Supported Environment Variables

IFORTCFG

INTEL_LICENSE_FILE

PATH

TMP

TMPDIR

TEMP

CPATH (Linux* and OS X*)

GCCROOT (Linux* and OS X*)

GXX_INCLUDE (Linux* and OS X*)

GXX_ROOT (Linux* and OS X*)

LIBRARY_PATH (Linux* and OS X*)

LD_LIBRARY_PATH (Linux*)

DYLD_LIBRARY_PATH (OS X*)

INCLUDE (Windows*)

LIB (Windows*)

MIC_LIBRARY_PATH

decfort_dump_flag

F_UFMTENDIAN

This variable specifies the numbers of the units to be used for little-endian-to-big-endian conversion purposes. See F_UFMTENDIAN Method Environment Variable.

FOR_COARRAY_CONFIG_FILE (Windows* and Linux*)

This variable specifies the coarray configuration file and path to be used at execution time. It overrides, at runtime, the value specified by the [Q]coarray-config-file=value qualifier.

FOR_COARRAY_NUM_IMAGES (Windows* and Linux*)

This variable specifies the number of images created for coarrays. It overrides, at runtime, the value specified by the [Q]coarray-num-images=value qualifier.

FOR_FMT_TERMINATOR

This variable specifies the numbers of the units to have a specific record terminator. See Record Types.

FORT_FMT_NO_WRAP_MARGIN

Disables column wrapping in Fortran list-directed output when the record being written gets longer than 80 characters.

Default: NO (Wrap margin)

FOR_ACCEPT

The ACCEPT statement does not include an explicit logical unit number. Instead, it uses an implicit internal logical unit number and the FOR_ACCEPT environment variable. If FOR_ACCEPT is not defined, the code ACCEPT f,iolist reads from CONIN$ (standard input). If FOR_ACCEPT is defined (as a file name optionally containing a path), the specified file would be read.

FOR_DEBUGGER_IS_PRESENT

This variable tells the Fortran run-time library that your program is executing under a debugger. If set to True, it generates debug exceptions whenever severe or continuous errors are detected.

Under normal conditions, this variable can be extracted from the operating system for Windows*.

On Linux* and OS X*, this variable must be set for debug exceptions. Setting this variable to True when a program is not executing under a debugger will cause unpredictable behavior.

FOR_DEFAULT_PRINT_DEVICE (Windows* only)

This variable lets you specify the print device other than the default print device PRN (LPT1) for files closed (CLOSE statement) with the DISPOSE='PRINT' specifier. To specify a different print device for the file associated with the CLOSE statement DISPOSE='PRINT' specifier, set FOR_DEFAULT_PRINT_DEVICE to any legal DOS print device before executing the program.

FOR_DIAGNOSTIC_LOG_FILE

If this variable is set to the name of a file, diagnostic output is written to the specified file.

The Fortran run-time system attempts to open that file (append output) and write the error information (ASCII text) to the file.

The setting of FOR_DIAGNOSTIC_LOG_FILE is independent of FOR_DISABLE_DIAGNOSTIC_DISPLAY, so you can disable the screen display of information but still capture the error information in a file. The text string you assign for the file name is used literally, so you must specify the full name. If the file open fails, no error is reported and the run-time system continues diagnostic processing.

See also Locating Run-Time Errors and Using Traceback Information Overview.

FOR_DISABLE_DIAGNOSTIC_DISPLAY

This variable disables the display of all error information. This variable is helpful if you just want to test the error status of your program and do not want the Fortran run-time system to display any information about an abnormal program termination.

See also Using Traceback Information Overview.

FOR_FORCE_STACK_TRACE

If defined, forces a traceback to follow any run-time diagnostic message.

If FOR_DISABLE_STACK_TRACE is also set, FOR_FORCE_STACK_TRACE takes precedence over FOR_DISABLE_STACK_TRACE .

FOR_DISABLE_STACK_TRACE

Specifies that the call stack trace information that typically follows the displayed severe error message text should be disabled.

The Fortran run-time error message is displayed regardless of whether FOR_DISABLE_STACK_TRACE is set to true. If the program is executing under a debugger, the automatic output of the stack trace information by the Fortran library will be disabled to reduce noise. Use the debugger's stack trace facility to view the stack trace.

See also Locating Run-Time Errors and Using Traceback Information Overview.

