2018-09-24 16:06:59 -07:00
|
|
|
/*************************************************************************
|
2022-01-07 06:39:55 -08:00
|
|
|
* Copyright (c) 2015-2022, NVIDIA CORPORATION. All rights reserved.
|
2018-09-24 16:06:59 -07:00
|
|
|
*
|
|
|
|
|
* See LICENSE.txt for license information
|
|
|
|
|
************************************************************************/
|
|
|
|
|
|
|
|
|
|
#ifndef NCCL_ENQUEUE_H_
|
|
|
|
|
#define NCCL_ENQUEUE_H_
|
|
|
|
|
|
2019-11-19 14:57:39 -08:00
|
|
|
#include "comm.h"
|
2018-09-24 16:06:59 -07:00
|
|
|
#include "group.h"
|
2019-11-19 14:57:39 -08:00
|
|
|
#include "collectives.h"
|
2022-05-24 02:02:31 -07:00
|
|
|
#include "utils.h"
|
2019-03-14 19:39:20 -07:00
|
|
|
|
2021-07-08 14:12:04 -07:00
|
|
|
#define NCCL_MIN_CHANNEL_SIZE (NCCL_LL_THREAD_THRESHOLD*64)
|
|
|
|
|
#define NCCL_AGG_CHANNEL_SIZE (1LL << 21) /* 2 MiB, ideal per-channel size to fully utilize bandwidth */
|
|
|
|
|
|
2023-02-27 02:48:21 -08:00
|
|
|
ncclResult_t ncclInitKernelsForDevice(int cudaArch, size_t* maxStackSize);
|
2018-12-13 15:56:12 -08:00
|
|
|
ncclResult_t ncclEnqueueCheck(struct ncclInfo* info);
|
2022-05-24 02:02:31 -07:00
|
|
|
ncclResult_t ncclLaunchPrepare(struct ncclComm* comm);
|
|
|
|
|
ncclResult_t ncclLaunchKernelBefore_NoUncapturedCuda(struct ncclComm* comm, struct ncclKernelPlan* plan);
|
|
|
|
|
ncclResult_t ncclLaunchKernel(struct ncclComm* comm, struct ncclKernelPlan* plan);
|
|
|
|
|
ncclResult_t ncclLaunchKernelAfter_NoCuda(struct ncclComm* comm, struct ncclKernelPlan* plan);
|
|
|
|
|
ncclResult_t ncclLaunchFinish(struct ncclComm* comm);
|
2018-09-24 16:06:59 -07:00
|
|
|
|
|
|
|
|
#endif // End include guard
|
