Move combine over to statistics_counter_event.

[thirdparty/gcc.git] / libgomp / libgomp.texi

\input texinfo @c -*-texinfo-*-

@c %**start of header
@setfilename libgomp.info
@settitle GNU libgomp
@c %**end of header


@copying
Copyright @copyright{} 2006-2023 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.
@end copying

@ifinfo
@dircategory GNU Libraries
@direntry
* libgomp: (libgomp).          GNU Offloading and Multi Processing Runtime Library.
@end direntry

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

@insertcopying
@end ifinfo


@setchapternewpage odd

@titlepage
@title GNU Offloading and Multi Processing Runtime Library
@subtitle The GNU OpenMP and OpenACC Implementation
@page
@vskip 0pt plus 1filll
@comment For the @value{version-GCC} Version*
@sp 1
Published by the Free Software Foundation @*
51 Franklin Street, Fifth Floor@*
Boston, MA 02110-1301, USA@*
@sp 1
@insertcopying
@end titlepage

@summarycontents
@contents
@page


@node Top, Enabling OpenMP
@top Introduction
@cindex Introduction

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



@comment
@comment  When you add a new menu item, please keep the right hand
@comment  aligned to the same column.  Do not use tabs.  This provides
@comment  better formatting.
@comment
@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.
@end menu


@c ---------------------------------------------------------------------
@c Enabling OpenMP
@c ---------------------------------------------------------------------

@node Enabling OpenMP
@chapter Enabling OpenMP

To activate the OpenMP extensions for C/C++ and Fortran, the compile-time 
flag @command{-fopenmp} must be specified.  This enables the OpenMP directive
@code{#pragma omp} in C/C++ and @code{!$omp} directives in free form, 
@code{c$omp}, @code{*$omp} and @code{!$omp} directives in fixed form, 
@code{!$} conditional compilation sentinels in free form and @code{c$},
@code{*$} and @code{!$} sentinels in fixed form, for Fortran.  The flag also
arranges for automatic linking of the OpenMP runtime library 
(@ref{Runtime Library Routines}).

A complete description of all OpenMP directives may be found in the
@uref{https://www.openmp.org, OpenMP Application Program Interface} manuals.
See also @ref{OpenMP Implementation Status}.


@c ---------------------------------------------------------------------
@c OpenMP Implementation Status
@c ---------------------------------------------------------------------

@node OpenMP Implementation Status
@chapter 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 11:: Feature completion status to first 6.0 preview
@end menu

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

@node OpenMP 4.5
@section OpenMP 4.5

The OpenMP 4.5 specification is fully supported.

@node OpenMP 5.0
@section OpenMP 5.0

@unnumberedsubsec New features listed in Appendix B of the OpenMP specification
@c This list is sorted as in OpenMP 5.1's B.3 not as in OpenMP 5.0's B.2

@multitable @columnfractions .60 .10 .25
@headitem Description @tab Status @tab Comments
@item Array shaping @tab N @tab
@item Array sections with non-unit strides in C and C++ @tab N @tab
@item Iterators @tab Y @tab
@item @code{metadirective} directive @tab N @tab
@item @code{declare variant} directive
      @tab P @tab @emph{simd} traits not handled correctly
@item @var{target-offload-var} ICV and @code{OMP_TARGET_OFFLOAD}
      env variable @tab Y @tab
@item Nested-parallel changes to @var{max-active-levels-var} ICV @tab Y @tab
@item @code{requires} directive @tab P
      @tab complete but no non-host device provides @code{unified_shared_memory}
@item @code{teams} construct outside an enclosing target region @tab Y @tab
@item Non-rectangular loop nests @tab P
      @tab Full support for C/C++, partial for Fortran
           (@uref{https://gcc.gnu.org/PR110735,PR110735})
@item @code{!=} as relational-op in canonical loop form for C/C++ @tab Y @tab
@item @code{nonmonotonic} as default loop schedule modifier for worksharing-loop
      constructs @tab Y @tab
@item Collapse of associated loops that are imperfectly nested loops @tab N @tab
@item Clauses @code{if}, @code{nontemporal} and @code{order(concurrent)} in
      @code{simd} construct @tab Y @tab
@item @code{atomic} constructs in @code{simd} @tab Y @tab
@item @code{loop} construct @tab Y @tab
@item @code{order(concurrent)} clause @tab Y @tab
@item @code{scan} directive and @code{in_scan} modifier for the
      @code{reduction} clause @tab Y @tab
@item @code{in_reduction} clause on @code{task} constructs @tab Y @tab
@item @code{in_reduction} clause on @code{target} constructs @tab P
      @tab @code{nowait} only stub
@item @code{task_reduction} clause with @code{taskgroup} @tab Y @tab
@item @code{task} modifier to @code{reduction} clause @tab Y @tab
@item @code{affinity} clause to @code{task} construct @tab Y @tab Stub only
@item @code{detach} clause to @code{task} construct @tab Y @tab
@item @code{omp_fulfill_event} runtime routine @tab Y @tab
@item @code{reduction} and @code{in_reduction} clauses on @code{taskloop}
      and @code{taskloop simd} constructs @tab Y @tab
@item @code{taskloop} construct cancelable by @code{cancel} construct
      @tab Y @tab
@item @code{mutexinoutset} @emph{dependence-type} for @code{depend} clause
      @tab Y @tab
@item Predefined memory spaces, memory allocators, allocator traits
      @tab Y @tab See also @ref{Memory allocation}
@item Memory management routines @tab Y @tab
@item @code{allocate} directive @tab N @tab
@item @code{allocate} clause @tab P @tab Initial support
@item @code{use_device_addr} clause on @code{target data} @tab Y @tab
@item @code{ancestor} modifier on @code{device} clause @tab Y @tab
@item Implicit declare target directive @tab Y @tab
@item Discontiguous array section with @code{target update} construct
      @tab N @tab
@item C/C++'s lvalue expressions in @code{to}, @code{from}
      and @code{map} clauses @tab N @tab
@item C/C++'s lvalue expressions in @code{depend} clauses @tab Y @tab
@item Nested @code{declare target} directive @tab Y @tab
@item Combined @code{master} constructs @tab Y @tab
@item @code{depend} clause on @code{taskwait} @tab Y @tab
@item Weak memory ordering clauses on @code{atomic} and @code{flush} construct
      @tab Y @tab
@item @code{hint} clause on the @code{atomic} construct @tab Y @tab Stub only
@item @code{depobj} construct and depend objects  @tab Y @tab
@item Lock hints were renamed to synchronization hints @tab Y @tab
@item @code{conditional} modifier to @code{lastprivate} clause @tab Y @tab
@item Map-order clarifications @tab P @tab
@item @code{close} @emph{map-type-modifier} @tab Y @tab
@item Mapping C/C++ pointer variables and to assign the address of
      device memory mapped by an array section @tab P @tab
@item Mapping of Fortran pointer and allocatable variables, including pointer
      and allocatable components of variables
      @tab P @tab Mapping of vars with allocatable components unsupported
@item @code{defaultmap} extensions @tab Y @tab
@item @code{declare mapper} directive @tab N @tab
@item @code{omp_get_supported_active_levels} routine @tab Y @tab
@item Runtime routines and environment variables to display runtime thread
      affinity information @tab Y @tab
@item @code{omp_pause_resource} and @code{omp_pause_resource_all} runtime
      routines @tab Y @tab
@item @code{omp_get_device_num} runtime routine @tab Y @tab
@item OMPT interface @tab N @tab
@item OMPD interface @tab N @tab
@end multitable

@unnumberedsubsec Other new OpenMP 5.0 features

@multitable @columnfractions .60 .10 .25
@headitem Description @tab Status @tab Comments
@item Supporting C++'s range-based for loop @tab Y @tab
@end multitable


@node OpenMP 5.1
@section OpenMP 5.1

@unnumberedsubsec New features listed in Appendix B of the OpenMP specification

@multitable @columnfractions .60 .10 .25
@headitem Description @tab Status @tab Comments
@item OpenMP directive as C++ attribute specifiers @tab Y @tab
@item @code{omp_all_memory} reserved locator @tab Y @tab
@item @emph{target_device trait} in OpenMP Context @tab N @tab
@item @code{target_device} selector set in context selectors @tab N @tab
@item C/C++'s @code{declare variant} directive: elision support of
      preprocessed code @tab N @tab
@item @code{declare variant}: new clauses @code{adjust_args} and
      @code{append_args} @tab N @tab
@item @code{dispatch} construct @tab N @tab
@item device-specific ICV settings with environment variables @tab Y @tab
@item @code{assume} and @code{assumes} directives @tab Y @tab
@item @code{nothing} directive @tab Y @tab
@item @code{error} directive @tab Y @tab
@item @code{masked} construct @tab Y @tab
@item @code{scope} directive @tab Y @tab
@item Loop transformation constructs @tab N @tab
@item @code{strict} modifier in the @code{grainsize} and @code{num_tasks}
      clauses of the @code{taskloop} construct @tab Y @tab
@item @code{align} clause in @code{allocate} directive @tab N @tab
@item @code{align} modifier in @code{allocate} clause @tab Y @tab
@item @code{thread_limit} clause to @code{target} construct @tab Y @tab
@item @code{has_device_addr} clause to @code{target} construct @tab Y @tab
@item Iterators in @code{target update} motion clauses and @code{map}
      clauses @tab N @tab
@item Indirect calls to the device version of a procedure or function in
      @code{target} regions @tab N @tab
@item @code{interop} directive @tab N @tab
@item @code{omp_interop_t} object support in runtime routines @tab N @tab
@item @code{nowait} clause in @code{taskwait} directive @tab Y @tab
@item Extensions to the @code{atomic} directive @tab Y @tab
@item @code{seq_cst} clause on a @code{flush} construct @tab Y @tab
@item @code{inoutset} argument to the @code{depend} clause @tab Y @tab
@item @code{private} and @code{firstprivate} argument to @code{default}
      clause in C and C++ @tab Y @tab
@item @code{present} argument to @code{defaultmap} clause @tab Y @tab
@item @code{omp_set_num_teams}, @code{omp_set_teams_thread_limit},
      @code{omp_get_max_teams}, @code{omp_get_teams_thread_limit} runtime
      routines @tab Y @tab
@item @code{omp_target_is_accessible} runtime routine @tab Y @tab
@item @code{omp_target_memcpy_async} and @code{omp_target_memcpy_rect_async}
      runtime routines @tab Y @tab
@item @code{omp_get_mapped_ptr} runtime routine @tab Y @tab
@item @code{omp_calloc}, @code{omp_realloc}, @code{omp_aligned_alloc} and
      @code{omp_aligned_calloc} runtime routines @tab Y @tab
@item @code{omp_alloctrait_key_t} enum: @code{omp_atv_serialized} added,
      @code{omp_atv_default} changed @tab Y @tab
@item @code{omp_display_env} runtime routine @tab Y @tab
@item @code{ompt_scope_endpoint_t} enum: @code{ompt_scope_beginend} @tab N @tab
@item @code{ompt_sync_region_t} enum additions @tab N @tab
@item @code{ompt_state_t} enum: @code{ompt_state_wait_barrier_implementation}
      and @code{ompt_state_wait_barrier_teams} @tab N @tab
@item @code{ompt_callback_target_data_op_emi_t},
      @code{ompt_callback_target_emi_t}, @code{ompt_callback_target_map_emi_t}
      and @code{ompt_callback_target_submit_emi_t} @tab N @tab
@item @code{ompt_callback_error_t} type @tab N @tab
@item @code{OMP_PLACES} syntax extensions @tab Y @tab
@item @code{OMP_NUM_TEAMS} and @code{OMP_TEAMS_THREAD_LIMIT} environment
      variables @tab Y @tab
@end multitable

@unnumberedsubsec Other new OpenMP 5.1 features

@multitable @columnfractions .60 .10 .25
@headitem Description @tab Status @tab Comments
@item Support of strictly structured blocks in Fortran @tab Y @tab
@item Support of structured block sequences in C/C++ @tab Y @tab
@item @code{unconstrained} and @code{reproducible} modifiers on @code{order}
      clause @tab Y @tab
@item Support @code{begin/end declare target} syntax in C/C++ @tab Y @tab
@item Pointer predetermined firstprivate getting initialized
to address of matching mapped list item per 5.1, Sect. 2.21.7.2 @tab N @tab
@item For Fortran, diagnose placing declarative before/between @code{USE},
      @code{IMPORT}, and @code{IMPLICIT} as invalid @tab N @tab
@item Optional comma between directive and clause in the @code{#pragma} form @tab Y @tab
@item @code{indirect} clause in @code{declare target} @tab N @tab
@item @code{device_type(nohost)}/@code{device_type(host)} for variables @tab N @tab
@item @code{present} modifier to the @code{map}, @code{to} and @code{from}
      clauses @tab Y @tab
@end multitable


@node OpenMP 5.2
@section OpenMP 5.2

@unnumberedsubsec New features listed in Appendix B of the OpenMP specification

@multitable @columnfractions .60 .10 .25
@headitem Description @tab Status @tab Comments
@item @code{omp_in_explicit_task} routine and @var{explicit-task-var} ICV
      @tab Y @tab
@item @code{omp}/@code{ompx}/@code{omx} sentinels and @code{omp_}/@code{ompx_}
      namespaces @tab N/A
      @tab warning for @code{ompx/omx} sentinels@footnote{The @code{ompx}
      sentinel as C/C++ pragma and C++ attributes are warned for with
      @code{-Wunknown-pragmas} (implied by @code{-Wall}) and @code{-Wattributes}
      (enabled by default), respectively; for Fortran free-source code, there is
      a warning enabled by default and, for fixed-source code, the @code{omx}
      sentinel is warned for with with @code{-Wsurprising} (enabled by
      @code{-Wall}).  Unknown clauses are always rejected with an error.}
@item Clauses on @code{end} directive can be on directive @tab Y @tab
@item Deprecation of no-argument @code{destroy} clause on @code{depobj}
      @tab N @tab
@item @code{linear} clause syntax changes and @code{step} modifier @tab Y @tab
@item Deprecation of minus operator for reductions @tab N @tab
@item Deprecation of separating @code{map} modifiers without comma @tab N @tab
@item @code{declare mapper} with iterator and @code{present} modifiers
      @tab N @tab
@item If a matching mapped list item is not found in the data environment, the
      pointer retains its original value @tab Y @tab
@item New @code{enter} clause as alias for @code{to} on declare target directive
      @tab Y @tab
@item Deprecation of @code{to} clause on declare target directive @tab N @tab
@item Extended list of directives permitted in Fortran pure procedures
      @tab Y @tab
@item New @code{allocators} directive for Fortran @tab N @tab
@item Deprecation of @code{allocate} directive for Fortran
      allocatables/pointers @tab N @tab
@item Optional paired @code{end} directive with @code{dispatch} @tab N @tab
@item New @code{memspace} and @code{traits} modifiers for @code{uses_allocators}
      @tab N @tab
@item Deprecation of traits array following the allocator_handle expression in
      @code{uses_allocators} @tab N @tab
@item New @code{otherwise} clause as alias for @code{default} on metadirectives
      @tab N @tab
@item Deprecation of @code{default} clause on metadirectives @tab N @tab
@item Deprecation of delimited form of @code{declare target} @tab N @tab
@item Reproducible semantics changed for @code{order(concurrent)} @tab N @tab
@item @code{allocate} and @code{firstprivate} clauses on @code{scope}
      @tab Y @tab
@item @code{ompt_callback_work} @tab N @tab
@item Default map-type for the @code{map} clause in @code{target enter/exit data}
      @tab Y @tab
@item New @code{doacross} clause as alias for @code{depend} with
      @code{source}/@code{sink} modifier @tab Y @tab
@item Deprecation of @code{depend} with @code{source}/@code{sink} modifier
      @tab N @tab
@item @code{omp_cur_iteration} keyword @tab Y @tab
@end multitable

@unnumberedsubsec Other new OpenMP 5.2 features

@multitable @columnfractions .60 .10 .25
@headitem Description @tab Status @tab Comments
@item For Fortran, optional comma between directive and clause @tab N @tab
@item Conforming device numbers and @code{omp_initial_device} and
      @code{omp_invalid_device} enum/PARAMETER @tab Y @tab
@item Initial value of @var{default-device-var} ICV with
      @code{OMP_TARGET_OFFLOAD=mandatory} @tab Y @tab
@item @emph{interop_types} in any position of the modifier list for the @code{init} clause
      of the @code{interop} construct @tab N @tab
@end multitable


@node OpenMP Technical Report 11
@section OpenMP Technical Report 11

Technical Report (TR) 11 is the first preview for OpenMP 6.0.