FOR_IGNORE_EXCEPTIONS

Specifies that default run-time exception handling, for example, to allow just-in-time debugging, should be disabled. The run-time system exception handler returns EXCEPTION_CONTINUE_SEARCH to the operating system, which looks for other handlers to service the exception.

FOR_NOERROR_DIALOGS

Specifies that the display of dialog boxes should be disabled when certain exceptions or errors occur. This is useful when running many test programs in batch mode to prevent a failure from stopping execution of the entire test stream.

FOR_PRINT

Neither the PRINT statement nor a WRITE statement with an asterisk (*) in place of a unit number includes an explicit logical unit number. Instead, both use an implicit internal logical unit number and the FOR_PRINT environment variable. If FOR_PRINT is not defined, the code PRINT f,iolist or WRITE (*,f) iolist writes to CONOUT$ (standard output). If FOR_PRINT is defined (as a file name optionally containing a path), the specified file would be written to.

FOR_READ

A READ statement that uses an asterisk (*) in place of a unit number does not include an explicit logical unit number. Instead, it uses an implicit internal logical unit number and the FOR_READ environment variable. If FOR_READ is not defined, the code READ (*,f) iolist or READ f,iolist reads from CONIN$ (standard input). If FOR_READ is defined (as a file name optionally containing a path), the specified file would be read.

FOR_TYPE

The TYPE statement does not include an explicit logical unit number. Instead, it uses an implicit internal logical unit number and the FOR_TYPE environment variable. If FOR_TYPE is not defined, the code TYPE f,iolist writes to CONOUT$ (standard output). If FOR_TYPE is defined (as a file name optionally containing a path), the specified file would be written to.

FORT_BLOCKSIZE

Specifies the default BLOCKSIZE value to be used when BLOCKSIZE= is omitted on the OPEN statement. Valid values are 0 to 2147467264. Sizes are rounded up to the next 512-byte boundary.

FORT_BUFFERCOUNT

Specifies the default BUFFERCOUNTvalue to be used when BUFFERCOUNT= is omitted on the OPEN statement. Valid values are 0 to 127. If 0 is specified, the default value of 1 will be used.

FORT_BUFFERED

Specifies that buffered I/O should be used at run time for output of all Fortran I/O units, except those with output to the terminal. This provides a run-time mechanism to support the buffered_io keyword for the assume compiler option.

FORT_CONVERTn

Specifies the data format for an unformatted file associated with a particular unit number (n), as described in Methods of Specifying the Data Format.

FORT_CONVERT.ext and FORT_CONVERT_ext

Specifies the data format for unformatted files with a particular file extension suffix (.ext), as described in Methods of Specifying the Data Format.

FORT_FMT_RECL

Specifies the default record length (normally 132 bytes) for formatted files.

FORT_UFMT_RECL

Specifies the default record length (normally 2040 bytes) for unformatted files.

FORTn

Specifies the file name for a particular unit number n, when a file name is not specified in the OPEN statement or an implicit OPEN is used, and the compiler option fpscomp with option keyword filesfromcmd was not specified. Preconnected files attached to units 0, 5, and 6 are by default associated with system standard I/O files.

INTEL_PRE_FFLAGS

INTEL_POST_FFLAGS

Specifies to add a set of Intel® Fortran Compiler command line options onto the end of the compile line.

This is an extension to the facility already provided in the compiler configuration file ifort.cfg.

Command line options can be inserted in the “prefix” position using the __INTEL_PRE_CFLAGS environment variable or in the postfix position using the __INTEL_POST_CFLAGS environment variable. The command line will be built as follows:

Syntax: ifort <PRE flags> <flags from configuration file> <flags from the compiler invocation> <POST flags>

NLSPATH (Linux* and OS X* only)

Specifies the path for the Intel® Fortran run-time error message catalog.

TBK_ENABLE_VERBOSE_STACK_TRACE

Displays more detailed call stack information in the event of an error.

The default brief output is usually sufficient to determine where an error occurred. Brief output includes up to twenty stack frames, reported one line per stack frame. For each frame, the image name containing the PC, routine name, line number, and source file are given.

The verbose output, if selected, will provide (in addition to the information in brief output) the exception context record if the error was a machine exception (machine register dump), and for each frame, the return address, frame pointer and stack pointer and possible parameters to the routine. This output can be quite long (but limited to 16K bytes) and use of the environment variable FOR_DIAGNOSTIC_LOG_FILE is recommended if you want to capture the output accurately. Most situations should not require the use of verbose output.

