HcommReadReduceOnThread

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	x
Atlas inference product	x
Atlas training product	x

For Atlas A2 training product/Atlas A2 inference product, only the Atlas 800T A2 training server, Atlas 900 A2 PoD cluster basic unit, and Atlas 200T A2 Box16 heterogeneous subrack are supported.

Function

Reads data from the specified memory on a channel. It performs a reduceOp operation on the memory data of length count × sizeof(dataType) pointed to by src and the memory data of the same length pointed to by dst, and outputs the result to dst. The API caller is the node where dst resides.

Prototype

int32_t HcommReadReduceOnThread(ThreadHandle thread, ChannelHandle channel, void *dst, const void *src, uint64_t count, HcommDataType dataType, HcommReduceOp reduceOp)

Parameters

Parameter	Input/Output	Description
thread	Input	Communication thread handle, which is threads returned by the HcclThreadAcquire call. For details, see ThreadHandle.
channel	Input	Communication channel handle, which is channels returned by the HcclChannelAcquire call. For details, see ChannelHandle.
dst	Output	Destination memory address, which is obtained using HcclGetHcclBuffer or HcclChannelGetHcclBuffer.
src	Input	Source memory address, which is obtained using HcclGetHcclBuffer or HcclChannelGetHcclBuffer.
count	Input	Number of elements.
dataType	Input	Data type. For details, see HcommDataType. Atlas 350 Accelerator Card: The supported data types are int8, int16, int32, uint8, uint16, uint32, fp16, fp32, and bfp16. Atlas A3 training product/Atlas A3 inference product: The supported data types are int8, int16, int32, float16, float32, and bfp16. Atlas A2 training product/Atlas A2 inference product: The supported data types are int8, int16, int32, float16, float32, and bfp16.
reduceOp	Input	Reduction operation type. For details, see HcommReduceOp. Atlas 350 Accelerator Card: The supported reduction types are sum, max, and min. Atlas A3 training product/Atlas A3 inference product: The supported reduction types are sum, max, and min. Atlas A2 training product/Atlas A2 inference product: The supported reduction types are sum, max, and min.

Returns

int32_t: 0 on success; else, failure.

Restrictions

None

Example

// Allocate communication threads.
CommEngine engine = CommEngine::COMM_ENGINE_CPU_TS; // Used by the Atlas A3 training product/Atlas A3 inference product
CommEngine engine = CommEngine::COMM_ENGINE_AICPU_TS; // Used by the Atlas 350 Accelerator Card
uint32_t threadNum = 1;
uint32_t notifyNumPerThread = 1;
ThreadHandle thread;
HcclThreadAcquire(engine, threadNum, notifyNumPerThread, &thread);

// Allocate communication channels.
uint32_t channelNum = 1;
HcclChannelDesc channelDesc;
HcclChannelDescInit(&channelDesc, channelNum);
HcclComm comm;
ChannelHandle channel;
HcclChannelAcquire(comm, engine, &channelDesc, channelNum, &channel);

// Obtain the local communication memory information.
void * localBuffer;
uint64_t localBufferSize;
HcclGetHcclBuffer(comm, &localBuffer, &localBufferSize);

// Obtain the remote communication memory information.
void * remoteBuffer;
uint64_t remoteBufferSize;
HcclChannelGetHcclBuffer(comm, channel, &remoteBuffer, &remoteBufferSize);
uint64_t count = std::min(localBufferSize/sizeof(uint64_t), remoteBufferSize/sizeof(uint64_t));

// Perform the reduction operation on the local and remote memory data and output the result to the local memory.
HcommReadReduceOnThread(thread, channel, localBuffer, remoteBuffer, count, HCOMM_DATA_TYPE_INT32, HCOMM_REDUCE_SUM);

Parent topic: Communication Operations