@unnumberedsubsec New features listed in Appendix B of the OpenMP specification
@multitable @columnfractions .60 .10 .25
@item Features deprecated in versions 5.2, 5.1 and 5.0 were removed
      @tab N/A @tab Backward compatibility
@item The @code{decl} attribute was added to the C++ attribute syntax
      @tab N @tab
@item @code{_ALL} suffix to the device-scope environment variables
      @tab P @tab Host device number wrongly accepted
@item For Fortran, @emph{locator list} can be also function reference with
      data pointer result @tab N @tab
@item Ref-count change for @code{use_device_ptr}/@code{use_device_addr}
      @tab N @tab
@item Implicit reduction identifiers of C++ classes
      @tab N @tab
@item Change of the @emph{map-type} property from @emph{ultimate} to
      @emph{default} @tab N @tab
@item Concept of @emph{assumed-size arrays} in C and C++
      @tab N @tab
@item Mapping of @emph{assumed-size arrays} in C, C++ and Fortran
      @tab N @tab
@item @code{groupprivate} directive @tab N @tab
@item @code{local} clause to declare target directive @tab N @tab
@item @code{part_size} allocator trait @tab N @tab
@item @code{pin_device}, @code{preferred_device} and @code{target_access}
      allocator traits
      @tab N @tab
@item @code{access} allocator trait changes @tab N @tab
@item Extension of @code{interop} operation of @code{append_args}, allowing all
      modifiers of the @code{init} clause
      @tab N @tab
@item @code{interop} clause to @code{dispatch} @tab N @tab
@item @code{apply} code to loop-transforming constructs @tab N @tab
@item @code{omp_curr_progress_width} identifier @tab N @tab
@item @code{safesync} clause to the @code{parallel} construct @tab N @tab
@item @code{omp_get_max_progress_width} runtime routine @tab N @tab
@item @code{strict} modifier keyword to @code{num_threads} @tab N @tab
@item @code{memscope} clause to @code{atomic} and @code{flush} @tab N @tab
@item Routines for obtaining memory spaces/allocators for shared/device memory
      @tab N @tab
@item @code{omp_get_memspace_num_resources} routine @tab N @tab
@item @code{omp_get_submemspace} routine @tab N @tab
@item @code{ompt_get_buffer_limits} OMPT routine @tab N @tab
@item Extension of @code{OMP_DEFAULT_DEVICE} and new
      @code{OMP_AVAILABLE_DEVICES} environment vars @tab N @tab
@item Supporting increments with abstract names in @code{OMP_PLACES} @tab N @tab
@end multitable

@unnumberedsubsec Other new TR 11 features
@multitable @columnfractions .60 .10 .25
@item Relaxed Fortran restrictions to the @code{aligned} clause @tab N @tab
@item Mapping lambda captures @tab N @tab
@item For Fortran, atomic compare with storing the comparison result
      @tab N @tab
@end multitable



@c ---------------------------------------------------------------------
@c OpenMP Runtime Library Routines
@c ---------------------------------------------------------------------

@node Runtime Library Routines
@chapter OpenMP Runtime Library Routines

The runtime routines described here are defined by Section 3 of the OpenMP
specification in version 4.5.  The routines are structured in following
three parts:

@menu
Control threads, processors and the parallel environment.  They have C
linkage, and do not throw exceptions.

* omp_get_active_level::        Number of active parallel regions
* omp_get_ancestor_thread_num:: Ancestor thread ID
* omp_get_cancellation::        Whether cancellation support is enabled
* omp_get_default_device::      Get the default device for target regions
* omp_get_device_num::          Get device that current thread is running on
* omp_get_dynamic::             Dynamic teams setting
* omp_get_initial_device::      Device number of host device
* omp_get_level::               Number of parallel regions
* omp_get_max_active_levels::   Current maximum number of active regions
* omp_get_max_task_priority::   Maximum task priority value that can be set
* omp_get_max_teams::           Maximum number of teams for teams region
* omp_get_max_threads::         Maximum number of threads of parallel region
* omp_get_nested::              Nested parallel regions
* omp_get_num_devices::         Number of target devices
* omp_get_num_procs::           Number of processors online
* omp_get_num_teams::           Number of teams
* omp_get_num_threads::         Size of the active team
* omp_get_proc_bind::           Whether threads may be moved between CPUs
* omp_get_schedule::            Obtain the runtime scheduling method
* omp_get_supported_active_levels:: Maximum number of active regions supported
* omp_get_team_num::            Get team number
* omp_get_team_size::           Number of threads in a team
* omp_get_teams_thread_limit::  Maximum number of threads imposed by teams
* omp_get_thread_limit::        Maximum number of threads
* omp_get_thread_num::          Current thread ID
* omp_in_parallel::             Whether a parallel region is active
* omp_in_final::                Whether in final or included task region
* omp_is_initial_device::       Whether executing on the host device
* omp_set_default_device::      Set the default device for target regions
* omp_set_dynamic::             Enable/disable dynamic teams
* omp_set_max_active_levels::   Limits the number of active parallel regions
* omp_set_nested::              Enable/disable nested parallel regions
* omp_set_num_teams::           Set upper teams limit for teams region
* omp_set_num_threads::         Set upper team size limit
* omp_set_schedule::            Set the runtime scheduling method
* omp_set_teams_thread_limit::  Set upper thread limit for teams construct

Initialize, set, test, unset and destroy simple and nested locks.

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

Portable, thread-based, wall clock timer.

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

Support for event objects.

* omp_fulfill_event::        Fulfill and destroy an OpenMP event.
@end menu



@node omp_get_active_level
@section @code{omp_get_active_level} -- Number of parallel regions
@table @asis
@item @emph{Description}:
This function returns the nesting level for the active parallel blocks,
which enclose the calling call.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_active_level(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_active_level()}
@end multitable

@item @emph{See also}:
@ref{omp_get_level}, @ref{omp_get_max_active_levels}, @ref{omp_set_max_active_levels}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.20.
@end table



@node omp_get_ancestor_thread_num
@section @code{omp_get_ancestor_thread_num} -- Ancestor thread ID
@table @asis
@item @emph{Description}:
This function returns the thread identification number for the given
nesting level of the current thread.  For values of @var{level} outside
zero to @code{omp_get_level} -1 is returned; if @var{level} is
@code{omp_get_level} the result is identical to @code{omp_get_thread_num}.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_ancestor_thread_num(int level);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_ancestor_thread_num(level)}
@item                   @tab @code{integer level}
@end multitable

@item @emph{See also}:
@ref{omp_get_level}, @ref{omp_get_thread_num}, @ref{omp_get_team_size}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.18.
@end table



@node omp_get_cancellation
@section @code{omp_get_cancellation} -- Whether cancellation support is enabled
@table @asis
@item @emph{Description}:
This function returns @code{true} if cancellation is activated, @code{false}
otherwise.  Here, @code{true} and @code{false} represent their language-specific
counterparts.  Unless @env{OMP_CANCELLATION} is set true, cancellations are
deactivated.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_cancellation(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{logical function omp_get_cancellation()}
@end multitable

@item @emph{See also}:
@ref{OMP_CANCELLATION}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.9.
@end table



@node omp_get_default_device
@section @code{omp_get_default_device} -- Get the default device for target regions
@table @asis
@item @emph{Description}:
Get the default device for target regions without device clause.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_default_device(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_default_device()}
@end multitable

@item @emph{See also}:
@ref{OMP_DEFAULT_DEVICE}, @ref{omp_set_default_device}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.30.
@end table



@node omp_get_device_num
@section @code{omp_get_device_num} -- Return device number of current device
@table @asis
@item @emph{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 @code{omp_get_initial_device} function when called
from the host.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_device_num(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_device_num()}
@end multitable

@item @emph{See also}:
@ref{omp_get_initial_device}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.0}, Section 3.2.37.
@end table



@node omp_get_dynamic
@section @code{omp_get_dynamic} -- Dynamic teams setting
@table @asis
@item @emph{Description}:
This function returns @code{true} if enabled, @code{false} otherwise. 
Here, @code{true} and @code{false} represent their language-specific 
counterparts.

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

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_dynamic(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{logical function omp_get_dynamic()}
@end multitable

@item @emph{See also}:
@ref{omp_set_dynamic}, @ref{OMP_DYNAMIC}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.8.
@end table



@node omp_get_initial_device
@section @code{omp_get_initial_device} -- Return device number of initial device
@table @asis
@item @emph{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
@code{omp_get_num_devices} function.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_initial_device(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_initial_device()}
@end multitable

@item @emph{See also}:
@ref{omp_get_num_devices}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.35.
@end table



@node omp_get_level
@section @code{omp_get_level} -- Obtain the current nesting level
@table @asis
@item @emph{Description}:
This function returns the nesting level for the parallel blocks,
which enclose the calling call.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_level(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_level()}
@end multitable

@item @emph{See also}:
@ref{omp_get_active_level}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.17.
@end table



@node omp_get_max_active_levels
@section @code{omp_get_max_active_levels} -- Current maximum number of active regions
@table @asis
@item @emph{Description}:
This function obtains the maximum allowed number of nested, active parallel regions.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_max_active_levels(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_max_active_levels()}
@end multitable

@item @emph{See also}:
@ref{omp_set_max_active_levels}, @ref{omp_get_active_level}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.16.
@end table


@node omp_get_max_task_priority
@section @code{omp_get_max_task_priority} -- Maximum priority value
that can be set for tasks.
@table @asis
@item @emph{Description}:
This function obtains the maximum allowed priority number for tasks.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_max_task_priority(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_max_task_priority()}
@end multitable

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.29.
@end table


@node omp_get_max_teams
@section @code{omp_get_max_teams} -- Maximum number of teams of teams region
@table @asis
@item @emph{Description}:
Return the maximum number of teams used for the teams region
that does not use the clause @code{num_teams}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_max_teams(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_max_teams()}
@end multitable

@item @emph{See also}:
@ref{omp_set_num_teams}, @ref{omp_get_num_teams}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.1}, Section 3.4.4.
@end table



@node omp_get_max_threads
@section @code{omp_get_max_threads} -- Maximum number of threads of parallel region
@table @asis
@item @emph{Description}:
Return the maximum number of threads used for the current parallel region
that does not use the clause @code{num_threads}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_max_threads(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_max_threads()}
@end multitable

@item @emph{See also}:
@ref{omp_set_num_threads}, @ref{omp_set_dynamic}, @ref{omp_get_thread_limit}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.3.
@end table



@node omp_get_nested
@section @code{omp_get_nested} -- Nested parallel regions
@table @asis
@item @emph{Description}:
This function returns @code{true} if nested parallel regions are
enabled, @code{false} otherwise.  Here, @code{true} and @code{false}
represent their language-specific counterparts.

The state of nested parallel regions at startup depends on several
environment variables.  If @env{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 @env{OMP_NESTED}
environment variable will be followed if defined.  If neither are
defined, then if either @env{OMP_NUM_THREADS} or @env{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
@code{omp_set_nested}, or by setting the maximum number of nested
regions with @code{omp_set_max_active_levels} to one to disable, or
above one to enable.

Note that the @code{omp_get_nested} API routine was deprecated
in the OpenMP specification 5.2 in favor of @code{omp_get_max_active_levels}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_nested(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{logical function omp_get_nested()}
@end multitable

@item @emph{See also}:
@ref{omp_get_max_active_levels}, @ref{omp_set_nested},
@ref{OMP_MAX_ACTIVE_LEVELS}, @ref{OMP_NESTED}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.11.
@end table



@node omp_get_num_devices
@section @code{omp_get_num_devices} -- Number of target devices
@table @asis
@item @emph{Description}:
Returns the number of target devices.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_num_devices(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_num_devices()}
@end multitable

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.31.
@end table



@node omp_get_num_procs
@section @code{omp_get_num_procs} -- Number of processors online
@table @asis
@item @emph{Description}:
Returns the number of processors online on that device.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_num_procs(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_num_procs()}
@end multitable

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.5.
@end table



@node omp_get_num_teams
@section @code{omp_get_num_teams} -- Number of teams
@table @asis
@item @emph{Description}:
Returns the number of teams in the current team region.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_num_teams(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_num_teams()}
@end multitable

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.32.
@end table



@node omp_get_num_threads
@section @code{omp_get_num_threads} -- Size of the active team
@table @asis
@item @emph{Description}:
Returns the number of threads in the current team.  In a sequential section of
the program @code{omp_get_num_threads} returns 1.