The variable FOR_ENABLE_VERBOSE_STACK_TRACE is also recognized for compatibility with Compaq* Visual Fortran.

See also Locating Run-Time Errors and Using Traceback Information Overview.

TBK_FULL_SRC_FILE_SPEC

By default, the traceback output displays only the file name and extension in the source file field. To display complete file name information including the path, set the environment variable TBK_FULL_SRC_FILE_SPEC to true.

The variable FOR_FULL_SRC_FILE_SPEC is also recognized for compatibility with Compaq* Visual Fortran.

See also Locating Run-Time Errors and Using Traceback Information Overview.

FORT_TMPDIR

TMP

TMPDIR

TEMP

Specifies an alternate working directory where scratch files are created.

GOMP_STACKSIZE (Linux* and OS X* only)

GNU extension recognized by the Intel® OpenMP compatibility library. Same as OMP_STACKSIZE. KMP_STACKSIZE overrides GOMP_STACKSIZE, which overrides OMP_STACKSIZE.

Default: See OMP_STACKSIZE description

GOMP_CPU_AFFINITY (Linux* and OS X* only)

GNU extension recognized by the Intel® OpenMP compatibility library. Specifies a list of OS processor IDs.

Default: Affinity is disabled

OpenMP* Environment Variables (OMP_) and Extensions (KMP_)

OMP_DYNAMIC

Enables (.TRUE.) or disables (.FALSE.) the dynamic adjustment of the number of threads.

Default: .FALSE.

Syntax: export OMP_DYNAMIC=value

OMP_MAX_ACTIVE_LEVELS

The maximum number of levels of parallel nesting for the program.

Default: 1

Syntax: export OMP_MAX_ACTIVE_LEVEL=value

OMP_NESTED

Enables (.TRUE.) or disables (.FALSE.)nested parallelism.

Default: .FALSE.

Syntax: export OMP_NESTED=value

OMP_NUM_THREADS

Sets the maximum number of threads to use for OpenMP* parallel regions if no other value is specified in the application. The value can be a single integer, in which case it specifies the number of threads for all parallel regions. Or it can be a comma-separated list of integers, in which case each integer specifies the number of threads for a parallel region at a nesting level.

The first position in the list represents the outer-most parallel nesting level, the second position represents the next-inner parallel nesting level, and so on. At any level, the integer can be left out of the list. If the first integer in a list is left out, it implies the normal default value for threads is used at the outer-most level. If the integer is left out of any other level, the number of threads for that level is inherited from the previous level.

This environment variable applies to both [Q]openmp or [Q]parallel options.

Default: Number of processors visible to the operating system.

Syntax: export OMP_NUM_THREADS=value[,value]*

OMP_PLACES

An explicit ordered list of places specified as either an abstract name describing a set of places or an explicit list of places described by nonnegative numbers. An exclusion operator “!” can also be used to exclude the number or place immediately following the operator.

For explicit lists, the meaning of the numbers and how the numbering is done for a list of nonnegative numbers are implementation defined. Generally, the numbers represent the smallest unit of execution exposed by the execution environment, typically a hardware thread.

Intervals can be specified using the <lower-bound> : <length> : <stride> notation to represent the following list of numbers:

“<lower-bound>, 
<lower-bound> + <stride>, 
..., 
<lower-bound> +(<length>-1)*<stride>.”

# EXPLICIT LIST EXAMPLE
setenv OMP_PLACES "{0,1,2,3},{4,5,6,7},{8,9,10,11},{12,13,14,15}"
setenv OMP_PLACES "{0:4},{4:4},{8:4},{12:4}"
setenv OMP_PLACES "{0:4}:4:4"

The abstract names listed below should be understood by the execution and runtime environment:

• threads: Each place corresponds to a single hardware thread on the target machine.
• cores: Each place corresponds to a single core (having one or more hardware threads) on the target machine.
• sockets: Each place corresponds to a single socket (consisting of one or more cores) on the target machine.

When requesting fewer places or more resources than available on the system, the determination of which resources of type abstract_name are to be included in the place list is implementation defined. The precise definitions of the abstract names are implementation defined. An implementation may also add abstract names as appropriate for the target platform. The abstract name may be appended by a positive number in parentheses to denote the length of the place list to be created, that is abstract_name(num-places).

