Supported Environment Variables

CILK_NWORKERS (Windows*, Linux*)

Specifies the number of worker threads to use in an Intel® Cilk™ Plus program.

See Set Worker Count.

CL (Windows)

_CL_ (Windows)

Specifies file locations for the Microsoft* compiler. Define the files and options you use most often with the CL variable. Note: You cannot set the CL environment variable to a string that contains an equal sign. You can use the pound sign instead. In the following example, the pound sign (#) is used as a substitute for an equal sign in the assigned string: SET CL=/Dtest#100

COV_DIR (Windows)

Same as PROF_DIR.

COV_DPI (Windows)

Same as PROF_DPI.

IA32ROOT (IA-32 architecture and Intel® 64 architecture)

Points to the directories containing the include and library files for a non-standard installation structure.

ICCCFG

Specifies the configuration file for customizing compilations when invoking the compiler using icc.

ICPCCFG

Specifies the configuration file for customizing compilations when invoking the compiler using icpc.

ICLCFG (OS X*)

Specifies the configuration file for customizing compilations when invoking the compiler using icl. Valid for OS X* v10.9 or later only.

ICLXXCFG (OS X*)

Specifies the configuration file for customizing compilations when invoking the compiler using icl++. Valid for OS X* v10.9 or later only.

ICLCFG (Windows)

Specifies a configuration file, which the compiler should use instead of the default configuration file.

INTEL_LICENSE_FILE

Specifies the location for the Intel license file.

__INTEL_PRE_CFLAGS

__INTEL_POST_CFLAGS

Specifies a set of compiler options to add to the compile line.

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

You can insert command line options in the prefix position using __INTEL_PRE_CFLAGS , or in the suffix position using __INTEL_POST_CFLAGS . The command line is built as follows:

Syntax:(On Windows, use icl. On Linux or OS X use icc)iclicc <PRE flags> <flags from configuration file> <flags from the compiler invocation> <POST flags>

PATH

Specifies the directories the system searches for binary executable files.

TMP

TMPDIR

TEMP

Specifies the location for temporary files. If none of these are specified, or writeable, or found, the compiler stores temporary files in /tmp (Linux, OS X) or the current directory (Windows).

On Windows* and OS X*, The compiler searches for these variables in the following order: TMP, TMPDIR, and TEMP.

On Linux*, The compiler searches for these variables in the following order: TMPDIR, TMP, and TEMP.

LD_LIBRARY_PATH (Linux*)

Specifies the location for shared objects (.so files).

DYLD_LIBRARY_PATH (OS X*)

Specifies the path for dynamic libraries.

INCLUDE (Windows*)

Specifies the directories for the source header files (include files).

LIB

Specifies the directories for all libraries used by the compiler and linker.

MIC_LIBRARY_PATH

Specifies the location on the host for target-specific static libraries. LIBRARY_PATH specifies the location on the host for host-specific static libraries.

CPATH (Linux* and OS X*)

Specifies the path to include directory for C/C++ compilations.

C_INCLUDE_PATH

Specifies path to include directory for C compilations.

CPLUS_INCLUDE_PATH

Specifies path to include directory for C++ compilations.

DEPENDENCIES_OUTPUT

Specifies how to output dependencies for make based on the non-system header files processed by the compiler. System header files are ignored in the dependency output.

GCC_EXEC_PREFIX

Specifies alternative names for the linker (ld) and assembler (as).

GCCROOT (Linux*)

Specifies the location of the gcc* binaries.

Set this variable only when the compiler cannot locate the gcc binaries when using the -gcc-name option.

GXX_INCLUDE (Linux*)

Specifies the location of the gcc headers. Set this variable to specify the locations of the gcc installed files when the compiler does not find the needed values as specified by the use of -gcc-name=directory-name/gcc or -gxx-name=directory-name/g++.

GXX_ROOT (Linux*)

Specifies the location of the gcc binaries. Set this variable to specify the locations of the gcc installed files when the compiler does not find the needed values as specified by the use of -gcc-name=directory-name/gcc or -gxx-name=directory-name/g++.

LIBRARY_PATH (Linux* and OS X*)

Specifies the path for libraries to be used during the link phase.

SUNPRO_DEPENDENCIES

This variable is the same as DEPENDENCIES_OUTPUT, except that system header files are not ignored.