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

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_num_threads(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_num_threads()}
@end multitable

@item @emph{See also}:
@ref{omp_get_max_threads}, @ref{omp_set_num_threads}, @ref{OMP_NUM_THREADS}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.2.
@end table



@node omp_get_proc_bind
@section @code{omp_get_proc_bind} -- Whether threads may be moved between CPUs
@table @asis
@item @emph{Description}:
This functions returns the currently active thread affinity policy, which is
set via @env{OMP_PROC_BIND}.  Possible values are @code{omp_proc_bind_false},
@code{omp_proc_bind_true}, @code{omp_proc_bind_primary},
@code{omp_proc_bind_master}, @code{omp_proc_bind_close} and @code{omp_proc_bind_spread},
where @code{omp_proc_bind_master} is an alias for @code{omp_proc_bind_primary}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{omp_proc_bind_t omp_get_proc_bind(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer(kind=omp_proc_bind_kind) function omp_get_proc_bind()}
@end multitable

@item @emph{See also}:
@ref{OMP_PROC_BIND}, @ref{OMP_PLACES}, @ref{GOMP_CPU_AFFINITY},

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.22.
@end table



@node omp_get_schedule
@section @code{omp_get_schedule} -- Obtain the runtime scheduling method
@table @asis
@item @emph{Description}:
Obtain the runtime scheduling method.  The @var{kind} argument will be
set to the value @code{omp_sched_static}, @code{omp_sched_dynamic},
@code{omp_sched_guided} or @code{omp_sched_auto}.  The second argument,
@var{chunk_size}, is set to the chunk size.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_get_schedule(omp_sched_t *kind, int *chunk_size);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_get_schedule(kind, chunk_size)}
@item                   @tab @code{integer(kind=omp_sched_kind) kind}
@item                   @tab @code{integer chunk_size}
@end multitable

@item @emph{See also}:
@ref{omp_set_schedule}, @ref{OMP_SCHEDULE}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.13.
@end table


@node omp_get_supported_active_levels
@section @code{omp_get_supported_active_levels} -- Maximum number of active regions supported
@table @asis
@item @emph{Description}:
This function returns the maximum number of nested, active parallel regions
supported by this implementation.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_supported_active_levels(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_supported_active_levels()}
@end multitable

@item @emph{See also}:
@ref{omp_get_max_active_levels}, @ref{omp_set_max_active_levels}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.0}, Section 3.2.15.
@end table



@node omp_get_team_num
@section @code{omp_get_team_num} -- Get team number
@table @asis
@item @emph{Description}:
Returns the team number of the calling thread.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_team_num(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_team_num()}
@end multitable

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.33.
@end table



@node omp_get_team_size
@section @code{omp_get_team_size} -- Number of threads in a team
@table @asis
@item @emph{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 @var{level}
outside zero to @code{omp_get_level}, -1 is returned; if @var{level} is zero,
1 is returned, and for @code{omp_get_level}, the result is identical
to @code{omp_get_num_threads}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_team_size(int level);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_team_size(level)}
@item                   @tab @code{integer level}
@end multitable

@item @emph{See also}:
@ref{omp_get_num_threads}, @ref{omp_get_level}, @ref{omp_get_ancestor_thread_num}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.19.
@end table



@node omp_get_teams_thread_limit
@section @code{omp_get_teams_thread_limit} -- Maximum number of threads imposed by teams
@table @asis
@item @emph{Description}:
Return the maximum number of threads that will be able to participate in
each team created by a teams construct.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_teams_thread_limit(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_teams_thread_limit()}
@end multitable

@item @emph{See also}:
@ref{omp_set_teams_thread_limit}, @ref{OMP_TEAMS_THREAD_LIMIT}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.1}, Section 3.4.6.
@end table



@node omp_get_thread_limit
@section @code{omp_get_thread_limit} -- Maximum number of threads
@table @asis
@item @emph{Description}:
Return the maximum number of threads of the program.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_thread_limit(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_thread_limit()}
@end multitable

@item @emph{See also}:
@ref{omp_get_max_threads}, @ref{OMP_THREAD_LIMIT}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.14.
@end table



@node omp_get_thread_num
@section @code{omp_get_thread_num} -- Current thread ID
@table @asis
@item @emph{Description}:
Returns a unique thread identification number within the current team.
In a sequential parts of the program, @code{omp_get_thread_num}
always returns 0.  In parallel regions the return value varies
from 0 to @code{omp_get_num_threads}-1 inclusive.  The return
value of the primary thread of a team is always 0.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_get_thread_num(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function omp_get_thread_num()}
@end multitable

@item @emph{See also}:
@ref{omp_get_num_threads}, @ref{omp_get_ancestor_thread_num}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.4.
@end table



@node omp_in_parallel
@section @code{omp_in_parallel} -- Whether a parallel region is active
@table @asis
@item @emph{Description}:
This function returns @code{true} if currently running in parallel,
@code{false} otherwise.  Here, @code{true} and @code{false} represent
their language-specific counterparts.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_in_parallel(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{logical function omp_in_parallel()}
@end multitable

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.6.
@end table


@node omp_in_final
@section @code{omp_in_final} -- Whether in final or included task region
@table @asis
@item @emph{Description}:
This function returns @code{true} if currently running in a final
or included task region, @code{false} otherwise.  Here, @code{true}
and @code{false} represent their language-specific counterparts.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_in_final(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{logical function omp_in_final()}
@end multitable

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.21.
@end table



@node omp_is_initial_device
@section @code{omp_is_initial_device} -- Whether executing on the host device
@table @asis
@item @emph{Description}:
This function returns @code{true} if currently running on the host device,
@code{false} otherwise.  Here, @code{true} and @code{false} represent
their language-specific counterparts.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_is_initial_device(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{logical function omp_is_initial_device()}
@end multitable

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.34.
@end table



@node omp_set_default_device
@section @code{omp_set_default_device} -- Set the default device for target regions
@table @asis
@item @emph{Description}:
Set the default device for target regions without device clause.  The argument
shall be a nonnegative device number.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_set_default_device(int device_num);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_set_default_device(device_num)}
@item                   @tab @code{integer device_num}
@end multitable

@item @emph{See also}:
@ref{OMP_DEFAULT_DEVICE}, @ref{omp_get_default_device}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.29.
@end table



@node omp_set_dynamic
@section @code{omp_set_dynamic} -- Enable/disable dynamic teams
@table @asis
@item @emph{Description}:
Enable or disable the dynamic adjustment of the number of threads 
within a team.  The function takes the language-specific equivalent
of @code{true} and @code{false}, where @code{true} enables dynamic 
adjustment of team sizes and @code{false} disables it.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_set_dynamic(int dynamic_threads);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_set_dynamic(dynamic_threads)}
@item                   @tab @code{logical, intent(in) :: dynamic_threads}
@end multitable

@item @emph{See also}:
@ref{OMP_DYNAMIC}, @ref{omp_get_dynamic}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.7.
@end table



@node omp_set_max_active_levels
@section @code{omp_set_max_active_levels} -- Limits the number of active parallel regions
@table @asis
@item @emph{Description}:
This function limits the maximum allowed number of nested, active
parallel regions.  @var{max_levels} must be less or equal to
the value returned by @code{omp_get_supported_active_levels}.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_set_max_active_levels(int max_levels);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_set_max_active_levels(max_levels)}
@item                   @tab @code{integer max_levels}
@end multitable

@item @emph{See also}:
@ref{omp_get_max_active_levels}, @ref{omp_get_active_level},
@ref{omp_get_supported_active_levels}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.15.
@end table



@node omp_set_nested
@section @code{omp_set_nested} -- Enable/disable nested parallel regions
@table @asis
@item @emph{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 @code{true} and @code{false}, where @code{true} enables 
dynamic adjustment of team sizes and @code{false} disables it.

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

Note that the @code{omp_set_nested} API routine was deprecated
in the OpenMP specification 5.2 in favor of @code{omp_set_max_active_levels}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_set_nested(int nested);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_set_nested(nested)}
@item                   @tab @code{logical, intent(in) :: nested}
@end multitable

@item @emph{See also}:
@ref{omp_get_nested}, @ref{omp_set_max_active_levels},
@ref{OMP_MAX_ACTIVE_LEVELS}, @ref{OMP_NESTED}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.10.
@end table



@node omp_set_num_teams
@section @code{omp_set_num_teams} -- Set upper teams limit for teams construct
@table @asis
@item @emph{Description}:
Specifies the upper bound for number of teams created by the teams construct
which does not specify a @code{num_teams} clause.  The
argument of @code{omp_set_num_teams} shall be a positive integer.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_set_num_teams(int num_teams);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_set_num_teams(num_teams)}
@item                   @tab @code{integer, intent(in) :: num_teams}
@end multitable

@item @emph{See also}:
@ref{OMP_NUM_TEAMS}, @ref{omp_get_num_teams}, @ref{omp_get_max_teams}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.1}, Section 3.4.3.
@end table



@node omp_set_num_threads
@section @code{omp_set_num_threads} -- Set upper team size limit
@table @asis
@item @emph{Description}:
Specifies the number of threads used by default in subsequent parallel 
sections, if those do not specify a @code{num_threads} clause.  The
argument of @code{omp_set_num_threads} shall be a positive integer.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_set_num_threads(int num_threads);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_set_num_threads(num_threads)}
@item                   @tab @code{integer, intent(in) :: num_threads}
@end multitable

@item @emph{See also}:
@ref{OMP_NUM_THREADS}, @ref{omp_get_num_threads}, @ref{omp_get_max_threads}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.1.
@end table



@node omp_set_schedule
@section @code{omp_set_schedule} -- Set the runtime scheduling method
@table @asis
@item @emph{Description}:
Sets the runtime scheduling method.  The @var{kind} argument can have the
value @code{omp_sched_static}, @code{omp_sched_dynamic},
@code{omp_sched_guided} or @code{omp_sched_auto}.  Except for
@code{omp_sched_auto}, the chunk size is set to the value of
@var{chunk_size} if positive, or to the default value if zero or negative.
For @code{omp_sched_auto} the @var{chunk_size} argument is ignored.

@item @emph{C/C++}
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_set_schedule(omp_sched_t kind, int chunk_size);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_set_schedule(kind, chunk_size)}
@item                   @tab @code{integer(kind=omp_sched_kind) kind}
@item                   @tab @code{integer chunk_size}
@end multitable

@item @emph{See also}:
@ref{omp_get_schedule}
@ref{OMP_SCHEDULE}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.12.
@end table



@node omp_set_teams_thread_limit
@section @code{omp_set_teams_thread_limit} -- Set upper thread limit for teams construct
@table @asis
@item @emph{Description}:
Specifies the upper bound for number of threads that will be available
for each team created by the teams construct which does not specify a
@code{thread_limit} clause.  The argument of
@code{omp_set_teams_thread_limit} shall be a positive integer.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_set_teams_thread_limit(int thread_limit);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_set_teams_thread_limit(thread_limit)}
@item                   @tab @code{integer, intent(in) :: thread_limit}
@end multitable

@item @emph{See also}:
@ref{OMP_TEAMS_THREAD_LIMIT}, @ref{omp_get_teams_thread_limit}, @ref{omp_get_thread_limit}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.1}, Section 3.4.5.
@end table



@node omp_init_lock
@section @code{omp_init_lock} -- Initialize simple lock
@table @asis
@item @emph{Description}:
Initialize a simple lock.  After initialization, the lock is in
an unlocked state.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_init_lock(omp_lock_t *lock);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_init_lock(svar)}
@item                   @tab @code{integer(omp_lock_kind), intent(out) :: svar}
@end multitable

@item @emph{See also}:
@ref{omp_destroy_lock}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.1.
@end table



@node omp_set_lock
@section @code{omp_set_lock} -- Wait for and set simple lock
@table @asis
@item @emph{Description}:
Before setting a simple lock, the lock variable must be initialized by 
@code{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.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_set_lock(omp_lock_t *lock);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_set_lock(svar)}
@item                   @tab @code{integer(omp_lock_kind), intent(inout) :: svar}
@end multitable

@item @emph{See also}:
@ref{omp_init_lock}, @ref{omp_test_lock}, @ref{omp_unset_lock}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.4.
@end table



@node omp_test_lock
@section @code{omp_test_lock} -- Test and set simple lock if available
@table @asis
@item @emph{Description}:
Before setting a simple lock, the lock variable must be initialized by 
@code{omp_init_lock}.  Contrary to @code{omp_set_lock}, @code{omp_test_lock} 
does not block if the lock is not available.  This function returns
@code{true} upon success, @code{false} otherwise.  Here, @code{true} and
@code{false} represent their language-specific counterparts.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_test_lock(omp_lock_t *lock);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{logical function omp_test_lock(svar)}
@item                   @tab @code{integer(omp_lock_kind), intent(inout) :: svar}
@end multitable

@item @emph{See also}:
@ref{omp_init_lock}, @ref{omp_set_lock}, @ref{omp_set_lock}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.6.
@end table



@node omp_unset_lock
@section @code{omp_unset_lock} -- Unset simple lock
@table @asis
@item @emph{Description}:
A simple lock about to be unset must have been locked by @code{omp_set_lock}
or @code{omp_test_lock} before.  In addition, the lock must be held by the
thread calling @code{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.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_unset_lock(omp_lock_t *lock);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_unset_lock(svar)}
@item                   @tab @code{integer(omp_lock_kind), intent(inout) :: svar}
@end multitable

@item @emph{See also}:
@ref{omp_set_lock}, @ref{omp_test_lock}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.5.
@end table



@node omp_destroy_lock
@section @code{omp_destroy_lock} -- Destroy simple lock
@table @asis
@item @emph{Description}:
Destroy a simple lock.  In order to be destroyed, a simple lock must be
in the unlocked state. 

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_destroy_lock(omp_lock_t *lock);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_destroy_lock(svar)}
@item                   @tab @code{integer(omp_lock_kind), intent(inout) :: svar}
@end multitable

@item @emph{See also}:
@ref{omp_init_lock}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.3.
@end table



@node omp_init_nest_lock
@section @code{omp_init_nest_lock} -- Initialize nested lock
@table @asis
@item @emph{Description}:
Initialize a nested lock.  After initialization, the lock is in
an unlocked state and the nesting count is set to zero.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_init_nest_lock(omp_nest_lock_t *lock);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_init_nest_lock(nvar)}
@item                   @tab @code{integer(omp_nest_lock_kind), intent(out) :: nvar}
@end multitable

@item @emph{See also}:
@ref{omp_destroy_nest_lock}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.1.
@end table


@node omp_set_nest_lock
@section @code{omp_set_nest_lock} -- Wait for and set nested lock
@table @asis
@item @emph{Description}:
Before setting a nested lock, the lock variable must be initialized by 
@code{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.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_set_nest_lock(omp_nest_lock_t *lock);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_set_nest_lock(nvar)}
@item                   @tab @code{integer(omp_nest_lock_kind), intent(inout) :: nvar}
@end multitable

@item @emph{See also}:
@ref{omp_init_nest_lock}, @ref{omp_unset_nest_lock}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.4.
@end table



@node omp_test_nest_lock
@section @code{omp_test_nest_lock} -- Test and set nested lock if available
@table @asis
@item @emph{Description}:
Before setting a nested lock, the lock variable must be initialized by 
@code{omp_init_nest_lock}.  Contrary to @code{omp_set_nest_lock},
@code{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.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int omp_test_nest_lock(omp_nest_lock_t *lock);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{logical function omp_test_nest_lock(nvar)}
@item                   @tab @code{integer(omp_nest_lock_kind), intent(inout) :: nvar}
@end multitable


@item @emph{See also}:
@ref{omp_init_lock}, @ref{omp_set_lock}, @ref{omp_set_lock}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.6.
@end table



@node omp_unset_nest_lock
@section @code{omp_unset_nest_lock} -- Unset nested lock
@table @asis
@item @emph{Description}:
A nested lock about to be unset must have been locked by @code{omp_set_nested_lock}
or @code{omp_test_nested_lock} before.  In addition, the lock must be held by the
thread calling @code{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.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_unset_nest_lock(omp_nest_lock_t *lock);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_unset_nest_lock(nvar)}
@item                   @tab @code{integer(omp_nest_lock_kind), intent(inout) :: nvar}
@end multitable

@item @emph{See also}:
@ref{omp_set_nest_lock}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.5.
@end table



@node omp_destroy_nest_lock
@section @code{omp_destroy_nest_lock} -- Destroy nested lock
@table @asis
@item @emph{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.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_destroy_nest_lock(omp_nest_lock_t *);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_destroy_nest_lock(nvar)}
@item                   @tab @code{integer(omp_nest_lock_kind), intent(inout) :: nvar}
@end multitable

@item @emph{See also}:
@ref{omp_init_lock}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.3.
@end table



@node omp_get_wtick
@section @code{omp_get_wtick} -- Get timer precision
@table @asis
@item @emph{Description}:
Gets the timer precision, i.e., the number of seconds between two 
successive clock ticks.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{double omp_get_wtick(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{double precision function omp_get_wtick()}
@end multitable

@item @emph{See also}:
@ref{omp_get_wtime}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.4.2.
@end table



@node omp_get_wtime
@section @code{omp_get_wtime} -- Elapsed wall clock time
@table @asis
@item @emph{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.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{double omp_get_wtime(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{double precision function omp_get_wtime()}
@end multitable

@item @emph{See also}:
@ref{omp_get_wtick}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.4.1.
@end table



@node omp_fulfill_event
@section @code{omp_fulfill_event} -- Fulfill and destroy an OpenMP event
@table @asis
@item @emph{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 @code{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.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void omp_fulfill_event(omp_event_handle_t event);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine omp_fulfill_event(event)}
@item                   @tab @code{integer (kind=omp_event_handle_kind) :: event}
@end multitable

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.0}, Section 3.5.1.
@end table



@c ---------------------------------------------------------------------
@c OpenMP Environment Variables
@c ---------------------------------------------------------------------

@node Environment Variables
@chapter OpenMP Environment Variables

The environment variables which beginning with @env{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 @env{GOMP_} are GNU extensions.
Most @env{OMP_} environment variables have an associated internal control
variable (ICV).