# ABSTRACT NAMES EXAMPLE
setenv OMP_PLACES threads
setenv OMP_PLACES threads(4)

OMP_PROC_BIND

Sets the thread affinity policy to be used for parallel regions at the corresponding nested level. Enables (true) or disables (false) the binding of threads to processor contexts. If enabled, this is the same as specifying KMP_AFFINITY=scatter. If disabled, this is the same as specifying KMP_AFFINITY=none.

Acceptable values: true, false, or a comma separated list, each element of which is one of the following values: master, close, spread.

Default: false

If set to false, the execution environment may move OpenMP* threads between OpenMP* places, thread affinity is disabled, and proc_bind clauses on parallel constructs are ignored. Otherwise, the execution environment should not move OpenMP* threads between OpenMP* places, thread affinity is enabled, and the initial thread is bound to the first place in the OpenMP* place list.

If set to master, all threads are bound to the same place as the master thread. If set to close, threads will be bound to successive places, close to where the master thread is bound. If set to spread, the master thread's partition is subdivided and threads are bound to a single place successive sub-partitions.

OMP_SCHEDULE

Sets the run-time schedule type and an optional chunk size.

Default: STATIC, no chunk size specified

Syntax: export OMP_SCHEDULE="kind[,chunk_size]"

This environment variable is available for both Intel® and non-Intel microprocessors but it may perform additional optimizations for Intel microprocessors than it performs for non-Intel microprocessors.

OMP_STACKSIZE

Sets the number of bytes to allocate for each OpenMP* thread to use as the private stack for the thread. Recommended size is 16M.

Use the optional suffixes: B (bytes), K (Kilobytes), M (Megabytes), G (Gigabytes), or T (Terabytes) to specify the units. If only value is specified, the size is assumed to be K (Kilobytes).

This variable does not affect the native operating system threads created by the user program nor the thread executing the sequential part of an OpenMP* program or parallel programs created using the[Q]parallel option.

The kmp_{set,get}_stacksize_s() routines set/retrieve the value. The kmp_set_stacksize_s() routine must be called from sequential part, before first parallel region is created. Otherwise, calling kmp_set_stacksize_s() has no effect.

Default (IA-32 architecture): 2M

Default (Intel® 64 architecture): 4M

Syntax: export OMP_STACKSIZE=value

Related environment variables: KMP_STACKSIZE. KMP_STACKSIZE overrides OMP_STACKSIZE.

OMP_THREAD_LIMIT

Limits the number of simultaneously executing threads in an OpenMP* program.

If this limit is reached and another native operating system thread encounters OpenMP* API calls or constructs, the program can abort with an error message. If this limit is reached when an OpenMP* parallel region begins, a one-time warning message might be generated indicating that the number of threads in the team was reduced, but the program will continue.

This environment variable is only used for programs compiled with the [Q]openmp or [Q]parallel options.

The omp_get_thread_limit() routine returns the value of the limit.

Default: No enforced limit

Syntax: export OMP_THREAD_LIMIT=value

Related environment variable: KMP_ALL_THREADS. Its value overrides OMP_THREAD_LIMIT.

OMP_WAIT_POLICY

Decides whether threads spin (active) or yield (passive) while they are waiting. OMP_WAIT_POLICY=ACTIVE is an alias for KMP_LIBRARY=turnaround, and OMP_WAIT_POLICY=PASSIVE is an alias for KMP_LIBRARY=throughtput.

Default: Passive

Syntax: export OMP_WAIT_POLICY=value

KMP_AFFINITY

Enables run-time library to bind threads to physical processing units.

Default: None

This environment variable is available for both Intel® and non-Intel microprocessors but it may perform additional optimizations for Intel microprocessors than it performs for non-Intel microprocessors.

KMP_ALL_THREADS

Limits the number of simultaneously executing threads in an OpenMP* program. If this limit is reached and another native operating system thread encounters OpenMP* API calls or constructs, then the program may abort with an error message. If this limit is reached at the time an OpenMP* parallel region begins, a one-time warning message may be generated indicating that the number of threads in the team was reduced, but the program will continue execution. This environment variable is only used for programs compiled with -openmp or -openmp-profile.

Default: No enforced limit

KMP_BLOCKTIME

Sets the time, in milliseconds, that a thread should wait, after completing the execution of a parallel region, before sleeping.

Use the optional character suffixes: s (seconds), m (minutes), h (hours), or d (days) to specify the units.

