* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
* \file ops_proto_hccl.h
* \brief huawei collective communication library ops.
*/
#ifndef OPS_BUILT_IN_OP_PROTO_INC_OPS_PROTO_HCCL_H_
#define OPS_BUILT_IN_OP_PROTO_INC_OPS_PROTO_HCCL_H_
#ifndef OPS_BUILT_IN_OP_PROTO_INC_OPS_PROTO_HCCL_H_WARN_SHOWN
#define OPS_BUILT_IN_OP_PROTO_INC_OPS_PROTO_HCCL_H_WARN_SHOWN
#warning "ops_proto_hccl.h is scheduled to be deprecated in December 2026, and will be replaced by the \
ops_proto_math.h, ops_proto_cv.h, ops_proto_nn.h, ops_proto_transformer.h, ops_proto_legacy.h. \
We apologize for any inconvenience caused and appreciate your timely migration to the new interface."
#include "graph/operator_reg.h"
namespace ge {
* @brief Outputs a tensor gathering all input tensors.
* @par Inputs:
* x: A tensor. Must be one of the following types: int8, int16, int32, int64, float16, bfloat16,
float32, uint8, uint16, uint32, float64.
* @par Attributes:
* @li rank_size: A required integer identifying the number of ranks
participating in the op.
* @li group: A required string identifying the group name of ranks
participating in the op.
* @li fusion: An optional integer identifying the fusion flag of the op.
0: no fusion; 2: fusion the ops by fusion id.
* @li fusion_id: An optional integer identifying the fusion id of the op.
* The HcomAllGather ops with the same fusion id will be fused.
* @par Outputs:
* y: A Tensor. Has the same type as "x".
* @attention Constraints:
"group" is limited to 128 characters. Use "hccl_world_group"
as the name of a world group.
*/
REG_OP(HcomAllGather)
.INPUT(x, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_BFLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.OUTPUT(y, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_BFLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.REQUIRED_ATTR(rank_size, Int)
.REQUIRED_ATTR(group, String)
.ATTR(fusion, Int, 0)
.ATTR(fusion_id, Int, -1)
.OP_END_FACTORY_REG(HcomAllGather)
* @brief Outputs a tensor gathering all input tensors.
* @par Inputs:
* @li x: A tensor. Must be one of the following types: int8, int16, int32, int64, float16, bfloat16,
float32, uint8, uint16, uint32, uint64, float64.
* @li send_count: A data. specifies current rank the the number of
elements to receive to send, only support int64.
* @li recv_counts: A list, where entry i specifies the first dimension of
elements to receive from rank i, only support int64.
* @li recv_displacements: A list, where entry i specifies the displacement
(offset from recv_data) to which data from rank i should be written, only support int64.
* @par Attributes:
* group: A required string identifying the group name of ranks
participating in the op.
* @par Outputs:
* y: A Tensor. Has the same type as "x".
* @attention Constraints:
* @li "group" is limited to 128 characters. Use "hccl_world_group"
as the name of a world group.
* @li Only the single-node system scenario of the Atlas A2 Training Series Product is supported.
*/
REG_OP(HcomAllGatherV)
.INPUT(x, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_BFLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.INPUT(send_count, TensorType({DT_INT64}))
.OUTPUT(y, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_BFLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.INPUT(recv_counts, TensorType({DT_INT64}))
.OPTIONAL_INPUT(recv_displacements, TensorType({DT_INT64}))
.REQUIRED_ATTR(group, String)
.OP_END_FACTORY_REG(HcomAllGatherV)
* @brief Outputs a tensor containing the reduction across all input tensors
passed to op.
* @par Inputs:
* x: A tensor. Must be one of the following types: int8, int16, int32, int64, float16,
float32.
* @par Attributes:
* @li reduction: A required string identifying the reduction operation to
perform.The supported operation are: "sum", "max", "min", "prod".
* @li group: A required string identifying the group name of ranks
participating in the op.
* @li fusion: An optional integer identifying the fusion flag of the op.
0: no fusion; 1 (default): fusion the ops by gradient segmentation strategy; 2: fusion the ops by fusion id.
