_tnum      value returned by
                                'omp_get_ancestor_thread_num(omp_get_level()-1)'
     H       host               name of the host that executes the thread
     P       process_id         process identifier
     i       native_thread_id   native thread identifier
     A       thread_affinity    comma separated list of integer values or
                                ranges, representing the processors on
                                which a process might execute, subject to
                                affinity mechanisms

     For instance, after setting

          OMP_AFFINITY_FORMAT="%0.2a!%n!%.4L!%N;%.2t;%0.2T;%{team_num};%{num_teams};%A"

     with either 'OMP_DISPLAY_AFFINITY' being set or when calling
     'omp_display_affinity' with 'NULL' or an empty string, the program
     might display the following:

          00!0!   1!4; 0;01;0;1;0-11
          00!3!   1!4; 0;01;0;1;0-11
          00!2!   1!4; 0;01;0;1;0-11
          00!1!   1!4; 0;01;0;1;0-11

_See also_:
     *note OMP_DISPLAY_AFFINITY::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 6.14


File: libgomp.info,  Node: OMP_CANCELLATION,  Next: OMP_DISPLAY_AFFINITY,  Prev: OMP_AFFINITY_FORMAT,  Up: Environment Variables

4.3 'OMP_CANCELLATION' - Set whether cancellation is activated
==============================================================

_ICV:_ CANCEL-VAR
_Scope:_ global
_Description_:
     If set to 'TRUE', the cancellation is activated.  If set to 'FALSE'
     or if unset, cancellation is disabled and the 'cancel' construct is
     ignored.

_See also_:
     *note omp_get_cancellation::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.11


File: libgomp.info,  Node: OMP_DISPLAY_AFFINITY,  Next: OMP_DISPLAY_ENV,  Prev: OMP_CANCELLATION,  Up: Environment Variables

4.4 'OMP_DISPLAY_AFFINITY' - Display thread affinity information
================================================================

_ICV:_ DISPLAY-AFFINITY-VAR
_Scope:_ global
_Description_:
     If set to 'FALSE' or if unset, affinity displaying is disabled.  If
     set to 'TRUE', the runtime displays affinity information about
     OpenMP threads in a parallel region upon entering the region and
     every time any change occurs.

_See also_:
     *note OMP_AFFINITY_FORMAT::

_Reference_:
     OpenMP specification v5.0 (https://www.openmp.org), Section 6.13


File: libgomp.info,  Node: OMP_DISPLAY_ENV,  Next: OMP_DEFAULT_DEVICE,  Prev: OMP_DISPLAY_AFFINITY,  Up: Environment Variables

4.5 'OMP_DISPLAY_ENV' - Show OpenMP version and environment variables
=====================================================================

_ICV:_ none
_Scope:_ not applicable
_Description_:
     If set to 'TRUE', the runtime displays the same information to
     'stderr' as shown by the 'omp_display_env' routine invoked with
     VERBOSE argument set to 'false'.  If set to 'VERBOSE', the same
     information is shown as invoking the routine with VERBOSE set to
     'true'.  If unset or set to 'FALSE', this information is not shown.
     The result for any other value is unspecified.

_See also_:
     *note omp_display_env::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.12


File: libgomp.info,  Node: OMP_DEFAULT_DEVICE,  Next: OMP_DYNAMIC,  Prev: OMP_DISPLAY_ENV,  Up: Environment Variables

4.6 'OMP_DEFAULT_DEVICE' - Set the device used in target regions
================================================================

_ICV:_ DEFAULT-DEVICE-VAR
_Scope:_ data environment
_Description_:
     Set to choose the device which is used in a 'target' region, unless
     the value is overridden by 'omp_set_default_device' or by a
     'device' clause.  The value shall be the nonnegative device number.
     If no device with the given device number exists, the code is
     executed on the host.  If unset, 'OMP_TARGET_OFFLOAD' is
     'mandatory' and no non-host devices are available, it is set to
     'omp_invalid_device'.  Otherwise, if unset, device number 0 is
     used.

_See also_:
     *note omp_get_default_device::, *note omp_set_default_device::,
     *note OMP_TARGET_OFFLOAD::

_Reference_:
     OpenMP specification v5.2 (https://www.openmp.org), Section 21.2.7


File: libgomp.info,  Node: OMP_DYNAMIC,  Next: OMP_MAX_ACTIVE_LEVELS,  Prev: OMP_DEFAULT_DEVICE,  Up: Environment Variables

4.7 'OMP_DYNAMIC' - Dynamic adjustment of threads
=================================================

_ICV:_ DYN-VAR
_Scope:_ global
_Description_:
     Enable or disable the dynamic adjustment of the number of threads
     within a team.  The value of this environment variable shall be
     'TRUE' or 'FALSE'.  If undefined, dynamic adjustment is disabled by
     default.

_See also_:
     *note omp_set_dynamic::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.3


File: libgomp.info,  Node: OMP_MAX_ACTIVE_LEVELS,  Next: OMP_MAX_TASK_PRIORITY,  Prev: OMP_DYNAMIC,  Up: Environment Variables

4.8 'OMP_MAX_ACTIVE_LEVELS' - Set the maximum number of nested parallel regions
===============================================================================

_ICV:_ MAX-ACTIVE-LEVELS-VAR
_Scope:_ data environment
_Description_:
     Specifies the initial value for the maximum number of nested
     parallel regions.  The value of this variable shall be a positive
     integer.  If undefined, then if 'OMP_NESTED' is defined and set to
     true, or if 'OMP_NUM_THREADS' or 'OMP_PROC_BIND' are defined and
     set to a list with more than one item, the maximum number of nested
     parallel regions is initialized to the largest number supported,
     otherwise it is set to one.

_See also_:
     *note omp_set_max_active_levels::, *note OMP_NESTED::, *note
     OMP_PROC_BIND::, *note OMP_NUM_THREADS::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.9


File: libgomp.info,  Node: OMP_MAX_TASK_PRIORITY,  Next: OMP_NESTED,  Prev: OMP_MAX_ACTIVE_LEVELS,  Up: Environment Variables

4.9 'OMP_MAX_TASK_PRIORITY' - Set the maximum priority
======================================================

number that can be set for a task.
_ICV:_ MAX-TASK-PRIORITY-VAR
_Scope:_ global
_Description_:
     Specifies the initial value for the maximum priority value that can
     be set for a task.  The value of this variable shall be a
     non-negative integer, and zero is allowed.  If undefined, the
     default priority is 0.

_See also_:
     *note omp_get_max_task_priority::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.14


File: libgomp.info,  Node: OMP_NESTED,  Next: OMP_NUM_TEAMS,  Prev: OMP_MAX_TASK_PRIORITY,  Up: Environment Variables

4.10 'OMP_NESTED' - Nested parallel regions
===========================================

_ICV:_ MAX-ACTIVE-LEVELS-VAR
_Scope:_ data environment
_Description_:
     Enable or disable nested parallel regions, i.e., whether team
     members are allowed to create new teams.  The value of this
     environment variable shall be 'TRUE' or 'FALSE'.  If set to 'TRUE',
     the number of maximum active nested regions supported is by default
     set to the maximum supported, otherwise it is set to one.  If
     'OMP_MAX_ACTIVE_LEVELS' is defined, its setting overrides this
     setting.  If both are undefined, nested parallel regions are
     enabled if 'OMP_NUM_THREADS' or 'OMP_PROC_BINDS' are defined to a
     list with more than one item, otherwise they are disabled by
     default.

     Note that the 'OMP_NESTED' environment variable was deprecated in
     the OpenMP specification 5.2 in favor of 'OMP_MAX_ACTIVE_LEVELS'.

_See also_:
     *note omp_set_max_active_levels::, *note omp_set_nested::, *note
     OMP_MAX_ACTIVE_LEVELS::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.6


File: libgomp.info,  Node: OMP_NUM_TEAMS,  Next: OMP_NUM_THREADS,  Prev: OMP_NESTED,  Up: Environment Variables

4.11 'OMP_NUM_TEAMS' - Specifies the number of teams to use by teams region
===========================================================================

_ICV:_ NTEAMS-VAR
_Scope:_ device
_Description_:
     Specifies the upper bound for number of teams to use in teams
     regions without explicit 'num_teams' clause.  The value of this
     variable shall be a positive integer.  If undefined it defaults to
     0 which means implementation defined upper bound.

_See also_:
     *note omp_set_num_teams::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 6.23


File: libgomp.info,  Node: OMP_NUM_THREADS,  Next: OMP_PROC_BIND,  Prev: OMP_NUM_TEAMS,  Up: Environment Variables

4.12 'OMP_NUM_THREADS' - Specifies the number of threads to use
===============================================================

_ICV:_ NTHREADS-VAR
_Scope:_ data environment
_Description_:
     Specifies the default number of threads to use in parallel regions.
     The value of this variable shall be a comma-separated list of
     positive integers; the value specifies the number of threads to use
     for the corresponding nested level.  Specifying more than one item
     in the list automatically enables nesting by default.  If undefined
     one thread per CPU is used.

     When a list with more than value is specified, it also affects the
     MAX-ACTIVE-LEVELS-VAR ICV as described in *note
     OMP_MAX_ACTIVE_LEVELS::.

_See also_:
     *note omp_set_num_threads::, *note OMP_MAX_ACTIVE_LEVELS::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.2


File: libgomp.info,  Node: OMP_PROC_BIND,  Next: OMP_PLACES,  Prev: OMP_NUM_THREADS,  Up: Environment Variables

4.13 'OMP_PROC_BIND' - Whether threads may be moved between CPUs
================================================================

_ICV:_ BIND-VAR
_Scope:_ data environment
_Description_:
     Specifies whether threads may be moved between processors.  If set
     to 'TRUE', OpenMP threads should not be moved; if set to 'FALSE'
     they may be moved.  Alternatively, a comma separated list with the
     values 'PRIMARY', 'MASTER', 'CLOSE' and 'SPREAD' can be used to
     specify the thread affinity policy for the corresponding nesting
     level.  With 'PRIMARY' and 'MASTER' the worker threads are in the
     same place partition as the primary thread.  With 'CLOSE' those are
     kept close to the primary thread in contiguous place partitions.
     And with 'SPREAD' a sparse distribution across the place partitions
     is used.  Specifying more than one item in the list automatically
     enables nesting by default.

     When a list is specified, it also affects the MAX-ACTIVE-LEVELS-VAR
     ICV as described in *note OMP_MAX_ACTIVE_LEVELS::.

     When undefined, 'OMP_PROC_BIND' defaults to 'TRUE' when
     'OMP_PLACES' or 'GOMP_CPU_AFFINITY' is set and 'FALSE' otherwise.

_See also_:
     *note omp_get_proc_bind::, *note GOMP_CPU_AFFINITY::, *note
     OMP_PLACES::, *note OMP_MAX_ACTIVE_LEVELS::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.4


File: libgomp.info,  Node: OMP_PLACES,  Next: OMP_STACKSIZE,  Prev: OMP_PROC_BIND,  Up: Environment Variables

4.14 'OMP_PLACES' - Specifies on which CPUs the threads should be placed
========================================================================

_ICV:_ PLACE-PARTITION-VAR
_Scope:_ implicit tasks
_Description_:
     The thread placement can be either specified using an abstract name
     or by an explicit list of the places.  The abstract names
     'threads', 'cores', 'sockets', 'll_caches' and 'numa_domains' can
     be optionally followed by a positive number in parentheses, which
     denotes the how many places shall be created.  With 'threads' each
     place corresponds to a single hardware thread; 'cores' to a single
     core with the corresponding number of hardware threads; with
     'sockets' the place corresponds to a single socket; with
     'll_caches' to a set of cores that shares the last level cache on
     the device; and 'numa_domains' to a set of cores for which their
     closest memory on the device is the same memory and at a similar
     distance from the cores.  The resulting placement can be shown by
     setting the 'OMP_DISPLAY_ENV' environment variable.

     Alternatively, the placement can be specified explicitly as
     comma-separated list of places.  A place is specified by set of
     nonnegative numbers in curly braces, denoting the hardware threads.
     The curly braces can be omitted when only a single number has been
     specified.  The hardware threads belonging to a place can either be
     specified as comma-separated list of nonnegative thread numbers or
     using an interval.  Multiple places can also be either specified by
     a comma-separated list of places or by an interval.  To specify an
     interval, a colon followed by the count is placed after the
     hardware thread number or the place.  Optionally, the length can be
     followed by a colon and the stride number - otherwise a unit stride
     is assumed.  Placing an exclamation mark ('!') directly before a
     curly brace or numbers inside the curly braces (excluding
     intervals) excludes those hardware threads.

     For instance, the following specifies the same places list:
     '"{0,1,2}, {3,4,6}, {7,8,9}, {10,11,12}"'; '"{0:3}, {3:3}, {7:3},
     {10:3}"'; and '"{0:2}:4:3"'.

     If 'OMP_PLACES' and 'GOMP_CPU_AFFINITY' are unset and
     'OMP_PROC_BIND' is either unset or 'false', threads may be moved
     between CPUs following no placement policy.

_See also_:
     *note OMP_PROC_BIND::, *note GOMP_CPU_AFFINITY::, *note
     omp_get_proc_bind::, *note OMP_DISPLAY_ENV::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.5


File: libgomp.info,  Node: OMP_STACKSIZE,  Next: OMP_SCHEDULE,  Prev: OMP_PLACES,  Up: Environment Variables

4.15 'OMP_STACKSIZE' - Set default thread stack size
====================================================

_ICV:_ STACKSIZE-VAR
_Scope:_ device
_Description_:
     Set the default thread stack size in kilobytes, unless the number
     is suffixed by 'B', 'K', 'M' or 'G', in which case the size is,
     respectively, in bytes, kilobytes, megabytes or gigabytes.  This is
     different from 'pthread_attr_setstacksize' which gets the number of
     bytes as an argument.  If the stack size cannot be set due to
     system constraints, an error is reported and the initial stack size
     is left unchanged.  If undefined, the stack size is system
     dependent.

_See also_:
     *note GOMP_STACKSIZE::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.7


File: libgomp.info,  Node: OMP_SCHEDULE,  Next: OMP_TARGET_OFFLOAD,  Prev: OMP_STACKSIZE,  Up: Environment Variables

4.16 'OMP_SCHEDULE' - How threads are scheduled
===============================================

_ICV:_ RUN-SCHED-VAR
_Scope:_ data environment
_Description_:
     Allows to specify 'schedule type' and 'chunk size'.  The value of
     the variable shall have the form: 'type[,chunk]' where 'type' is
     one of 'static', 'dynamic', 'guided' or 'auto' The optional 'chunk'
     size shall be a positive integer.  If undefined, dynamic scheduling
     and a chunk size of 1 is used.

_See also_:
     *note omp_set_schedule::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Sections
     2.7.1.1 and 4.1


File: libgomp.info,  Node: OMP_TARGET_OFFLOAD,  Next: OMP_TEAMS_THREAD_LIMIT,  Prev: OMP_SCHEDULE,  Up: Environment Variables

4.17 'OMP_TARGET_OFFLOAD' - Controls offloading behavior
========================================================

_ICV:_ TARGET-OFFLOAD-VAR
_Scope:_ global
_Description_:
     Specifies the behavior with regard to offloading code to a device.
     This variable can be set to one of three values - 'MANDATORY',
     'DISABLED' or 'DEFAULT'.

     If set to 'MANDATORY', the program terminates with an error if any
     device construct or device memory routine uses a device that is
     unavailable or not supported by the implementation, or uses a
     non-conforming device number.  If set to 'DISABLED', then
     offloading is disabled and all code runs on the host.  If set to
     'DEFAULT', the program tries offloading to the device first, then
     falls back to running code on the host if it cannot.

     If undefined, then the program behaves as if 'DEFAULT' was set.

     Note: Even with 'MANDATORY', no run-time termination is performed
     when the device number in a 'device' clause or argument to a device
     memory routine is for host, which includes using the device number
     in the DEFAULT-DEVICE-VAR ICV. However, the initial value of the
     DEFAULT-DEVICE-VAR ICV is affected by 'MANDATORY'.

_See also_:
     *note OMP_DEFAULT_DEVICE::

_Reference_:
     OpenMP specification v5.2 (https://www.openmp.org), Section 21.2.8


File: libgomp.info,  Node: OMP_TEAMS_THREAD_LIMIT,  Next: OMP_THREAD_LIMIT,  Prev: OMP_TARGET_OFFLOAD,  Up: Environment Variables

4.18 'OMP_TEAMS_THREAD_LIMIT' - Set the maximum number of threads imposed by teams
==================================================================================

_ICV:_ TEAMS-THREAD-LIMIT-VAR
_Scope:_ device
_Description_:
     Specifies an upper bound for the number of threads to use by each
     contention group created by a teams construct without explicit
     'thread_limit' clause.  The value of this variable shall be a
     positive integer.  If undefined, the value of 0 is used which
     stands for an implementation defined upper limit.

_See also_:
     *note OMP_THREAD_LIMIT::, *note omp_set_teams_thread_limit::

_Reference_:
     OpenMP specification v5.1 (https://www.openmp.org), Section 6.24


File: libgomp.info,  Node: OMP_THREAD_LIMIT,  Next: OMP_WAIT_POLICY,  Prev: OMP_TEAMS_THREAD_LIMIT,  Up: Environment Variables

4.19 'OMP_THREAD_LIMIT' - Set the maximum number of threads
===========================================================

_ICV:_ THREAD-LIMIT-VAR
_Scope:_ data environment
_Description_:
     Specifies the number of threads to use for the whole program.  The
     value of this variable shall be a positive integer.  If undefined,
     the number of threads is not limited.

_See also_:
     *note OMP_NUM_THREADS::, *note omp_get_thread_limit::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.10


File: libgomp.info,  Node: OMP_WAIT_POLICY,  Next: GOMP_CPU_AFFINITY,  Prev: OMP_THREAD_LIMIT,  Up: Environment Variables

4.20 'OMP_WAIT_POLICY' - How waiting threads are handled
========================================================

_Description_:
     Specifies whether waiting threads should be active or passive.  If
     the value is 'PASSIVE', waiting threads should not consume CPU
     power while waiting; while the value is 'ACTIVE' specifies that
     they should.  If undefined, threads wait actively for a short time
     before waiting passively.

_See also_:
     *note GOMP_SPINCOUNT::

_Reference_:
     OpenMP specification v4.5 (https://www.openmp.org), Section 4.8


File: libgomp.info,  Node: GOMP_CPU_AFFINITY,  Next: GOMP_DEBUG,  Prev: OMP_WAIT_POLICY,  Up: Environment Variables

4.21 'GOMP_CPU_AFFINITY' - Bind threads to specific CPUs
========================================================

_Description_:
     Binds threads to specific CPUs.  The variable should contain a
     space-separated or comma-separated list of CPUs.  This list may
     contain different kinds of entries: either single CPU numbers in
     any order, a range of CPUs (M-N) or a range with some stride
     (M-N:S). CPU numbers are zero based.  For example,
     'GOMP_CPU_AFFINITY="0 3 1-2 4-15:2"' binds the initial thread to
     CPU 0, the second to CPU 3, the third to CPU 1, the fourth to CPU
     2, the fifth to CPU 4, the sixth through tenth to CPUs 6, 8, 10,
     12, and 14 respectively and then starts assigning back from the
     beginning of the list.  'GOMP_CPU_AFFINITY=0' binds all threads to
     CPU 0.

     There is no libgomp library routine to determine whether a CPU
     affinity specification is in effect.  As a workaround,
     language-specific library functions, e.g., 'getenv' in C or
     'GET_ENVIRONMENT_VARIABLE' in Fortran, may be used to query the
     setting of the 'GOMP_CPU_AFFINITY' environment variable.  A defined
     CPU affinity on startup cannot be changed or disabled during the
     runtime of the application.

     If both 'GOMP_CPU_AFFINITY' and 'OMP_PROC_BIND' are set,
     'OMP_PROC_BIND' has a higher precedence.  If neither has been set
     and 'OMP_PROC_BIND' is unset, or when 'OMP_PROC_BIND' is set to
     'FALSE', the host system handles the assignment of threads to CPUs.

_See also_:
     *note OMP_PLACES::, *note OMP_PROC_BIND::


File: libgomp.info,  Node: GOMP_DEBUG,  Next: GOMP_STACKSIZE,  Prev: GOMP_CPU_AFFINITY,  Up: Environment Variables

4.22 'GOMP_DEBUG' - Enable debugging output
===========================================

_Description_:
     Enable debugging output.  The variable should be set to '0'
     (disabled, also the default if not set), or '1' (enabled).