Default: 200 milliseconds

KMP_CPUINFO_FILE

Specifies an alternate file name for file containing machine topology description. The file must be in the same format as /proc/cpuinfo.

Default: None

This environment variable is available for both Intel® and non-Intel microprocessors but it may perform additional optimizations for Intel microprocessors than it performs for non-Intel microprocessors.

KMP_DETERMINISTIC_REDUCTION

Enables (1) or disables (0) the use of a specific ordering of the reduction operations for implementing the reduction clause for an OpenMP* parallel region. This has the effect that, for a given number of threads, in a given parallel region, for a given data set and reduction operation, a floating point reduction done for an OpenMP* reduction clause will have a consistent floating point result from run to run, since round-off errors will be identical.

Default: 0

KMP_DYNAMIC_MODE

Selects the method used to determine the number of threads to use for a parallel region when OMP_DYNAMIC=true. Possible values: (asat | load_balance | thread_limit), where,

• asat: estimates number of threads based on parallel start time;

• load_balance: tries to avoid using more threads than available execution units on the machine;

• thread_limit: tries to avoid using more threads than total execution units on the machine.

Default (IA-32 architecture): load_balance (on all supported OSes)

Default (Intel® 64 architecture): load_balance (on all supported OSes)

Default (Intel® MIC architecture): thread_limit (on supported OSes)

KMP_INHERIT_FP_CONTROL

Enables (1) or disables (0) the copying of the floating point control settings of the master thread to the floating point control settings of the OpenMP* worker threads at the start of each parallel region.

Default: 1

KMP_LIBRARY

Selects the OpenMP* run-time library execution mode. The values for this variable are serial, turnaround, or throughput.

Default: throughput

KMP_MONITOR_STACKSIZE

Sets the number of bytes to allocate for the monitor thread, which is used for book-keeping during program execution.

Use the optional suffixes: b (bytes), k (kilobytes), m (megabytes), g (gigabytes), or t (terabytes) to specify the units.

Default: max (32k, system minimum thread stack size)

KMP_PLACE_THREADS

This environment variable only applies to Intel® Many Integrated Core Architecture (Intel® MIC Architecture).

Specifies the number of cores, with optional offset value and number of threads per core to use with an OpenMP application, as an alternative to writing explicit affinity settings or a process affinity mask.

value = (  int [ "C" | "T" ] [ delim ] | delim ) 
[ int [ "T" ] [ delim ] ] 
[ int [ "O" ] ];


delim = "," | "x";

"C" indicates cores, "T" indicates threads, "O" indicates an offset.

Default: If omitted, the default value is the available number of cores (threads).

Examples:

• 5C,3T,1O: use 5 cores with offset 1, 3 threads per core.

• 5,3,1: same as above.

• 24: use first 24 cores, all available threads per core.

• 2T: use all cores, 2 threads per core.

• ,2: same as above.

• 3x2: use 3 cores, 2 threads per core

• 4C12O: use 4 cores with offset 12, all available threads per core.

KMP_SETTINGS

Enables (1) or disables (0) the printing of OpenMP* run-time library environment variables during program execution. Two lists of variables are printed: user-defined environment variables settings and effective values of variables used by OpenMP* run-time library.

Default: 0

KMP_STACKSIZE

Sets the number of bytes to allocate for each OpenMP* thread to use as its private stack. Recommended size is 16m. Use the optional suffix b, k, m, g, or t, to specify bytes, kilobytes, megabytes, gigabytes, or terabytes. This variable does not affect the native operating system threads created by the user program nor the thread executing the sequential part of an OpenMP* program or parallel programs created using [Q]parallel option.

Default (IA-32 architecture): 2m

Default (Intel® 64 architecture): 4m

KMP_VERSION

Enables (1) or disables (0) the printing of OpenMP* run-time library version information during program execution.

Default: 0

INTEL_PROF_DUMP_CUMULATIVE

When using interval profile dumping (initiated by INTEL_PROF_DUMP_INTERVAL or the function _PGOPTI_Set_Interval_Prof_Dump) during the execution of an instrumented user application, allows creation of a single .dyn file to contain profiling information instead of multiple .dyn files.

If this environment variable is not set, executing an instrumented user application creates a new .dyn file for each interval. Setting this environment variable is useful for applications that do not terminate or those that terminate abnormally (bypass the normal exit code).

INTEL_PROF_DUMP_INTERVAL

Initiates interval profile dumping in an instrumented user application. This environment variable may be used to initiate Interval Profile Dumping in an instrumented application.