INTEL_CHKP_REPORT_MODE (Linux*)

Changes the pointer checker reporting mode at runtime.

See Finding and Reporting Out-of-Bounds Errors.

GOMP_CPU_AFFINITY (Linux*)

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

You must set this environment variable before the first parallel region or before certain API calls including omp_get_max_threads(), omp_get_num_procs() and any affinity API calls. For detailed information on this environment variable, see Thread Affinity Interface.

Default: Affinity is disabled

GOMP_STACKSIZE (Linux*)

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

Default: See the description for OMP_STACKSIZE.

OMP_CANCELLATION

Activates cancellation of the innermost enclosing region of the type specified. If set to true, the effects of the cancel construct and of cancellation points are enabled and cancellation is activated. If set to false, cancellation is disabled and the cancel construct and cancellation points are effectively ignored.

Enables (true) or disables (false) cancellation of the innermost enclosing region of the type specified.

Default: FALSE

Example: OMP_CANCELLATION=true

OMP_DYNAMIC

Enables (true) or disables (false) the dynamic adjustment of the number of threads.

Default: FALSE

Example: OMP_DYNAMIC=true

OMP_MAX_ACTIVE_LEVELS

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

Default: 1

Syntax: OMP_MAX_ACTIVE_LEVELS=true

OMP_NESTED

Enables (true) or disables (false) nested parallelism.

Default: FALSE

Example: OMP_NESTED=true

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. The value can also 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 the options Qopenmp (Windows) or qopenmp (Linux and OS X), or Qparallel (Windows) or qparallel (Linux and OS X) .

Default: The number of processors visible to the operating system on which the program is executed.

Syntax: OMP_NUM_THREADS=value[,value]*

OMP_PLACES

Specifies an explicit ordered list of places, either as an abstract name describing a set of places or as 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 (Windows, Linux)

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 are 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 single place successive sub-partitions.

OMP_SCHEDULE

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

Default: STATIC, no chunk size specified

Example syntax: OMP_SCHEDULE="kind[,chunk_size]"

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 to specify byte units: B (bytes), K (Kilobytes), M (Megabytes), G (Gigabytes), or T (Terabytes) to specify the units. If you specify a value without a suffix, the byte unit is assumed to be K (Kilobytes).

This variable does not affect the native operating system threads created by the user program, or the thread executing the sequential part of an OpenMP* program or parallel programs created using the option Qparallel (Windows) or qparallel (Linux and OS X) .

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

Default (Intel® MIC architecture):4M (on supported OSes)

Related environment variables: KMP_STACKSIZE (overrides OMP_STACKSIZE).

Syntax: OMP_STACKSIZE=value

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 following options: Qopenmp (Windows) or qopenmp (Linux and OS X), or Qparallel (Windows) or qparallel (Linux and OS X) .

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

Default: No enforced limit

Related environment variable: KMP_ALL_THREADS (overrides OMP_THREAD_LIMIT).

Example syntax: OMP_THREAD_LIMIT=value

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=throughput.

Default: Passive

Syntax: OMP_WAIT_POLICY=value

KMP_AFFINITY (Windows, Linux)

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

You must set this environment variable before the first parallel region, or certain API calls including omp_get_max_threads(), omp_get_num_procs() and any affinity API calls. For detailed information on this environment variable, see Thread Affinity Interface.

Default: noverbose,warnings,respect,granularity=core,duplicates,none

Default (Intel® MIC architecture): noverbose,warnings,respect,granularity=fine,duplicates,scatter,0,0

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 the Qopenmp(Windows) or qopenmp (Linux and OS X) option.

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.

Specify infinite for an unlimited wait time.

Default: 200 milliseconds

Related Environment Variable: KMP_LIBRARY environment variable.

KMP_CPUINFO_FILE

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

Default: None

KMP_DETERMINISTIC_REDUCTION

Enables (true) or disables (false) 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 has a consistent floating point result from run to run, since round-off errors are identical.

Default: false

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;

  Note

  Support for asat (automatic self-allocating threads) is now deprecated and will be removed in a future release.

• 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_HOT_TEAMS_MAX_LEVEL

Sets the maximum nested level to which teams of threads will be hot.

For values of 2 and above, nested parallelism should be enabled.

Default: 1

KMP_HOT_TEAMS_MODE

Specifies the run-time behavior when the number of threads in a hot team is reduced.

Possible values:

• 0: Extra threads are freed and put into a common pool of threads.