For any OpenMP environment variable that sets an ICV and is neither
@code{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 @code{_DEV_} followed
by a non-negative device number less that the number of available devices sets
the ICV for the corresponding device.  The suffix @code{_DEV} sets the ICV
of all non-host devices for which a device-specific corresponding environment
variable has not been set while the @code{_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 behaviour
* 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
@end menu


@node OMP_ALLOCATOR
@section @env{OMP_ALLOCATOR} -- Set the default allocator
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{available-devices-var}
@item @emph{Scope:} data environment
@item @emph{Description}:
Sets the default allocator that is used when no allocator has been specified
in the @code{allocate} or @code{allocator} clause or if an OpenMP memory
routine is invoked with the @code{omp_null_allocator} allocator.
If unset, @code{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 @code{=}.

Note: The corresponding device environment variables are currently not
supported.  Therefore, the non-host @var{def-allocator-var} ICVs are always
initialized to @code{omp_default_mem_alloc}.  However, on all devices,
the @code{omp_set_default_allocator} API routine can be used to change
value.

@multitable @columnfractions .45 .45
@headitem Predefined allocators @tab Associated predefined memory spaces
@item omp_default_mem_alloc     @tab omp_default_mem_space
@item omp_large_cap_mem_alloc   @tab omp_large_cap_mem_space
@item omp_const_mem_alloc       @tab omp_const_mem_space
@item omp_high_bw_mem_alloc     @tab omp_high_bw_mem_space
@item omp_low_lat_mem_alloc     @tab omp_low_lat_mem_space
@item omp_cgroup_mem_alloc      @tab --
@item omp_pteam_mem_alloc       @tab --
@item omp_thread_mem_alloc      @tab --
@end multitable

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

@multitable @columnfractions .25 .40 .25
@headitem Trait @tab Allowed values @tab Default value
@item @code{sync_hint} @tab @code{contended}, @code{uncontended},
                            @code{serialized}, @code{private}
                       @tab @code{contended}
@item @code{alignment} @tab Positive integer being a power of two
                       @tab 1 byte
@item @code{access}    @tab @code{all}, @code{cgroup},
                            @code{pteam}, @code{thread}
                       @tab @code{all}
@item @code{pool_size} @tab Positive integer
                       @tab See @ref{Memory allocation}
@item @code{fallback}  @tab @code{default_mem_fb}, @code{null_fb},
                            @code{abort_fb}, @code{allocator_fb}
                       @tab See below
@item @code{fb_data}   @tab @emph{unsupported as it needs an allocator handle}
                       @tab (none)
@item @code{pinned}    @tab @code{true}, @code{false}
                       @tab @code{false}
@item @code{partition} @tab @code{environment}, @code{nearest},
                            @code{blocked}, @code{interleaved}
                       @tab @code{environment}
@end multitable

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

Examples:
@smallexample
OMP_ALLOCATOR=omp_high_bw_mem_alloc
OMP_ALLOCATOR=omp_large_cap_mem_space
OMP_ALLOCATR=omp_low_lat_mem_space:pinned=true,partition=nearest
@end smallexample

@item @emph{See also}:
@ref{Memory allocation}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.0}, Section 6.21
@end table



@node OMP_AFFINITY_FORMAT
@section @env{OMP_AFFINITY_FORMAT} -- Set the format string used for affinity display
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{affinity-format-var}
@item @emph{Scope:} device
@item @emph{Description}:
Sets the format string used when displaying OpenMP thread affinity information.
Special values are output using @code{%} followed by an optional size
specification and then either the single-character field type or its long
name enclosed in curly braces; using @code{%%} will display a literal percent.
The size specification consists of an optional @code{0.} or @code{.} followed
by a positive integer, specifying the minimal width of the output.  With
@code{0.} and numerical values, the output is padded with zeros on the left;
with @code{.}, the output is padded by spaces on the left; otherwise, the
output is padded by spaces on the right.  If unset, the value is
``@code{level %L thread %i affinity %A}''.

Supported field types are:

@multitable @columnfractions .10 .25 .60
@item t @tab team_num @tab value returned by @code{omp_get_team_num}
@item T @tab num_teams @tab value returned by @code{omp_get_num_teams}
@item L @tab nesting_level @tab value returned by @code{omp_get_level}
@item n @tab thread_num @tab value returned by @code{omp_get_thread_num}
@item N @tab num_threads @tab value returned by @code{omp_get_num_threads}
@item a @tab ancestor_tnum
      @tab value returned by
           @code{omp_get_ancestor_thread_num(omp_get_level()-1)}
@item H @tab host @tab name of the host that executes the thread
@item P @tab process_id @tab process identifier
@item i @tab native_thread_id @tab native thread identifier
@item A @tab thread_affinity
      @tab comma separated list of integer values or ranges, representing the
           processors on which a process might execute, subject to affinity
           mechanisms
@end multitable

For instance, after setting

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

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

@smallexample
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
@end smallexample

@item @emph{See also}:
@ref{OMP_DISPLAY_AFFINITY}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.0}, Section 6.14
@end table



@node OMP_CANCELLATION
@section @env{OMP_CANCELLATION} -- Set whether cancellation is activated
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{cancel-var}
@item @emph{Scope:} global
@item @emph{Description}:
If set to @code{TRUE}, the cancellation is activated.  If set to @code{FALSE} or
if unset, cancellation is disabled and the @code{cancel} construct is ignored.

@item @emph{See also}:
@ref{omp_get_cancellation}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.11
@end table



@node OMP_DISPLAY_AFFINITY
@section @env{OMP_DISPLAY_AFFINITY} -- Display thread affinity information
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{display-affinity-var}
@item @emph{Scope:} global
@item @emph{Description}:
If set to @code{FALSE} or if unset, affinity displaying is disabled.
If set to @code{TRUE}, the runtime will display affinity information about
OpenMP threads in a parallel region upon entering the region and every time
any change occurs.

@item @emph{See also}:
@ref{OMP_AFFINITY_FORMAT}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.0}, Section 6.13
@end table




@node OMP_DISPLAY_ENV
@section @env{OMP_DISPLAY_ENV} -- Show OpenMP version and environment variables
@cindex Environment Variable
@table @asis
@item @emph{ICV:} none
@item @emph{Scope:} not applicable
@item @emph{Description}:
If set to @code{TRUE}, the OpenMP version number and the values
associated with the OpenMP environment variables are printed to @code{stderr}.
If set to @code{VERBOSE}, it additionally shows the value of the environment
variables which are GNU extensions.  If undefined or set to @code{FALSE},
this information will not be shown.


@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.12
@end table



@node OMP_DEFAULT_DEVICE
@section @env{OMP_DEFAULT_DEVICE} -- Set the device used in target regions
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{default-device-var}
@item @emph{Scope:} data environment
@item @emph{Description}:
Set to choose the device which is used in a @code{target} region, unless the
value is overridden by @code{omp_set_default_device} or by a @code{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,
@env{OMP_TARGET_OFFLOAD} is @code{mandatory} and no non-host devices are
available, it is set to @code{omp_invalid_device}.  Otherwise, if unset,
device number 0 will be used.


@item @emph{See also}:
@ref{omp_get_default_device}, @ref{omp_set_default_device},

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.13
@end table



@node OMP_DYNAMIC
@section @env{OMP_DYNAMIC} -- Dynamic adjustment of threads
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{dyn-var}
@item @emph{Scope:} global
@item @emph{Description}:
Enable or disable the dynamic adjustment of the number of threads 
within a team.  The value of this environment variable shall be 
@code{TRUE} or @code{FALSE}.  If undefined, dynamic adjustment is
disabled by default.

@item @emph{See also}:
@ref{omp_set_dynamic}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.3
@end table



@node OMP_MAX_ACTIVE_LEVELS
@section @env{OMP_MAX_ACTIVE_LEVELS} -- Set the maximum number of nested parallel regions
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{max-active-levels-var}
@item @emph{Scope:} data environment
@item @emph{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 @env{OMP_NESTED} is defined and set to true, or
if @env{OMP_NUM_THREADS} or @env{OMP_PROC_BIND} are defined and set to
a list with more than one item, the maximum number of nested parallel
regions will be initialized to the largest number supported, otherwise
it will be set to one.

@item @emph{See also}:
@ref{omp_set_max_active_levels}, @ref{OMP_NESTED}, @ref{OMP_PROC_BIND},
@ref{OMP_NUM_THREADS}


@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.9
@end table



@node OMP_MAX_TASK_PRIORITY
@section @env{OMP_MAX_TASK_PRIORITY} -- Set the maximum priority
number that can be set for a task.
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{max-task-priority-var}
@item @emph{Scope:} global
@item @emph{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.

@item @emph{See also}:
@ref{omp_get_max_task_priority}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.14
@end table



@node OMP_NESTED
@section @env{OMP_NESTED} -- Nested parallel regions
@cindex Environment Variable
@cindex Implementation specific setting
@table @asis
@item @emph{ICV:} @var{max-active-levels-var}
@item @emph{Scope:} data environment
@item @emph{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 @code{TRUE} or @code{FALSE}.  If set to @code{TRUE}, the number
of maximum active nested regions supported will by default be set to the
maximum supported, otherwise it will be set to one.  If
@env{OMP_MAX_ACTIVE_LEVELS} is defined, its setting will override this
setting.  If both are undefined, nested parallel regions are enabled if
@env{OMP_NUM_THREADS} or @env{OMP_PROC_BINDS} are defined to a list with
more than one item, otherwise they are disabled by default.

Note that the @code{OMP_NESTED} environment variable was deprecated in
the OpenMP specification 5.2 in favor of @code{OMP_MAX_ACTIVE_LEVELS}.

@item @emph{See also}:
@ref{omp_set_max_active_levels}, @ref{omp_set_nested},
@ref{OMP_MAX_ACTIVE_LEVELS}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.6
@end table



@node OMP_NUM_TEAMS
@section @env{OMP_NUM_TEAMS} -- Specifies the number of teams to use by teams region
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{nteams-var}
@item @emph{Scope:} device
@item @emph{Description}:
Specifies the upper bound for number of teams to use in teams regions
without explicit @code{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.

@item @emph{See also}:
@ref{omp_set_num_teams}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v5.1}, Section 6.23
@end table



@node OMP_NUM_THREADS
@section @env{OMP_NUM_THREADS} -- Specifies the number of threads to use
@cindex Environment Variable
@cindex Implementation specific setting
@table @asis
@item @emph{ICV:} @var{nthreads-var}
@item @emph{Scope:} data environment
@item @emph{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 will automatically enable
nesting by default.  If undefined one thread per CPU is used.

When a list with more than value is specified, it also affects the
@var{max-active-levels-var} ICV as described in @ref{OMP_MAX_ACTIVE_LEVELS}.

@item @emph{See also}:
@ref{omp_set_num_threads}, @ref{OMP_MAX_ACTIVE_LEVELS}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.2
@end table



@node OMP_PROC_BIND
@section @env{OMP_PROC_BIND} -- Whether threads may be moved between CPUs
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{bind-var}
@item @emph{Scope:} data environment
@item @emph{Description}:
Specifies whether threads may be moved between processors.  If set to
@code{TRUE}, OpenMP threads should not be moved; if set to @code{FALSE}
they may be moved.  Alternatively, a comma separated list with the
values @code{PRIMARY}, @code{MASTER}, @code{CLOSE} and @code{SPREAD} can
be used to specify the thread affinity policy for the corresponding nesting
level.  With @code{PRIMARY} and @code{MASTER} the worker threads are in the
same place partition as the primary thread.  With @code{CLOSE} those are
kept close to the primary thread in contiguous place partitions.  And
with @code{SPREAD} a sparse distribution
across the place partitions is used.  Specifying more than one item in the
list will automatically enable nesting by default.

When a list is specified, it also affects the @var{max-active-levels-var} ICV
as described in @ref{OMP_MAX_ACTIVE_LEVELS}.

When undefined, @env{OMP_PROC_BIND} defaults to @code{TRUE} when
@env{OMP_PLACES} or @env{GOMP_CPU_AFFINITY} is set and @code{FALSE} otherwise.

@item @emph{See also}:
@ref{omp_get_proc_bind}, @ref{GOMP_CPU_AFFINITY}, @ref{OMP_PLACES},
@ref{OMP_MAX_ACTIVE_LEVELS}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.4
@end table



@node OMP_PLACES
@section @env{OMP_PLACES} -- Specifies on which CPUs the threads should be placed
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{place-partition-var}
@item @emph{Scope:} implicit tasks
@item @emph{Description}:
The thread placement can be either specified using an abstract name or by an
explicit list of the places.  The abstract names @code{threads}, @code{cores},
@code{sockets}, @code{ll_caches} and @code{numa_domains} can be optionally
followed by a positive number in parentheses, which denotes the how many places
shall be created.  With @code{threads} each place corresponds to a single
hardware thread; @code{cores} to a single core with the corresponding number of
hardware threads; with @code{sockets} the place corresponds to a single
socket; with @code{ll_caches} to a set of cores that shares the last level
cache on the device; and @code{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
@env{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 (@code{!}) directly before a curly
brace or numbers inside the curly braces (excluding intervals) will
exclude those hardware threads.

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

If @env{OMP_PLACES} and @env{GOMP_CPU_AFFINITY} are unset and
@env{OMP_PROC_BIND} is either unset or @code{false}, threads may be moved
between CPUs following no placement policy.

@item @emph{See also}:
@ref{OMP_PROC_BIND}, @ref{GOMP_CPU_AFFINITY}, @ref{omp_get_proc_bind},
@ref{OMP_DISPLAY_ENV}

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.5
@end table



@node OMP_STACKSIZE
@section @env{OMP_STACKSIZE} -- Set default thread stack size
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{stacksize-var}
@item @emph{Scope:} device
@item @emph{Description}:
Set the default thread stack size in kilobytes, unless the number
is suffixed by @code{B}, @code{K}, @code{M} or @code{G}, in which
case the size is, respectively, in bytes, kilobytes, megabytes
or gigabytes.  This is different from @code{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.

@item @emph{See also}:
@ref{GOMP_STACKSIZE}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.7
@end table



@node OMP_SCHEDULE
@section @env{OMP_SCHEDULE} -- How threads are scheduled
@cindex Environment Variable
@cindex Implementation specific setting
@table @asis
@item @emph{ICV:} @var{run-sched-var}
@item @emph{Scope:} data environment
@item @emph{Description}:
Allows to specify @code{schedule type} and @code{chunk size}. 
The value of the variable shall have the form: @code{type[,chunk]} where
@code{type} is one of @code{static}, @code{dynamic}, @code{guided} or @code{auto}
The optional @code{chunk} size shall be a positive integer.  If undefined,
dynamic scheduling and a chunk size of 1 is used.