* @li fusion_id: An optional integer identifying the fusion id of the op.
* The HcomAllReduce ops with the same fusion id will be fused.
* @par Outputs:
* y: A Tensor. Has the same type as "x".
* @attention Constraints:
* @li "group" is limited to 128 characters. Use "hccl_world_group"
as the name of a world group.
* @li For Altas 300I Duo, "prod"/"max"/"min" does not support int16
* @li For Altas A2, "prod" does not support int16/bfp16
* @li For Altas A3, "prod" does not support int16/bfp16
*/
REG_OP(HcomAllReduce)
.INPUT(x, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64}))
.OUTPUT(y, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64}))
.REQUIRED_ATTR(reduction, String)
.REQUIRED_ATTR(group, String)
.ATTR(fusion, Int, 1)
.ATTR(fusion_id, Int, -1)
.OP_END_FACTORY_REG(HcomAllReduce)
* @brief Broadcasts the input tensor in root rank to all ranks.
* @par Inputs:
* x: A list of dynamic input tensor. Must be one of the following types:
int8, int16, int32, float16, float32. It's a dynamic input.
* @par Attributes:
* @li root_rank: A required integer identifying the root rank in the op
input of this rank will be broadcast to other ranks.
* @li fusion: A required integer identifying if the op need to fusion,
0: no fusion; 2(default): fusion the ops by fusion id.
* @li fusion_id: A required integer identifying the fusion id if para fusion
is set.
* @li group: A required string identifying the group name of ranks
participating in the op.
* @par Outputs:
* y: A list of dynamic output tensor. Has the same type and length as "x".
* It's a dynamic output.
* @attention Constraints:
"group" is limited to 128 characters. Use "hccl_world_group"
as the name of a world group.
*/
REG_OP(HcomBroadcast)
.DYNAMIC_INPUT(x, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.DYNAMIC_OUTPUT(y, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.REQUIRED_ATTR(root_rank, Int)
.REQUIRED_ATTR(group, String)
.ATTR(fusion, Int, 0)
.ATTR(fusion_id, Int, -1)
.OP_END_FACTORY_REG(HcomBroadcast)
* @brief preforms reduction from others rank to rootrank
* @par Inputs:
* @li root_rank: A required integer identifying the root rank in the op
the reduction result will be on this root rank
* x: A tensor. Must be one of the following types: int8, int16, int32, int64, float16,
float32.
* @par Attributes:
* @li reduction: A required string identifying the reduction operation to
perform.The supported operation are: "sum", "max", "min", "prod".
* @li group: A required string identifying the group name of ranks
participating in the op.
* @li fusion: An optional integer identifying the fusion flag of the op.
0: no fusion; 1 (default): fusion; 2: fusion the ops by fusion id.
* @li fusion_id: An optional integer identifying the fusion id of the op.
* The HcomReduce ops with the same fusion id will be fused.
* @par Outputs:
* y: A Tensor. Has the same type as "x".
* @attention Constraints:
*"group" is limited to 128 characters. Use "hccl_world_group"
as the name of a world group.
*/
REG_OP(HcomReduce)
.INPUT(x, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64}))
.OUTPUT(y, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64}))
.REQUIRED_ATTR(root_rank, Int)
.REQUIRED_ATTR(reduction, String)
.REQUIRED_ATTR(group, String)
.ATTR(fusion, Int, 0)
.ATTR(fusion_id, Int, -1)
.OP_END_FACTORY_REG(HcomReduce)
* @brief Performs reduction across all input tensors, scattering in equal
blocks among ranks, each rank getting a chunk of data based on its rank
index.
* @par Inputs:
* x: A tensor. Must be one of the following types: int8, int16, int32, int64, float16,
float32.
* @par Attributes:
* @li reduction: A required string identifying the reduction operation to
perform. The supported operation are: "sum", "max", "min", "prod".
* @li fusion: An optional integer identifying the fusion flag of the op.
0: no fusion; 2: fusion the ops by fusion id.
* @li fusion_id: An optional integer identifying the fusion id of the op.
* The HcomReduceScatter ops with the same fusion id will be fused.