See Interval Profile Dumping for more information.

PROF_DIR

Specifies the directory in which dynamic information files are created. This variable applies to all three phases of the profiling process.

PROF_DUMP_INTERVAL

Deprecated; use INTEL_PROF_DUMP_INTERVAL instead.

PROF_NO_CLOBBER

Alters the feedback compilation phase slightly. By default, during the feedback compilation phase, the compiler merges data from all dynamic information files and creates a new pgopti.dpi file if the .dyn files are newer than an existing pgopti.dpi file.

When this variable is set the compiler does not overwrite the existing pgopti.dpi file. Instead, the compiler issues a warning. You must remove the pgopti.dpi file if you want to use additional dynamic information files.

MIC_ENV_PREFIX

Used to control environment variables passed to the target.

By default, all environment variables set on the host are passed to the target. Setting MIC_ENV_PREFIX passes only environment variables that have a prefix of the value of this variable. For example, setting MIC_ENV_PREFIX=ABC passes only environment variables that have a prefix of ABC.

For information on passing environment variables to a particular coprocessor, and on passing multiple environment variables to one or all coprocessors, see Setting Environment Variables on the CPU to Modify the Coprocessor's Execution Environment.

MIC_HOST_LOG

A string containing the file name to output traces from the host.

Default: None

Example: MIC_HOST_LOG=/home/user/app/host.log

MIC_LD_LIBRARY_PATH

Specifies the location for target shared objects. LD_LIBRARY_PATH specifies the location for host objects.

MIC_PROXY_IO

Enables (1) or disables (0) the proxy of stderr and stdout, and specifies whether the process wants I/O proxy support between the host and target.

Default: 1

Example: MIC_PROXY_IO=0

MIC_SEP_MONITOR

Enables (1) or disables (0) SEP at offload regions.

Default: 0

Example: MIC_SEP_MONITOR=1

MIC_STACKSIZE

Specifies the stack size of the main thread for the offload. This value is typically much larger than the stack size for individual threads. It corresponds to what you would set with ulimit -s (BASH shell) or limit stacksize (C shell) if you were running natively on either the target or the host.

Set this variable to integerB|K|M|G|T, where:

• B = bytes

• K = kilobytes

• M = megabytes

• G = gigabytes

• T = terabytes

Default: 12M

Example: MIC_STACKSIZE=16M

OFFLOAD_DEVICES

Restricts the offload process to use only the coprocessors specified as the value of the variable.

Set this variable to a comma separated list of target device numbers in the range 0 to number_of_devices_in_the_system-1, where 0 is the first coprocessor in the system, and number_of_devices_in_the_system-1 is the last coprocessor in the system.

Coprocessors available for offloading are numbered logically. The function OFFLOAD_NUMBER_OF_DEVICES returns the number of available coprocessors. Coprocessor indices that you use in the target specifier of the offload directives are in the range 0 to number_of_devices_in_the_system-1.

Default: The offload process uses all devices.

Example: OFFLOAD_DEVICES = 1,2

On a system with more than two coprocessors installed, this setting enables the application to use only coprocessors 1 and 2. Offloads to coprocessors 0 or 1 are performed on the first and second physical coprocessors. Offloads to target numbers higher than 1 wrap-around so that all offloads remain within coprocessors 0 and 1. The function OFFLOAD_NUMBER_OF_DEVICES executed on a coprocessor return 0 or 1, when the offload is running on the first or second coprocessor.

OFFLOAD_INIT

Specifies a hint to the offload runtime when to initialize targets.

Supported values:

• on_start: All available coprocessors are initialized before entering main.

• on_offload: The target is initialized immediately before the first offload to it. Initialization is done only on the target that handles the offload.

• on_offload_all: All available targets are initialized immediately before the first offload in an application.

Default: on_offload_all

Example: OFFLOAD_INIT=on_start

OFFLOAD_REPORT

Controls printing offload execution time, in seconds, and the amount of data transferred, in bytes. This environment variable is equivalent to using the __Offload_report API.

Supported values:

• 1: Prints the offload computation time, in seconds.

• 2: In addition to the information produced at value 1, adds the amount of data transferred between the CPU and the target, in bytes.

• 3: In addition to the information produced at value 2, gives additional details on offload activity, including device initialization, and individual variable transfers.

Default: None

Example: OFFLOAD_REPORT=1