aclrtLaunchKernelWithHostArgs

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product / Atlas A3 inference product	√
Atlas A2 training product / Atlas A2 inference product	√
Atlas 200I/500 A2 inference product	√
Atlas inference product	√
Atlas training product	√

Description

Specifies the configuration to be delivered by a task and starts the compute task of the corresponding operator. This API is asynchronous.

Prototype

      
           aclError aclrtLaunchKernelWithHostArgs(aclrtFuncHandle funcHandle, uint32_t numBlocks, aclrtStream stream, aclrtLaunchKernelCfg *cfg, void *hostArgs, size_t argsSize, aclrtPlaceHolderInfo *placeHolderArray, size_t placeHolderNum)

Parameters

Parameter	Input/Output	Description
funcHandle	Input	Kernel function handle. For details about the type definition, see aclrtFuncHandle.
numBlocks	Input	Number of cores on which the kernel function will be executed.
stream	Input	Stream for executing the task. For details about the type definition, see aclrtStream.
cfg	Input	Configuration to be delivered by the task. For details about the type definition, see aclrtLaunchKernelCfg. If no configuration is specified, NULL can be passed.
hostArgs	Input	Pointer to the host memory address that stores all input parameters of the kernel function.
argsSize	Input	Size of the hostArgs parameter's value, in bytes.
placeHolderArray	Input	placeholder parameter array. aclrtPlaceHolderInfo is defined as follows: typedef struct { uint32_t addrOffset; uint32_t dataOffset; } aclrtPlaceHolderInfo; The member variables are described as follows: addrOffset: After the data area pointed to by placeholder is copied to the device, the actual device memory address needs to be updated to hostArgs during launch. This parameter specifies the location offset to be updated. dataOffset: The data area pointed to by placeholder needs to be copied to the device. This parameter specifies the address offset of the data area based on hostArgs.
placeHolderNum	Input	Size of the placeholder parameter array.

Returns

0 on success; otherwise, failure. For details, see aclError.

API Call Example

For the API call example, see AI Core custom operator loading.

Reference

All the APIs listed in the following table enable the compute tasks of the corresponding operators. However, their functions and usage modes are different.

API	How Kernel Function Parameter Values Are Passed	Where Kernel Function Parameter Values Are Stored	Configuration Customizable to Deliver
aclrtLaunchKernel	Specify the device memory address pointer for storing all input arguments of the kernel function.	Device memory	No
aclrtLaunchKernelV2	Specify the device memory address pointer for storing all input arguments of the kernel function.	Device memory	Yes
aclrtLaunchKernelWithConfig	Specify the parameter list handle aclrtArgsHandle.	Host memory	Yes
aclrtLaunchKernelWithHostArgs	Specify the host memory address pointer for storing all input arguments of the kernel function.	Host memory	Yes

Parent topic: Kernel Loading and Execution