     If enabled, some debugging output is printed during execution.
     This is currently not specified in more detail, and subject to
     change.


File: libgomp.info,  Node: GOMP_STACKSIZE,  Next: GOMP_SPINCOUNT,  Prev: GOMP_DEBUG,  Up: Environment Variables

4.23 'GOMP_STACKSIZE' - Set default thread stack size
=====================================================

_Description_:
     Set the default thread stack size in kilobytes.  This is different
     from 'pthread_attr_setstacksize' which gets the number of bytes as
     an argument.  If the stack size cannot be set due to system
     constraints, an error is reported and the initial stack size is
     left unchanged.  If undefined, the stack size is system dependent.

_See also_:
     *note OMP_STACKSIZE::

_Reference_:
     GCC Patches Mailinglist
     (https://gcc.gnu.org/ml/gcc-patches/2006-06/msg00493.html), GCC
     Patches Mailinglist
     (https://gcc.gnu.org/ml/gcc-patches/2006-06/msg00496.html)


File: libgomp.info,  Node: GOMP_SPINCOUNT,  Next: GOMP_RTEMS_THREAD_POOLS,  Prev: GOMP_STACKSIZE,  Up: Environment Variables

4.24 'GOMP_SPINCOUNT' - Set the busy-wait spin count
====================================================

_Description_:
     Determines how long a threads waits actively with consuming CPU
     power before waiting passively without consuming CPU power.  The
     value may be either 'INFINITE', 'INFINITY' to always wait actively
     or an integer which gives the number of spins of the busy-wait
     loop.  The integer may optionally be followed by the following
     suffixes acting as multiplication factors: 'k' (kilo, thousand),
     'M' (mega, million), 'G' (giga, billion), or 'T' (tera, trillion).
     If undefined, 0 is used when 'OMP_WAIT_POLICY' is 'PASSIVE',
     300,000 is used when 'OMP_WAIT_POLICY' is undefined and 30 billion
     is used when 'OMP_WAIT_POLICY' is 'ACTIVE'.  If there are more
     OpenMP threads than available CPUs, 1000 and 100 spins are used for
     'OMP_WAIT_POLICY' being 'ACTIVE' or undefined, respectively; unless
     the 'GOMP_SPINCOUNT' is lower or 'OMP_WAIT_POLICY' is 'PASSIVE'.

_See also_:
     *note OMP_WAIT_POLICY::


File: libgomp.info,  Node: GOMP_RTEMS_THREAD_POOLS,  Prev: GOMP_SPINCOUNT,  Up: Environment Variables

4.25 'GOMP_RTEMS_THREAD_POOLS' - Set the RTEMS specific thread pools
====================================================================

_Description_:
     This environment variable is only used on the RTEMS real-time
     operating system.  It determines the scheduler instance specific
     thread pools.  The format for 'GOMP_RTEMS_THREAD_POOLS' is a list
     of optional '<thread-pool-count>[$<priority>]@<scheduler-name>'
     configurations separated by ':' where:
        * '<thread-pool-count>' is the thread pool count for this
          scheduler instance.
        * '$<priority>' is an optional priority for the worker threads
          of a thread pool according to 'pthread_setschedparam'.  In
          case a priority value is omitted, then a worker thread
          inherits the priority of the OpenMP primary thread that
          created it.  The priority of the worker thread is not changed
          after creation, even if a new OpenMP primary thread using the
          worker has a different priority.
        * '@<scheduler-name>' is the scheduler instance name according
          to the RTEMS application configuration.
     In case no thread pool configuration is specified for a scheduler
     instance, then each OpenMP primary thread of this scheduler
     instance uses its own dynamically allocated thread pool.  To limit
     the worker thread count of the thread pools, each OpenMP primary
     thread must call 'omp_set_num_threads'.
_Example_:
     Lets suppose we have three scheduler instances 'IO', 'WRK0', and
     'WRK1' with 'GOMP_RTEMS_THREAD_POOLS' set to '"1@WRK0:3$4@WRK1"'.
     Then there are no thread pool restrictions for scheduler instance
     'IO'.  In the scheduler instance 'WRK0' there is one thread pool
     available.  Since no priority is specified for this scheduler
     instance, the worker thread inherits the priority of the OpenMP
     primary thread that created it.  In the scheduler instance 'WRK1'
     there are three thread pools available and their worker threads run
     at priority four.


File: libgomp.info,  Node: Enabling OpenACC,  Next: OpenACC Runtime Library Routines,  Prev: Environment Variables,  Up: Top

5 Enabling OpenACC
******************

To activate the OpenACC extensions for C/C++ and Fortran, the
compile-time flag '-fopenacc' must be specified.  This enables the
OpenACC directive '#pragma acc' in C/C++ and, in Fortran, the '!$acc'
sentinel in free source form and the 'c$acc', '*$acc' and '!$acc'
sentinels in fixed source form.  The flag also arranges for automatic
linking of the OpenACC runtime library (*note OpenACC Runtime Library
Routines::).

   See <https://gcc.gnu.org/wiki/OpenACC> for more information.

   A complete description of all OpenACC directives accepted may be
found in the OpenACC (https://www.openacc.org) Application Programming
Interface manual, version 2.6.


File: libgomp.info,  Node: OpenACC Runtime Library Routines,  Next: OpenACC Environment Variables,  Prev: Enabling OpenACC,  Up: Top

6 OpenACC Runtime Library Routines
**********************************

The runtime routines described here are defined by section 3 of the
OpenACC specifications in version 2.6.  They have C linkage, and do not
throw exceptions.  Generally, they are available only for the host, with
the exception of 'acc_on_device', which is available for both the host
and the acceleration device.

* Menu:

* acc_get_num_devices::         Get number of devices for the given device
                                type.
* acc_set_device_type::         Set type of device accelerator to use.
* acc_get_device_type::         Get type of device accelerator to be used.
* acc_set_device_num::          Set device number to use.
* acc_get_device_num::          Get device number to be used.
* acc_get_property::            Get device property.
* acc_async_test::              Tests for completion of a specific asynchronous
                                operation.
* acc_async_test_all::          Tests for completion of all asynchronous
                                operations.
* acc_wait::                    Wait for completion of a specific asynchronous
                                operation.
* acc_wait_all::                Waits for completion of all asynchronous
                                operations.
* acc_wait_all_async::          Wait for completion of all asynchronous
                                operations.
* acc_wait_async::              Wait for completion of asynchronous operations.
* acc_init::                    Initialize runtime for a specific device type.
* acc_shutdown::                Shuts down the runtime for a specific device
                                type.
* acc_on_device::               Whether executing on a particular device
* acc_malloc::                  Allocate device memory.
* acc_free::                    Free device memory.
* acc_copyin::                  Allocate device memory and copy host memory to
                                it.
* acc_present_or_copyin::       If the data is not present on the device,
                                allocate device memory and copy from host
                                memory.
* acc_create::                  Allocate device memory and map it to host
                                memory.
* acc_present_or_create::       If the data is not present on the device,
                                allocate device memory and map it to host
                                memory.
* acc_copyout::                 Copy device memory to host memory.
* acc_delete::                  Free device memory.
* acc_update_device::           Update device memory from mapped host memory.
* acc_update_self::             Update host memory from mapped device memory.
* acc_map_data::                Map previously allocated device memory to host
                                memory.
* acc_unmap_data::              Unmap device memory from host memory.
* acc_deviceptr::               Get device pointer associated with specific
                                host address.
* acc_hostptr::                 Get host pointer associated with specific
                                device address.
* acc_is_present::              Indicate whether host variable / array is
                                present on device.
* acc_memcpy_to_device::        Copy host memory to device memory.
* acc_memcpy_from_device::      Copy device memory to host memory.
* acc_attach::                  Let device pointer point to device-pointer target.
* acc_detach::                  Let device pointer point to host-pointer target.

API routines for target platforms.

* acc_get_current_cuda_device:: Get CUDA device handle.
* acc_get_current_cuda_context::Get CUDA context handle.
* acc_get_cuda_stream::         Get CUDA stream handle.
* acc_set_cuda_stream::         Set CUDA stream handle.

API routines for the OpenACC Profiling Interface.

* acc_prof_register::           Register callbacks.
* acc_prof_unregister::         Unregister callbacks.
* acc_prof_lookup::             Obtain inquiry functions.
* acc_register_library::        Library registration.


File: libgomp.info,  Node: acc_get_num_devices,  Next: acc_set_device_type,  Up: OpenACC Runtime Library Routines

6.1 'acc_get_num_devices' - Get number of devices for given device type
=======================================================================

_Description_
     This function returns a value indicating the number of devices
     available for the device type specified in DEVICETYPE.

_C/C++_:
     _Prototype_:   'int acc_get_num_devices(acc_device_t devicetype);'

_Fortran_:
     _Interface_:   'integer function acc_get_num_devices(devicetype)'
                    'integer(kind=acc_device_kind) devicetype'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.1.


File: libgomp.info,  Node: acc_set_device_type,  Next: acc_get_device_type,  Prev: acc_get_num_devices,  Up: OpenACC Runtime Library Routines

6.2 'acc_set_device_type' - Set type of device accelerator to use.
==================================================================

_Description_
     This function indicates to the runtime library which device type,
     specified in DEVICETYPE, to use when executing a parallel or
     kernels region.

_C/C++_:
     _Prototype_:   'acc_set_device_type(acc_device_t devicetype);'

_Fortran_:
     _Interface_:   'subroutine acc_set_device_type(devicetype)'
                    'integer(kind=acc_device_kind) devicetype'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.2.


File: libgomp.info,  Node: acc_get_device_type,  Next: acc_set_device_num,  Prev: acc_set_device_type,  Up: OpenACC Runtime Library Routines

6.3 'acc_get_device_type' - Get type of device accelerator to be used.
======================================================================

_Description_
     This function returns what device type will be used when executing
     a parallel or kernels region.

     This function returns 'acc_device_none' if 'acc_get_device_type' is
     called from 'acc_ev_device_init_start', 'acc_ev_device_init_end'
     callbacks of the OpenACC Profiling Interface (*note OpenACC
     Profiling Interface::), that is, if the device is currently being
     initialized.

_C/C++_:
     _Prototype_:   'acc_device_t acc_get_device_type(void);'

_Fortran_:
     _Interface_:   'function acc_get_device_type(void)'
                    'integer(kind=acc_device_kind) acc_get_device_type'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.3.


File: libgomp.info,  Node: acc_set_device_num,  Next: acc_get_device_num,  Prev: acc_get_device_type,  Up: OpenACC Runtime Library Routines

6.4 'acc_set_device_num' - Set device number to use.
====================================================

_Description_
     This function will indicate to the runtime which device number,
     specified by DEVICENUM, associated with the specified device type
     DEVICETYPE.

_C/C++_:
     _Prototype_:   'acc_set_device_num(int devicenum, acc_device_t
                    devicetype);'

_Fortran_:
     _Interface_:   'subroutine acc_set_device_num(devicenum, devicetype)'
                    'integer devicenum'
                    'integer(kind=acc_device_kind) devicetype'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.4.


File: libgomp.info,  Node: acc_get_device_num,  Next: acc_get_property,  Prev: acc_set_device_num,  Up: OpenACC Runtime Library Routines

6.5 'acc_get_device_num' - Get device number to be used.
========================================================

_Description_
     This function returns which device number associated with the
     specified device type DEVICETYPE, will be used when executing a
     parallel or kernels region.

_C/C++_:
     _Prototype_:   'int acc_get_device_num(acc_device_t devicetype);'

_Fortran_:
     _Interface_:   'function acc_get_device_num(devicetype)'
                    'integer(kind=acc_device_kind) devicetype'
                    'integer acc_get_device_num'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.5.


File: libgomp.info,  Node: acc_get_property,  Next: acc_async_test,  Prev: acc_get_device_num,  Up: OpenACC Runtime Library Routines

6.6 'acc_get_property' - Get device property.
=============================================

_Description_
     These routines return the value of the specified PROPERTY for the
     device being queried according to DEVICENUM and DEVICETYPE.
     Integer-valued and string-valued properties are returned by
     'acc_get_property' and 'acc_get_property_string' respectively.  The
     Fortran 'acc_get_property_string' subroutine returns the string
     retrieved in its fourth argument while the remaining entry points
     are functions, which pass the return value as their result.

     Note for Fortran, only: the OpenACC technical committee corrected
     and, hence, modified the interface introduced in OpenACC 2.6.  The
     kind-value parameter 'acc_device_property' has been renamed to
     'acc_device_property_kind' for consistency and the return type of
     the 'acc_get_property' function is now a 'c_size_t' integer instead
     of a 'acc_device_property' integer.  The parameter
     'acc_device_property' is still provided, but might be removed in a
     future version of GCC.

_C/C++_:
     _Prototype_:   'size_t acc_get_property(int devicenum, acc_device_t
                    devicetype, acc_device_property_t property);'
     _Prototype_:   'const char *acc_get_property_string(int devicenum,
                    acc_device_t devicetype, acc_device_property_t
                    property);'

_Fortran_:
     _Interface_:   'function acc_get_property(devicenum, devicetype,
                    property)'
     _Interface_:   'subroutine acc_get_property_string(devicenum,
                    devicetype, property, string)'
                    'use ISO_C_Binding, only: c_size_t'
                    'integer devicenum'
                    'integer(kind=acc_device_kind) devicetype'
                    'integer(kind=acc_device_property_kind) property'
                    'integer(kind=c_size_t) acc_get_property'
                    'character(*) string'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.6.


File: libgomp.info,  Node: acc_async_test,  Next: acc_async_test_all,  Prev: acc_get_property,  Up: OpenACC Runtime Library Routines

6.7 'acc_async_test' - Test for completion of a specific asynchronous operation.
================================================================================

_Description_
     This function tests for completion of the asynchronous operation
     specified in ARG.  In C/C++, a non-zero value is returned to
     indicate the specified asynchronous operation has completed while
     Fortran returns 'true'.  If the asynchronous operation has not
     completed, C/C++ returns zero and Fortran returns 'false'.

_C/C++_:
     _Prototype_:   'int acc_async_test(int arg);'

_Fortran_:
     _Interface_:   'function acc_async_test(arg)'
                    'integer(kind=acc_handle_kind) arg'
                    'logical acc_async_test'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.9.


File: libgomp.info,  Node: acc_async_test_all,  Next: acc_wait,  Prev: acc_async_test,  Up: OpenACC Runtime Library Routines

6.8 'acc_async_test_all' - Tests for completion of all asynchronous operations.
===============================================================================

_Description_
     This function tests for completion of all asynchronous operations.
     In C/C++, a non-zero value is returned to indicate all asynchronous
     operations have completed while Fortran returns 'true'.  If any
     asynchronous operation has not completed, C/C++ returns zero and
     Fortran returns 'false'.

_C/C++_:
     _Prototype_:   'int acc_async_test_all(void);'

_Fortran_:
     _Interface_:   'function acc_async_test()'
                    'logical acc_get_device_num'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.10.


File: libgomp.info,  Node: acc_wait,  Next: acc_wait_all,  Prev: acc_async_test_all,  Up: OpenACC Runtime Library Routines

6.9 'acc_wait' - Wait for completion of a specific asynchronous operation.
==========================================================================

_Description_
     This function waits for completion of the asynchronous operation
     specified in ARG.

_C/C++_:
     _Prototype_:   'acc_wait(arg);'
     _Prototype     'acc_async_wait(arg);'
     (OpenACC 1.0
     compatibility)_:

_Fortran_:
     _Interface_:   'subroutine acc_wait(arg)'
                    'integer(acc_handle_kind) arg'
     _Interface     'subroutine acc_async_wait(arg)'
     (OpenACC 1.0
     compatibility)_:
                    'integer(acc_handle_kind) arg'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.11.


File: libgomp.info,  Node: acc_wait_all,  Next: acc_wait_all_async,  Prev: acc_wait,  Up: OpenACC Runtime Library Routines

6.10 'acc_wait_all' - Waits for completion of all asynchronous operations.
==========================================================================

_Description_
     This function waits for the completion of all asynchronous
     operations.

_C/C++_:
     _Prototype_:   'acc_wait_all(void);'
     _Prototype     'acc_async_wait_all(void);'
     (OpenACC 1.0
     compatibility)_:

_Fortran_:
     _Interface_:   'subroutine acc_wait_all()'
     _Interface     'subroutine acc_async_wait_all()'
     (OpenACC 1.0
     compatibility)_:

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.13.


File: libgomp.info,  Node: acc_wait_all_async,  Next: acc_wait_async,  Prev: acc_wait_all,  Up: OpenACC Runtime Library Routines

6.11 'acc_wait_all_async' - Wait for completion of all asynchronous operations.
===============================================================================

_Description_
     This function enqueues a wait operation on the queue ASYNC for any
     and all asynchronous operations that have been previously enqueued
     on any queue.

_C/C++_:
     _Prototype_:   'acc_wait_all_async(int async);'

_Fortran_:
     _Interface_:   'subroutine acc_wait_all_async(async)'
                    'integer(acc_handle_kind) async'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.14.


File: libgomp.info,  Node: acc_wait_async,  Next: acc_init,  Prev: acc_wait_all_async,  Up: OpenACC Runtime Library Routines

6.12 'acc_wait_async' - Wait for completion of asynchronous operations.
=======================================================================

_Description_
     This function enqueues a wait operation on queue ASYNC for any and
     all asynchronous operations enqueued on queue ARG.

_C/C++_:
     _Prototype_:   'acc_wait_async(int arg, int async);'

_Fortran_:
     _Interface_:   'subroutine acc_wait_async(arg, async)'
                    'integer(acc_handle_kind) arg, async'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.12.


File: libgomp.info,  Node: acc_init,  Next: acc_shutdown,  Prev: acc_wait_async,  Up: OpenACC Runtime Library Routines

6.13 'acc_init' - Initialize runtime for a specific device type.
================================================================

_Description_
     This function initializes the runtime for the device type specified
     in DEVICETYPE.

_C/C++_:
     _Prototype_:   'acc_init(acc_device_t devicetype);'

_Fortran_:
     _Interface_:   'subroutine acc_init(devicetype)'
                    'integer(acc_device_kind) devicetype'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.7.


File: libgomp.info,  Node: acc_shutdown,  Next: acc_on_device,  Prev: acc_init,  Up: OpenACC Runtime Library Routines

6.14 'acc_shutdown' - Shuts down the runtime for a specific device type.
========================================================================

_Description_
     This function shuts down the runtime for the device type specified
     in DEVICETYPE.

_C/C++_:
     _Prototype_:   'acc_shutdown(acc_device_t devicetype);'

_Fortran_:
     _Interface_:   'subroutine acc_shutdown(devicetype)'
                    'integer(acc_device_kind) devicetype'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.8.


File: libgomp.info,  Node: acc_on_device,  Next: acc_malloc,  Prev: acc_shutdown,  Up: OpenACC Runtime Library Routines

6.15 'acc_on_device' - Whether executing on a particular device
===============================================================

_Description_:
     This function returns whether the program is executing on a
     particular device specified in DEVICETYPE.  In C/C++ a non-zero
     value is returned to indicate the device is executing on the
     specified device type.  In Fortran, 'true' is returned.  If the
     program is not executing on the specified device type C/C++ returns
     zero, while Fortran returns 'false'.

_C/C++_:
     _Prototype_:   'acc_on_device(acc_device_t devicetype);'

_Fortran_:
     _Interface_:   'function acc_on_device(devicetype)'
                    'integer(acc_device_kind) devicetype'
                    'logical acc_on_device'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.17.


File: libgomp.info,  Node: acc_malloc,  Next: acc_free,  Prev: acc_on_device,  Up: OpenACC Runtime Library Routines

6.16 'acc_malloc' - Allocate device memory.
===========================================

_Description_
     This function allocates BYTES bytes of device memory.  It returns
     the device address of the allocated memory.

_C/C++_:
     _Prototype_:   'd_void* acc_malloc(size_t bytes);'

_Fortran_:
     _Interface_:   'type(c_ptr) function acc_malloc(bytes)'
                    'integer(c_size_t), value :: bytes'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.18.  openacc specification v3.3 (https://www.openacc.org),
     section 3.2.16.


File: libgomp.info,  Node: acc_free,  Next: acc_copyin,  Prev: acc_malloc,  Up: OpenACC Runtime Library Routines

6.17 'acc_free' - Free device memory.
=====================================

_Description_
     Free previously allocated device memory at the device address
     'data_dev'.