• 1: Extra threads are kept in the team in reserve, for faster reuse in subsequent parallel regions.

Default: 0

KMP_INHERIT_FP_CONTROL

Enables (true) or disables (false) 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: true

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 to specify byte units: B (bytes), K (Kilobytes), M (Megabytes), G (Gigabytes), or T (Terabytes) to specify the units. If you specify a value without a suffix, the byte unit is assumed to be K (Kilobytes).

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

KMP_PLACE_THREADS

Specifies the number of sockets, cores per socket, and the number of threads per core, to use with an OpenMP* application, as an alternative to writing explicit affinity settings or a process affinity mask. You can also specify an offset value to set which sockets and cores to use.

Syntax: socketsS[@offset],coresC[@offset],threadsT

S, C and T are not case-sensitive.

sockets

The number of sockets to use.

cores

The number of cores to use per socket.

threads

The number of threads to use per core.

offset

(Optional) The number of sockets or cores to skip.

Default: If omitted, the default value is the available number of sockets, cores, or threads.

Examples:

• 2s,4c,2t: Use the first 2 sockets (s0 and s1), the first 4 cores on each socket (c0 - c3), and 2 threads per core.

• 2s@2,4c@8,2t: Skip the first 2 sockets (s0 and s1) and use 2 sockets (s2-s3), skip the first 8 cores (c0-c7) and use 4 cores on each socket (c8-c11), and use 2 threads per core.

• 5C@1,3T: Use all available sockets, skip the first core and use 5 cores, and use 3 threads per core.

• 2T: Use all cores on all sockets, 2 threads per core.

• 4C@12: Use 4 cores with offset 12, all available threads per core.

KMP_SETTINGS

Enables (true) or disables (false) 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: false

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 suffixes to specify byte units: B (bytes), K (Kilobytes), M (Megabytes), G (Gigabytes), or T (Terabytes) to specify the units. If you specify a value without a suffix, the byte unit 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 option Qparallel (Windows) or qparallel (Linux and OS X) .

Default (IA-32 architecture): 2m

Default (Intel® 64 architecture): 4m

Default (Intel® MIC architecture):4M (on supported OSes)

KMP_VERSION

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

Default: false

KMP_WARNINGS

Enables (true) or disables (false) displaying warnings from the OpenMP* run-time library during program execution.

Default: true

GFX_CPU_BACKUP

Controls whether heterogeneous code is executed on the host when the target is not available.

If the if clause of the offload pragma evaluates to false then the code is executed on the host regardless of the target's availability.

If the if clause of the offload pragma is not provided or evaluates to true and the target and the Intel® Graphics Technology driver supporting offload are available and correctly initialized, then the code is executed on the target. Otherwise, back-up execution on the host depends on the value of GFX_CPU_BACKUP.

To prevent execution from occurring on the host, which causes the application to fail when the target is not available, set GFX_CPU_BACKUP to 0.

To execute on the host when the target is not available, set GFX_CPU_BACKUP to 1. By default, this variable is set to 1, so the code is executed on the host.

Default: 1

GFX_MAX_THREAD_COUNT

Controls the maximum number of target threads to parallelize loop nests offloaded to the graphics cores. Use this variable to override the default maximum value.

Two values have special meaning:

• - 1 (the default): The runtime creates one thread for each parallel loop nest iteration. Loops can be vectorized, and one iteration executed by a thread may include several original loop iterations.

• 0: The threadcount is N*hardware_thread_count, where N is:

  • 4 for one-dimensional loop nests.

  • 8 for loop nests of two or more dimensions.

  Try using this value when tuning your program for performance.

Scheduling many threads (e.g. 512) helps hide the latency of some operations, but a large number of threads can increase the cost of offload or negatively affect performance. By default, the runtime automatically determines the maximum thread count.

The real thread count for a particular offload execution can be lower than the maximum, and is determined by the offload runtime, depending on the real iteration space for that execution. For more information, see Fully Using Parallelism Specific to Intel® Graphics Technology.

Default: -1

Example: GFX_MAX_THREAD_COUNT=32

GFX_OFFLOAD_TIMEOUT

By default, the system will not allow for an offload task to execute longer than system recovery timeout period (usually two seconds). The task appears to hang and abnormally terminate after another timeout, defined by this variable. To enable longer execution of your offload tasks, disable recovery on timeout in the system registry (refer to Microsoft* documentation for instructions) and set GFX_OFFLOAD_TIMEOUT to a higher value, if 60 seconds are insufficient.