@item @emph{See also}:
@ref{omp_set_schedule}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Sections 2.7.1.1 and 4.1
@end table



@node OMP_TARGET_OFFLOAD
@section @env{OMP_TARGET_OFFLOAD} -- Controls offloading behaviour
@cindex Environment Variable
@cindex Implementation specific setting
@table @asis
@item @emph{ICV:} @var{target-offload-var}
@item @emph{Scope:} global
@item @emph{Description}:
Specifies the behaviour with regard to offloading code to a device.  This
variable can be set to one of three values - @code{MANDATORY}, @code{DISABLED}
or @code{DEFAULT}.

If set to @code{MANDATORY}, the program will terminate with an error if
the offload device is not present or is not supported.  If set to
@code{DISABLED}, then offloading is disabled and all code will run on the
host. If set to @code{DEFAULT}, the program will try offloading to the
device first, then fall back to running code on the host if it cannot.

If undefined, then the program will behave as if @code{DEFAULT} was set.

@item @emph{Reference}:
@uref{https://www.openmp.org, OpenMP specification v5.0}, Section 6.17
@end table



@node OMP_TEAMS_THREAD_LIMIT
@section @env{OMP_TEAMS_THREAD_LIMIT} -- Set the maximum number of threads imposed by teams
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{teams-thread-limit-var}
@item @emph{Scope:} device
@item @emph{Description}:
Specifies an upper bound for the number of threads to use by each contention
group created by a teams construct without explicit @code{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.

@item @emph{See also}:
@ref{OMP_THREAD_LIMIT}, @ref{omp_set_teams_thread_limit}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v5.1}, Section 6.24
@end table



@node OMP_THREAD_LIMIT
@section @env{OMP_THREAD_LIMIT} -- Set the maximum number of threads
@cindex Environment Variable
@table @asis
@item @emph{ICV:} @var{thread-limit-var}
@item @emph{Scope:} data environment
@item @emph{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.

@item @emph{See also}:
@ref{OMP_NUM_THREADS}, @ref{omp_get_thread_limit}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.10
@end table



@node OMP_WAIT_POLICY
@section @env{OMP_WAIT_POLICY} -- How waiting threads are handled
@cindex Environment Variable
@table @asis
@item @emph{Description}:
Specifies whether waiting threads should be active or passive.  If
the value is @code{PASSIVE}, waiting threads should not consume CPU
power while waiting; while the value is @code{ACTIVE} specifies that
they should.  If undefined, threads wait actively for a short time
before waiting passively.

@item @emph{See also}:
@ref{GOMP_SPINCOUNT}

@item @emph{Reference}: 
@uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.8
@end table



@node GOMP_CPU_AFFINITY
@section @env{GOMP_CPU_AFFINITY} -- Bind threads to specific CPUs
@cindex Environment Variable
@table @asis
@item @emph{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,
@code{GOMP_CPU_AFFINITY="0 3 1-2 4-15:2"} will bind 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 start assigning back from the beginning of
the list.  @code{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., @code{getenv} in C or @code{GET_ENVIRONMENT_VARIABLE} in 
Fortran, may be used to query the setting of the @code{GOMP_CPU_AFFINITY} 
environment variable.  A defined CPU affinity on startup cannot be changed 
or disabled during the runtime of the application.

If both @env{GOMP_CPU_AFFINITY} and @env{OMP_PROC_BIND} are set,
@env{OMP_PROC_BIND} has a higher precedence.  If neither has been set and
@env{OMP_PROC_BIND} is unset, or when @env{OMP_PROC_BIND} is set to
@code{FALSE}, the host system will handle the assignment of threads to CPUs.

@item @emph{See also}:
@ref{OMP_PLACES}, @ref{OMP_PROC_BIND}
@end table



@node GOMP_DEBUG
@section @env{GOMP_DEBUG} -- Enable debugging output
@cindex Environment Variable
@table @asis
@item @emph{Description}:
Enable debugging output.  The variable should be set to @code{0}
(disabled, also the default if not set), or @code{1} (enabled).

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



@node GOMP_STACKSIZE
@section @env{GOMP_STACKSIZE} -- Set default thread stack size
@cindex Environment Variable
@cindex Implementation specific setting
@table @asis
@item @emph{Description}:
Set the default thread stack size in kilobytes.  This is different from
@code{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.

@item @emph{See also}:
@ref{OMP_STACKSIZE}

@item @emph{Reference}: 
@uref{https://gcc.gnu.org/ml/gcc-patches/2006-06/msg00493.html,
GCC Patches Mailinglist}, 
@uref{https://gcc.gnu.org/ml/gcc-patches/2006-06/msg00496.html,
GCC Patches Mailinglist}
@end table



@node GOMP_SPINCOUNT
@section @env{GOMP_SPINCOUNT} -- Set the busy-wait spin count
@cindex Environment Variable
@cindex Implementation specific setting
@table @asis
@item @emph{Description}:
Determines how long a threads waits actively with consuming CPU power
before waiting passively without consuming CPU power.  The value may be
either @code{INFINITE}, @code{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: @code{k} (kilo, thousand), @code{M} (mega,
million), @code{G} (giga, billion), or @code{T} (tera, trillion).
If undefined, 0 is used when @env{OMP_WAIT_POLICY} is @code{PASSIVE},
300,000 is used when @env{OMP_WAIT_POLICY} is undefined and
30 billion is used when @env{OMP_WAIT_POLICY} is @code{ACTIVE}.
If there are more OpenMP threads than available CPUs, 1000 and 100
spins are used for @env{OMP_WAIT_POLICY} being @code{ACTIVE} or
undefined, respectively; unless the @env{GOMP_SPINCOUNT} is lower
or @env{OMP_WAIT_POLICY} is @code{PASSIVE}.

@item @emph{See also}:
@ref{OMP_WAIT_POLICY}
@end table



@node GOMP_RTEMS_THREAD_POOLS
@section @env{GOMP_RTEMS_THREAD_POOLS} -- Set the RTEMS specific thread pools
@cindex Environment Variable
@cindex Implementation specific setting
@table @asis
@item @emph{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
@env{GOMP_RTEMS_THREAD_POOLS} is a list of optional
@code{<thread-pool-count>[$<priority>]@@<scheduler-name>} configurations
separated by @code{:} where:
@itemize @bullet
@item @code{<thread-pool-count>} is the thread pool count for this scheduler
instance.
@item @code{$<priority>} is an optional priority for the worker threads of a
thread pool according to @code{pthread_setschedparam}.  In case a priority
value is omitted, then a worker thread will inherit 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.
@item @code{@@<scheduler-name>} is the scheduler instance name according to the
RTEMS application configuration.
@end itemize
In case no thread pool configuration is specified for a scheduler instance,
then each OpenMP primary thread of this scheduler instance will use its own
dynamically allocated thread pool.  To limit the worker thread count of the
thread pools, each OpenMP primary thread must call @code{omp_set_num_threads}.
@item @emph{Example}:
Lets suppose we have three scheduler instances @code{IO}, @code{WRK0}, and
@code{WRK1} with @env{GOMP_RTEMS_THREAD_POOLS} set to
@code{"1@@WRK0:3$4@@WRK1"}.  Then there are no thread pool restrictions for
scheduler instance @code{IO}.  In the scheduler instance @code{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 @code{WRK1} there are three thread
pools available and their worker threads run at priority four.
@end table



@c ---------------------------------------------------------------------
@c Enabling OpenACC
@c ---------------------------------------------------------------------

@node Enabling OpenACC
@chapter Enabling OpenACC

To activate the OpenACC extensions for C/C++ and Fortran, the compile-time 
flag @option{-fopenacc} must be specified.  This enables the OpenACC directive
@code{#pragma acc} in C/C++ and @code{!$acc} directives in free form,
@code{c$acc}, @code{*$acc} and @code{!$acc} directives in fixed form,
@code{!$} conditional compilation sentinels in free form and @code{c$},
@code{*$} and @code{!$} sentinels in fixed form, for Fortran.  The flag also
arranges for automatic linking of the OpenACC runtime library 
(@ref{OpenACC Runtime Library Routines}).

See @uref{https://gcc.gnu.org/wiki/OpenACC} for more information.

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



@c ---------------------------------------------------------------------
@c OpenACC Runtime Library Routines
@c ---------------------------------------------------------------------

@node OpenACC Runtime Library Routines
@chapter 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
@code{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.
@end menu



@node acc_get_num_devices
@section @code{acc_get_num_devices} -- Get number of devices for given device type
@table @asis
@item @emph{Description}
This function returns a value indicating the number of devices available
for the device type specified in @var{devicetype}. 

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int acc_get_num_devices(acc_device_t devicetype);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{integer function acc_get_num_devices(devicetype)}
@item                  @tab @code{integer(kind=acc_device_kind) devicetype}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.1.
@end table



@node acc_set_device_type
@section @code{acc_set_device_type} -- Set type of device accelerator to use.
@table @asis
@item @emph{Description}
This function indicates to the runtime library which device type, specified
in @var{devicetype}, to use when executing a parallel or kernels region. 

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_set_device_type(acc_device_t devicetype);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_set_device_type(devicetype)}
@item                   @tab @code{integer(kind=acc_device_kind) devicetype}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.2.
@end table



@node acc_get_device_type
@section @code{acc_get_device_type} -- Get type of device accelerator to be used.
@table @asis
@item @emph{Description}
This function returns what device type will be used when executing a
parallel or kernels region.

This function returns @code{acc_device_none} if
@code{acc_get_device_type} is called from
@code{acc_ev_device_init_start}, @code{acc_ev_device_init_end}
callbacks of the OpenACC Profiling Interface (@ref{OpenACC Profiling
Interface}), that is, if the device is currently being initialized.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_device_t acc_get_device_type(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{function acc_get_device_type(void)}
@item                  @tab @code{integer(kind=acc_device_kind) acc_get_device_type}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.3.
@end table



@node acc_set_device_num
@section @code{acc_set_device_num} -- Set device number to use.
@table @asis
@item @emph{Description}
This function will indicate to the runtime which device number,
specified by @var{devicenum}, associated with the specified device
type @var{devicetype}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_set_device_num(int devicenum, acc_device_t devicetype);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_set_device_num(devicenum, devicetype)}
@item                   @tab @code{integer devicenum}
@item                   @tab @code{integer(kind=acc_device_kind) devicetype}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.4.
@end table



@node acc_get_device_num
@section @code{acc_get_device_num} -- Get device number to be used.
@table @asis
@item @emph{Description}
This function returns which device number associated with the specified device
type @var{devicetype}, will be used when executing a parallel or kernels
region.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int acc_get_device_num(acc_device_t devicetype);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{function acc_get_device_num(devicetype)}
@item                   @tab @code{integer(kind=acc_device_kind) devicetype}
@item                   @tab @code{integer acc_get_device_num}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.5.
@end table



@node acc_get_property
@section @code{acc_get_property} -- Get device property.
@cindex acc_get_property
@cindex acc_get_property_string
@table @asis
@item @emph{Description}
These routines return the value of the specified @var{property} for the
device being queried according to @var{devicenum} and @var{devicetype}.
Integer-valued and string-valued properties are returned by
@code{acc_get_property} and @code{acc_get_property_string} respectively.
The Fortran @code{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
@code{acc_device_property} has been renamed to @code{acc_device_property_kind}
for consistency and the return type of the @code{acc_get_property} function is
now a @code{c_size_t} integer instead of a @code{acc_device_property} integer.
The parameter @code{acc_device_property} will continue to be provided,
but might be removed in a future version of GCC.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{size_t acc_get_property(int devicenum, acc_device_t devicetype, acc_device_property_t property);}
@item @emph{Prototype}: @tab @code{const char *acc_get_property_string(int devicenum, acc_device_t devicetype, acc_device_property_t property);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{function acc_get_property(devicenum, devicetype, property)}
@item @emph{Interface}: @tab @code{subroutine acc_get_property_string(devicenum, devicetype, property, string)}
@item                   @tab @code{use ISO_C_Binding, only: c_size_t}
@item                   @tab @code{integer devicenum}
@item                   @tab @code{integer(kind=acc_device_kind) devicetype}
@item                   @tab @code{integer(kind=acc_device_property_kind) property}
@item                   @tab @code{integer(kind=c_size_t) acc_get_property}
@item                   @tab @code{character(*) string}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.6.
@end table



@node acc_async_test
@section @code{acc_async_test} -- Test for completion of a specific asynchronous operation.
@table @asis
@item @emph{Description}
This function tests for completion of the asynchronous operation specified
in @var{arg}. In C/C++, a non-zero value will be returned to indicate
the specified asynchronous operation has completed. While Fortran will return
a @code{true}. If the asynchronous operation has not completed, C/C++ returns
a zero and Fortran returns a @code{false}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int acc_async_test(int arg);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{function acc_async_test(arg)}
@item                   @tab @code{integer(kind=acc_handle_kind) arg}
@item                   @tab @code{logical acc_async_test}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.9.
@end table



@node acc_async_test_all
@section @code{acc_async_test_all} -- Tests for completion of all asynchronous operations.
@table @asis
@item @emph{Description}
This function tests for completion of all asynchronous operations.
In C/C++, a non-zero value will be returned to indicate all asynchronous
operations have completed. While Fortran will return a @code{true}. If
any asynchronous operation has not completed, C/C++ returns a zero and
Fortran returns a @code{false}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int acc_async_test_all(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{function acc_async_test()}
@item                   @tab @code{logical acc_get_device_num}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.10.
@end table



@node acc_wait
@section @code{acc_wait} -- Wait for completion of a specific asynchronous operation.
@table @asis
@item @emph{Description}
This function waits for completion of the asynchronous operation
specified in @var{arg}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_wait(arg);}
@item @emph{Prototype (OpenACC 1.0 compatibility)}: @tab @code{acc_async_wait(arg);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_wait(arg)}
@item                   @tab @code{integer(acc_handle_kind) arg}
@item @emph{Interface (OpenACC 1.0 compatibility)}: @tab @code{subroutine acc_async_wait(arg)}
@item                                               @tab @code{integer(acc_handle_kind) arg}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.11.
@end table



@node acc_wait_all
@section @code{acc_wait_all} -- Waits for completion of all asynchronous operations.
@table @asis
@item @emph{Description}
This function waits for the completion of all asynchronous operations.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_wait_all(void);}
@item @emph{Prototype (OpenACC 1.0 compatibility)}: @tab @code{acc_async_wait_all(void);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_wait_all()}
@item @emph{Interface (OpenACC 1.0 compatibility)}: @tab @code{subroutine acc_async_wait_all()}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.13.
@end table



@node acc_wait_all_async
@section @code{acc_wait_all_async} -- Wait for completion of all asynchronous operations.
@table @asis
@item @emph{Description}
This function enqueues a wait operation on the queue @var{async} for any
and all asynchronous operations that have been previously enqueued on
any queue.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_wait_all_async(int async);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_wait_all_async(async)}
@item                   @tab @code{integer(acc_handle_kind) async}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.14.
@end table