_C/C++_:
     _Prototype_:   'void acc_free(d_void *data_dev);'

_Fortran_:
     _Interface_:   'subroutine acc_free(data_dev)'
                    'type(c_ptr), value :: data_dev'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.19.  openacc specification v3.3 (https://www.openacc.org),
     section 3.2.17.


File: libgomp.info,  Node: acc_copyin,  Next: acc_present_or_copyin,  Prev: acc_free,  Up: OpenACC Runtime Library Routines

6.18 'acc_copyin' - Allocate device memory and copy host memory to it.
======================================================================

_Description_
     In C/C++, this function allocates LEN bytes of device memory and
     maps it to the specified host address in A.  The device address of
     the newly allocated device memory is returned.

     In Fortran, two (2) forms are supported.  In the first form, A
     specifies a contiguous array section.  The second form A specifies
     a variable or array element and LEN specifies the length in bytes.

_C/C++_:
     _Prototype_:   'void *acc_copyin(h_void *a, size_t len);'
     _Prototype_:   'void *acc_copyin_async(h_void *a, size_t len, int
                    async);'

_Fortran_:
     _Interface_:   'subroutine acc_copyin(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_copyin(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
     _Interface_:   'subroutine acc_copyin_async(a, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer(acc_handle_kind) :: async'
     _Interface_:   'subroutine acc_copyin_async(a, len, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
                    'integer(acc_handle_kind) :: async'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.20.


File: libgomp.info,  Node: acc_present_or_copyin,  Next: acc_create,  Prev: acc_copyin,  Up: OpenACC Runtime Library Routines

6.19 'acc_present_or_copyin' - If the data is not present on the device, allocate device memory and copy from host memory.
==========================================================================================================================

_Description_
     This function tests if the host data specified by A and of length
     LEN is present or not.  If it is not present, device memory is
     allocated and the host memory copied.  The device address of the
     newly allocated device memory is returned.

     In Fortran, two (2) forms are supported.  In the first form, A
     specifies a contiguous array section.  The second form A specifies
     a variable or array element and LEN specifies the length in bytes.

     Note that 'acc_present_or_copyin' and 'acc_pcopyin' exist for
     backward compatibility with OpenACC 2.0; use *note acc_copyin::
     instead.

_C/C++_:
     _Prototype_:   'void *acc_present_or_copyin(h_void *a, size_t len);'
     _Prototype_:   'void *acc_pcopyin(h_void *a, size_t len);'

_Fortran_:
     _Interface_:   'subroutine acc_present_or_copyin(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_present_or_copyin(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
     _Interface_:   'subroutine acc_pcopyin(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_pcopyin(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.20.


File: libgomp.info,  Node: acc_create,  Next: acc_present_or_create,  Prev: acc_present_or_copyin,  Up: OpenACC Runtime Library Routines

6.20 'acc_create' - Allocate device memory and map it to host memory.
=====================================================================

_Description_
     This function allocates device memory and maps it to host memory
     specified by the host address A with a length of LEN bytes.  In
     C/C++, the function returns the device address of the allocated
     device memory.

     In Fortran, two (2) forms are supported.  In the first form, A
     specifies a contiguous array section.  The second form A specifies
     a variable or array element and LEN specifies the length in bytes.

_C/C++_:
     _Prototype_:   'void *acc_create(h_void *a, size_t len);'
     _Prototype_:   'void *acc_create_async(h_void *a, size_t len, int
                    async);'

_Fortran_:
     _Interface_:   'subroutine acc_create(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_create(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
     _Interface_:   'subroutine acc_create_async(a, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer(acc_handle_kind) :: async'
     _Interface_:   'subroutine acc_create_async(a, len, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
                    'integer(acc_handle_kind) :: async'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.21.


File: libgomp.info,  Node: acc_present_or_create,  Next: acc_copyout,  Prev: acc_create,  Up: OpenACC Runtime Library Routines

6.21 'acc_present_or_create' - If the data is not present on the device, allocate device memory and map it to host memory.
==========================================================================================================================

_Description_
     This function tests if the host data specified by A and of length
     LEN is present or not.  If it is not present, device memory is
     allocated and mapped to host memory.  In C/C++, the device address
     of the newly allocated device memory is returned.

     In Fortran, two (2) forms are supported.  In the first form, A
     specifies a contiguous array section.  The second form A specifies
     a variable or array element and LEN specifies the length in bytes.

     Note that 'acc_present_or_create' and 'acc_pcreate' exist for
     backward compatibility with OpenACC 2.0; use *note acc_create::
     instead.

_C/C++_:
     _Prototype_:   'void *acc_present_or_create(h_void *a, size_t len)'
     _Prototype_:   'void *acc_pcreate(h_void *a, size_t len)'

_Fortran_:
     _Interface_:   'subroutine acc_present_or_create(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_present_or_create(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
     _Interface_:   'subroutine acc_pcreate(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_pcreate(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.21.


File: libgomp.info,  Node: acc_copyout,  Next: acc_delete,  Prev: acc_present_or_create,  Up: OpenACC Runtime Library Routines

6.22 'acc_copyout' - Copy device memory to host memory.
=======================================================

_Description_
     This function copies mapped device memory to host memory which is
     specified by host address A for a length LEN bytes in C/C++.

     In Fortran, two (2) forms are supported.  In the first form, A
     specifies a contiguous array section.  The second form A specifies
     a variable or array element and LEN specifies the length in bytes.

_C/C++_:
     _Prototype_:   'acc_copyout(h_void *a, size_t len);'
     _Prototype_:   'acc_copyout_async(h_void *a, size_t len, int async);'
     _Prototype_:   'acc_copyout_finalize(h_void *a, size_t len);'
     _Prototype_:   'acc_copyout_finalize_async(h_void *a, size_t len, int
                    async);'

_Fortran_:
     _Interface_:   'subroutine acc_copyout(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_copyout(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
     _Interface_:   'subroutine acc_copyout_async(a, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer(acc_handle_kind) :: async'
     _Interface_:   'subroutine acc_copyout_async(a, len, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
                    'integer(acc_handle_kind) :: async'
     _Interface_:   'subroutine acc_copyout_finalize(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_copyout_finalize(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
     _Interface_:   'subroutine acc_copyout_finalize_async(a, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer(acc_handle_kind) :: async'
     _Interface_:   'subroutine acc_copyout_finalize_async(a, len, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
                    'integer(acc_handle_kind) :: async'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.22.


File: libgomp.info,  Node: acc_delete,  Next: acc_update_device,  Prev: acc_copyout,  Up: OpenACC Runtime Library Routines

6.23 'acc_delete' - Free device memory.
=======================================

_Description_
     This function frees previously allocated device memory specified by
     the device address A and the length of LEN bytes.

     In Fortran, two (2) forms are supported.  In the first form, A
     specifies a contiguous array section.  The second form A specifies
     a variable or array element and LEN specifies the length in bytes.

_C/C++_:
     _Prototype_:   'acc_delete(h_void *a, size_t len);'
     _Prototype_:   'acc_delete_async(h_void *a, size_t len, int async);'
     _Prototype_:   'acc_delete_finalize(h_void *a, size_t len);'
     _Prototype_:   'acc_delete_finalize_async(h_void *a, size_t len, int
                    async);'

_Fortran_:
     _Interface_:   'subroutine acc_delete(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_delete(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
     _Interface_:   'subroutine acc_delete_async(a, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer(acc_handle_kind) :: async'
     _Interface_:   'subroutine acc_delete_async(a, len, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
                    'integer(acc_handle_kind) :: async'
     _Interface_:   'subroutine acc_delete_finalize(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_delete_finalize(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
     _Interface_:   'subroutine acc_delete_async_finalize(a, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer(acc_handle_kind) :: async'
     _Interface_:   'subroutine acc_delete_async_finalize(a, len, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
                    'integer(acc_handle_kind) :: async'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.23.


File: libgomp.info,  Node: acc_update_device,  Next: acc_update_self,  Prev: acc_delete,  Up: OpenACC Runtime Library Routines

6.24 'acc_update_device' - Update device memory from mapped host memory.
========================================================================

_Description_
     This function updates the device copy from the previously mapped
     host memory.  The host memory is specified with the host address A
     and a length of LEN bytes.

     In Fortran, two (2) forms are supported.  In the first form, A
     specifies a contiguous array section.  The second form A specifies
     a variable or array element and LEN specifies the length in bytes.

_C/C++_:
     _Prototype_:   'acc_update_device(h_void *a, size_t len);'
     _Prototype_:   'acc_update_device(h_void *a, size_t len, async);'

_Fortran_:
     _Interface_:   'subroutine acc_update_device(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_update_device(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
     _Interface_:   'subroutine acc_update_device_async(a, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer(acc_handle_kind) :: async'
     _Interface_:   'subroutine acc_update_device_async(a, len, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
                    'integer(acc_handle_kind) :: async'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.24.


File: libgomp.info,  Node: acc_update_self,  Next: acc_map_data,  Prev: acc_update_device,  Up: OpenACC Runtime Library Routines

6.25 'acc_update_self' - Update host memory from mapped device memory.
======================================================================

_Description_
     This function updates the host copy from the previously mapped
     device memory.  The host memory is specified with the host address
     A and a length of LEN bytes.

     In Fortran, two (2) forms are supported.  In the first form, A
     specifies a contiguous array section.  The second form A specifies
     a variable or array element and LEN specifies the length in bytes.

_C/C++_:
     _Prototype_:   'acc_update_self(h_void *a, size_t len);'
     _Prototype_:   'acc_update_self_async(h_void *a, size_t len, int
                    async);'

_Fortran_:
     _Interface_:   'subroutine acc_update_self(a)'
                    'type, dimension(:[,:]...) :: a'
     _Interface_:   'subroutine acc_update_self(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
     _Interface_:   'subroutine acc_update_self_async(a, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer(acc_handle_kind) :: async'
     _Interface_:   'subroutine acc_update_self_async(a, len, async)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
                    'integer(acc_handle_kind) :: async'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.25.


File: libgomp.info,  Node: acc_map_data,  Next: acc_unmap_data,  Prev: acc_update_self,  Up: OpenACC Runtime Library Routines

6.26 'acc_map_data' - Map previously allocated device memory to host memory.
============================================================================

_Description_
     This function maps previously allocated device and host memory.
     The device memory is specified with the device address DATA_DEV.
     The host memory is specified with the host address DATA_ARG and a
     length of BYTES.

_C/C++_:
     _Prototype_:   'void acc_map_data(h_void *data_arg, d_void *data_dev,
                    size_t bytes);'

_Fortran_:
     _Interface_:   'subroutine acc_map_data(data_arg, data_dev, bytes)'
                    'type(*), dimension(*) :: data_arg'
                    'type(c_ptr), value :: data_dev'
                    'integer(c_size_t), value :: bytes'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.26.  OpenACC specification v3.3 (https://www.openacc.org),
     section 3.2.21.


File: libgomp.info,  Node: acc_unmap_data,  Next: acc_deviceptr,  Prev: acc_map_data,  Up: OpenACC Runtime Library Routines

6.27 'acc_unmap_data' - Unmap device memory from host memory.
=============================================================

_Description_
     This function unmaps previously mapped device and host memory.  The
     latter specified by DATA_ARG.

_C/C++_:
     _Prototype_:   'void acc_unmap_data(h_void *data_arg);'

_Fortran_:
     _Interface_:   'subroutine acc_unmap_data(data_arg)'
                    'type(*), dimension(*) :: data_arg'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.27.  OpenACC specification v3.3 (https://www.openacc.org),
     section 3.2.22.


File: libgomp.info,  Node: acc_deviceptr,  Next: acc_hostptr,  Prev: acc_unmap_data,  Up: OpenACC Runtime Library Routines

6.28 'acc_deviceptr' - Get device pointer associated with specific host address.
================================================================================

_Description_
     This function returns the device address that has been mapped to
     the host address specified by DATA_ARG.

_C/C++_:
     _Prototype_:   'void *acc_deviceptr(h_void *data_arg);'

_Fortran_:
     _Interface_:   'type(c_ptr) function acc_deviceptr(data_arg)'
                    'type(*), dimension(*) :: data_arg'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.28.  OpenACC specification v3.3 (https://www.openacc.org),
     section 3.2.23.


File: libgomp.info,  Node: acc_hostptr,  Next: acc_is_present,  Prev: acc_deviceptr,  Up: OpenACC Runtime Library Routines

6.29 'acc_hostptr' - Get host pointer associated with specific device address.
==============================================================================

_Description_
     This function returns the host address that has been mapped to the
     device address specified by DATA_DEV.

_C/C++_:
     _Prototype_:   'void *acc_hostptr(d_void *data_dev);'

_Fortran_:
     _Interface_:   'type(c_ptr) function acc_hostptr(data_dev)'
                    'type(c_ptr), value :: data_dev'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.29.  OpenACC specification v3.3 (https://www.openacc.org),
     section 3.2.24.


File: libgomp.info,  Node: acc_is_present,  Next: acc_memcpy_to_device,  Prev: acc_hostptr,  Up: OpenACC Runtime Library Routines

6.30 'acc_is_present' - Indicate whether host variable / array is present on device.
====================================================================================

_Description_
     This function indicates whether the specified host address in A and
     a length of LEN bytes is present on the device.  In C/C++, a
     non-zero value is returned to indicate the presence of the mapped
     memory on the device.  A zero is returned to indicate the memory is
     not mapped on the device.

     In Fortran, two (2) forms are supported.  In the first form, A
     specifies a contiguous array section.  The second form A specifies
     a variable or array element and LEN specifies the length in bytes.
     If the host memory is mapped to device memory, then a 'true' is
     returned.  Otherwise, a 'false' is return to indicate the mapped
     memory is not present.

_C/C++_:
     _Prototype_:   'int acc_is_present(h_void *a, size_t len);'

_Fortran_:
     _Interface_:   'function acc_is_present(a)'
                    'type, dimension(:[,:]...) :: a'
                    'logical acc_is_present'
     _Interface_:   'function acc_is_present(a, len)'
                    'type, dimension(:[,:]...) :: a'
                    'integer len'
                    'logical acc_is_present'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.30.


File: libgomp.info,  Node: acc_memcpy_to_device,  Next: acc_memcpy_from_device,  Prev: acc_is_present,  Up: OpenACC Runtime Library Routines

6.31 'acc_memcpy_to_device' - Copy host memory to device memory.
================================================================

_Description_
     This function copies host memory specified by host address of
     DATA_HOST_SRC to device memory specified by the device address
     DATA_DEV_DEST for a length of BYTES bytes.

_C/C++_:
     _Prototype_:   'void acc_memcpy_to_device(d_void* data_dev_dest,'
                    'h_void* data_host_src, size_t bytes);'
     _Prototype_:   'void acc_memcpy_to_device_async(d_void* data_dev_dest,'
                    'h_void* data_host_src, size_t bytes, int async_arg);'

_Fortran_:
     _Interface_:   'subroutine acc_memcpy_to_device(data_dev_dest, &'
                    'data_host_src, bytes)'
     _Interface_:   'subroutine acc_memcpy_to_device_async(data_dev_dest, &'
                    'data_host_src, bytes, async_arg)'
                    'type(c_ptr), value :: data_dev_dest'
                    'type(*), dimension(*) :: data_host_src'
                    'integer(c_size_t), value :: bytes'
                    'integer(acc_handle_kind), value :: async_arg'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.31 OpenACC specification v3.3 (https://www.openacc.org),
     section 3.2.26.


File: libgomp.info,  Node: acc_memcpy_from_device,  Next: acc_attach,  Prev: acc_memcpy_to_device,  Up: OpenACC Runtime Library Routines

6.32 'acc_memcpy_from_device' - Copy device memory to host memory.
==================================================================

_Description_
     This function copies device memory specified by device address of
     DATA_DEV_SRC to host memory specified by the host address
     DATA_HOST_DEST for a length of BYTES bytes.

_C/C++_:
     _Prototype_:   'void acc_memcpy_from_device(h_void* data_host_dest,'
                    'd_void* data_dev_src, size_t bytes);'
     _Prototype_:   'void acc_memcpy_from_device_async(h_void*
                    data_host_dest,'
                    'd_void* data_dev_src, size_t bytes, int async_arg);'

_Fortran_:
     _Interface_:   'subroutine acc_memcpy_from_device(data_host_dest, &'
                    'data_dev_src, bytes)'
     _Interface_:   'subroutine acc_memcpy_from_device_async(data_host_dest,
                    &'
                    'data_dev_src, bytes, async_arg)'
                    'type(*), dimension(*) :: data_host_dest'
                    'type(c_ptr), value :: data_dev_src'
                    'integer(c_size_t), value :: bytes'
                    'integer(acc_handle_kind), value :: async_arg'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.32.  OpenACC specification v3.3 (https://www.openacc.org),
     section 3.2.27.


File: libgomp.info,  Node: acc_attach,  Next: acc_detach,  Prev: acc_memcpy_from_device,  Up: OpenACC Runtime Library Routines

6.33 'acc_attach' - Let device pointer point to device-pointer target.
======================================================================

_Description_
     This function updates a pointer on the device from pointing to a
     host-pointer address to pointing to the corresponding device data.

_C/C++_:
     _Prototype_:   'void acc_attach(h_void **ptr_addr);'
     _Prototype_:   'void acc_attach_async(h_void **ptr_addr, int async);'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.34.


File: libgomp.info,  Node: acc_detach,  Next: acc_get_current_cuda_device,  Prev: acc_attach,  Up: OpenACC Runtime Library Routines

6.34 'acc_detach' - Let device pointer point to host-pointer target.
====================================================================

_Description_
     This function updates a pointer on the device from pointing to a
     device-pointer address to pointing to the corresponding host data.

_C/C++_:
     _Prototype_:   'void acc_detach(h_void **ptr_addr);'
     _Prototype_:   'void acc_detach_async(h_void **ptr_addr, int async);'
     _Prototype_:   'void acc_detach_finalize(h_void **ptr_addr);'
     _Prototype_:   'void acc_detach_finalize_async(h_void **ptr_addr, int
                    async);'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     3.2.35.


File: libgomp.info,  Node: acc_get_current_cuda_device,  Next: acc_get_current_cuda_context,  Prev: acc_detach,  Up: OpenACC Runtime Library Routines

6.35 'acc_get_current_cuda_device' - Get CUDA device handle.
============================================================

_Description_
     This function returns the CUDA device handle.  This handle is the
     same as used by the CUDA Runtime or Driver API's.

_C/C++_:
     _Prototype_:   'void *acc_get_current_cuda_device(void);'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     A.2.1.1.


File: libgomp.info,  Node: acc_get_current_cuda_context,  Next: acc_get_cuda_stream,  Prev: acc_get_current_cuda_device,  Up: OpenACC Runtime Library Routines

6.36 'acc_get_current_cuda_context' - Get CUDA context handle.
==============================================================

_Description_
     This function returns the CUDA context handle.  This handle is the
     same as used by the CUDA Runtime or Driver API's.

_C/C++_:
     _Prototype_:   'void *acc_get_current_cuda_context(void);'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     A.2.1.2.


File: libgomp.info,  Node: acc_get_cuda_stream,  Next: acc_set_cuda_stream,  Prev: acc_get_current_cuda_context,  Up: OpenACC Runtime Library Routines

6.37 'acc_get_cuda_stream' - Get CUDA stream handle.
====================================================

_Description_
     This function returns the CUDA stream handle for the queue ASYNC.
     This handle is the same as used by the CUDA Runtime or Driver
     API's.

_C/C++_:
     _Prototype_:   'void *acc_get_cuda_stream(int async);'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     A.2.1.3.


File: libgomp.info,  Node: acc_set_cuda_stream,  Next: acc_prof_register,  Prev: acc_get_cuda_stream,  Up: OpenACC Runtime Library Routines

6.38 'acc_set_cuda_stream' - Set CUDA stream handle.
====================================================

_Description_
     This function associates the stream handle specified by STREAM with
     the queue ASYNC.

     This cannot be used to change the stream handle associated with
     'acc_async_sync'.

     The return value is not specified.

_C/C++_:
     _Prototype_:   'int acc_set_cuda_stream(int async, void *stream);'

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section
     A.2.1.4.


File: libgomp.info,  Node: acc_prof_register,  Next: acc_prof_unregister,  Prev: acc_set_cuda_stream,  Up: OpenACC Runtime Library Routines

6.39 'acc_prof_register' - Register callbacks.
==============================================

_Description_:
     This function registers callbacks.

_C/C++_:
     _Prototype_:   'void acc_prof_register (acc_event_t, acc_prof_callback,
                    acc_register_t);'

_See also_:
     *note OpenACC Profiling Interface::

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section 5.3.