* @li group: A required string identifying the group name of ranks
participating in the op.
* @li rank_size: A required integer identifying the number of ranks
participating in the op.
* @par Outputs:
* y: A Tensor. Has the same type as "x".
* @attention Constraints:
"group" is limited to 128 characters. Use "hccl_world_group"
as the name of a world group.
*/
REG_OP(HcomReduceScatter)
.INPUT(x, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64}))
.OUTPUT(y, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64}))
.REQUIRED_ATTR(reduction, String)
.ATTR(fusion, Int, 0)
.ATTR(fusion_id, Int, -1)
.REQUIRED_ATTR(group, String)
.REQUIRED_ATTR(rank_size, Int)
.OP_END_FACTORY_REG(HcomReduceScatter)
* @brief Performs reduction across all input tensors, scattering in vary size
blocks among ranks, each rank getting a chunk of data based on its rank
index.
* @par Inputs:
* @li x: A tensor. Must be one of the following types: int8, int16, int32, float16, float32.
* @li send_counts: int64 array, where entry i specifies the first dimension number of elements to send to rank i.
* @li send_displacesments: int64 array, optional, where entry i specifies the displacement from which to send data to rank i.
If not provided, it is assumed to be contiguous memory by default.
* @li recv_count: int64 array, only one entry which specifies the first dimension number of elements of the output data.
* @par Attributes:
* @li reduction: A required string identifying the reduction operation to
perform. The supported operation are: "sum", "max", "min".
* @li group: A required string identifying the group name of ranks
participating in the op.
* @par Outputs:
* y: A Tensor. Has the same type as "x".
* @attention Constraints:
* @li "group" is limited to 128 characters. Use "hccl_world_group"
as the name of a world group.
* @li Only the single-node system scenario of the Atlas A2 Training Series Product is supported.
*/
REG_OP(HcomReduceScatterV)
.INPUT(x, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32, DT_INT16, DT_INT8}))
.INPUT(recv_count, TensorType({DT_INT64}))
.INPUT(send_counts, TensorType({DT_INT64}))
.OPTIONAL_INPUT(send_displacements, TensorType({DT_INT64}))
.OUTPUT(y, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32, DT_INT16, DT_INT8}))
.REQUIRED_ATTR(reduction, String)
.REQUIRED_ATTR(group, String)
.OP_END_FACTORY_REG(HcomReduceScatterV)
* @brief Sends the input tensor to destination rank.
* @par Inputs:
* x: A tensor. Must be one of the following types: int8, int16, int32, float16,
float32.
* @par Attributes:
* @li sr_tag: A required integer identifying the send/recv message tag. The
message will be received by the HcomReceive op with the same "sr_tag".
* @li dest_rank: A required integer identifying the destination rank.
* @li group: A string identifying the group name of ranks participating in
the op.
* @par Outputs:
* None.
* @attention Constraints:
@li "group" is limited to 128 characters. Use
"hccl_world_group" as the name of a world group.
* @li Operators HcomSend and HcomReceive have the same "sr_tag".
* @see HcomReceive
*/
REG_OP(HcomSend)
.INPUT(x, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.REQUIRED_ATTR(group, String)
.REQUIRED_ATTR(sr_tag, Int)
.REQUIRED_ATTR(dest_rank, Int)
.OP_END_FACTORY_REG(HcomSend)
* @brief Receives the tensor from source rank.
* @par Inputs:
* None.
* @par Attributes:
* @li sr_tag: A required integer identifying the send/recv message tag. The
message will be send by the HcomSend op with the same "sr_tag".
* @li src_rank: A required integer identifying the source rank.
* @li group: A required string identifying the group name of ranks
* participating in the op.
* @li shape: A required list identifying the shape of the tensor to be
received.
* @li dtype: A required integer identifying the type of the tensor to be
received. The supported types are: int8, int16, int32, float16, float32.
* @par Outputs:
* y: A tensor with type identified in "dtype".
* @attention Constraints:
@li "group" is limited to 128 characters. Use
"hccl_world_group" as the name of a world group.