@node acc_wait_async
@section @code{acc_wait_async} -- Wait for completion of asynchronous operations.
@table @asis
@item @emph{Description}
This function enqueues a wait operation on queue @var{async} for any and all
asynchronous operations enqueued on queue @var{arg}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_wait_async(int arg, int async);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_wait_async(arg, async)}
@item                   @tab @code{integer(acc_handle_kind) arg, async}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.12.
@end table



@node acc_init
@section @code{acc_init} -- Initialize runtime for a specific device type.
@table @asis
@item @emph{Description}
This function initializes the runtime for the device type specified in
@var{devicetype}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_init(acc_device_t devicetype);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_init(devicetype)}
@item                   @tab @code{integer(acc_device_kind) devicetype}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.7.
@end table



@node acc_shutdown
@section @code{acc_shutdown} -- Shuts down the runtime for a specific device type.
@table @asis
@item @emph{Description}
This function shuts down the runtime for the device type specified in
@var{devicetype}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_shutdown(acc_device_t devicetype);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_shutdown(devicetype)}
@item                   @tab @code{integer(acc_device_kind) devicetype}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.8.
@end table



@node acc_on_device
@section @code{acc_on_device} -- Whether executing on a particular device
@table @asis
@item @emph{Description}:
This function returns whether the program is executing on a particular
device specified in @var{devicetype}. In C/C++ a non-zero value is
returned to indicate the device is executing on the specified device type.
In Fortran, @code{true} will be returned. If the program is not executing
on the specified device type C/C++ will return a zero, while Fortran will
return @code{false}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_on_device(acc_device_t devicetype);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{function acc_on_device(devicetype)}
@item                   @tab @code{integer(acc_device_kind) devicetype}
@item                   @tab @code{logical acc_on_device}
@end multitable


@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.17.
@end table



@node acc_malloc
@section @code{acc_malloc} -- Allocate device memory.
@table @asis
@item @emph{Description}
This function allocates @var{len} bytes of device memory. It returns
the device address of the allocated memory.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{d_void* acc_malloc(size_t len);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.18.
@end table



@node acc_free
@section @code{acc_free} -- Free device memory.
@table @asis
@item @emph{Description}
Free previously allocated device memory at the device address @code{a}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_free(d_void *a);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.19.
@end table



@node acc_copyin
@section @code{acc_copyin} -- Allocate device memory and copy host memory to it.
@table @asis
@item @emph{Description}
In C/C++, this function allocates @var{len} bytes of device memory
and maps it to the specified host address in @var{a}. The device
address of the newly allocated device memory is returned.

In Fortran, two (2) forms are supported. In the first form, @var{a} specifies
a contiguous array section. The second form @var{a} specifies a
variable or array element and @var{len} specifies the length in bytes.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void *acc_copyin(h_void *a, size_t len);}
@item @emph{Prototype}: @tab @code{void *acc_copyin_async(h_void *a, size_t len, int async);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_copyin(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_copyin(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item @emph{Interface}: @tab @code{subroutine acc_copyin_async(a, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@item @emph{Interface}: @tab @code{subroutine acc_copyin_async(a, len, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.20.
@end table



@node acc_present_or_copyin
@section @code{acc_present_or_copyin} -- If the data is not present on the device, allocate device memory and copy from host memory.
@table @asis
@item @emph{Description}
This function tests if the host data specified by @var{a} and of length
@var{len} is present or not. If it is not present, then device memory
will be 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, @var{a} specifies
a contiguous array section. The second form @var{a} specifies a variable or
array element and @var{len} specifies the length in bytes.

Note that @code{acc_present_or_copyin} and @code{acc_pcopyin} exist for
backward compatibility with OpenACC 2.0; use @ref{acc_copyin} instead.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void *acc_present_or_copyin(h_void *a, size_t len);}
@item @emph{Prototype}: @tab @code{void *acc_pcopyin(h_void *a, size_t len);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_present_or_copyin(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_present_or_copyin(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item @emph{Interface}: @tab @code{subroutine acc_pcopyin(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_pcopyin(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.20.
@end table



@node acc_create
@section @code{acc_create} -- Allocate device memory and map it to host memory.
@table @asis
@item @emph{Description}
This function allocates device memory and maps it to host memory specified
by the host address @var{a} with a length of @var{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, @var{a} specifies
a contiguous array section. The second form @var{a} specifies a variable or
array element and @var{len} specifies the length in bytes.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void *acc_create(h_void *a, size_t len);}
@item @emph{Prototype}: @tab @code{void *acc_create_async(h_void *a, size_t len, int async);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_create(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_create(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item @emph{Interface}: @tab @code{subroutine acc_create_async(a, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@item @emph{Interface}: @tab @code{subroutine acc_create_async(a, len, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.21.
@end table



@node acc_present_or_create
@section @code{acc_present_or_create} -- If the data is not present on the device, allocate device memory and map it to host memory.
@table @asis
@item @emph{Description}
This function tests if the host data specified by @var{a} and of length
@var{len} is present or not. If it is not present, then device memory
will be 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, @var{a} specifies
a contiguous array section. The second form @var{a} specifies a variable or
array element and @var{len} specifies the length in bytes.

Note that @code{acc_present_or_create} and @code{acc_pcreate} exist for
backward compatibility with OpenACC 2.0; use @ref{acc_create} instead.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void *acc_present_or_create(h_void *a, size_t len)}
@item @emph{Prototype}: @tab @code{void *acc_pcreate(h_void *a, size_t len)}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_present_or_create(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_present_or_create(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item @emph{Interface}: @tab @code{subroutine acc_pcreate(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_pcreate(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.21.
@end table



@node acc_copyout
@section @code{acc_copyout} -- Copy device memory to host memory.
@table @asis
@item @emph{Description}
This function copies mapped device memory to host memory which is specified
by host address @var{a} for a length @var{len} bytes in C/C++.

In Fortran, two (2) forms are supported. In the first form, @var{a} specifies
a contiguous array section. The second form @var{a} specifies a variable or
array element and @var{len} specifies the length in bytes.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_copyout(h_void *a, size_t len);}
@item @emph{Prototype}: @tab @code{acc_copyout_async(h_void *a, size_t len, int async);}
@item @emph{Prototype}: @tab @code{acc_copyout_finalize(h_void *a, size_t len);}
@item @emph{Prototype}: @tab @code{acc_copyout_finalize_async(h_void *a, size_t len, int async);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_copyout(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_copyout(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item @emph{Interface}: @tab @code{subroutine acc_copyout_async(a, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@item @emph{Interface}: @tab @code{subroutine acc_copyout_async(a, len, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@item @emph{Interface}: @tab @code{subroutine acc_copyout_finalize(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_copyout_finalize(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item @emph{Interface}: @tab @code{subroutine acc_copyout_finalize_async(a, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@item @emph{Interface}: @tab @code{subroutine acc_copyout_finalize_async(a, len, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.22.
@end table



@node acc_delete
@section @code{acc_delete} -- Free device memory.
@table @asis
@item @emph{Description}
This function frees previously allocated device memory specified by
the device address @var{a} and the length of @var{len} bytes.

In Fortran, two (2) forms are supported. In the first form, @var{a} specifies
a contiguous array section. The second form @var{a} specifies a variable or
array element and @var{len} specifies the length in bytes.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_delete(h_void *a, size_t len);}
@item @emph{Prototype}: @tab @code{acc_delete_async(h_void *a, size_t len, int async);}
@item @emph{Prototype}: @tab @code{acc_delete_finalize(h_void *a, size_t len);}
@item @emph{Prototype}: @tab @code{acc_delete_finalize_async(h_void *a, size_t len, int async);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_delete(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_delete(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item @emph{Interface}: @tab @code{subroutine acc_delete_async(a, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@item @emph{Interface}: @tab @code{subroutine acc_delete_async(a, len, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@item @emph{Interface}: @tab @code{subroutine acc_delete_finalize(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_delete_finalize(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item @emph{Interface}: @tab @code{subroutine acc_delete_async_finalize(a, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@item @emph{Interface}: @tab @code{subroutine acc_delete_async_finalize(a, len, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.23.
@end table



@node acc_update_device
@section @code{acc_update_device} -- Update device memory from mapped host memory.
@table @asis
@item @emph{Description}
This function updates the device copy from the previously mapped host memory.
The host memory is specified with the host address @var{a} and a length of
@var{len} bytes.

In Fortran, two (2) forms are supported. In the first form, @var{a} specifies
a contiguous array section. The second form @var{a} specifies a variable or
array element and @var{len} specifies the length in bytes.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_update_device(h_void *a, size_t len);}
@item @emph{Prototype}: @tab @code{acc_update_device(h_void *a, size_t len, async);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_update_device(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_update_device(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item @emph{Interface}: @tab @code{subroutine acc_update_device_async(a, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@item @emph{Interface}: @tab @code{subroutine acc_update_device_async(a, len, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.24.
@end table



@node acc_update_self
@section @code{acc_update_self} -- Update host memory from mapped device memory.
@table @asis
@item @emph{Description}
This function updates the host copy from the previously mapped device memory.
The host memory is specified with the host address @var{a} and a length of
@var{len} bytes.

In Fortran, two (2) forms are supported. In the first form, @var{a} specifies
a contiguous array section. The second form @var{a} specifies a variable or
array element and @var{len} specifies the length in bytes.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_update_self(h_void *a, size_t len);}
@item @emph{Prototype}: @tab @code{acc_update_self_async(h_void *a, size_t len, int async);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{subroutine acc_update_self(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item @emph{Interface}: @tab @code{subroutine acc_update_self(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item @emph{Interface}: @tab @code{subroutine acc_update_self_async(a, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@item @emph{Interface}: @tab @code{subroutine acc_update_self_async(a, len, async)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item                   @tab @code{integer(acc_handle_kind) :: async}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.25.
@end table



@node acc_map_data
@section @code{acc_map_data} -- Map previously allocated device memory to host memory.
@table @asis
@item @emph{Description}
This function maps previously allocated device and host memory. The device
memory is specified with the device address @var{d}. The host memory is
specified with the host address @var{h} and a length of @var{len}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_map_data(h_void *h, d_void *d, size_t len);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.26.
@end table



@node acc_unmap_data
@section @code{acc_unmap_data} -- Unmap device memory from host memory.
@table @asis
@item @emph{Description}
This function unmaps previously mapped device and host memory. The latter
specified by @var{h}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_unmap_data(h_void *h);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.27.
@end table



@node acc_deviceptr
@section @code{acc_deviceptr} -- Get device pointer associated with specific host address.
@table @asis
@item @emph{Description}
This function returns the device address that has been mapped to the
host address specified by @var{h}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void *acc_deviceptr(h_void *h);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.28.
@end table



@node acc_hostptr
@section @code{acc_hostptr} -- Get host pointer associated with specific device address.
@table @asis
@item @emph{Description}
This function returns the host address that has been mapped to the
device address specified by @var{d}.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void *acc_hostptr(d_void *d);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.29.
@end table



@node acc_is_present
@section @code{acc_is_present} -- Indicate whether host variable / array is present on device.
@table @asis
@item @emph{Description}
This function indicates whether the specified host address in @var{a} and a
length of @var{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, @var{a} specifies
a contiguous array section. The second form @var{a} specifies a variable or
array element and @var{len} specifies the length in bytes. If the host
memory is mapped to device memory, then a @code{true} is returned. Otherwise,
a @code{false} is return to indicate the mapped memory is not present.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int acc_is_present(h_void *a, size_t len);}
@end multitable

@item @emph{Fortran}:
@multitable @columnfractions .20 .80
@item @emph{Interface}: @tab @code{function acc_is_present(a)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{logical acc_is_present}
@item @emph{Interface}: @tab @code{function acc_is_present(a, len)}
@item                   @tab @code{type, dimension(:[,:]...) :: a}
@item                   @tab @code{integer len}
@item                   @tab @code{logical acc_is_present}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.30.
@end table



@node acc_memcpy_to_device
@section @code{acc_memcpy_to_device} -- Copy host memory to device memory.
@table @asis
@item @emph{Description}
This function copies host memory specified by host address of @var{src} to
device memory specified by the device address @var{dest} for a length of
@var{bytes} bytes.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_memcpy_to_device(d_void *dest, h_void *src, size_t bytes);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.31.
@end table



@node acc_memcpy_from_device
@section @code{acc_memcpy_from_device} -- Copy device memory to host memory.
@table @asis
@item @emph{Description}
This function copies host memory specified by host address of @var{src} from
device memory specified by the device address @var{dest} for a length of
@var{bytes} bytes.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_memcpy_from_device(d_void *dest, h_void *src, size_t bytes);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.32.
@end table



@node acc_attach
@section @code{acc_attach} -- Let device pointer point to device-pointer target.
@table @asis
@item @emph{Description}
This function updates a pointer on the device from pointing to a host-pointer
address to pointing to the corresponding device data.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_attach(h_void **ptr);}
@item @emph{Prototype}: @tab @code{acc_attach_async(h_void **ptr, int async);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.34.
@end table



@node acc_detach
@section @code{acc_detach} -- Let device pointer point to host-pointer target.
@table @asis
@item @emph{Description}
This function updates a pointer on the device from pointing to a device-pointer
address to pointing to the corresponding host data.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_detach(h_void **ptr);}
@item @emph{Prototype}: @tab @code{acc_detach_async(h_void **ptr, int async);}
@item @emph{Prototype}: @tab @code{acc_detach_finalize(h_void **ptr);}
@item @emph{Prototype}: @tab @code{acc_detach_finalize_async(h_void **ptr, int async);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
3.2.35.
@end table



@node acc_get_current_cuda_device
@section @code{acc_get_current_cuda_device} -- Get CUDA device handle.
@table @asis
@item @emph{Description}
This function returns the CUDA device handle. This handle is the same
as used by the CUDA Runtime or Driver API's.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void *acc_get_current_cuda_device(void);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
A.2.1.1.
@end table



@node acc_get_current_cuda_context
@section @code{acc_get_current_cuda_context} -- Get CUDA context handle.
@table @asis
@item @emph{Description}
This function returns the CUDA context handle. This handle is the same
as used by the CUDA Runtime or Driver API's.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void *acc_get_current_cuda_context(void);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
A.2.1.2.
@end table



@node acc_get_cuda_stream
@section @code{acc_get_cuda_stream} -- Get CUDA stream handle.
@table @asis
@item @emph{Description}
This function returns the CUDA stream handle for the queue @var{async}.
This handle is the same as used by the CUDA Runtime or Driver API's.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void *acc_get_cuda_stream(int async);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
A.2.1.3.
@end table



@node acc_set_cuda_stream
@section @code{acc_set_cuda_stream} -- Set CUDA stream handle.
@table @asis
@item @emph{Description}
This function associates the stream handle specified by @var{stream} with
the queue @var{async}.

This cannot be used to change the stream handle associated with
@code{acc_async_sync}.