File: libgomp.info,  Node: acc_prof_unregister,  Next: acc_prof_lookup,  Prev: acc_prof_register,  Up: OpenACC Runtime Library Routines

6.40 'acc_prof_unregister' - Unregister callbacks.
==================================================

_Description_:
     This function unregisters callbacks.

_C/C++_:
     _Prototype_:   'void acc_prof_unregister (acc_event_t,
                    acc_prof_callback, acc_register_t);'

_See also_:
     *note OpenACC Profiling Interface::

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section 5.3.


File: libgomp.info,  Node: acc_prof_lookup,  Next: acc_register_library,  Prev: acc_prof_unregister,  Up: OpenACC Runtime Library Routines

6.41 'acc_prof_lookup' - Obtain inquiry functions.
==================================================

_Description_:
     Function to obtain inquiry functions.

_C/C++_:
     _Prototype_:   'acc_query_fn acc_prof_lookup (const char *);'

_See also_:
     *note OpenACC Profiling Interface::

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section 5.3.


File: libgomp.info,  Node: acc_register_library,  Prev: acc_prof_lookup,  Up: OpenACC Runtime Library Routines

6.42 'acc_register_library' - Library registration.
===================================================

_Description_:
     Function for library registration.

_C/C++_:
     _Prototype_:   'void acc_register_library (acc_prof_reg, acc_prof_reg,
                    acc_prof_lookup_func);'

_See also_:
     *note OpenACC Profiling Interface::, *note ACC_PROFLIB::

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section 5.3.


File: libgomp.info,  Node: OpenACC Environment Variables,  Next: CUDA Streams Usage,  Prev: OpenACC Runtime Library Routines,  Up: Top

7 OpenACC Environment Variables
*******************************

The variables 'ACC_DEVICE_TYPE' and 'ACC_DEVICE_NUM' are defined by
section 4 of the OpenACC specification in version 2.0.  The variable
'ACC_PROFLIB' is defined by section 4 of the OpenACC specification in
version 2.6.

* Menu:

* ACC_DEVICE_TYPE::
* ACC_DEVICE_NUM::
* ACC_PROFLIB::


File: libgomp.info,  Node: ACC_DEVICE_TYPE,  Next: ACC_DEVICE_NUM,  Up: OpenACC Environment Variables

7.1 'ACC_DEVICE_TYPE'
=====================

_Description_:
     Control the default device type to use when executing compute
     regions.  If unset, the code can be run on any device type,
     favoring a non-host device type.

     Supported values in GCC (if compiled in) are
        * 'host'
        * 'nvidia'
        * 'radeon'
_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section 4.1.


File: libgomp.info,  Node: ACC_DEVICE_NUM,  Next: ACC_PROFLIB,  Prev: ACC_DEVICE_TYPE,  Up: OpenACC Environment Variables

7.2 'ACC_DEVICE_NUM'
====================

_Description_:
     Control which device, identified by device number, is the default
     device.  The value must be a nonnegative integer less than the
     number of devices.  If unset, device number zero is used.
_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section 4.2.


File: libgomp.info,  Node: ACC_PROFLIB,  Prev: ACC_DEVICE_NUM,  Up: OpenACC Environment Variables

7.3 'ACC_PROFLIB'
=================

_Description_:
     Semicolon-separated list of dynamic libraries that are loaded as
     profiling libraries.  Each library must provide at least the
     'acc_register_library' routine.  Each library file is found as
     described by the documentation of 'dlopen' of your operating
     system.
_See also_:
     *note acc_register_library::, *note OpenACC Profiling Interface::

_Reference_:
     OpenACC specification v2.6 (https://www.openacc.org), section 4.3.


File: libgomp.info,  Node: CUDA Streams Usage,  Next: OpenACC Library Interoperability,  Prev: OpenACC Environment Variables,  Up: Top

8 CUDA Streams Usage
********************

This applies to the 'nvptx' plugin only.

   The library provides elements that perform asynchronous movement of
data and asynchronous operation of computing constructs.  This
asynchronous functionality is implemented by making use of CUDA
streams(1).

   The primary means by that the asynchronous functionality is accessed
is through the use of those OpenACC directives which make use of the
'async' and 'wait' clauses.  When the 'async' clause is first used with
a directive, it creates a CUDA stream.  If an 'async-argument' is used
with the 'async' clause, then the stream is associated with the
specified 'async-argument'.

   Following the creation of an association between a CUDA stream and
the 'async-argument' of an 'async' clause, both the 'wait' clause and
the 'wait' directive can be used.  When either the clause or directive
is used after stream creation, it creates a rendezvous point whereby
execution waits until all operations associated with the
'async-argument', that is, stream, have completed.

   Normally, the management of the streams that are created as a result
of using the 'async' clause, is done without any intervention by the
caller.  This implies the association between the 'async-argument' and
the CUDA stream is maintained for the lifetime of the program.  However,
this association can be changed through the use of the library function
'acc_set_cuda_stream'.  When the function 'acc_set_cuda_stream' is
called, the CUDA stream that was originally associated with the 'async'
clause is destroyed.  Caution should be taken when changing the
association as subsequent references to the 'async-argument' refer to a
different CUDA stream.

   ---------- Footnotes ----------

   (1) See "Stream Management" in "CUDA Driver API", TRM-06703-001,
Version 5.5, for additional information


File: libgomp.info,  Node: OpenACC Library Interoperability,  Next: OpenACC Profiling Interface,  Prev: CUDA Streams Usage,  Up: Top

9 OpenACC Library Interoperability
**********************************

9.1 Introduction
================

The OpenACC library uses the CUDA Driver API, and may interact with
programs that use the Runtime library directly, or another library based
on the Runtime library, e.g., CUBLAS(1). This chapter describes the use
cases and what changes are required in order to use both the OpenACC
library and the CUBLAS and Runtime libraries within a program.

9.2 First invocation: NVIDIA CUBLAS library API
===============================================

In this first use case (see below), a function in the CUBLAS library is
called prior to any of the functions in the OpenACC library.  More
specifically, the function 'cublasCreate()'.

   When invoked, the function initializes the library and allocates the
hardware resources on the host and the device on behalf of the caller.
Once the initialization and allocation has completed, a handle is
returned to the caller.  The OpenACC library also requires
initialization and allocation of hardware resources.  Since the CUBLAS
library has already allocated the hardware resources for the device, all
that is left to do is to initialize the OpenACC library and acquire the
hardware resources on the host.

   Prior to calling the OpenACC function that initializes the library
and allocate the host hardware resources, you need to acquire the device
number that was allocated during the call to 'cublasCreate()'.  The
invoking of the runtime library function 'cudaGetDevice()' accomplishes
this.  Once acquired, the device number is passed along with the device
type as parameters to the OpenACC library function
'acc_set_device_num()'.

   Once the call to 'acc_set_device_num()' has completed, the OpenACC
library uses the context that was created during the call to
'cublasCreate()'.  In other words, both libraries share the same
context.

         /* Create the handle */
         s = cublasCreate(&h);
         if (s != CUBLAS_STATUS_SUCCESS)
         {
             fprintf(stderr, "cublasCreate failed %d\n", s);
             exit(EXIT_FAILURE);
         }

         /* Get the device number */
         e = cudaGetDevice(&dev);
         if (e != cudaSuccess)
         {
             fprintf(stderr, "cudaGetDevice failed %d\n", e);
             exit(EXIT_FAILURE);
         }

         /* Initialize OpenACC library and use device 'dev' */
         acc_set_device_num(dev, acc_device_nvidia);

                              Use Case 1

9.3 First invocation: OpenACC library API
=========================================

In this second use case (see below), a function in the OpenACC library
is called prior to any of the functions in the CUBLAS library.  More
specifically, the function 'acc_set_device_num()'.

   In the use case presented here, the function 'acc_set_device_num()'
is used to both initialize the OpenACC library and allocate the hardware
resources on the host and the device.  In the call to the function, the
call parameters specify which device to use and what device type to use,
i.e., 'acc_device_nvidia'.  It should be noted that this is but one
method to initialize the OpenACC library and allocate the appropriate
hardware resources.  Other methods are available through the use of
environment variables and these is discussed in the next section.

   Once the call to 'acc_set_device_num()' has completed, other OpenACC
functions can be called as seen with multiple calls being made to
'acc_copyin()'.  In addition, calls can be made to functions in the
CUBLAS library.  In the use case a call to 'cublasCreate()' is made
subsequent to the calls to 'acc_copyin()'.  As seen in the previous use
case, a call to 'cublasCreate()' initializes the CUBLAS library and
allocates the hardware resources on the host and the device.  However,
since the device has already been allocated, 'cublasCreate()' only
initializes the CUBLAS library and allocates the appropriate hardware
resources on the host.  The context that was created as part of the
OpenACC initialization is shared with the CUBLAS library, similarly to
the first use case.

         dev = 0;

         acc_set_device_num(dev, acc_device_nvidia);

         /* Copy the first set to the device */
         d_X = acc_copyin(&h_X[0], N * sizeof (float));
         if (d_X == NULL)
         {
             fprintf(stderr, "copyin error h_X\n");
             exit(EXIT_FAILURE);
         }

         /* Copy the second set to the device */
         d_Y = acc_copyin(&h_Y1[0], N * sizeof (float));
         if (d_Y == NULL)
         {
             fprintf(stderr, "copyin error h_Y1\n");
             exit(EXIT_FAILURE);
         }

         /* Create the handle */
         s = cublasCreate(&h);
         if (s != CUBLAS_STATUS_SUCCESS)
         {
             fprintf(stderr, "cublasCreate failed %d\n", s);
             exit(EXIT_FAILURE);
         }

         /* Perform saxpy using CUBLAS library function */
         s = cublasSaxpy(h, N, &alpha, d_X, 1, d_Y, 1);
         if (s != CUBLAS_STATUS_SUCCESS)
         {
             fprintf(stderr, "cublasSaxpy failed %d\n", s);
             exit(EXIT_FAILURE);
         }

         /* Copy the results from the device */
         acc_memcpy_from_device(&h_Y1[0], d_Y, N * sizeof (float));

                              Use Case 2

9.4 OpenACC library and environment variables
=============================================

There are two environment variables associated with the OpenACC library
that may be used to control the device type and device number:
'ACC_DEVICE_TYPE' and 'ACC_DEVICE_NUM', respectively.  These two
environment variables can be used as an alternative to calling
'acc_set_device_num()'.  As seen in the second use case, the device type
and device number were specified using 'acc_set_device_num()'.  If
however, the aforementioned environment variables were set, then the
call to 'acc_set_device_num()' would not be required.

   The use of the environment variables is only relevant when an OpenACC
function is called prior to a call to 'cudaCreate()'.  If 'cudaCreate()'
is called prior to a call to an OpenACC function, then you must call
'acc_set_device_num()'(2)

   ---------- Footnotes ----------

   (1) See section 2.26, "Interactions with the CUDA Driver API" in
"CUDA Runtime API", Version 5.5, and section 2.27, "VDPAU
Interoperability", in "CUDA Driver API", TRM-06703-001, Version 5.5, for
additional information on library interoperability.

   (2) More complete information about 'ACC_DEVICE_TYPE' and
'ACC_DEVICE_NUM' can be found in sections 4.1 and 4.2 of the OpenACC
(https://www.openacc.org) Application Programming Interface”, Version
2.6.


File: libgomp.info,  Node: OpenACC Profiling Interface,  Next: OpenMP-Implementation Specifics,  Prev: OpenACC Library Interoperability,  Up: Top

10 OpenACC Profiling Interface
******************************

10.1 Implementation Status and Implementation-Defined Behavior
==============================================================

We're implementing the OpenACC Profiling Interface as defined by the
OpenACC 2.6 specification.  We're clarifying some aspects here as
_implementation-defined behavior_, while they're still under discussion
within the OpenACC Technical Committee.

   This implementation is tuned to keep the performance impact as low as
possible for the (very common) case that the Profiling Interface is not
enabled.  This is relevant, as the Profiling Interface affects all the
_hot_ code paths (in the target code, not in the offloaded code).  Users
of the OpenACC Profiling Interface can be expected to understand that
performance is impacted to some degree once the Profiling Interface is
enabled: for example, because of the _runtime_ (libgomp) calling into a
third-party _library_ for every event that has been registered.

   We're not yet accounting for the fact that 'OpenACC events may occur
during event processing'.  We just handle one case specially, as
required by CUDA 9.0 'nvprof', that 'acc_get_device_type' (*note
acc_get_device_type::)) may be called from 'acc_ev_device_init_start',
'acc_ev_device_init_end' callbacks.

   We're not yet implementing initialization via a
'acc_register_library' function that is either statically linked in, or
dynamically via 'LD_PRELOAD'.  Initialization via 'acc_register_library'
functions dynamically loaded via the 'ACC_PROFLIB' environment variable
does work, as does directly calling 'acc_prof_register',
'acc_prof_unregister', 'acc_prof_lookup'.

   As currently there are no inquiry functions defined, calls to
'acc_prof_lookup' always returns 'NULL'.

   There aren't separate _start_, _stop_ events defined for the event
types 'acc_ev_create', 'acc_ev_delete', 'acc_ev_alloc', 'acc_ev_free'.
It's not clear if these should be triggered before or after the actual
device-specific call is made.  We trigger them after.

   Remarks about data provided to callbacks:

'acc_prof_info.event_type'
     It's not clear if for _nested_ event callbacks (for example,
     'acc_ev_enqueue_launch_start' as part of a parent compute
     construct), this should be set for the nested event
     ('acc_ev_enqueue_launch_start'), or if the value of the parent
     construct should remain ('acc_ev_compute_construct_start').  In
     this implementation, the value generally corresponds to the
     innermost nested event type.

'acc_prof_info.device_type'

        * For 'acc_ev_compute_construct_start', and in presence of an
          'if' clause with _false_ argument, this still refers to the
          offloading device type.  It's not clear if that's the expected
          behavior.

        * Complementary to the item before, for
          'acc_ev_compute_construct_end', this is set to
          'acc_device_host' in presence of an 'if' clause with _false_
          argument.  It's not clear if that's the expected behavior.

'acc_prof_info.thread_id'
     Always '-1'; not yet implemented.

'acc_prof_info.async'

        * Not yet implemented correctly for
          'acc_ev_compute_construct_start'.

        * In a compute construct, for host-fallback
          execution/'acc_device_host' it always is 'acc_async_sync'.  It
          is unclear if that is the expected behavior.

        * For 'acc_ev_device_init_start' and 'acc_ev_device_init_end',
          it will always be 'acc_async_sync'.  It is unclear if that is
          the expected behavior.

'acc_prof_info.async_queue'
     There is no 'limited number of asynchronous queues' in libgomp.
     This always has the same value as 'acc_prof_info.async'.

'acc_prof_info.src_file'
     Always 'NULL'; not yet implemented.

'acc_prof_info.func_name'
     Always 'NULL'; not yet implemented.

'acc_prof_info.line_no'
     Always '-1'; not yet implemented.

'acc_prof_info.end_line_no'
     Always '-1'; not yet implemented.

'acc_prof_info.func_line_no'
     Always '-1'; not yet implemented.

'acc_prof_info.func_end_line_no'
     Always '-1'; not yet implemented.

'acc_event_info.event_type', 'acc_event_info.*.event_type'
     Relating to 'acc_prof_info.event_type' discussed above, in this
     implementation, this will always be the same value as
     'acc_prof_info.event_type'.

'acc_event_info.*.parent_construct'

        * Will be 'acc_construct_parallel' for all OpenACC compute
          constructs as well as many OpenACC Runtime API calls; should
          be the one matching the actual construct, or
          'acc_construct_runtime_api', respectively.

        * Will be 'acc_construct_enter_data' or
          'acc_construct_exit_data' when processing variable mappings
          specified in OpenACC _declare_ directives; should be
          'acc_construct_declare'.

        * For implicit 'acc_ev_device_init_start',
          'acc_ev_device_init_end', and explicit as well as implicit
          'acc_ev_alloc', 'acc_ev_free', 'acc_ev_enqueue_upload_start',
          'acc_ev_enqueue_upload_end', 'acc_ev_enqueue_download_start',
          and 'acc_ev_enqueue_download_end', will be
          'acc_construct_parallel'; should reflect the real parent
          construct.

'acc_event_info.*.implicit'
     For 'acc_ev_alloc', 'acc_ev_free', 'acc_ev_enqueue_upload_start',
     'acc_ev_enqueue_upload_end', 'acc_ev_enqueue_download_start', and
     'acc_ev_enqueue_download_end', this currently will be '1' also for
     explicit usage.

'acc_event_info.data_event.var_name'
     Always 'NULL'; not yet implemented.

'acc_event_info.data_event.host_ptr'
     For 'acc_ev_alloc', and 'acc_ev_free', this is always 'NULL'.

'typedef union acc_api_info'
     ... as printed in '5.2.3. Third Argument: API-Specific
     Information'.  This should obviously be 'typedef _struct_
     acc_api_info'.

'acc_api_info.device_api'
     Possibly not yet implemented correctly for
     'acc_ev_compute_construct_start', 'acc_ev_device_init_start',
     'acc_ev_device_init_end': will always be 'acc_device_api_none' for
     these event types.  For 'acc_ev_enter_data_start', it will be
     'acc_device_api_none' in some cases.

'acc_api_info.device_type'
     Always the same as 'acc_prof_info.device_type'.

'acc_api_info.vendor'
     Always '-1'; not yet implemented.

'acc_api_info.device_handle'
     Always 'NULL'; not yet implemented.

'acc_api_info.context_handle'
     Always 'NULL'; not yet implemented.

'acc_api_info.async_handle'
     Always 'NULL'; not yet implemented.

   Remarks about certain event types:

'acc_ev_device_init_start', 'acc_ev_device_init_end'

        * When a compute construct triggers implicit
          'acc_ev_device_init_start' and 'acc_ev_device_init_end'
          events, they currently aren't _nested within_ the
          corresponding 'acc_ev_compute_construct_start' and
          'acc_ev_compute_construct_end', but they're currently observed
          _before_ 'acc_ev_compute_construct_start'.  It's not clear
          what to do: the standard asks us provide a lot of details to
          the 'acc_ev_compute_construct_start' callback, without
          (implicitly) initializing a device before?

        * Callbacks for these event types will not be invoked for calls
          to the 'acc_set_device_type' and 'acc_set_device_num'
          functions.  It's not clear if they should be.

'acc_ev_enter_data_start', 'acc_ev_enter_data_end', 'acc_ev_exit_data_start', 'acc_ev_exit_data_end'

        * Callbacks for these event types will also be invoked for
          OpenACC _host_data_ constructs.  It's not clear if they should
          be.

        * Callbacks for these event types will also be invoked when
          processing variable mappings specified in OpenACC _declare_
          directives.  It's not clear if they should be.