* @li Operators HcomSend and HcomReceive have the same "sr_tag".
* @li "shape" should be same as the input tensor of HcomSend.
* @li "dtype" should be same as the input tensor of HcomSend.
* @see HcomSend
*/
REG_OP(HcomReceive)
.OUTPUT(y, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.REQUIRED_ATTR(group, String)
.REQUIRED_ATTR(sr_tag, Int)
.REQUIRED_ATTR(src_rank, Int)
.REQUIRED_ATTR(shape, ListInt)
.REQUIRED_ATTR(dtype, Type)
.OP_END_FACTORY_REG(HcomReceive)
* @brief All ranks send different amount of data to, and receive different
amount of data from, all ranks.
* @par Inputs:
* Five inputs, including:
* @li send_data: A tensor. the memory to send.
* @li send_counts: A list, where entry i specifies the number of elements in
send_data to send to rank i.
* @li send_displacements: A list, where entry i specifies the displacement
(offset from sendbuf) from which to send data to rank i.
* @li recv_counts: A list, where entry i specifies the number of
elements to receive from rank i.
* @li recv_displacements: A list, , where entry i specifies the displacement
(offset from recv_data) to which data from rank i should be written.
* @par Outputs:
* recv_data: A Tensor has same element type as send_data.
* @par Attributes:
* @li group: A string identifying the group name of ranks participating in
the op.
* @attention all ranks participating in the op should be full-mesh networking
using the RDMA.
*/
REG_OP(HcomAllToAllV)
.INPUT(send_data, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.INPUT(send_counts, TensorType({DT_INT64}))
.INPUT(send_displacements, TensorType({DT_INT64}))
.INPUT(recv_counts, TensorType({DT_INT64}))
.INPUT(recv_displacements, TensorType({DT_INT64}))
.OUTPUT(recv_data, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.REQUIRED_ATTR(group, String)
.OP_END_FACTORY_REG(HcomAllToAllV)
* @brief All ranks send the same amount of data to each other, and receive the same amount of data from each other.
* @par Inputs:
* @li x: A tensor. Must be one of the following types: float32, int32, int8, int16, float16,
int64, uint64, uint8, uint16, uint32, float64.
* @par Outputs:
* @li y: A Tensor. Has the same type as "x".
* @par Attributes:
* @li group: A string identifying the group name of ranks participating in
the op.
* @attention all ranks participating in the op should be full-mesh networking
using the RDMA.
*/
REG_OP(HcomAllToAll)
.INPUT(x, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.OUTPUT(y, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.REQUIRED_ATTR(group, String)
.OP_END_FACTORY_REG(HcomAllToAll)
* @brief All ranks send different amount of data to, and receive different
amount of data from, all ranks.
* @par Inputs:
* Two inputs, including:
* @li send_data: A tensor. the memory to send.
* @li send_count_matrix: A two dimensional matrix, where entry [i][j] specifies
* the number of elements in the send_data that rank i to rank j.
* @li fusion: An optional integer identifying the fusion flag of the op.
0(default): no fusion; 2: fusion the ops by fusion id.
* @li fusion_id: An optional integer identifying the fusion id of the op.
* The HcomAllToAllVC ops with the same fusion id will be fused.
* @par Outputs:
* recv_data: A Tensor has same element type as send_data.
* @par Attributes:
* @li rank: A required integer identifying the self rank.
* @li group: A string identifying the group name of ranks participating in
the op.
*/
REG_OP(HcomAllToAllVC)
.INPUT(send_data, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.INPUT(send_count_matrix, TensorType({DT_INT64}))
.OUTPUT(recv_data, TensorType({DT_FLOAT, DT_INT32, DT_INT8, DT_INT16, DT_FLOAT16, DT_INT64, DT_UINT64,
DT_UINT8, DT_UINT16, DT_UINT32, DT_FLOAT64}))
.REQUIRED_ATTR(rank, Int)
.REQUIRED_ATTR(group, String)
.ATTR(fusion, Int, 0)
.ATTR(fusion_id, Int, -1)
.OP_END_FACTORY_REG(HcomAllToAllVC)
}
#endif
#endif