The return value is not specified.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{int acc_set_cuda_stream(int async, void *stream);}
@end multitable

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
A.2.1.4.
@end table



@node acc_prof_register
@section @code{acc_prof_register} -- Register callbacks.
@table @asis
@item @emph{Description}:
This function registers callbacks.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void acc_prof_register (acc_event_t, acc_prof_callback, acc_register_t);}
@end multitable

@item @emph{See also}:
@ref{OpenACC Profiling Interface}

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
5.3.
@end table



@node acc_prof_unregister
@section @code{acc_prof_unregister} -- Unregister callbacks.
@table @asis
@item @emph{Description}:
This function unregisters callbacks.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void acc_prof_unregister (acc_event_t, acc_prof_callback, acc_register_t);}
@end multitable

@item @emph{See also}:
@ref{OpenACC Profiling Interface}

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
5.3.
@end table



@node acc_prof_lookup
@section @code{acc_prof_lookup} -- Obtain inquiry functions.
@table @asis
@item @emph{Description}:
Function to obtain inquiry functions.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{acc_query_fn acc_prof_lookup (const char *);}
@end multitable

@item @emph{See also}:
@ref{OpenACC Profiling Interface}

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
5.3.
@end table



@node acc_register_library
@section @code{acc_register_library} -- Library registration.
@table @asis
@item @emph{Description}:
Function for library registration.

@item @emph{C/C++}:
@multitable @columnfractions .20 .80
@item @emph{Prototype}: @tab @code{void acc_register_library (acc_prof_reg, acc_prof_reg, acc_prof_lookup_func);}
@end multitable

@item @emph{See also}:
@ref{OpenACC Profiling Interface}, @ref{ACC_PROFLIB}

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
5.3.
@end table



@c ---------------------------------------------------------------------
@c OpenACC Environment Variables
@c ---------------------------------------------------------------------

@node OpenACC Environment Variables
@chapter OpenACC Environment Variables

The variables @env{ACC_DEVICE_TYPE} and @env{ACC_DEVICE_NUM}
are defined by section 4 of the OpenACC specification in version 2.0.
The variable @env{ACC_PROFLIB}
is defined by section 4 of the OpenACC specification in version 2.6.
The variable @env{GCC_ACC_NOTIFY} is used for diagnostic purposes.

@menu
* ACC_DEVICE_TYPE::
* ACC_DEVICE_NUM::
* ACC_PROFLIB::
* GCC_ACC_NOTIFY::
@end menu



@node ACC_DEVICE_TYPE
@section @code{ACC_DEVICE_TYPE}
@table @asis
@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
4.1.
@end table



@node ACC_DEVICE_NUM
@section @code{ACC_DEVICE_NUM}
@table @asis
@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
4.2.
@end table



@node ACC_PROFLIB
@section @code{ACC_PROFLIB}
@table @asis
@item @emph{See also}:
@ref{acc_register_library}, @ref{OpenACC Profiling Interface}

@item @emph{Reference}:
@uref{https://www.openacc.org, OpenACC specification v2.6}, section
4.3.
@end table



@node GCC_ACC_NOTIFY
@section @code{GCC_ACC_NOTIFY}
@table @asis
@item @emph{Description}:
Print debug information pertaining to the accelerator.
@end table



@c ---------------------------------------------------------------------
@c CUDA Streams Usage
@c ---------------------------------------------------------------------

@node CUDA Streams Usage
@chapter CUDA Streams Usage

This applies to the @code{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@footnote{See "Stream Management" in "CUDA Driver API",
TRM-06703-001, Version 5.5, for additional information}.

The primary means by that the asynchronous functionality is accessed
is through the use of those OpenACC directives which make use of the
@code{async} and @code{wait} clauses.  When the @code{async} clause is
first used with a directive, it creates a CUDA stream.  If an
@code{async-argument} is used with the @code{async} clause, then the
stream is associated with the specified @code{async-argument}.

Following the creation of an association between a CUDA stream and the
@code{async-argument} of an @code{async} clause, both the @code{wait}
clause and the @code{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 @code{async-argument}, that is, stream, have
completed.

Normally, the management of the streams that are created as a result of
using the @code{async} clause, is done without any intervention by the
caller.  This implies the association between the @code{async-argument}
and the CUDA stream will be maintained for the lifetime of the program.
However, this association can be changed through the use of the library
function @code{acc_set_cuda_stream}.  When the function
@code{acc_set_cuda_stream} is called, the CUDA stream that was
originally associated with the @code{async} clause will be destroyed.
Caution should be taken when changing the association as subsequent
references to the @code{async-argument} refer to a different
CUDA stream.



@c ---------------------------------------------------------------------
@c OpenACC Library Interoperability
@c ---------------------------------------------------------------------

@node OpenACC Library Interoperability
@chapter OpenACC Library Interoperability

@section 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@footnote{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.}.
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.

@section 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 @code{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 @code{cublasCreate()}. The invoking of the
runtime library function @code{cudaGetDevice()} accomplishes this. Once
acquired, the device number is passed along with the device type as
parameters to the OpenACC library function @code{acc_set_device_num()}.

Once the call to @code{acc_set_device_num()} has completed, the OpenACC
library uses the  context that was created during the call to
@code{cublasCreate()}. In other words, both libraries will be sharing the
same context.

@smallexample
    /* 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);

@end smallexample
@center Use Case 1 

@section 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 @code{acc_set_device_num()}.

In the use case presented here, the function @code{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., @code{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 will be discussed in the next section.

Once the call to @code{acc_set_device_num()} has completed, other OpenACC
functions can be called as seen with multiple calls being made to
@code{acc_copyin()}. In addition, calls can be made to functions in the
CUBLAS library. In the use case a call to @code{cublasCreate()} is made
subsequent to the calls to @code{acc_copyin()}.
As seen in the previous use case, a call to @code{cublasCreate()}
initializes the CUBLAS library and allocates the hardware resources on the
host and the device.  However, since the device has already been allocated,
@code{cublasCreate()} will only initialize the CUBLAS library and allocate
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.

@smallexample
    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));

@end smallexample
@center Use Case 2

@section 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:
@env{ACC_DEVICE_TYPE} and @env{ACC_DEVICE_NUM}, respectively. These two
environment variables can be used as an alternative to calling
@code{acc_set_device_num()}. As seen in the second use case, the device
type and device number were specified using @code{acc_set_device_num()}.
If however, the aforementioned environment variables were set, then the
call to @code{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 @code{cudaCreate()}. If @code{cudaCreate()}
is called prior to a call to an OpenACC function, then you must call
@code{acc_set_device_num()}@footnote{More complete information
about @env{ACC_DEVICE_TYPE} and @env{ACC_DEVICE_NUM} can be found in
sections 4.1 and 4.2 of the @uref{https://www.openacc.org, OpenACC}
Application Programming Interface”, Version 2.6.}



@c ---------------------------------------------------------------------
@c OpenACC Profiling Interface
@c ---------------------------------------------------------------------

@node OpenACC Profiling Interface
@chapter OpenACC Profiling Interface

@section 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
@emph{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 @emph{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 will be impacted to some degree once the
Profiling Interface has gotten enabled: for example, because of the
@emph{runtime} (libgomp) calling into a third-party @emph{library} for
every event that has been registered.

We're not yet accounting for the fact that @cite{OpenACC events may
occur during event processing}.
We just handle one case specially, as required by CUDA 9.0
@command{nvprof}, that @code{acc_get_device_type}
(@ref{acc_get_device_type})) may be called from
@code{acc_ev_device_init_start}, @code{acc_ev_device_init_end}
callbacks.

We're not yet implementing initialization via a
@code{acc_register_library} function that is either statically linked
in, or dynamically via @env{LD_PRELOAD}.
Initialization via @code{acc_register_library} functions dynamically
loaded via the @env{ACC_PROFLIB} environment variable does work, as
does directly calling @code{acc_prof_register},
@code{acc_prof_unregister}, @code{acc_prof_lookup}.

As currently there are no inquiry functions defined, calls to
@code{acc_prof_lookup} will always return @code{NULL}.

There aren't separate @emph{start}, @emph{stop} events defined for the
event types @code{acc_ev_create}, @code{acc_ev_delete},
@code{acc_ev_alloc}, @code{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:

@table @asis

@item @code{acc_prof_info.event_type}
It's not clear if for @emph{nested} event callbacks (for example,
@code{acc_ev_enqueue_launch_start} as part of a parent compute
construct), this should be set for the nested event
(@code{acc_ev_enqueue_launch_start}), or if the value of the parent
construct should remain (@code{acc_ev_compute_construct_start}).  In
this implementation, the value will generally correspond to the
innermost nested event type.

@item @code{acc_prof_info.device_type}
@itemize

@item
For @code{acc_ev_compute_construct_start}, and in presence of an
@code{if} clause with @emph{false} argument, this will still refer to
the offloading device type.
It's not clear if that's the expected behavior.

@item
Complementary to the item before, for
@code{acc_ev_compute_construct_end}, this is set to
@code{acc_device_host} in presence of an @code{if} clause with
@emph{false} argument.
It's not clear if that's the expected behavior.

@end itemize

@item @code{acc_prof_info.thread_id}
Always @code{-1}; not yet implemented.

@item @code{acc_prof_info.async}
@itemize

@item
Not yet implemented correctly for
@code{acc_ev_compute_construct_start}.

@item
In a compute construct, for host-fallback
execution/@code{acc_device_host} it will always be
@code{acc_async_sync}.
It's not clear if that's the expected behavior.

@item
For @code{acc_ev_device_init_start} and @code{acc_ev_device_init_end},
it will always be @code{acc_async_sync}.
It's not clear if that's the expected behavior.

@end itemize

@item @code{acc_prof_info.async_queue}
There is no @cite{limited number of asynchronous queues} in libgomp.
This will always have the same value as @code{acc_prof_info.async}.

@item @code{acc_prof_info.src_file}
Always @code{NULL}; not yet implemented.

@item @code{acc_prof_info.func_name}
Always @code{NULL}; not yet implemented.

@item @code{acc_prof_info.line_no}
Always @code{-1}; not yet implemented.

@item @code{acc_prof_info.end_line_no}
Always @code{-1}; not yet implemented.

@item @code{acc_prof_info.func_line_no}
Always @code{-1}; not yet implemented.

@item @code{acc_prof_info.func_end_line_no}
Always @code{-1}; not yet implemented.

@item @code{acc_event_info.event_type}, @code{acc_event_info.*.event_type}
Relating to @code{acc_prof_info.event_type} discussed above, in this
implementation, this will always be the same value as
@code{acc_prof_info.event_type}.

@item @code{acc_event_info.*.parent_construct}
@itemize

@item
Will be @code{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
@code{acc_construct_runtime_api}, respectively.

@item
Will be @code{acc_construct_enter_data} or
@code{acc_construct_exit_data} when processing variable mappings
specified in OpenACC @emph{declare} directives; should be
@code{acc_construct_declare}.

@item
For implicit @code{acc_ev_device_init_start},
@code{acc_ev_device_init_end}, and explicit as well as implicit
@code{acc_ev_alloc}, @code{acc_ev_free},
@code{acc_ev_enqueue_upload_start}, @code{acc_ev_enqueue_upload_end},
@code{acc_ev_enqueue_download_start}, and
@code{acc_ev_enqueue_download_end}, will be
@code{acc_construct_parallel}; should reflect the real parent
construct.

@end itemize

@item @code{acc_event_info.*.implicit}
For @code{acc_ev_alloc}, @code{acc_ev_free},
@code{acc_ev_enqueue_upload_start}, @code{acc_ev_enqueue_upload_end},
@code{acc_ev_enqueue_download_start}, and
@code{acc_ev_enqueue_download_end}, this currently will be @code{1}
also for explicit usage.

@item @code{acc_event_info.data_event.var_name}
Always @code{NULL}; not yet implemented.

@item @code{acc_event_info.data_event.host_ptr}
For @code{acc_ev_alloc}, and @code{acc_ev_free}, this is always
@code{NULL}.

@item @code{typedef union acc_api_info}
@dots{} as printed in @cite{5.2.3. Third Argument: API-Specific
Information}.  This should obviously be @code{typedef @emph{struct}
acc_api_info}.

@item @code{acc_api_info.device_api}
Possibly not yet implemented correctly for
@code{acc_ev_compute_construct_start},
@code{acc_ev_device_init_start}, @code{acc_ev_device_init_end}:
will always be @code{acc_device_api_none} for these event types.
For @code{acc_ev_enter_data_start}, it will be
@code{acc_device_api_none} in some cases.

@item @code{acc_api_info.device_type}
Always the same as @code{acc_prof_info.device_type}.

@item @code{acc_api_info.vendor}
Always @code{-1}; not yet implemented.

@item @code{acc_api_info.device_handle}
Always @code{NULL}; not yet implemented.

@item @code{acc_api_info.context_handle}
Always @code{NULL}; not yet implemented.

@item @code{acc_api_info.async_handle}
Always @code{NULL}; not yet implemented.

@end table

Remarks about certain event types:

@table @asis

@item @code{acc_ev_device_init_start}, @code{acc_ev_device_init_end}
@itemize

@item
@c See 'DEVICE_INIT_INSIDE_COMPUTE_CONSTRUCT' in
@c 'libgomp.oacc-c-c++-common/acc_prof-kernels-1.c',
@c 'libgomp.oacc-c-c++-common/acc_prof-parallel-1.c'.
When a compute construct triggers implicit
@code{acc_ev_device_init_start} and @code{acc_ev_device_init_end}
events, they currently aren't @emph{nested within} the corresponding
@code{acc_ev_compute_construct_start} and
@code{acc_ev_compute_construct_end}, but they're currently observed
@emph{before} @code{acc_ev_compute_construct_start}.
It's not clear what to do: the standard asks us provide a lot of
details to the @code{acc_ev_compute_construct_start} callback, without
(implicitly) initializing a device before?

@item
Callbacks for these event types will not be invoked for calls to the
@code{acc_set_device_type} and @code{acc_set_device_num} functions.
It's not clear if they should be.

@end itemize

@item @code{acc_ev_enter_data_start}, @code{acc_ev_enter_data_end}, @code{acc_ev_exit_data_start}, @code{acc_ev_exit_data_end}
@itemize

@item
Callbacks for these event types will also be invoked for OpenACC
@emph{host_data} constructs.
It's not clear if they should be.

@item
Callbacks for these event types will also be invoked when processing
variable mappings specified in OpenACC @emph{declare} directives.
It's not clear if they should be.