   Callbacks for the following event types will be invoked, but dispatch
and information provided therein has not yet been thoroughly reviewed:

   * 'acc_ev_alloc'
   * 'acc_ev_free'
   * 'acc_ev_update_start', 'acc_ev_update_end'
   * 'acc_ev_enqueue_upload_start', 'acc_ev_enqueue_upload_end'
   * 'acc_ev_enqueue_download_start', 'acc_ev_enqueue_download_end'

   During device initialization, and finalization, respectively,
callbacks for the following event types will not yet be invoked:

   * 'acc_ev_alloc'
   * 'acc_ev_free'

   Callbacks for the following event types have not yet been
implemented, so currently won't be invoked:

   * 'acc_ev_device_shutdown_start', 'acc_ev_device_shutdown_end'
   * 'acc_ev_runtime_shutdown'
   * 'acc_ev_create', 'acc_ev_delete'
   * 'acc_ev_wait_start', 'acc_ev_wait_end'

   For the following runtime library functions, not all expected
callbacks will be invoked (mostly concerning implicit device
initialization):

   * 'acc_get_num_devices'
   * 'acc_set_device_type'
   * 'acc_get_device_type'
   * 'acc_set_device_num'
   * 'acc_get_device_num'
   * 'acc_init'
   * 'acc_shutdown'

   Aside from implicit device initialization, for the following runtime
library functions, no callbacks will be invoked for shared-memory
offloading devices (it's not clear if they should be):

   * 'acc_malloc'
   * 'acc_free'
   * 'acc_copyin', 'acc_present_or_copyin', 'acc_copyin_async'
   * 'acc_create', 'acc_present_or_create', 'acc_create_async'
   * 'acc_copyout', 'acc_copyout_async', 'acc_copyout_finalize',
     'acc_copyout_finalize_async'
   * 'acc_delete', 'acc_delete_async', 'acc_delete_finalize',
     'acc_delete_finalize_async'
   * 'acc_update_device', 'acc_update_device_async'
   * 'acc_update_self', 'acc_update_self_async'
   * 'acc_map_data', 'acc_unmap_data'
   * 'acc_memcpy_to_device', 'acc_memcpy_to_device_async'
   * 'acc_memcpy_from_device', 'acc_memcpy_from_device_async'


File: libgomp.info,  Node: OpenMP-Implementation Specifics,  Next: Offload-Target Specifics,  Prev: OpenACC Profiling Interface,  Up: Top

11 OpenMP-Implementation Specifics
**********************************

* Menu:

* Implementation-defined ICV Initialization::
* OpenMP Context Selectors::
* Memory allocation::


File: libgomp.info,  Node: Implementation-defined ICV Initialization,  Next: OpenMP Context Selectors,  Up: OpenMP-Implementation Specifics

11.1 Implementation-defined ICV Initialization
==============================================

AFFINITY-FORMAT-VAR    See *note OMP_AFFINITY_FORMAT::.
DEF-ALLOCATOR-VAR      See *note OMP_ALLOCATOR::.
MAX-ACTIVE-LEVELS-VAR  See *note OMP_MAX_ACTIVE_LEVELS::.
DYN-VAR                See *note OMP_DYNAMIC::.
NTHREADS-VAR           See *note OMP_NUM_THREADS::.
NUM-DEVICES-VAR        Number of non-host devices found by GCC's
                       run-time library
NUM-PROCS-VAR          The number of CPU cores on the initial device,
                       except that affinity settings might lead to a
                       smaller number.  On non-host devices, the value
                       of the NTHREADS-VAR ICV.
PLACE-PARTITION-VAR    See *note OMP_PLACES::.
RUN-SCHED-VAR          See *note OMP_SCHEDULE::.
STACKSIZE-VAR          See *note OMP_STACKSIZE::.
THREAD-LIMIT-VAR       See *note OMP_TEAMS_THREAD_LIMIT::
WAIT-POLICY-VAR        See *note OMP_WAIT_POLICY:: and
                       *note GOMP_SPINCOUNT::


File: libgomp.info,  Node: OpenMP Context Selectors,  Next: Memory allocation,  Prev: Implementation-defined ICV Initialization,  Up: OpenMP-Implementation Specifics

11.2 OpenMP Context Selectors
=============================

'vendor' is always 'gnu'.  References are to the GCC manual.

   For the host compiler, 'kind' always matches 'host'; for the
offloading architectures AMD GCN and Nvidia PTX, 'kind' always matches
'gpu'.  For the x86 family of computers, AMD GCN and Nvidia PTX the
following traits are supported in addition; while OpenMP is supported on
more architectures, GCC currently does not match any 'arch' or 'isa'
traits for those.

'arch'                                          'isa'
-----------------------------------------------------------------------
'x86', 'x86_64', 'i386', 'i486', 'i586',        See '-m...' flags in
'i686', 'ia32'                                  "x86 Options"
                                                (without '-m')
'amdgcn', 'gcn'                                 See '-march=' in
                                                "AMD GCN Options"(1)
'nvptx', 'nvptx64'                              See '-march=' in
                                                "Nvidia PTX Options"

   ---------- Footnotes ----------

   (1) Additionally, 'gfx803' is supported as an alias for 'fiji'.


File: libgomp.info,  Node: Memory allocation,  Prev: OpenMP Context Selectors,  Up: OpenMP-Implementation Specifics

11.3 Memory allocation
======================

The description below applies to:

   * Explicit use of the OpenMP API routines, see *note Memory
     Management Routines::.
   * The 'allocate' clause, except when the 'allocator' modifier is a
     constant expression with value 'omp_default_mem_alloc' and no
     'align' modifier has been specified.  (In that case, the normal
     'malloc' allocation is used.)
   * Using the 'allocate' directive for automatic/stack variables,
     except when the 'allocator' clause is a constant expression with
     value 'omp_default_mem_alloc' and no 'align' clause has been
     specified.  (In that case, the normal allocation is used: stack
     allocation and, sometimes for Fortran, also 'malloc' [depending on
     flags such as '-fstack-arrays'].)
   * Using the 'allocate' directive for variable in static memory is
     currently not supported (compile time error).
   * In Fortran, the 'allocators' directive and the executable
     'allocate' directive for Fortran pointers and allocatables is
     supported, but requires that files containing those directives has
     to be compiled with '-fopenmp-allocators'.  Additionally, all files
     that might explicitly or implicitly deallocate memory allocated
     that way must also be compiled with that option.

   For the available predefined allocators and, as applicable, their
associated predefined memory spaces and for the available traits and
their default values, see *note OMP_ALLOCATOR::.  Predefined allocators
without an associated memory space use the 'omp_default_mem_space'
memory space.

   For the memory spaces, the following applies:
   * 'omp_default_mem_space' is supported
   * 'omp_const_mem_space' maps to 'omp_default_mem_space'
   * 'omp_low_lat_mem_space' is only available on supported devices, and
     maps to 'omp_default_mem_space' otherwise.
   * 'omp_large_cap_mem_space' maps to 'omp_default_mem_space', unless
     the memkind library is available
   * 'omp_high_bw_mem_space' maps to 'omp_default_mem_space', unless the
     memkind library is available

   On Linux systems, where the memkind library
(https://github.com/memkind/memkind) ('libmemkind.so.0') is available at
runtime, it is used when creating memory allocators requesting

   * the memory space 'omp_high_bw_mem_space'
   * the memory space 'omp_large_cap_mem_space'
   * the 'partition' trait 'interleaved'; note that for
     'omp_large_cap_mem_space' the allocation will not be interleaved

   On Linux systems, where the numa library
(https://github.com/numactl/numactl) ('libnuma.so.1') is available at
runtime, it used when creating memory allocators requesting

   * the 'partition' trait 'nearest', except when both the libmemkind
     library is available and the memory space is either
     'omp_large_cap_mem_space' or 'omp_high_bw_mem_space'

   Note that the numa library will round up the allocation size to a
multiple of the system page size; therefore, consider using it only with
large data or by sharing allocations via the 'pool_size' trait.
Furthermore, the Linux kernel does not guarantee that an allocation will
always be on the nearest NUMA node nor that after reallocation the same
node will be used.  Note additionally that, on Linux, the default
setting of the memory placement policy is to use the current node;
therefore, unless the memory placement policy has been overridden, the
'partition' trait 'environment' (the default) will be effectively a
'nearest' allocation.

   Additional notes regarding the traits:
   * The 'pinned' trait is supported on Linux hosts, but is subject to
     the OS 'ulimit'/'rlimit' locked memory settings.
   * The default for the 'pool_size' trait is no pool and for every
     (re)allocation the associated library routine is called, which
     might internally use a memory pool.
   * For the 'partition' trait, the partition part size will be the same
     as the requested size (i.e.  'interleaved' or 'blocked' has no
     effect), except for 'interleaved' when the memkind library is
     available.  Furthermore, for 'nearest' and unless the numa library
     is available, the memory might not be on the same NUMA node as
     thread that allocated the memory; on Linux, this is in particular
     the case when the memory placement policy is set to preferred.
   * The 'access' trait has no effect such that memory is always
     accessible by all threads.
   * The 'sync_hint' trait has no effect.

   See also: *note Offload-Target Specifics::


File: libgomp.info,  Node: Offload-Target Specifics,  Next: The libgomp ABI,  Prev: OpenMP-Implementation Specifics,  Up: Top

12 Offload-Target Specifics
***************************

The following sections present notes on the offload-target specifics

* Menu:

* AMD Radeon::
* nvptx::


File: libgomp.info,  Node: AMD Radeon,  Next: nvptx,  Up: Offload-Target Specifics

12.1 AMD Radeon (GCN)
=====================

On the hardware side, there is the hierarchy (fine to coarse):
   * work item (thread)
   * wavefront
   * work group
   * compute unit (CU)

   All OpenMP and OpenACC levels are used, i.e.
   * OpenMP's simd and OpenACC's vector map to work items (thread)
   * OpenMP's threads ("parallel") and OpenACC's workers map to
     wavefronts
   * OpenMP's teams and OpenACC's gang use a threadpool with the size of
     the number of teams or gangs, respectively.

   The used sizes are
   * Number of teams is the specified 'num_teams' (OpenMP) or
     'num_gangs' (OpenACC) or otherwise the number of CU. It is limited
     by two times the number of CU.
   * Number of wavefronts is 4 for gfx900 and 16 otherwise;
     'num_threads' (OpenMP) and 'num_workers' (OpenACC) overrides this
     if smaller.
   * The wavefront has 102 scalars and 64 vectors
   * Number of workitems is always 64
   * The hardware permits maximally 40 workgroups/CU and 16
     wavefronts/workgroup up to a limit of 40 wavefronts in total per
     CU.
   * 80 scalars registers and 24 vector registers in non-kernel
     functions (the chosen procedure-calling API).
   * For the kernel itself: as many as register pressure demands (number
     of teams and number of threads, scaled down if registers are
     exhausted)

   The implementation remark:
   * I/O within OpenMP target regions and OpenACC parallel/kernels is
     supported using the C library 'printf' functions and the Fortran
     'print'/'write' statements.
   * Reverse offload regions (i.e.  'target' regions with
     'device(ancestor:1)') are processed serially per 'target' region
     such that the next reverse offload region is only executed after
     the previous one returned.
   * OpenMP code that has a 'requires' directive with
     'unified_shared_memory' will remove any GCN device from the list of
     available devices ("host fallback").
   * The available stack size can be changed using the 'GCN_STACK_SIZE'
     environment variable; the default is 32 kiB per thread.
   * Low-latency memory ('omp_low_lat_mem_space') is supported when the
     the 'access' trait is set to 'cgroup'.  The default pool size is
     automatically scaled to share the 64 kiB LDS memory between the
     number of teams configured to run on each compute-unit, but may be
     adjusted at runtime by setting environment variable
     'GOMP_GCN_LOWLAT_POOL=BYTES'.
   * 'omp_low_lat_mem_alloc' cannot be used with true low-latency memory
     because the definition implies the 'omp_atv_all' trait; main
     graphics memory is used instead.
   * 'omp_cgroup_mem_alloc', 'omp_pteam_mem_alloc', and
     'omp_thread_mem_alloc', all use low-latency memory as first
     preference, and fall back to main graphics memory when the
     low-latency pool is exhausted.


File: libgomp.info,  Node: nvptx,  Prev: AMD Radeon,  Up: Offload-Target Specifics

12.2 nvptx
==========

On the hardware side, there is the hierarchy (fine to coarse):
   * thread
   * warp
   * thread block
   * streaming multiprocessor

   All OpenMP and OpenACC levels are used, i.e.
   * OpenMP's simd and OpenACC's vector map to threads
   * OpenMP's threads ("parallel") and OpenACC's workers map to warps
   * OpenMP's teams and OpenACC's gang use a threadpool with the size of
     the number of teams or gangs, respectively.

   The used sizes are
   * The 'warp_size' is always 32
   * CUDA kernel launched: 'dim={#teams,1,1},
     blocks={#threads,warp_size,1}'.
   * The number of teams is limited by the number of blocks the device
     can host simultaneously.

   Additional information can be obtained by setting the environment
variable to 'GOMP_DEBUG=1' (very verbose; grep for 'kernel.*launch' for
launch parameters).

   GCC generates generic PTX ISA code, which is just-in-time compiled by
CUDA, which caches the JIT in the user's directory (see CUDA
documentation; can be tuned by the environment variables
'CUDA_CACHE_{DISABLE,MAXSIZE,PATH}'.

   Note: While PTX ISA is generic, the '-mptx=' and '-march='
commandline options still affect the used PTX ISA code and, thus, the
requirements on CUDA version and hardware.

   The implementation remark:
   * I/O within OpenMP target regions and OpenACC parallel/kernels is
     supported using the C library 'printf' functions.  Note that the
     Fortran 'print'/'write' statements are not supported, yet.
   * Compilation OpenMP code that contains 'requires reverse_offload'
     requires at least '-march=sm_35', compiling for '-march=sm_30' is
     not supported.
   * For code containing reverse offload (i.e.  'target' regions with
     'device(ancestor:1)'), there is a slight performance penalty for
     _all_ target regions, consisting mostly of shutdown delay Per
     device, reverse offload regions are processed serially such that
     the next reverse offload region is only executed after the previous
     one returned.
   * OpenMP code that has a 'requires' directive with
     'unified_shared_memory' will remove any nvptx device from the list
     of available devices ("host fallback").
   * The default per-warp stack size is 128 kiB; see also '-msoft-stack'
     in the GCC manual.
   * The OpenMP routines 'omp_target_memcpy_rect' and
     'omp_target_memcpy_rect_async' and the 'target update' directive
     for non-contiguous list items will use the 2D and 3D memory-copy
     functions of the CUDA library.  Higher dimensions will call those
     functions in a loop and are therefore supported.
   * Low-latency memory ('omp_low_lat_mem_space') is supported when the
     the 'access' trait is set to 'cgroup', the ISA is at least 'sm_53',
     and the PTX version is at least 4.1.  The default pool size is 8
     kiB per team, but may be adjusted at runtime by setting environment
     variable 'GOMP_NVPTX_LOWLAT_POOL=BYTES'.  The maximum value is
     limited by the available hardware, and care should be taken that
     the selected pool size does not unduly limit the number of teams
     that can run simultaneously.
   * 'omp_low_lat_mem_alloc' cannot be used with true low-latency memory
     because the definition implies the 'omp_atv_all' trait; main
     graphics memory is used instead.
   * 'omp_cgroup_mem_alloc', 'omp_pteam_mem_alloc', and
     'omp_thread_mem_alloc', all use low-latency memory as first
     preference, and fall back to main graphics memory when the
     low-latency pool is exhausted.


File: libgomp.info,  Node: The libgomp ABI,  Next: Reporting Bugs,  Prev: Offload-Target Specifics,  Up: Top

13 The libgomp ABI
******************

The following sections present notes on the external ABI as presented by
libgomp.  Only maintainers should need them.

* Menu:

* Implementing MASTER construct::
* Implementing CRITICAL construct::
* Implementing ATOMIC construct::
* Implementing FLUSH construct::
* Implementing BARRIER construct::
* Implementing THREADPRIVATE construct::
* Implementing PRIVATE clause::
* Implementing FIRSTPRIVATE LASTPRIVATE COPYIN and COPYPRIVATE clauses::
* Implementing REDUCTION clause::
* Implementing PARALLEL construct::
* Implementing FOR construct::
* Implementing ORDERED construct::
* Implementing SECTIONS construct::
* Implementing SINGLE construct::
* Implementing OpenACC's PARALLEL construct::


File: libgomp.info,  Node: Implementing MASTER construct,  Next: Implementing CRITICAL construct,  Up: The libgomp ABI

13.1 Implementing MASTER construct
==================================

     if (omp_get_thread_num () == 0)
       block

   Alternately, we generate two copies of the parallel subfunction and
only include this in the version run by the primary thread.  Surely this
is not worthwhile though...


File: libgomp.info,  Node: Implementing CRITICAL construct,  Next: Implementing ATOMIC construct,  Prev: Implementing MASTER construct,  Up: The libgomp ABI

13.2 Implementing CRITICAL construct
====================================

Without a specified name,

       void GOMP_critical_start (void);
       void GOMP_critical_end (void);

   so that we don't get COPY relocations from libgomp to the main
application.

   With a specified name, use omp_set_lock and omp_unset_lock with name
being transformed into a variable declared like

       omp_lock_t gomp_critical_user_<name> __attribute__((common))

   Ideally the ABI would specify that all zero is a valid unlocked
state, and so we wouldn't need to initialize this at startup.


File: libgomp.info,  Node: Implementing ATOMIC construct,  Next: Implementing FLUSH construct,  Prev: Implementing CRITICAL construct,  Up: The libgomp ABI

13.3 Implementing ATOMIC construct
==================================

The target should implement the '__sync' builtins.

   Failing that we could add

       void GOMP_atomic_enter (void)
       void GOMP_atomic_exit (void)

   which reuses the regular lock code, but with yet another lock object
private to the library.


File: libgomp.info,  Node: Implementing FLUSH construct,  Next: Implementing BARRIER construct,  Prev: Implementing ATOMIC construct,  Up: The libgomp ABI

13.4 Implementing FLUSH construct
=================================

Expands to the '__sync_synchronize' builtin.


File: libgomp.info,  Node: Implementing BARRIER construct,  Next: Implementing THREADPRIVATE construct,  Prev: Implementing FLUSH construct,  Up: The libgomp ABI

13.5 Implementing BARRIER construct
===================================

       void GOMP_barrier (void)


File: libgomp.info,  Node: Implementing THREADPRIVATE construct,  Next: Implementing PRIVATE clause,  Prev: Implementing BARRIER construct,  Up: The libgomp ABI

13.6 Implementing THREADPRIVATE construct
=========================================

In _most_ cases we can map this directly to '__thread'.  Except that OMP
allows constructors for C++ objects.  We can either refuse to support
this (how often is it used?)  or we can implement something akin to
.ctors.

   Even more ideally, this ctor feature is handled by extensions to the
main pthreads library.  Failing that, we can have a set of entry points
to register ctor functions to be called.


File: libgomp.info,  Node: Implementing PRIVATE clause,  Next: Implementing FIRSTPRIVATE LASTPRIVATE COPYIN and COPYPRIVATE clauses,  Prev: Implementing THREADPRIVATE construct,  Up: The libgomp ABI

13.7 Implementing PRIVATE clause
================================

In association with a PARALLEL, or within the lexical extent of a
PARALLEL block, the variable becomes a local variable in the parallel
subfunction.

   In association with FOR or SECTIONS blocks, create a new automatic
variable within the current function.  This preserves the semantic of
new variable creation.


File: libgomp.info,  Node: Implementing FIRSTPRIVATE LASTPRIVATE COPYIN and COPYPRIVATE clauses,  Next: Implementing REDUCTION clause,  Prev: Implementing PRIVATE clause,  Up: The libgomp ABI

13.8 Implementing FIRSTPRIVATE LASTPRIVATE COPYIN and COPYPRIVATE clauses
=========================================================================

This seems simple enough for PARALLEL blocks.  Create a private struct
for communicating between the parent and subfunction.  In the parent,
copy in values for scalar and "small" structs; copy in addresses for
others TREE_ADDRESSABLE types.  In the subfunction, copy the value into
the local variable.

   It is not clear what to do with bare FOR or SECTION blocks.  The only
thing I can figure is that we do something like:

     #pragma omp for firstprivate(x) lastprivate(y)
     for (int i = 0; i < n; ++i)
       body;

   which becomes

     {
       int x = x, y;

       // for stuff

       if (i == n)
         y = y;
     }

   where the "x=x" and "y=y" assignments actually have different uids
for the two variables, i.e.  not something you could write directly in
C. Presumably this only makes sense if the "outer" x and y are global
variables.

   COPYPRIVATE would work the same way, except the structure broadcast
would have to happen via SINGLE machinery instead.


File: libgomp.info,  Node: Implementing REDUCTION clause,  Next: Implementing PARALLEL construct,  Prev: Implementing FIRSTPRIVATE LASTPRIVATE COPYIN and COPYPRIVATE clauses,  Up: The libgomp ABI

13.9 Implementing REDUCTION clause
==================================

The private struct mentioned in the previous section should have a
pointer to an array of the type of the variable, indexed by the thread's
TEAM_ID.  The thread stores its final value into the array, and after
the barrier, the primary thread iterates over the array to collect the
values.


File: libgomp.info,  Node: Implementing PARALLEL construct,  Next: Implementing FOR construct,  Prev: Implementing REDUCTION clause,  Up: The libgomp ABI

13.10 Implementing PARALLEL construct
=====================================

       #pragma omp parallel
       {
         body;
       }

   becomes

       void subfunction (void *data)
       {
         use data;
         body;
       }

       setup data;
       GOMP_parallel_start (subfunction, &data, num_threads);
       subfunction (&data);
       GOMP_parallel_end ();

       void GOMP_parallel_start (void (*fn)(void *), void *data, unsigned num_threads)

   The FN argument is the subfunction to be run in parallel.

   The DATA argument is a pointer to a structure used to communicate
data in and out of the subfunction, as discussed above with respect to
FIRSTPRIVATE et al.

   The NUM_THREADS argument is 1 if an IF clause is present and false,
or the value of the NUM_THREADS clause, if present, or 0.

