set_compute_device_paramT_set_compute_device_paramSetComputeDeviceParamSetComputeDeviceParam (Operator)

Name

set_compute_device_paramT_set_compute_device_paramSetComputeDeviceParamSetComputeDeviceParam — Set parameters of an compute device.

Signature

Description

set_compute_device_paramset_compute_device_paramSetComputeDeviceParamSetComputeDeviceParamSetComputeDeviceParam sets parameters of the compute device DeviceHandleDeviceHandleDeviceHandleDeviceHandledeviceHandle for the current HALCON thread.

The following Parameters can be set:

'alloc_pinned'

If 'true', the output image matrices of all operators (executed on the compute device or not) called in the current HALCON thread are created in page locked (so called 'pinned') memory. This accelerates the transfer between host and device memory. However, too excessive usage of page locked memory may have a negative impact on overall system performance. You should activate page locked memory allocation if the output image of of the next operator call is to be transferred to the device. Page locked memory allocation should be deactivated for all operator calls whose output images are not required on the compute device (see example).

value: 'true' or 'false'

default: 'true'

'asynchronous_execution'

If 'true', operators executed on the compute device do not wait for the device computations to finish but return after initiating the computations. All device computations are synchronized as soon as the output of a compute device operation is used on the CPU (e.g. by disp_imagedisp_imageDispImageDispImageDispImage).

value: 'true' or 'false'

default: 'true'

'buffer_cache_capacity'

Maximum size (in bytes) of the compute device buffer cache.

value: size in bytes

default: 1/3 of the available device memory

'image_cache_capacity'

Maximum size (in bytes) of the compute device image cache.

value: size in bytes

default: 1/3 of the available device memory

'pinned_mem_cache_capacity'

Maximum size (in bytes) of the page locked (pinned) memory cache.

value: size in bytes

default: 32 MBytes

Execution Information

• Multithreading type: reentrant (runs in parallel with non-exclusive operators).
• Multithreading scope: local (may only be called from the same thread in which the window, model, or tool instance was created).
• Processed without parallelization.

Parameters

Example (HDevelop)

activate_compute_device (DeviceHandle)
read_image (Image, 'fuse')
set_compute_device_param (DeviceHandle, 'alloc_pinned', 'true')
* filter on compute device, output image is page locked
derivate_gauss (Image, DerivGauss, 3, 'gradient')
* filter result on the CPU, output image should not be page locked
set_compute_device_param (DeviceHandle, 'alloc_pinned', 'false')
median_image (DerivGauss, ImageMedian, 'circle', 1, 'mirrored')

Result

The operator set_compute_device_paramset_compute_device_paramSetComputeDeviceParamSetComputeDeviceParamSetComputeDeviceParam returns the value 2 (H_MSG_TRUE) if the parameters are correct. Otherwise an exception will be raised.

Possible Predecessors

activate_compute_deviceactivate_compute_deviceActivateComputeDeviceActivateComputeDeviceActivateComputeDevice

See also

get_compute_device_paramget_compute_device_paramGetComputeDeviceParamGetComputeDeviceParamGetComputeDeviceParam

Module

Foundation