@end itemize

@end table

Callbacks for the following event types will be invoked, but dispatch
and information provided therein has not yet been thoroughly reviewed:

@itemize
@item @code{acc_ev_alloc}
@item @code{acc_ev_free}
@item @code{acc_ev_update_start}, @code{acc_ev_update_end}
@item @code{acc_ev_enqueue_upload_start}, @code{acc_ev_enqueue_upload_end}
@item @code{acc_ev_enqueue_download_start}, @code{acc_ev_enqueue_download_end}
@end itemize

During device initialization, and finalization, respectively,
callbacks for the following event types will not yet be invoked:

@itemize
@item @code{acc_ev_alloc}
@item @code{acc_ev_free}
@end itemize

Callbacks for the following event types have not yet been implemented,
so currently won't be invoked:

@itemize
@item @code{acc_ev_device_shutdown_start}, @code{acc_ev_device_shutdown_end}
@item @code{acc_ev_runtime_shutdown}
@item @code{acc_ev_create}, @code{acc_ev_delete}
@item @code{acc_ev_wait_start}, @code{acc_ev_wait_end}
@end itemize

For the following runtime library functions, not all expected
callbacks will be invoked (mostly concerning implicit device
initialization):

@itemize
@item @code{acc_get_num_devices}
@item @code{acc_set_device_type}
@item @code{acc_get_device_type}
@item @code{acc_set_device_num}
@item @code{acc_get_device_num}
@item @code{acc_init}
@item @code{acc_shutdown}
@end itemize

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):

@itemize
@item @code{acc_malloc}
@item @code{acc_free}
@item @code{acc_copyin}, @code{acc_present_or_copyin}, @code{acc_copyin_async}
@item @code{acc_create}, @code{acc_present_or_create}, @code{acc_create_async}
@item @code{acc_copyout}, @code{acc_copyout_async}, @code{acc_copyout_finalize}, @code{acc_copyout_finalize_async}
@item @code{acc_delete}, @code{acc_delete_async}, @code{acc_delete_finalize}, @code{acc_delete_finalize_async}
@item @code{acc_update_device}, @code{acc_update_device_async}
@item @code{acc_update_self}, @code{acc_update_self_async}
@item @code{acc_map_data}, @code{acc_unmap_data}
@item @code{acc_memcpy_to_device}, @code{acc_memcpy_to_device_async}
@item @code{acc_memcpy_from_device}, @code{acc_memcpy_from_device_async}
@end itemize

@c ---------------------------------------------------------------------
@c OpenMP-Implementation Specifics
@c ---------------------------------------------------------------------

@node OpenMP-Implementation Specifics
@chapter OpenMP-Implementation Specifics

@menu
* Implementation-defined ICV Initialization::
* OpenMP Context Selectors::
* Memory allocation::
@end menu

@node Implementation-defined ICV Initialization
@section Implementation-defined ICV Initialization
@cindex Implementation specific setting

@multitable @columnfractions .30 .70
@item @var{affinity-format-var} @tab See @ref{OMP_AFFINITY_FORMAT}.
@item @var{def-allocator-var} @tab See @ref{OMP_ALLOCATOR}.
@item @var{max-active-levels-var} @tab See @ref{OMP_MAX_ACTIVE_LEVELS}.
@item @var{dyn-var} @tab See @ref{OMP_DYNAMIC}.
@item @var{nthreads-var} @tab See @code{OMP_NUM_THREADS}.
@item @var{num-devices-var} @tab Number of non-host devices found
by GCC's run-time library
@item @var{num-procs-var} @tab 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
@var{nthreads-var} ICV.
@item @var{place-partition-var} @tab See @ref{OMP_PLACES}.
@item @var{run-sched-var} @tab See @ref{OMP_SCHEDULE}.
@item @var{stacksize-var} @tab See @ref{OMP_STACKSIZE}.
@item @var{thread-limit-var} @tab See @ref{OMP_TEAMS_THREAD_LIMIT}
@item @var{wait-policy-var} @tab See @ref{OMP_WAIT_POLICY} and
@ref{GOMP_SPINCOUNT}
@end multitable

@node OpenMP Context Selectors
@section OpenMP Context Selectors

@code{vendor} is always @code{gnu}. References are to the GCC manual.

@multitable @columnfractions .60 .10 .25
@headitem @code{arch} @tab @code{kind} @tab @code{isa}
@item @code{x86}, @code{x86_64}, @code{i386}, @code{i486},
      @code{i586}, @code{i686}, @code{ia32}
      @tab @code{host}
      @tab See @code{-m...} flags in ``x86 Options'' (without @code{-m})
@item @code{amdgcn}, @code{gcn}
      @tab @code{gpu}
      @tab See @code{-march=} in ``AMD GCN Options''@footnote{Additionally,
      @code{gfx803} is supported as an alias for @code{fiji}.}
@item @code{nvptx}
      @tab @code{gpu}
      @tab See @code{-march=} in ``Nvidia PTX Options''
@end multitable

@node Memory allocation
@section Memory allocation

For the available predefined allocators and, as applicable, their associated
predefined memory spaces and for the available traits and their default values,
see @ref{OMP_ALLOCATOR}.  Predefined allocators without an associated memory
space use the @code{omp_default_mem_space} memory space.

For the memory spaces, the following applies:
@itemize
@item @code{omp_default_mem_space} is supported
@item @code{omp_const_mem_space} maps to @code{omp_default_mem_space}
@item @code{omp_low_lat_mem_space} maps to @code{omp_default_mem_space}
@item @code{omp_large_cap_mem_space} maps to @code{omp_default_mem_space},
      unless the memkind library is available
@item @code{omp_high_bw_mem_space} maps to @code{omp_default_mem_space},
      unless the memkind library is available
@end itemize

On Linux systems, where the @uref{https://github.com/memkind/memkind, memkind
library} (@code{libmemkind.so.0}) is available at runtime, it is used when
creating memory allocators requesting

@itemize
@item the memory space @code{omp_high_bw_mem_space}
@item the memory space @code{omp_large_cap_mem_space}
@item the @code{partition} trait @code{interleaved}; note that for
      @code{omp_large_cap_mem_space} the allocation will not be interleaved
@end itemize

On Linux systems, where the @uref{https://github.com/numactl/numactl, numa
library} (@code{libnuma.so.1}) is available at runtime, it used when creating
memory allocators requesting

@itemize
@item the @code{partition} trait @code{nearest}, except when both the
libmemkind library is available and the memory space is either
@code{omp_large_cap_mem_space} or @code{omp_high_bw_mem_space}
@end itemize

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 @code{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 @code{partition} trait @code{environment} (the default) will be effectively
a @code{nearest} allocation.

Additional notes regarding the traits:
@itemize
@item The @code{pinned} trait is unsupported.
@item The default for the @code{pool_size} trait is no pool and for every
      (re)allocation the associated library routine is called, which might
      internally use a memory pool.
@item For the @code{partition} trait, the partition part size will be the same
      as the requested size (i.e. @code{interleaved} or @code{blocked} has no
      effect), except for @code{interleaved} when the memkind library is
      available.  Furthermore, for @code{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.
@item The @code{access} trait has no effect such that memory is always
      accessible by all threads.
@item The @code{sync_hint} trait has no effect.
@end itemize

@c ---------------------------------------------------------------------
@c Offload-Target Specifics
@c ---------------------------------------------------------------------

@node Offload-Target Specifics
@chapter Offload-Target Specifics

The following sections present notes on the offload-target specifics

@menu
* AMD Radeon::
* nvptx::
@end menu

@node AMD Radeon
@section AMD Radeon (GCN)

On the hardware side, there is the hierarchy (fine to coarse):
@itemize
@item work item (thread)
@item wavefront
@item work group
@item compute unit (CU)
@end itemize

All OpenMP and OpenACC levels are used, i.e.
@itemize
@item OpenMP's simd and OpenACC's vector map to work items (thread)
@item OpenMP's threads (``parallel'') and OpenACC's workers map
      to wavefronts
@item OpenMP's teams and OpenACC's gang use a threadpool with the
      size of the number of teams or gangs, respectively.
@end itemize

The used sizes are
@itemize
@item Number of teams is the specified @code{num_teams} (OpenMP) or
      @code{num_gangs} (OpenACC) or otherwise the number of CU. It is limited
      by two times the number of CU.
@item Number of wavefronts is 4 for gfx900 and 16 otherwise;
      @code{num_threads} (OpenMP) and @code{num_workers} (OpenACC)
      overrides this if smaller.
@item The wavefront has 102 scalars and 64 vectors
@item Number of workitems is always 64
@item The hardware permits maximally 40 workgroups/CU and
      16 wavefronts/workgroup up to a limit of 40 wavefronts in total per CU.
@item 80 scalars registers and 24 vector registers in non-kernel functions
      (the chosen procedure-calling API).
@item For the kernel itself: as many as register pressure demands (number of
      teams and number of threads, scaled down if registers are exhausted)
@end itemize

The implementation remark:
@itemize
@item I/O within OpenMP target regions and OpenACC parallel/kernels is supported
      using the C library @code{printf} functions and the Fortran
      @code{print}/@code{write} statements.
@item Reverse offload regions (i.e. @code{target} regions with
      @code{device(ancestor:1)}) are processed serially per @code{target} region
      such that the next reverse offload region is only executed after the previous
      one returned.
@item OpenMP code that has a @code{requires} directive with
      @code{unified_shared_memory} will remove any GCN device from the list of
      available devices (``host fallback'').
@item The available stack size can be changed using the @code{GCN_STACK_SIZE}
      environment variable; the default is 32 kiB per thread.
@end itemize



@node nvptx
@section nvptx

On the hardware side, there is the hierarchy (fine to coarse):
@itemize
@item thread
@item warp
@item thread block
@item streaming multiprocessor
@end itemize

All OpenMP and OpenACC levels are used, i.e.
@itemize
@item OpenMP's simd and OpenACC's vector map to threads
@item OpenMP's threads (``parallel'') and OpenACC's workers map to warps
@item OpenMP's teams and OpenACC's gang use a threadpool with the
      size of the number of teams or gangs, respectively.
@end itemize

The used sizes are
@itemize
@item The @code{warp_size} is always 32
@item CUDA kernel launched: @code{dim=@{#teams,1,1@}, blocks=@{#threads,warp_size,1@}}.
@item The number of teams is limited by the number of blocks the device can
      host simultaneously.
@end itemize

Additional information can be obtained by setting the environment variable to
@code{GOMP_DEBUG=1} (very verbose; grep for @code{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 @code{CUDA_CACHE_@{DISABLE,MAXSIZE,PATH@}}.

Note: While PTX ISA is generic, the @code{-mptx=} and @code{-march=} commandline
options still affect the used PTX ISA code and, thus, the requirements on
CUDA version and hardware.

The implementation remark:
@itemize
@item I/O within OpenMP target regions and OpenACC parallel/kernels is supported
      using the C library @code{printf} functions. Note that the Fortran
      @code{print}/@code{write} statements are not supported, yet.
@item Compilation OpenMP code that contains @code{requires reverse_offload}
      requires at least @code{-march=sm_35}, compiling for @code{-march=sm_30}
      is not supported.
@item For code containing reverse offload (i.e. @code{target} regions with
      @code{device(ancestor:1)}), there is a slight performance penalty
      for @emph{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.
@item OpenMP code that has a @code{requires} directive with
      @code{unified_shared_memory} will remove any nvptx device from the
      list of available devices (``host fallback'').
@item The default per-warp stack size is 128 kiB; see also @code{-msoft-stack}
      in the GCC manual.
@end itemize


@c ---------------------------------------------------------------------
@c The libgomp ABI
@c ---------------------------------------------------------------------

@node The libgomp ABI
@chapter 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::
@end menu


@node Implementing MASTER construct
@section Implementing MASTER construct

@smallexample
if (omp_get_thread_num () == 0)
  block
@end smallexample

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...



@node Implementing CRITICAL construct
@section Implementing CRITICAL construct

Without a specified name,

@smallexample
  void GOMP_critical_start (void);
  void GOMP_critical_end (void);
@end smallexample

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

@smallexample
  omp_lock_t gomp_critical_user_<name> __attribute__((common))
@end smallexample

Ideally the ABI would specify that all zero is a valid unlocked
state, and so we wouldn't need to initialize this at
startup.



@node Implementing ATOMIC construct
@section Implementing ATOMIC construct

The target should implement the @code{__sync} builtins.

Failing that we could add

@smallexample
  void GOMP_atomic_enter (void)
  void GOMP_atomic_exit (void)
@end smallexample

which reuses the regular lock code, but with yet another lock
object private to the library.



@node Implementing FLUSH construct
@section Implementing FLUSH construct

Expands to the @code{__sync_synchronize} builtin.



@node Implementing BARRIER construct
@section Implementing BARRIER construct

@smallexample
  void GOMP_barrier (void)
@end smallexample


@node Implementing THREADPRIVATE construct
@section Implementing THREADPRIVATE construct

In _most_ cases we can map this directly to @code{__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.



@node Implementing PRIVATE clause
@section 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.



@node Implementing FIRSTPRIVATE LASTPRIVATE COPYIN and COPYPRIVATE clauses
@section 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:

@smallexample
#pragma omp for firstprivate(x) lastprivate(y)
for (int i = 0; i < n; ++i)
  body;
@end smallexample

which becomes

@smallexample
@{
  int x = x, y;

  // for stuff

  if (i == n)
    y = y;
@}
@end smallexample

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.



@node Implementing REDUCTION clause
@section 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 @var{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.


@node Implementing PARALLEL construct
@section Implementing PARALLEL construct

@smallexample
  #pragma omp parallel
  @{
    body;
  @}
@end smallexample

becomes

@smallexample
  void subfunction (void *data)
  @{
    use data;
    body;
  @}

  setup data;
  GOMP_parallel_start (subfunction, &data, num_threads);
  subfunction (&data);
  GOMP_parallel_end ();
@end smallexample

@smallexample
  void GOMP_parallel_start (void (*fn)(void *), void *data, unsigned num_threads)
@end smallexample

The @var{FN} argument is the subfunction to be run in parallel.

The @var{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 @var{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.

@smallexample
  void GOMP_parallel_end (void)
@end smallexample

Tears down the team and returns us to the previous @code{omp_in_parallel()} state.



@node Implementing FOR construct
@section Implementing FOR construct

@smallexample
  #pragma omp parallel for
  for (i = lb; i <= ub; i++)
    body;
@end smallexample

becomes

@smallexample
  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 ();
@end smallexample

@smallexample
  #pragma omp for schedule(runtime)
  for (i = 0; i < n; i++)
    body;
@end smallexample

becomes

@smallexample
  @{
    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 ();
  @}
@end smallexample

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...



@node Implementing ORDERED construct
@section Implementing ORDERED construct

@smallexample
  void GOMP_ordered_start (void)
  void GOMP_ordered_end (void)
@end smallexample



@node Implementing SECTIONS construct
@section Implementing SECTIONS construct

A block as 

@smallexample
  #pragma omp sections
  @{
    #pragma omp section
    stmt1;
    #pragma omp section
    stmt2;
    #pragma omp section
    stmt3;
  @}
@end smallexample

becomes

@smallexample
  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 ();
@end smallexample


@node Implementing SINGLE construct
@section Implementing SINGLE construct

A block like 

@smallexample
  #pragma omp single
  @{
    body;
  @}
@end smallexample

becomes

@smallexample
  if (GOMP_single_start ())
    body;
  GOMP_barrier ();
@end smallexample

while 

@smallexample
  #pragma omp single copyprivate(x)
    body;
@end smallexample

becomes

@smallexample
  datap = GOMP_single_copy_start ();
  if (datap == NULL)
    @{
      body;
      data.x = x;
      GOMP_single_copy_end (&data);
    @}
  else
    x = datap->x;
  GOMP_barrier ();
@end smallexample



@node Implementing OpenACC's PARALLEL construct
@section Implementing OpenACC's PARALLEL construct

@smallexample
  void GOACC_parallel ()
@end smallexample



@c ---------------------------------------------------------------------
@c Reporting Bugs
@c ---------------------------------------------------------------------

@node Reporting Bugs
@chapter Reporting Bugs

Bugs in the GNU Offloading and Multi Processing Runtime Library should
be reported via @uref{https://gcc.gnu.org/bugzilla/, Bugzilla}.  Please add
"openacc", or "openmp", or both to the keywords field in the bug
report, as appropriate.



@c ---------------------------------------------------------------------
@c GNU General Public License
@c ---------------------------------------------------------------------

@include gpl_v3.texi



@c ---------------------------------------------------------------------
@c GNU Free Documentation License
@c ---------------------------------------------------------------------

@include fdl.texi



@c ---------------------------------------------------------------------
@c Funding Free Software
@c ---------------------------------------------------------------------

@include funding.texi

@c ---------------------------------------------------------------------
@c Index
@c ---------------------------------------------------------------------

@node Library Index
@unnumbered Library Index

@printindex cp

@bye