   The function needs to create the appropriate number of threads and/or
launch them from the dock.  It needs to create the team structure and
assign team ids.

       void GOMP_parallel_end (void)

   Tears down the team and returns us to the previous
'omp_in_parallel()' state.


File: libgomp.info,  Node: Implementing FOR construct,  Next: Implementing ORDERED construct,  Prev: Implementing PARALLEL construct,  Up: The libgomp ABI

13.11 Implementing FOR construct
================================

       #pragma omp parallel for
       for (i = lb; i <= ub; i++)
         body;

   becomes

       void subfunction (void *data)
       {
         long _s0, _e0;
         while (GOMP_loop_static_next (&_s0, &_e0))
         {
           long _e1 = _e0, i;
           for (i = _s0; i < _e1; i++)
             body;
         }
         GOMP_loop_end_nowait ();
       }

       GOMP_parallel_loop_static (subfunction, NULL, 0, lb, ub+1, 1, 0);
       subfunction (NULL);
       GOMP_parallel_end ();

       #pragma omp for schedule(runtime)
       for (i = 0; i < n; i++)
         body;

   becomes

       {
         long i, _s0, _e0;
         if (GOMP_loop_runtime_start (0, n, 1, &_s0, &_e0))
           do {
             long _e1 = _e0;
             for (i = _s0, i < _e0; i++)
               body;
           } while (GOMP_loop_runtime_next (&_s0, _&e0));
         GOMP_loop_end ();
       }

   Note that while it looks like there is trickiness to propagating a
non-constant STEP, there isn't really.  We're explicitly allowed to
evaluate it as many times as we want, and any variables involved should
automatically be handled as PRIVATE or SHARED like any other variables.
So the expression should remain evaluable in the subfunction.  We can
also pull it into a local variable if we like, but since its supposed to
remain unchanged, we can also not if we like.

   If we have SCHEDULE(STATIC), and no ORDERED, then we ought to be able
to get away with no work-sharing context at all, since we can simply
perform the arithmetic directly in each thread to divide up the
iterations.  Which would mean that we wouldn't need to call any of these
routines.

   There are separate routines for handling loops with an ORDERED
clause.  Bookkeeping for that is non-trivial...


File: libgomp.info,  Node: Implementing ORDERED construct,  Next: Implementing SECTIONS construct,  Prev: Implementing FOR construct,  Up: The libgomp ABI

13.12 Implementing ORDERED construct
====================================

       void GOMP_ordered_start (void)
       void GOMP_ordered_end (void)


File: libgomp.info,  Node: Implementing SECTIONS construct,  Next: Implementing SINGLE construct,  Prev: Implementing ORDERED construct,  Up: The libgomp ABI

13.13 Implementing SECTIONS construct
=====================================

A block as

       #pragma omp sections
       {
         #pragma omp section
         stmt1;
         #pragma omp section
         stmt2;
         #pragma omp section
         stmt3;
       }

   becomes

       for (i = GOMP_sections_start (3); i != 0; i = GOMP_sections_next ())
         switch (i)
           {
           case 1:
             stmt1;
             break;
           case 2:
             stmt2;
             break;
           case 3:
             stmt3;
             break;
           }
       GOMP_barrier ();


File: libgomp.info,  Node: Implementing SINGLE construct,  Next: Implementing OpenACC's PARALLEL construct,  Prev: Implementing SECTIONS construct,  Up: The libgomp ABI

13.14 Implementing SINGLE construct
===================================

A block like

       #pragma omp single
       {
         body;
       }

   becomes

       if (GOMP_single_start ())
         body;
       GOMP_barrier ();

   while

       #pragma omp single copyprivate(x)
         body;

   becomes

       datap = GOMP_single_copy_start ();
       if (datap == NULL)
         {
           body;
           data.x = x;
           GOMP_single_copy_end (&data);
         }
       else
         x = datap->x;
       GOMP_barrier ();


File: libgomp.info,  Node: Implementing OpenACC's PARALLEL construct,  Prev: Implementing SINGLE construct,  Up: The libgomp ABI

13.15 Implementing OpenACC's PARALLEL construct
===============================================

       void GOACC_parallel ()


File: libgomp.info,  Node: Reporting Bugs,  Next: Copying,  Prev: The libgomp ABI,  Up: Top

14 Reporting Bugs
*****************

Bugs in the GNU Offloading and Multi Processing Runtime Library should
be reported via Bugzilla (https://gcc.gnu.org/bugzilla/).  Please add
"openacc", or "openmp", or both to the keywords field in the bug report,
as appropriate.


File: libgomp.info,  Node: Copying,  Next: GNU Free Documentation License,  Prev: Reporting Bugs,  Up: Top

GNU General Public License
**************************

                        Version 3, 29 June 2007

     Copyright (C) 2007 Free Software Foundation, Inc. <https://www.fsf.org>

     Everyone is permitted to copy and distribute verbatim copies of this
     license document, but changing it is not allowed.

Preamble
========

The GNU General Public License is a free, copyleft license for software
and other kinds of works.

   The licenses for most software and other practical works are designed
to take away your freedom to share and change the works.  By contrast,
the GNU General Public License is intended to guarantee your freedom to
share and change all versions of a program-to make sure it remains free
software for all its users.  We, the Free Software Foundation, use the
GNU General Public License for most of our software; it applies also to
any other work released this way by its authors.  You can apply it to
your programs, too.

   When we speak of free software, we are referring to freedom, not
price.  Our General Public Licenses are designed to make sure that you
have the freedom to distribute copies of free software (and charge for
them if you wish), that you receive source code or can get it if you
want it, that you can change the software or use pieces of it in new
free programs, and that you know you can do these things.

   To protect your rights, we need to prevent others from denying you
these rights or asking you to surrender the rights.  Therefore, you have
certain responsibilities if you distribute copies of the software, or if
you modify it: responsibilities to respect the freedom of others.

   For example, if you distribute copies of such a program, whether
gratis or for a fee, you must pass on to the recipients the same
freedoms that you received.  You must make sure that they, too, receive
or can get the source code.  And you must show them these terms so they
know their rights.

   Developers that use the GNU GPL protect your rights with two steps:
(1) assert copyright on the software, and (2) offer you this License
giving you legal permission to copy, distribute and/or modify it.

   For the developers' and authors' protection, the GPL clearly explains
that there is no warranty for this free software.  For both users' and
authors' sake, the GPL requires that modified versions be marked as
changed, so that their problems will not be attributed erroneously to
authors of previous versions.

   Some devices are designed to deny users access to install or run
modified versions of the software inside them, although the manufacturer
can do so.  This is fundamentally incompatible with the aim of
protecting users' freedom to change the software.  The systematic
pattern of such abuse occurs in the area of products for individuals to
use, which is precisely where it is most unacceptable.  Therefore, we
have designed this version of the GPL to prohibit the practice for those
products.  If such problems arise substantially in other domains, we
stand ready to extend this provision to those domains in future versions
of the GPL, as needed to protect the freedom of users.

   Finally, every program is threatened constantly by software patents.
States should not allow patents to restrict development and use of
software on general-purpose computers, but in those that do, we wish to
avoid the special danger that patents applied to a free program could
make it effectively proprietary.  To prevent this, the GPL assures that
patents cannot be used to render the program non-free.

   The precise terms and conditions for copying, distribution and
modification follow.

TERMS AND CONDITIONS
====================

  0. Definitions.

     "This License" refers to version 3 of the GNU General Public
     License.

     "Copyright" also means copyright-like laws that apply to other
     kinds of works, such as semiconductor masks.

     "The Program" refers to any copyrightable work licensed under this
     License.  Each licensee is addressed as "you".  "Licensees" and
     "recipients" may be individuals or organizations.

     To "modify" a work means to copy from or adapt all or part of the
     work in a fashion requiring copyright permission, other than the
     making of an exact copy.  The resulting work is called a "modified
     version" of the earlier work or a work "based on" the earlier work.

     A "covered work" means either the unmodified Program or a work
     based on the Program.

     To "propagate" a work means to do anything with it that, without
     permission, would make you directly or secondarily liable for
     infringement under applicable copyright law, except executing it on
     a computer or modifying a private copy.  Propagation includes
     copying, distribution (with or without modification), making
     available to the public, and in some countries other activities as
     well.

     To "convey" a work means any kind of propagation that enables other
     parties to make or receive copies.  Mere interaction with a user
     through a computer network, with no transfer of a copy, is not
     conveying.

     An interactive user interface displays "Appropriate Legal Notices"
     to the extent that it includes a convenient and prominently visible
     feature that (1) displays an appropriate copyright notice, and (2)
     tells the user that there is no warranty for the work (except to
     the extent that warranties are provided), that licensees may convey
     the work under this License, and how to view a copy of this
     License.  If the interface presents a list of user commands or
     options, such as a menu, a prominent item in the list meets this
     criterion.

  1. Source Code.

     The "source code" for a work means the preferred form of the work
     for making modifications to it.  "Object code" means any non-source
     form of a work.

     A "Standard Interface" means an interface that either is an
     official standard defined by a recognized standards body, or, in
     the case of interfaces specified for a particular programming
     language, one that is widely used among developers working in that
     language.

     The "System Libraries" of an executable work include anything,
     other than the work as a whole, that (a) is included in the normal
     form of packaging a Major Component, but which is not part of that
     Major Component, and (b) serves only to enable use of the work with
     that Major Component, or to implement a Standard Interface for
     which an implementation is available to the public in source code
     form.  A "Major Component", in this context, means a major
     essential component (kernel, window system, and so on) of the
     specific operating system (if any) on which the executable work
     runs, or a compiler used to produce the work, or an object code
     interpreter used to run it.

     The "Corresponding Source" for a work in object code form means all
     the source code needed to generate, install, and (for an executable
     work) run the object code and to modify the work, including scripts
     to control those activities.  However, it does not include the
     work's System Libraries, or general-purpose tools or generally
     available free programs which are used unmodified in performing
     those activities but which are not part of the work.  For example,
     Corresponding Source includes interface definition files associated
     with source files for the work, and the source code for shared
     libraries and dynamically linked subprograms that the work is
     specifically designed to require, such as by intimate data
     communication or control flow between those subprograms and other
     parts of the work.

     The Corresponding Source need not include anything that users can
     regenerate automatically from other parts of the Corresponding
     Source.

     The Corresponding Source for a work in source code form is that
     same work.

  2. Basic Permissions.

     All rights granted under this License are granted for the term of
     copyright on the Program, and are irrevocable provided the stated
     conditions are met.  This License explicitly affirms your unlimited
     permission to run the unmodified Program.  The output from running
     a covered work is covered by this License only if the output, given
     its content, constitutes a covered work.  This License acknowledges
     your rights of fair use or other equivalent, as provided by
     copyright law.

     You may make, run and propagate covered works that you do not
     convey, without conditions so long as your license otherwise
     remains in force.  You may convey covered works to others for the
     sole purpose of having them make modifications exclusively for you,
     or provide you with facilities for running those works, provided
     that you comply with the terms of this License in conveying all
     material for which you do not control copyright.  Those thus making
     or running the covered works for you must do so exclusively on your
     behalf, under your direction and control, on terms that prohibit
     them from making any copies of your copyrighted material outside
     their relationship with you.

     Conveying under any other circumstances is permitted solely under
     the conditions stated below.  Sublicensing is not allowed; section
     10 makes it unnecessary.

  3. Protecting Users' Legal Rights From Anti-Circumvention Law.

     No covered work shall be deemed part of an effective technological
     measure under any applicable law fulfilling obligations under
     article 11 of the WIPO copyright treaty adopted on 20 December
     1996, or similar laws prohibiting or restricting circumvention of
     such measures.

     When you convey a covered work, you waive any legal power to forbid
     circumvention of technological measures to the extent such
     circumvention is effected by exercising rights under this License
     with respect to the covered work, and you disclaim any intention to
     limit operation or modification of the work as a means of
     enforcing, against the work's users, your or third parties' legal
     rights to forbid circumvention of technological measures.

  4. Conveying Verbatim Copies.

     You may convey verbatim copies of the Program's source code as you
     receive it, in any medium, provided that you conspicuously and
     appropriately publish on each copy an appropriate copyright notice;
     keep intact all notices stating that this License and any
     non-permissive terms added in accord with section 7 apply to the
     code; keep intact all notices of the absence of any warranty; and
     give all recipients a copy of this License along with the Program.

     You may charge any price or no price for each copy that you convey,
     and you may offer support or warranty protection for a fee.

  5. Conveying Modified Source Versions.

     You may convey a work based on the Program, or the modifications to
     produce it from the Program, in the form of source code under the
     terms of section 4, provided that you also meet all of these
     conditions:

       a. The work must carry prominent notices stating that you
          modified it, and giving a relevant date.

       b. The work must carry prominent notices stating that it is
          released under this License and any conditions added under
          section 7.  This requirement modifies the requirement in
          section 4 to "keep intact all notices".

       c. You must license the entire work, as a whole, under this
          License to anyone who comes into possession of a copy.  This
          License will therefore apply, along with any applicable
          section 7 additional terms, to the whole of the work, and all
          its parts, regardless of how they are packaged.  This License
          gives no permission to license the work in any other way, but
          it does not invalidate such permission if you have separately
          received it.

       d. If the work has interactive user interfaces, each must display
          Appropriate Legal Notices; however, if the Program has
          interactive interfaces that do not display Appropriate Legal
          Notices, your work need not make them do so.

     A compilation of a covered work with other separate and independent
     works, which are not by their nature extensions of the covered
     work, and which are not combined with it such as to form a larger
     program, in or on a volume of a storage or distribution medium, is
     called an "aggregate" if the compilation and its resulting
     copyright are not used to limit the access or legal rights of the
     compilation's users beyond what the individual works permit.
     Inclusion of a covered work in an aggregate does not cause this
     License to apply to the other parts of the aggregate.

  6. Conveying Non-Source Forms.

     You may convey a covered work in object code form under the terms
     of sections 4 and 5, provided that you also convey the
     machine-readable Corresponding Source under the terms of this
     License, in one of these ways:

       a. Convey the object code in, or embodied in, a physical product
          (including a physical distribution medium), accompanied by the
          Corresponding Source fixed on a durable physical medium
          customarily used for software interchange.

       b. Convey the object code in, or embodied in, a physical product
          (including a physical distribution medium), accompanied by a
          written offer, valid for at least three years and valid for as
          long as you offer spare parts or customer support for that
          product model, to give anyone who possesses the object code
          either (1) a copy of the Corresponding Source for all the
          software in the product that is covered by this License, on a
          durable physical medium customarily used for software
          interchange, for a price no more than your reasonable cost of
          physically performing this conveying of source, or (2) access
          to copy the Corresponding Source from a network server at no
          charge.

       c. Convey individual copies of the object code with a copy of the
          written offer to provide the Corresponding Source.  This
          alternative is allowed only occasionally and noncommercially,
          and only if you received the object code with such an offer,
          in accord with subsection 6b.

       d. Convey the object code by offering access from a designated
          place (gratis or for a charge), and offer equivalent access to
          the Corresponding Source in the same way through the same
          place at no further charge.  You need not require recipients
          to copy the Corresponding Source along with the object code.
          If the place to copy the object code is a network server, the
          Corresponding Source may be on a different server (operated by
          you or a third party) that supports equivalent copying
          facilities, provided you maintain clear directions next to the
          object code saying where to find the Corresponding Source.
          Regardless of what server hosts the Corresponding Source, you
          remain obligated to ensure that it is available for as long as
          needed to satisfy these requirements.

       e. Convey the object code using peer-to-peer transmission,
          provided you inform other peers where the object code and
          Corresponding Source of the work are being offered to the
          general public at no charge under subsection 6d.

     A separable portion of the object code, whose source code is
     excluded from the Corresponding Source as a System Library, need
     not be included in conveying the object code work.

     A "User Product" is either (1) a "consumer product", which means
     any tangible personal property which is normally used for personal,
     family, or household purposes, or (2) anything designed or sold for
     incorporation into a dwelling.  In determining whether a product is
     a consumer product, doubtful cases shall be resolved in favor of
     coverage.  For a particular product received by a particular user,
     "normally used" refers to a typical or common use of that class of
     product, regardless of the status of the particular user or of the
     way in which the particular user actually uses, or expects or is
     expected to use, the product.  A product is a consumer product
     regardless of whether the product has substantial commercial,
     industrial or non-consumer uses, unless such uses represent the
     only significant mode of use of the product.

     "Installation Information" for a User Product means any methods,
     procedures, authorization keys, or other information required to
     install and execute modified versions of a covered work in that
     User Product from a modified version of its Corresponding Source.
     The information must suffice to ensure that the continued
     functioning of the modified object code is in no case prevented or
     interfered with solely because modification has been made.

     If you convey an object code work under this section in, or with,
     or specifically for use in, a User Product, and the conveying
     occurs as part of a transaction in which the right of possession
     and use of the User Product is transferred to the recipient in
     perpetuity or for a fixed term (regardless of how the transaction
     is characterized), the Corresponding Source conveyed under this
     section must be accompanied by the Installation Information.  But
     this requirement does not apply if neither you nor any third party
     retains the ability to install modified object code on the User
     Product (for example, the work has been installed in ROM).

     The requirement to provide Installation Information does not
     include a requirement to continue to provide support service,
     warranty, or updates for a work that has been modified or installed
     by the recipient, or for the User Product in which it has been
     modified or installed.  Access to a network may be denied when the
     modification itself materially and adversely affects the operation
     of the network or violates the rules and protocols for
     communication across the network.

     Corresponding Source conveyed, and Installation Information
     provided, in accord with this section must be in a format that is
     publicly documented (and with an implementation available to the
     public in source code form), and must require no special password
     or key for unpacking, reading or copying.

  7. Additional Terms.

     "Additional permissions" are terms that supplement the terms of
     this License by making exceptions from one or more of its
     conditions.  Additional permissions that are applicable to the
     entire Program shall be treated as though they were included in
     this License, to the extent that they are valid under applicable
     law.  If additional permissions apply only to part of the Program,
     that part may be used separately under those permissions, but the
     entire Program remains governed by this License without regard to
     the additional permissions.

     When you convey a copy of a covered work, you may at your option
     remove any additional permissions from that copy, or from any part
     of it.  (Additional permissions may be written to require their own
     removal in certain cases when you modify the work.)  You may place
     additional permissions on material, added by you to a covered work,
     for which you have or can give appropriate copyright permission.

     Notwithstanding any other provision of this License, for material
     you add to a covered work, you may (if authorized by the copyright
     holders of that material) supplement the terms of this License with
     terms:

       a. Disclaiming warranty or limiting liability differently from
          the terms of sections 15 and 16 of this License; or

       b. Requiring preservation of specified reasonable legal notices
          or author attributions in that material or in the Appropriate
          Legal Notices displayed by works containing it; or

       c. Prohibiting misrepresentation of the origin of that material,
          or requiring that modified versions of such material be marked
          in reasonable ways as different from the original version; or

       d. Limiting the use for publicity purposes of names of licensors
          or authors of the material; or

       e. Declining to grant rights under trademark law for use of some
          trade names, trademarks, or service marks; or

       f. Requiring indemnification of licensors and authors of that
          material by anyone who conveys the material (or modified
          versions of it) with contractual assumptions of liability to
          the recipient, for any liability that these contractual
          assumptions directly impose on those licensors and authors.

     All other non-permissive additional terms are considered "further
     restrictions" within the meaning of section 10.  If the Program as
     you received it, or any part of it, contains a notice stating that
     it is governed by this License along with a term that is a further
     restriction, you may remove that term.  If a license document
     contains a further restriction but permits relicensing or conveying
     under this License, you may add to a covered work material governed
     by the terms of that license document, provided that the further
     restriction does not survive such relicensing or conveying.

     If you add terms to a covered work in accord with this section, you
     must place, in the relevant source files, a statement of the
     additional terms that apply to those files, or a notice indicating
     where to find the applicable terms.

     Additional terms, permissive or non-permissive, may be stated in
     the form of a separately written license, or stated as exceptions;
     the above requirements apply either way.

  8. Termination.

     You may not propagate or modify a covered work except as expressly
     provided under this License.  Any attempt otherwise to propagate or
     modify it is void, and will automatically terminate your rights
     under this License (including any patent licenses granted under the
     third paragraph of section 11).

     However, if you cease all violation of this License, then your
     license from a particular copyright holder is reinstated (a)
     provisionally, unless and until the copyright holder explicitly and
     finally terminates your license, and (b) permanently, if the
     copyright holder fails to notify you of the violation by some
     reasonable means prior to 60 days after the cessation.

