HcclAllGather

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	√

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.

For the Atlas inference products, only the Atlas 300I Duo inference card is supported.

Function

Re-sorts the inputs of all ranks in the communicator by rank ID, combines the inputs, and sends the results to the outputs of all ranks.

For the AllGather operation, each rank receives a set of data that is resorted based on a rank ID, that is, AllGather outputs of all ranks are the same.

Prototype

HcclResult HcclAllGather(void *sendBuf, void *recvBuf, uint64_t sendCount, HcclDataType dataType, HcclComm comm, aclrtStream stream)

Parameters

Parameter	Input/Output	Description
sendBuf	Input	Address of the send buffer.
recvBuf	Output	Address of the buffer to receive collective communication result.
sendCount	Input	sendBuf size involved in the AllGather operation. The size of recvBuf data is calculated as: sendCount × rank size.
dataType	Input	Data type of the AllGather operation, which is of the HcclDataType type. Atlas 350 Accelerator Card: The supported data types are int8, uint8, int16, uint16, int32, uint32, int64, uint64, float8-e5m2, float8-e4m3, float8-e8m0, hifloat8, float16, float32, float64, and bfp16. Atlas A3 training product/Atlas A3 inference product: The supported data types are int8, uint8, int16, uint16, int32, uint32, int64, uint64, float16, float32, float64, and bfp16. Atlas A2 training product/Atlas A2 inference product: The supported data types are int8, uint8, int16, uint16, int32, uint32, int64, uint64, float16, float32, float64, and bfp16. Atlas training product: The supported data types are int8, uint8, int16, uint16, int32, uint32, int64, uint64, float16, float32, and float64. Atlas 300I Duo Inference Card: The supported data types are int8, uint8, int16, uint16, int32, uint32, int64, uint64, float16, float32, and float64.
comm	Input	Communicator where the operation is performed.
stream	Input	Stream of the rank.

Returns

HcclResult: HCCL_SUCCESS on success, or else failure.

Constraints

The ranks must have the same sendCount and dataType.
Atlas 300I Duo Inference Card: Only the single-server scenario is supported, and a maximum of 16 Atlas 300I Duo Inference Cards (32 NPUs) can be deployed on a single server.

Call Example

// Allocate device memory for collective communication.
void *sendBuf = nullptr, *recvBuf = nullptr;
uint32_t rankSize = 8;
uint64_t sendCount = 1;  // Number of data elements sent by each rank.
size_t sendSize = sendCount * sizeof(float);
size_t recvSize = rankSize * sendCount * sizeof(float);
aclrtMalloc(&sendBuf, sendSize, ACL_MEM_MALLOC_HUGE_ONLY);
aclrtMalloc(&recvBuf, recvSize, ACL_MEM_MALLOC_HUGE_ONLY);

// Initialize the communicator and streams.
HcclComm hcclComm;
HcclCommInitRootInfo(rankSize, &rootInfo, devId, &hcclComm);

// Create a task flow.
aclrtStream stream;
aclrtCreateStream(&stream);

// Execute AllGather to stitch sendBuf of all ranks in the communicator by the sequence of rank IDs, and send the result to recvBuf of all ranks.
HcclAllGather(sendBuf, recvBuf, sendCount, HCCL_DATA_TYPE_FP32, hcclComm, stream);
// Wait until the collective communication task in the task flow is complete.
aclrtSynchronizeStream(stream);

// Free resources.
aclrtFree(sendBuf);          // Free the device memory.
aclrtFree(recvBuf);          // Free the device memory.
aclrtDestroyStream(stream);  // Destroy the task flow.
HcclCommDestroy(hcclComm);   // Destroy the communicator.

Parent topic: Collective Communication