4_custom_kernel_launch

Description

This sample provides a minimum runnable example demonstrating how to use the <<<>>> kernel call operator to dispatch custom AscendC Kernel to complete 8-element vector addition. After successful execution, the result should be consistent with the vector addition result in 0_hello_cann.

Product Support

This sample supports the following products:

Product	Supported
Ascend 950PR/Ascend 950DT	Yes
Atlas A3 training series products/Atlas A3 inference series products	Yes
Atlas A2 training series products/Atlas A2 inference series products	Yes

Compile and Run

1. Download the sample code to the environment where CANN software is installed. Switch to the sample directory.

cd ${git_clone_path}/example/0_quickstart/4_custom_kernel_launch

2. Set environment variables.

# Replace ${install_root} with the CANN installation root directory. The default installation is in the `/usr/local/Ascend` directory.
source ${install_root}/cann/set_env.sh
export ASCEND_INSTALL_PATH=${install_root}/cann

# Set SOC_VERSION and ASCENDC_CMAKE_DIR
# -SOC_VERSION: Ascend AI processor model, such as Ascend910_9362, Ascend910B2, and so on
# -ASCENDC_CMAKE_DIR: The sample involves calling AscendC operators. Configure the AscendC compiler ascendc.cmake path, such as /usr/local/Ascend/cann/x86_64-linux/tikcpp/ascendc_kernel_cmake
source ${git_clone_path}/example/set_sample_env.sh

3. Run the following command to execute the sample.

bash run.sh

CANN RUNTIME API

The key functionality points and their key interfaces involved in this sample are as follows:

• Initialization
  • Call aclInit interface for initialization configuration.
  • Call aclFinalize interface for deinitialization.
• Device and Stream Management
  • Call aclrtSetDevice interface to specify the Device for computation.
  • Call aclrtCreateStream interface to create a Stream.
  • Call aclrtSynchronizeStream interface to block and wait for Stream tasks to complete.
  • Call aclrtDestroyStream interface to destroy a Stream.
  • Call aclrtResetDeviceForce interface to forcefully reset the current computation Device and reclaim Device resources.
• Memory Management and Data Transfer
  • Call aclrtMalloc interface to allocate memory on Device.
  • Call aclrtMemcpy interface to complete Host/Device data transfer.
  • Call aclrtFree interface to release memory on Device.
• Kernel Call
  • Use <<<>>> kernel call operator to dispatch custom AscendC Kernel.

Sample Output

The sample will print input vectors, <<<>>> call success information, and the final result. Typical output is as follows:

ACL init successfully
Set device 0 successfully
Create stream successfully
Allocate device buffers successfully
Copy input vectors to device successfully
Input vectors:
  self:   [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]
  other:  [0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0]
  alpha:  1.0
Custom AscendC kernel <<<>>> call successfully
Synchronize stream successfully

Vector addition result:
  result[0] = 1.5 (expected: 1.5)
  result[1] = 3.0 (expected: 3.0)
  result[2] = 4.5 (expected: 4.5)
  result[3] = 6.0 (expected: 6.0)
  result[4] = 7.5 (expected: 7.5)
  result[5] = 9.0 (expected: 9.0)
  result[6] = 10.5 (expected: 10.5)
  result[7] = 12.0 (expected: 12.0)

Sample run successfully with <<<>>> kernel call!

If the output result matches expectations, it indicates that the CANN custom Kernel <<<>>> call path is working properly.

Related Samples

• 0_hello_cann: Use aclnnAdd to complete the same vector addition.