     Moreover, your license from a particular copyright holder is
     reinstated permanently if the copyright holder notifies you of the
     violation by some reasonable means, this is the first time you have
     received notice of violation of this License (for any work) from
     that copyright holder, and you cure the violation prior to 30 days
     after your receipt of the notice.

     Termination of your rights under this section does not terminate
     the licenses of parties who have received copies or rights from you
     under this License.  If your rights have been terminated and not
     permanently reinstated, you do not qualify to receive new licenses
     for the same material under section 10.

  9. Acceptance Not Required for Having Copies.

     You are not required to accept this License in order to receive or
     run a copy of the Program.  Ancillary propagation of a covered work
     occurring solely as a consequence of using peer-to-peer
     transmission to receive a copy likewise does not require
     acceptance.  However, nothing other than this License grants you
     permission to propagate or modify any covered work.  These actions
     infringe copyright if you do not accept this License.  Therefore,
     by modifying or propagating a covered work, you indicate your
     acceptance of this License to do so.

  10. Automatic Licensing of Downstream Recipients.

     Each time you convey a covered work, the recipient automatically
     receives a license from the original licensors, to run, modify and
     propagate that work, subject to this License.  You are not
     responsible for enforcing compliance by third parties with this
     License.

     An "entity transaction" is a transaction transferring control of an
     organization, or substantially all assets of one, or subdividing an
     organization, or merging organizations.  If propagation of a
     covered work results from an entity transaction, each party to that
     transaction who receives a copy of the work also receives whatever
     licenses to the work the party's predecessor in interest had or
     could give under the previous paragraph, plus a right to possession
     of the Corresponding Source of the work from the predecessor in
     interest, if the predecessor has it or can get it with reasonable
     efforts.

     You may not impose any further restrictions on the exercise of the
     rights granted or affirmed under this License.  For example, you
     may not impose a license fee, royalty, or other charge for exercise
     of rights granted under this License, and you may not initiate
     litigation (including a cross-claim or counterclaim in a lawsuit)
     alleging that any patent claim is infringed by making, using,
     selling, offering for sale, or importing the Program or any portion
     of it.

  11. Patents.

     A "contributor" is a copyright holder who authorizes use under this
     License of the Program or a work on which the Program is based.
     The work thus licensed is called the contributor's "contributor
     version".

     A contributor's "essential patent claims" are all patent claims
     owned or controlled by the contributor, whether already acquired or
     hereafter acquired, that would be infringed by some manner,
     permitted by this License, of making, using, or selling its
     contributor version, but do not include claims that would be
     infringed only as a consequence of further modification of the
     contributor version.  For purposes of this definition, "control"
     includes the right to grant patent sublicenses in a manner
     consistent with the requirements of this License.

     Each contributor grants you a non-exclusive, worldwide,
     royalty-free patent license under the contributor's essential
     patent claims, to make, use, sell, offer for sale, import and
     otherwise run, modify and propagate the contents of its contributor
     version.

     In the following three paragraphs, a "patent license" is any
     express agreement or commitment, however denominated, not to
     enforce a patent (such as an express permission to practice a
     patent or covenant not to sue for patent infringement).  To "grant"
     such a patent license to a party means to make such an agreement or
     commitment not to enforce a patent against the party.

     If you convey a covered work, knowingly relying on a patent
     license, and the Corresponding Source of the work is not available
     for anyone to copy, free of charge and under the terms of this
     License, through a publicly available network server or other
     readily accessible means, then you must either (1) cause the
     Corresponding Source to be so available, or (2) arrange to deprive
     yourself of the benefit of the patent license for this particular
     work, or (3) arrange, in a manner consistent with the requirements
     of this License, to extend the patent license to downstream
     recipients.  "Knowingly relying" means you have actual knowledge
     that, but for the patent license, your conveying the covered work
     in a country, or your recipient's use of the covered work in a
     country, would infringe one or more identifiable patents in that
     country that you have reason to believe are valid.

     If, pursuant to or in connection with a single transaction or
     arrangement, you convey, or propagate by procuring conveyance of, a
     covered work, and grant a patent license to some of the parties
     receiving the covered work authorizing them to use, propagate,
     modify or convey a specific copy of the covered work, then the
     patent license you grant is automatically extended to all
     recipients of the covered work and works based on it.

     A patent license is "discriminatory" if it does not include within
     the scope of its coverage, prohibits the exercise of, or is
     conditioned on the non-exercise of one or more of the rights that
     are specifically granted under this License.  You may not convey a
     covered work if you are a party to an arrangement with a third
     party that is in the business of distributing software, under which
     you make payment to the third party based on the extent of your
     activity of conveying the work, and under which the third party
     grants, to any of the parties who would receive the covered work
     from you, a discriminatory patent license (a) in connection with
     copies of the covered work conveyed by you (or copies made from
     those copies), or (b) primarily for and in connection with specific
     products or compilations that contain the covered work, unless you
     entered into that arrangement, or that patent license was granted,
     prior to 28 March 2007.

     Nothing in this License shall be construed as excluding or limiting
     any implied license or other defenses to infringement that may
     otherwise be available to you under applicable patent law.

  12. No Surrender of Others' Freedom.

     If conditions are imposed on you (whether by court order, agreement
     or otherwise) that contradict the conditions of this License, they
     do not excuse you from the conditions of this License.  If you
     cannot convey a covered work so as to satisfy simultaneously your
     obligations under this License and any other pertinent obligations,
     then as a consequence you may not convey it at all.  For example,
     if you agree to terms that obligate you to collect a royalty for
     further conveying from those to whom you convey the Program, the
     only way you could satisfy both those terms and this License would
     be to refrain entirely from conveying the Program.

  13. Use with the GNU Affero General Public License.

     Notwithstanding any other provision of this License, you have
     permission to link or combine any covered work with a work licensed
     under version 3 of the GNU Affero General Public License into a
     single combined work, and to convey the resulting work.  The terms
     of this License will continue to apply to the part which is the
     covered work, but the special requirements of the GNU Affero
     General Public License, section 13, concerning interaction through
     a network will apply to the combination as such.

  14. Revised Versions of this License.

     The Free Software Foundation may publish revised and/or new
     versions of the GNU General Public License from time to time.  Such
     new versions will be similar in spirit to the present version, but
     may differ in detail to address new problems or concerns.

     Each version is given a distinguishing version number.  If the
     Program specifies that a certain numbered version of the GNU
     General Public License "or any later version" applies to it, you
     have the option of following the terms and conditions either of
     that numbered version or of any later version published by the Free
     Software Foundation.  If the Program does not specify a version
     number of the GNU General Public License, you may choose any
     version ever published by the Free Software Foundation.

     If the Program specifies that a proxy can decide which future
     versions of the GNU General Public License can be used, that
     proxy's public statement of acceptance of a version permanently
     authorizes you to choose that version for the Program.

     Later license versions may give you additional or different
     permissions.  However, no additional obligations are imposed on any
     author or copyright holder as a result of your choosing to follow a
     later version.

  15. Disclaimer of Warranty.

     THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
     APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE
     COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS"
     WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED,
     INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
     MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE
     RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU.
     SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL
     NECESSARY SERVICING, REPAIR OR CORRECTION.

  16. Limitation of Liability.

     IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN
     WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES
     AND/OR CONVEYS THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR
     DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR
     CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE
     THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA
     BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
     PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER
     PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF
     THE POSSIBILITY OF SUCH DAMAGES.

  17. Interpretation of Sections 15 and 16.

     If the disclaimer of warranty and limitation of liability provided
     above cannot be given local legal effect according to their terms,
     reviewing courts shall apply local law that most closely
     approximates an absolute waiver of all civil liability in
     connection with the Program, unless a warranty or assumption of
     liability accompanies a copy of the Program in return for a fee.

END OF TERMS AND CONDITIONS
===========================

How to Apply These Terms to Your New Programs
=============================================

If you develop a new program, and you want it to be of the greatest
possible use to the public, the best way to achieve this is to make it
free software which everyone can redistribute and change under these
terms.

   To do so, attach the following notices to the program.  It is safest
to attach them to the start of each source file to most effectively
state the exclusion of warranty; and each file should have at least the
"copyright" line and a pointer to where the full notice is found.

     ONE LINE TO GIVE THE PROGRAM'S NAME AND A BRIEF IDEA OF WHAT IT DOES.
     Copyright (C) YEAR NAME OF AUTHOR

     This program is free software: you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
     the Free Software Foundation, either version 3 of the License, or (at
     your option) any later version.

     This program is distributed in the hope that it will be useful, but
     WITHOUT ANY WARRANTY; without even the implied warranty of
     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
     General Public License for more details.

     You should have received a copy of the GNU General Public License
     along with this program.  If not, see <https://www.gnu.org/licenses/>.

   Also add information on how to contact you by electronic and paper
mail.

   If the program does terminal interaction, make it output a short
notice like this when it starts in an interactive mode:

     PROGRAM Copyright (C) YEAR NAME OF AUTHOR
     This program comes with ABSOLUTELY NO WARRANTY; for details type 'show w'.
     This is free software, and you are welcome to redistribute it
     under certain conditions; type 'show c' for details.

   The hypothetical commands 'show w' and 'show c' should show the
appropriate parts of the General Public License.  Of course, your
program's commands might be different; for a GUI interface, you would
use an "about box".

   You should also get your employer (if you work as a programmer) or
school, if any, to sign a "copyright disclaimer" for the program, if
necessary.  For more information on this, and how to apply and follow
the GNU GPL, see <https://www.gnu.org/licenses/>.

   The GNU General Public License does not permit incorporating your
program into proprietary programs.  If your program is a subroutine
library, you may consider it more useful to permit linking proprietary
applications with the library.  If this is what you want to do, use the
GNU Lesser General Public License instead of this License.  But first,
please read <https://www.gnu.org/licenses/why-not-lgpl.html>.


File: libgomp.info,  Node: GNU Free Documentation License,  Next: Funding,  Prev: Copying,  Up: Top

GNU Free Documentation License
******************************

                     Version 1.3, 3 November 2008

     Copyright (C) 2000, 2001, 2002, 2007, 2008 Free Software Foundation, Inc.
     <https://www.fsf.org>

     Everyone is permitted to copy and distribute verbatim copies
     of this license document, but changing it is not allowed.

  0. PREAMBLE

     The purpose of this License is to make a manual, textbook, or other
     functional and useful document "free" in the sense of freedom: to
     assure everyone the effective freedom to copy and redistribute it,
     with or without modifying it, either commercially or
     noncommercially.  Secondarily, this License preserves for the
     author and publisher a way to get credit for their work, while not
     being considered responsible for modifications made by others.

     This License is a kind of "copyleft", which means that derivative
     works of the document must themselves be free in the same sense.
     It complements the GNU General Public License, which is a copyleft
     license designed for free software.

     We have designed this License in order to use it for manuals for
     free software, because free software needs free documentation: a
     free program should come with manuals providing the same freedoms
     that the software does.  But this License is not limited to
     software manuals; it can be used for any textual work, regardless
     of subject matter or whether it is published as a printed book.  We
     recommend this License principally for works whose purpose is
     instruction or reference.

  1. APPLICABILITY AND DEFINITIONS

     This License applies to any manual or other work, in any medium,
     that contains a notice placed by the copyright holder saying it can
     be distributed under the terms of this License.  Such a notice
     grants a world-wide, royalty-free license, unlimited in duration,
     to use that work under the conditions stated herein.  The
     "Document", below, refers to any such manual or work.  Any member
     of the public is a licensee, and is addressed as "you".  You accept
     the license if you copy, modify or distribute the work in a way
     requiring permission under copyright law.

     A "Modified Version" of the Document means any work containing the
     Document or a portion of it, either copied verbatim, or with
     modifications and/or translated into another language.

     A "Secondary Section" is a named appendix or a front-matter section
     of the Document that deals exclusively with the relationship of the
     publishers or authors of the Document to the Document's overall
     subject (or to related matters) and contains nothing that could
     fall directly within that overall subject.  (Thus, if the Document
     is in part a textbook of mathematics, a Secondary Section may not
     explain any mathematics.)  The relationship could be a matter of
     historical connection with the subject or with related matters, or
     of legal, commercial, philosophical, ethical or political position
     regarding them.

     The "Invariant Sections" are certain Secondary Sections whose
     titles are designated, as being those of Invariant Sections, in the
     notice that says that the Document is released under this License.
     If a section does not fit the above definition of Secondary then it
     is not allowed to be designated as Invariant.  The Document may
     contain zero Invariant Sections.  If the Document does not identify
     any Invariant Sections then there are none.

     The "Cover Texts" are certain short passages of text that are
     listed, as Front-Cover Texts or Back-Cover Texts, in the notice
     that says that the Document is released under this License.  A
     Front-Cover Text may be at most 5 words, and a Back-Cover Text may
     be at most 25 words.

     A "Transparent" copy of the Document means a machine-readable copy,
     represented in a format whose specification is available to the
     general public, that is suitable for revising the document
     straightforwardly with generic text editors or (for images composed
     of pixels) generic paint programs or (for drawings) some widely
     available drawing editor, and that is suitable for input to text
     formatters or for automatic translation to a variety of formats
     suitable for input to text formatters.  A copy made in an otherwise
     Transparent file format whose markup, or absence of markup, has
     been arranged to thwart or discourage subsequent modification by
     readers is not Transparent.  An image format is not Transparent if
     used for any substantial amount of text.  A copy that is not
     "Transparent" is called "Opaque".

     Examples of suitable formats for Transparent copies include plain
     ASCII without markup, Texinfo input format, LaTeX input format,
     SGML or XML using a publicly available DTD, and standard-conforming
     simple HTML, PostScript or PDF designed for human modification.
     Examples of transparent image formats include PNG, XCF and JPG.
     Opaque formats include proprietary formats that can be read and
     edited only by proprietary word processors, SGML or XML for which
     the DTD and/or processing tools are not generally available, and
     the machine-generated HTML, PostScript or PDF produced by some word
     processors for output purposes only.

     The "Title Page" means, for a printed book, the title page itself,
     plus such following pages as are needed to hold, legibly, the
     material this License requires to appear in the title page.  For
     works in formats which do not have any title page as such, "Title
     Page" means the text near the most prominent appearance of the
     work's title, preceding the beginning of the body of the text.

     The "publisher" means any person or entity that distributes copies
     of the Document to the public.

     A section "Entitled XYZ" means a named subunit of the Document
     whose title either is precisely XYZ or contains XYZ in parentheses
     following text that translates XYZ in another language.  (Here XYZ
     stands for a specific section name mentioned below, such as
     "Acknowledgements", "Dedications", "Endorsements", or "History".)
     To "Preserve the Title" of such a section when you modify the
     Document means that it remains a section "Entitled XYZ" according
     to this definition.

     The Document may include Warranty Disclaimers next to the notice
     which states that this License applies to the Document.  These
     Warranty Disclaimers are considered to be included by reference in
     this License, but only as regards disclaiming warranties: any other
     implication that these Warranty Disclaimers may have is void and
     has no effect on the meaning of this License.

  2. VERBATIM COPYING

     You may copy and distribute the Document in any medium, either
     commercially or noncommercially, provided that this License, the
     copyright notices, and the license notice saying this License
     applies to the Document are reproduced in all copies, and that you
     add no other conditions whatsoever to those of this License.  You
     may not use technical measures to obstruct or control the reading
     or further copying of the copies you make or distribute.  However,
     you may accept compensation in exchange for copies.  If you
     distribute a large enough number of copies you must also follow the
     conditions in section 3.

     You may also lend copies, under the same conditions stated above,
     and you may publicly display copies.

  3. COPYING IN QUANTITY

     If you publish printed copies (or copies in media that commonly
     have printed covers) of the Document, numbering more than 100, and
     the Document's license notice requires Cover Texts, you must
     enclose the copies in covers that carry, clearly and legibly, all
     these Cover Texts: Front-Cover Texts on the front cover, and
     Back-Cover Texts on the back cover.  Both covers must also clearly
     and legibly identify you as the publisher of these copies.  The
     front cover must present the full title with all words of the title
     equally prominent and visible.  You may add other material on the
     covers in addition.  Copying with changes limited to the covers, as
     long as they preserve the title of the Document and satisfy these
     conditions, can be treated as verbatim copying in other respects.

     If the required texts for either cover are too voluminous to fit
     legibly, you should put the first ones listed (as many as fit
     reasonably) on the actual cover, and continue the rest onto
     adjacent pages.

     If you publish or distribute Opaque copies of the Document
     numbering more than 100, you must either include a machine-readable
     Transparent copy along with each Opaque copy, or state in or with
     each Opaque copy a computer-network location from which the general
     network-using public has access to download using public-standard
     network protocols a complete Transparent copy of the Document, free
     of added material.  If you use the latter option, you must take
     reasonably prudent steps, when you begin distribution of Opaque
     copies in quantity, to ensure that this Transparent copy will
     remain thus accessible at the stated location until at least one
     year after the last time you distribute an Opaque copy (directly or
     through your agents or retailers) of that edition to the public.

     It is requested, but not required, that you contact the authors of
     the Document well before redistributing any large number of copies,
     to give them a chance to provide you with an updated version of the
     Document.

  4. MODIFICATIONS

     You may copy and distribute a Modified Version of the Document
     under the conditions of sections 2 and 3 above, provided that you
     release the Modified Version under precisely this License, with the
     Modified Version filling the role of the Document, thus licensing
     distribution and modification of the Modified Version to whoever
     possesses a copy of it.  In addition, you must do these things in
     the Modified Version:

       A. Use in the Title Page (and on the covers, if any) a title
          distinct from that of the Document, and from those of previous
          versions (which should, if there were any, be listed in the
          History section of the Document).  You may use the same title
          as a previous version if the original publisher of that
          version gives permission.

       B. List on the Title Page, as authors, one or more persons or
          entities responsible for authorship of the modifications in
          the Modified Version, together with at least five of the
          principal authors of the Document (all of its principal
          authors, if it has fewer than five), unless they release you
          from this requirement.

       C. State on the Title page the name of the publisher of the
          Modified Version, as the publisher.

       D. Preserve all the copyright notices of the Document.

       E. Add an appropriate copyright notice for your modifications
          adjacent to the other copyright notices.

       F. Include, immediately after the copyright notices, a license
          notice giving the public permission to use the Modified
          Version under the terms of this License, in the form shown in
          the Addendum below.

       G. Preserve in that license notice the full lists of Invariant
          Sections and required Cover Texts given in the Document's
          license notice.

       H. Include an unaltered copy of this License.

       I. Preserve the section Entitled "History", Preserve its Title,
          and add to it an item stating at least the title, year, new
          authors, and publisher of the Modified Version as given on the
          Title Page.  If there is no section Entitled "History" in the
          Document, create one stating the title, year, authors, and
          publisher of the Document as given on its Title Page, then add
          an item describing the Modified Version as stated in the
          previous sentence.

       J. Preserve the network location, if any, given in the Document
          for public access to a Transparent copy of the Document, and
          likewise the network locations given in the Document for
          previous versions it was based on.  These may be placed in the
          "History" section.  You may omit a network location for a work
          that was published at least four years before the Document
          itself, or if the original publisher of the version it refers
          to gives permission.

       K. For any section Entitled "Acknowledgements" or "Dedications",
          Preserve the Title of the section, and preserve in the section
          all the substance and tone of each of the contributor
          acknowledgements and/or dedications given therein.

       L. Preserve all the Invariant Sections of the Document, unaltered
          in their text and in their titles.  Section numbers or the
          equivalent are not considered part of the section titles.

       M. Delete any section Entitled "Endorsements".  Such a section
          may not be included in the Modified Version.

       N. Do not retitle any existing section to be Entitled
          "Endorsements" or to conflict in title with any Invariant
          Section.

       O. Preserve any Warranty Disclaimers.

     If the Modified Version includes new front-matter sections or
     appendices that qualify as Secondary Sections and contain no
     material copied from the Document, you may at your option designate
     some or all of these sections as invariant.  To do this, add their
     titles to the list of Invariant Sections in the Modified Version's
     license notice.  These titles must be distinct from any other
     section titles.

     You may add a section Entitled "Endorsements", provided it contains
     nothing but endorsements of your Modified Version by various
     parties--for example, statements of peer review or that the text
     has been approved by an organization as the authoritative
     definition of a standard.

     You may add a passage of up to five words as a Front-Cover Text,
     and a passage of up to 25 words as a Back-Cover Text, to the end of
     the list of Cover Texts in the Modified Version.  Only one passage
     of Front-Cover Text and one of Back-Cover Text may be added by (or
     through arrangements made by) any one entity.  If the Document
     already includes a cover text for the same cover, previously added
     by you or by arrangement made by the same entity you are acting on
     behalf of, you may not add another; but you may replace the old
     one, on explicit permission from the previous publisher that added
     the old one.

