Function: create_kernel

Applicability

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

Function Usage

Creates the kernel function of a dynamic-shape operator for operator execution.

Prototype

C Prototype

aclError aclopCreateKernel(const char *opType,
const char *kernelId,
const char *kernelName,
void *binData,
int binSize,
aclopEngineType enginetype,
aclDataDeallocator deallocator)

Python Function

ret = acl.op.create_kernel(op_type, kernel_id, kernel_name, bin_data, bin_size, enginetype, deallocator)

Parameters

Parameter	Description
op_type	Str, operator type.
kernel_id	Str, kernel ID.
kernel_name	Str, operator kernel name, which must be the same as the kernelName argument in the operator binary file.
bin_data	Int, memory address of the operator kernel file.
bin_size	Int, memory size (in bytes) of the operator kernel file.
enginetype	Int, operator execution engine. This parameter is valid only when compile_flag of acl.op.update.params is set to ACL_COMPILE_SYS. 0: ACL_ENGINE_SYS, set when the execution target is not concerned. 1: ACL_ENGINE_AICORE, compiles the operator into an AI Core operator. 2: ACL_ENGINE_VECTOR, compiles the operator into a Vector Core operator.
deallocator	Int, whether to automatically release the bin_data memory. 0: Data is not automatically released. The caller compiles the code to release the data. 1: The bin_data memory is automatically released. A callback function is set internally to release the bin_data memory.

Returns

Return Value	Description
ret	Int, error code: 0 on success; else, failure.

Parent topic: Single-Operator Model Execution