Default: 60 seconds.

GFX_LOG_OFFLOAD

Enables or disables logging of the offload process by setting this to the desirable logging level, from 1 to 3, where 3 has the greatest amount of detail.

To see the number of available hardware threads, set this environment variable to 1.

To monitor the real thread count, set this environment variable to 2.

To monitor the real thread count as well as distribution of the parallel loop nest’s iteration space, set this environment variable to 3.

To disable logging, set this environment variable to either an empty string or 0.

GFX_SHOW_TIME

Enables or disables printing timing information at the end of execution.

To disable timer printing, set GFX_SHOW_TIME to either an empty string or 0.

See Offload Timers for GFX.

GFX_THREAD_GROUP_WIDTH

Sets the width of a thread group.

See Controlling Thread Group and Thread Space Shape.

GFX_THREAD_GROUP_HEIGHT

Sets the height of a thread group.

See Controlling Thread Group and Thread Space Shape.

GFX_THREAD_SPACE_WIDTH

Sets the width of the thread space, in threads.

When this variable is set to 0, or not set, an appropriate value is calculated automatically.

This variable overrides the function _GFX_set_thread_space_config.

See Controlling Thread Group and Thread Space Shape.

GFX_THREAD_SPACE_HEIGHT

Sets the height of the thread space, in threads.

When this variable is set to 0, or not set, an appropriate value is calculated automatically.

This variable overrides the function _GFX_set_thread_space_config.

See Controlling Thread Group and Thread Space Shape.

Intel® Many Integrated Core (Intel® MIC) Environment Variables

MIC_ENV_PREFIX

Controls 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.

MIC_ENV_PREFIX=ABC
ABC_LD_LIBRARY_PATH=/tmp/mylibs

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_LD_LIBRARY_PATH

Specifies the location on the host for target-specific shared objects that may be loaded to the target. The value is not passed to the target. See MIC_ENV_PREFIX for special handling of the value on the target.

LD_LIBRARY_PATH specifies the location on the host for host-specific shared 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

MIC_USE_2MB_BUFFERS

Use 2M pages for (size > MIC_USE_2MB_BUFFERS). Pointer-based variables whose runtime length exceeds the value of this environment variable are allocated in large pages.

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

• B = Bytes

• K = Kilobytes

• M = Megabytes

• G = Gigabytes

• T = Terabytes

Example: MIC_USE_2MB_BUFFERS=64K

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 pragmas 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 second and third physical coprocessors. Offloads to target numbers higher than 1 wrap around, and all offloads remain within coprocessors 0 and 1. The function _Offload_number_of_devices() , executed on a coprocessor, returns 0 or 1 when the offload is running on the first or second coprocessor.

Setting this variable to an empty value on Linux (export OFFLOAD_DEVICES= ) or the value none on Windows (set OFFLOAD_DEVICES=none) impacts the execution of offloaded code within your program as follows:

• When you use optional offloading, the code executes on the host CPU.

• When you use mandatory offloading, the program terminates with an offload error, reporting that a device is not available, when it reaches a point in the execution where the coprocessor requires initialization for offloading.

You can specify optional and mandatory offloading using the optional and mandatory clauses in either of the following:

• The compiler option qoffload (Linux) or Qoffload (Windows)

• The offload pragma

OFFLOAD_INIT

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

Possible 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.

Possible 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

See also _Offload_report.

Profile Guided Optimization (PGO_) Environment Variables

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 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

INTEL_PROF_DYN_PREFIX

Specifies the prefix to be used for the .dyn filename to distinguish it from the other .dyn files dumped by other PGO runs. Executing the instrumented application generates a .dyn filename as follows: <prefix>_<timestamp>_<pid>.dyn, where <prefix> is the identifier that you have specified.

PROF_DIR

Specifies the directory where profiling files (files with extensions .dyn, .dpi, .spi and so on) are stored. The default is to store the .dyn files in the source directory of the file containing the first executed instrumented routine in the binary compiled with [Q]prof-gen option.

This variable applies to all three phases of the profiling process:

• Instrumentation compilation and linking

• Instrumented execution

• Feedback compilation

PROF_DPI

Name for the .dpi file.

Default: pgopti.dpi

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.