     The author(s) and publisher(s) of the Document do not by this
     License give permission to use their names for publicity for or to
     assert or imply endorsement of any Modified Version.

  5. COMBINING DOCUMENTS

     You may combine the Document with other documents released under
     this License, under the terms defined in section 4 above for
     modified versions, provided that you include in the combination all
     of the Invariant Sections of all of the original documents,
     unmodified, and list them all as Invariant Sections of your
     combined work in its license notice, and that you preserve all
     their Warranty Disclaimers.

     The combined work need only contain one copy of this License, and
     multiple identical Invariant Sections may be replaced with a single
     copy.  If there are multiple Invariant Sections with the same name
     but different contents, make the title of each such section unique
     by adding at the end of it, in parentheses, the name of the
     original author or publisher of that section if known, or else a
     unique number.  Make the same adjustment to the section titles in
     the list of Invariant Sections in the license notice of the
     combined work.

     In the combination, you must combine any sections Entitled
     "History" in the various original documents, forming one section
     Entitled "History"; likewise combine any sections Entitled
     "Acknowledgements", and any sections Entitled "Dedications".  You
     must delete all sections Entitled "Endorsements."

  6. COLLECTIONS OF DOCUMENTS

     You may make a collection consisting of the Document and other
     documents released under this License, and replace the individual
     copies of this License in the various documents with a single copy
     that is included in the collection, provided that you follow the
     rules of this License for verbatim copying of each of the documents
     in all other respects.

     You may extract a single document from such a collection, and
     distribute it individually under this License, provided you insert
     a copy of this License into the extracted document, and follow this
     License in all other respects regarding verbatim copying of that
     document.

  7. AGGREGATION WITH INDEPENDENT WORKS

     A compilation of the Document or its derivatives with other
     separate and independent documents or works, in or on a volume of a
     storage or distribution medium, is called an "aggregate" if the
     copyright resulting from the compilation is not used to limit the
     legal rights of the compilation's users beyond what the individual
     works permit.  When the Document is included in an aggregate, this
     License does not apply to the other works in the aggregate which
     are not themselves derivative works of the Document.

     If the Cover Text requirement of section 3 is applicable to these
     copies of the Document, then if the Document is less than one half
     of the entire aggregate, the Document's Cover Texts may be placed
     on covers that bracket the Document within the aggregate, or the
     electronic equivalent of covers if the Document is in electronic
     form.  Otherwise they must appear on printed covers that bracket
     the whole aggregate.

  8. TRANSLATION

     Translation is considered a kind of modification, so you may
     distribute translations of the Document under the terms of section
     4.  Replacing Invariant Sections with translations requires special
     permission from their copyright holders, but you may include
     translations of some or all Invariant Sections in addition to the
     original versions of these Invariant Sections.  You may include a
     translation of this License, and all the license notices in the
     Document, and any Warranty Disclaimers, provided that you also
     include the original English version of this License and the
     original versions of those notices and disclaimers.  In case of a
     disagreement between the translation and the original version of
     this License or a notice or disclaimer, the original version will
     prevail.

     If a section in the Document is Entitled "Acknowledgements",
     "Dedications", or "History", the requirement (section 4) to
     Preserve its Title (section 1) will typically require changing the
     actual title.

  9. TERMINATION

     You may not copy, modify, sublicense, or distribute the Document
     except as expressly provided under this License.  Any attempt
     otherwise to copy, modify, sublicense, or distribute it is void,
     and will automatically terminate your rights under this License.

     However, if you cease all violation of this License, then your
     license from a particular copyright holder is reinstated (a)
     provisionally, unless and until the copyright holder explicitly and
     finally terminates your license, and (b) permanently, if the
     copyright holder fails to notify you of the violation by some
     reasonable means prior to 60 days after the cessation.

     Moreover, your license from a particular copyright holder is
     reinstated permanently if the copyright holder notifies you of the
     violation by some reasonable means, this is the first time you have
     received notice of violation of this License (for any work) from
     that copyright holder, and you cure the violation prior to 30 days
     after your receipt of the notice.

     Termination of your rights under this section does not terminate
     the licenses of parties who have received copies or rights from you
     under this License.  If your rights have been terminated and not
     permanently reinstated, receipt of a copy of some or all of the
     same material does not give you any rights to use it.

  10. FUTURE REVISIONS OF THIS LICENSE

     The Free Software Foundation may publish new, revised versions of
     the GNU Free Documentation License from time to time.  Such new
     versions will be similar in spirit to the present version, but may
     differ in detail to address new problems or concerns.  See
     <https://www.gnu.org/copyleft/>.

     Each version of the License is given a distinguishing version
     number.  If the Document specifies that a particular numbered
     version of this License "or any later version" applies to it, you
     have the option of following the terms and conditions either of
     that specified version or of any later version that has been
     published (not as a draft) by the Free Software Foundation.  If the
     Document does not specify a version number of this License, you may
     choose any version ever published (not as a draft) by the Free
     Software Foundation.  If the Document specifies that a proxy can
     decide which future versions of this License can be used, that
     proxy's public statement of acceptance of a version permanently
     authorizes you to choose that version for the Document.

  11. RELICENSING

     "Massive Multiauthor Collaboration Site" (or "MMC Site") means any
     World Wide Web server that publishes copyrightable works and also
     provides prominent facilities for anybody to edit those works.  A
     public wiki that anybody can edit is an example of such a server.
     A "Massive Multiauthor Collaboration" (or "MMC") contained in the
     site means any set of copyrightable works thus published on the MMC
     site.

     "CC-BY-SA" means the Creative Commons Attribution-Share Alike 3.0
     license published by Creative Commons Corporation, a not-for-profit
     corporation with a principal place of business in San Francisco,
     California, as well as future copyleft versions of that license
     published by that same organization.

     "Incorporate" means to publish or republish a Document, in whole or
     in part, as part of another Document.

     An MMC is "eligible for relicensing" if it is licensed under this
     License, and if all works that were first published under this
     License somewhere other than this MMC, and subsequently
     incorporated in whole or in part into the MMC, (1) had no cover
     texts or invariant sections, and (2) were thus incorporated prior
     to November 1, 2008.

     The operator of an MMC Site may republish an MMC contained in the
     site under CC-BY-SA on the same site at any time before August 1,
     2009, provided the MMC is eligible for relicensing.

ADDENDUM: How to use this License for your documents
====================================================

To use this License in a document you have written, include a copy of
the License in the document and put the following copyright and license
notices just after the title page:

       Copyright (C)  YEAR  YOUR NAME.
       Permission is granted to copy, distribute and/or modify this document
       under the terms of the GNU Free Documentation License, Version 1.3
       or any later version published by the Free Software Foundation;
       with no Invariant Sections, no Front-Cover Texts, and no Back-Cover
       Texts.  A copy of the license is included in the section entitled ``GNU
       Free Documentation License''.

   If you have Invariant Sections, Front-Cover Texts and Back-Cover
Texts, replace the "with...Texts."  line with this:

         with the Invariant Sections being LIST THEIR TITLES, with
         the Front-Cover Texts being LIST, and with the Back-Cover Texts
         being LIST.

   If you have Invariant Sections without Cover Texts, or some other
combination of the three, merge those two alternatives to suit the
situation.

   If your document contains nontrivial examples of program code, we
recommend releasing these examples in parallel under your choice of free
software license, such as the GNU General Public License, to permit
their use in free software.


File: libgomp.info,  Node: Funding,  Next: Library Index,  Prev: GNU Free Documentation License,  Up: Top

Funding Free Software
*********************

If you want to have more free software a few years from now, it makes
sense for you to help encourage people to contribute funds for its
development.  The most effective approach known is to encourage
commercial redistributors to donate.

   Users of free software systems can boost the pace of development by
encouraging for-a-fee distributors to donate part of their selling price
to free software developers--the Free Software Foundation, and others.

   The way to convince distributors to do this is to demand it and
expect it from them.  So when you compare distributors, judge them
partly by how much they give to free software development.  Show
distributors they must compete to be the one who gives the most.

   To make this approach work, you must insist on numbers that you can
compare, such as, "We will donate ten dollars to the Frobnitz project
for each disk sold."  Don't be satisfied with a vague promise, such as
"A portion of the profits are donated," since it doesn't give a basis
for comparison.

   Even a precise fraction "of the profits from this disk" is not very
meaningful, since creative accounting and unrelated business decisions
can greatly alter what fraction of the sales price counts as profit.  If
the price you pay is $50, ten percent of the profit is probably less
than a dollar; it might be a few cents, or nothing at all.

   Some redistributors do development work themselves.  This is useful
too; but to keep everyone honest, you need to inquire how much they do,
and what kind.  Some kinds of development make much more long-term
difference than others.  For example, maintaining a separate version of
a program contributes very little; maintaining the standard version of a
program for the whole community contributes much.  Easy new ports
contribute little, since someone else would surely do them; difficult
ports such as adding a new CPU to the GNU Compiler Collection contribute
more; major new features or packages contribute the most.

   By establishing the idea that supporting further development is "the
proper thing to do" when distributing free software for a fee, we can
assure a steady flow of resources into making more free software.

     Copyright (C) 1994 Free Software Foundation, Inc.
     Verbatim copying and redistribution of this section is permitted
     without royalty; alteration is not permitted.


File: libgomp.info,  Node: Library Index,  Prev: Funding,  Up: Top

Library Index
*************

�[index�]
* Menu:

* acc_get_property:                      acc_get_property.      (line 6)
* acc_get_property_string:               acc_get_property.      (line 6)
* Environment Variable:                  OMP_ALLOCATOR.         (line 6)
* Environment Variable <1>:              OMP_AFFINITY_FORMAT.   (line 6)
* Environment Variable <2>:              OMP_CANCELLATION.      (line 6)
* Environment Variable <3>:              OMP_DISPLAY_AFFINITY.  (line 6)
* Environment Variable <4>:              OMP_DISPLAY_ENV.       (line 6)
* Environment Variable <5>:              OMP_DEFAULT_DEVICE.    (line 6)
* Environment Variable <6>:              OMP_DYNAMIC.           (line 6)
* Environment Variable <7>:              OMP_MAX_ACTIVE_LEVELS. (line 6)
* Environment Variable <8>:              OMP_MAX_TASK_PRIORITY. (line 6)
* Environment Variable <9>:              OMP_NESTED.            (line 6)
* Environment Variable <10>:             OMP_NUM_TEAMS.         (line 6)
* Environment Variable <11>:             OMP_NUM_THREADS.       (line 6)
* Environment Variable <12>:             OMP_PROC_BIND.         (line 6)
* Environment Variable <13>:             OMP_PLACES.            (line 6)
* Environment Variable <14>:             OMP_STACKSIZE.         (line 6)
* Environment Variable <15>:             OMP_SCHEDULE.          (line 6)
* Environment Variable <16>:             OMP_TARGET_OFFLOAD.    (line 6)
* Environment Variable <17>:             OMP_TEAMS_THREAD_LIMIT.
                                                                (line 6)
* Environment Variable <18>:             OMP_THREAD_LIMIT.      (line 6)
* Environment Variable <19>:             OMP_WAIT_POLICY.       (line 6)
* Environment Variable <20>:             GOMP_CPU_AFFINITY.     (line 6)
* Environment Variable <21>:             GOMP_DEBUG.            (line 6)
* Environment Variable <22>:             GOMP_STACKSIZE.        (line 6)
* Environment Variable <23>:             GOMP_SPINCOUNT.        (line 6)
* Environment Variable <24>:             GOMP_RTEMS_THREAD_POOLS.
                                                                (line 6)
* FDL, GNU Free Documentation License:   GNU Free Documentation License.
                                                                (line 6)
* Implementation specific setting:       OMP_NESTED.            (line 6)
* Implementation specific setting <1>:   OMP_NUM_THREADS.       (line 6)
* Implementation specific setting <2>:   OMP_SCHEDULE.          (line 6)
* Implementation specific setting <3>:   OMP_TARGET_OFFLOAD.    (line 6)
* Implementation specific setting <4>:   GOMP_STACKSIZE.        (line 6)
* Implementation specific setting <5>:   GOMP_SPINCOUNT.        (line 6)
* Implementation specific setting <6>:   GOMP_RTEMS_THREAD_POOLS.
                                                                (line 6)
* Implementation specific setting <7>:   Implementation-defined ICV Initialization.
                                                                (line 6)
* Introduction:                          Top.                   (line 6)



Tag Table:
Node: Top2083
Node: Enabling OpenMP4912
Node: OpenMP Implementation Status6093
Node: OpenMP 4.56789
Node: OpenMP 5.06965
Node: OpenMP 5.112386
Node: OpenMP 5.216698
Ref: OpenMP 5.2-Footnote-119843
Node: OpenMP Technical Report 1220244
Node: Runtime Library Routines24771
Node: Thread Team Routines25404
Node: omp_set_num_threads26991
Node: omp_get_num_threads27837
Node: omp_get_max_threads28928
Node: omp_get_thread_num29683
Node: omp_in_parallel30552
Node: omp_set_dynamic31202
Node: omp_get_dynamic32069
Node: omp_get_cancellation32944
Node: omp_set_nested33738
Node: omp_get_nested35057
Node: omp_set_schedule36799
Node: omp_get_schedule37878
Node: omp_get_teams_thread_limit38826
Node: omp_get_supported_active_levels39594
Node: omp_set_max_active_levels40399
Node: omp_get_max_active_levels41348
Node: omp_get_level42071
Node: omp_get_ancestor_thread_num42705
Node: omp_get_team_size43627
Node: omp_get_active_level44598
Node: Thread Affinity Routines45290
Node: omp_get_proc_bind45690
Node: Teams Region Routines46662
Node: omp_get_num_teams47370
Node: omp_get_team_num47856
Node: omp_set_num_teams48355
Node: omp_get_max_teams49227
Node: omp_set_teams_thread_limit49919
Node: omp_get_thread_limit50924
Node: Tasking Routines51524
Node: omp_get_max_task_priority52035
Node: omp_in_explicit_task52618
Node: omp_in_final53508
Node: Resource Relinquishing Routines54151
Node: omp_pause_resource54623
Node: omp_pause_resource_all55716
Node: Device Information Routines56703
Node: omp_get_num_procs57521
Node: omp_set_default_device58042
Node: omp_get_default_device58841
Node: omp_get_num_devices59530
Node: omp_get_device_num60065
Node: omp_is_initial_device60880
Node: omp_get_initial_device61584
Node: Device Memory Routines62319
Node: omp_target_alloc63359
Node: omp_target_free64934
Node: omp_target_is_present66122
Node: omp_target_is_accessible67794
Node: omp_target_memcpy69608
Node: omp_target_memcpy_async71422
Node: omp_target_memcpy_rect73801
Node: omp_target_memcpy_rect_async77005
Node: omp_target_associate_ptr80777
Node: omp_target_disassociate_ptr83742
Node: omp_get_mapped_ptr85495
Node: Lock Routines87038
Node: omp_init_lock87886
Node: omp_init_nest_lock88511
Node: omp_destroy_lock89233
Node: omp_destroy_nest_lock89908
Node: omp_set_lock90646
Node: omp_set_nest_lock91504
Node: omp_unset_lock92400
Node: omp_unset_nest_lock93329
Node: omp_test_lock94323
Node: omp_test_nest_lock95301
Node: Timing Routines96289
Node: omp_get_wtick96666
Node: omp_get_wtime97221
Node: Event Routine97990
Node: omp_fulfill_event98323
Node: Memory Management Routines99315
Node: omp_init_allocator100231
Node: omp_destroy_allocator101860
Node: omp_set_default_allocator102844
Node: omp_get_default_allocator103878
Node: omp_alloc104819
Node: omp_aligned_alloc106922
Node: omp_free109452
Node: omp_calloc111107
Node: omp_aligned_calloc113326
Node: omp_realloc115878
Node: Environment Display Routine119264
Node: omp_display_env119640
Node: Environment Variables122474
Node: OMP_ALLOCATOR125255
Node: OMP_AFFINITY_FORMAT129065
Node: OMP_CANCELLATION131673
Node: OMP_DISPLAY_AFFINITY132273
Node: OMP_DISPLAY_ENV132968
Node: OMP_DEFAULT_DEVICE133818
Node: OMP_DYNAMIC134826
Node: OMP_MAX_ACTIVE_LEVELS135453
Node: OMP_MAX_TASK_PRIORITY136478
Node: OMP_NESTED137181
Node: OMP_NUM_TEAMS138430
Node: OMP_NUM_THREADS139139
Node: OMP_PROC_BIND140155
Node: OMP_PLACES141676
Node: OMP_STACKSIZE144412
Node: OMP_SCHEDULE145314
Node: OMP_TARGET_OFFLOAD146061
Node: OMP_TEAMS_THREAD_LIMIT147548
Node: OMP_THREAD_LIMIT148401
Node: OMP_WAIT_POLICY149061
Node: GOMP_CPU_AFFINITY149753
Node: GOMP_DEBUG151471
Node: GOMP_STACKSIZE151973
Node: GOMP_SPINCOUNT152804
Node: GOMP_RTEMS_THREAD_POOLS154008
Node: Enabling OpenACC156183
Node: OpenACC Runtime Library Routines157005
Node: acc_get_num_devices161286
Node: acc_set_device_type162012
Node: acc_get_device_type162776
Node: acc_set_device_num163789
Node: acc_get_device_num164606
Node: acc_get_property165405
Node: acc_async_test167617
Node: acc_async_test_all168588
Node: acc_wait169471
Node: acc_wait_all170334
Node: acc_wait_all_async171095
Node: acc_wait_async171847
Node: acc_init172555
Node: acc_shutdown173200
Node: acc_on_device173867
Node: acc_malloc174856
Node: acc_free175564
Node: acc_copyin176208
Node: acc_present_or_copyin177795
Node: acc_create179563
Node: acc_present_or_create181195
Node: acc_copyout182971
Node: acc_delete185275
Node: acc_update_device187522
Node: acc_update_self189096
Node: acc_map_data190686
Node: acc_unmap_data191758
Node: acc_deviceptr192500
Node: acc_hostptr193295
Node: acc_is_present194079
Node: acc_memcpy_to_device195606
Node: acc_memcpy_from_device197041
Node: acc_attach198526
Node: acc_detach199193
Node: acc_get_current_cuda_device200032
Node: acc_get_current_cuda_context200617
Node: acc_get_cuda_stream201217
Node: acc_set_cuda_stream201808
Node: acc_prof_register202479
Node: acc_prof_unregister203038
Node: acc_prof_lookup203605
Node: acc_register_library204126
Node: OpenACC Environment Variables204692
Node: ACC_DEVICE_TYPE205181
Node: ACC_DEVICE_NUM205708
Node: ACC_PROFLIB206179
Node: CUDA Streams Usage206785
Ref: CUDA Streams Usage-Footnote-1208677
Node: OpenACC Library Interoperability208786
Ref: OpenACC Library Interoperability-Footnote-1215135
Ref: OpenACC Library Interoperability-Footnote-2215387
Node: OpenACC Profiling Interface215595
Node: OpenMP-Implementation Specifics225593
Node: Implementation-defined ICV Initialization225912
Node: OpenMP Context Selectors227083
Ref: OpenMP Context Selectors-Footnote-1228364
Node: Memory allocation228432
Node: Offload-Target Specifics233077
Node: AMD Radeon233368
Node: nvptx236306
Node: The libgomp ABI239933
Node: Implementing MASTER construct240783
Node: Implementing CRITICAL construct241200
Node: Implementing ATOMIC construct241941
Node: Implementing FLUSH construct242424
Node: Implementing BARRIER construct242697
Node: Implementing THREADPRIVATE construct242968
Node: Implementing PRIVATE clause243623
Node: Implementing FIRSTPRIVATE LASTPRIVATE COPYIN and COPYPRIVATE clauses244206
Node: Implementing REDUCTION clause245532
Node: Implementing PARALLEL construct246092
Node: Implementing FOR construct247351
Node: Implementing ORDERED construct249351
Node: Implementing SECTIONS construct249659
Node: Implementing SINGLE construct250427
Node: Implementing OpenACC's PARALLEL construct251141
Node: Reporting Bugs251401
Node: Copying251764
Node: GNU Free Documentation License289316
Node: Funding314444
Node: Library Index316970

End Tag Table


Local Variables:
coding: utf-8
End: