David Addison
17c8317cb1
Added detection of IBM/Power NVLink bridge device.
Add NUMA support to PCI distance calculations.
Added NCCL_IGNORE_CPU_AFFINITY env var.
Fix memory leaks; GithubIssue#180
Compiler warning fix; GithubIssue#178
Replace non-standard variable length arrays. GithubIssue#171
Fix Tree+Shared Memory crash. GithubPR#185
Fix LL cleanup hang during long running DL jobs.
Fix NCCL_RINGS environment variable handling.
Added extra checks to catch repeat calls to ncclCommDestroy() GithubIssue#191
Improve bootstrap socket connection reliability at scale.
Fix hostname hashing issue. GithubIssue#187
Code cleanup to rename all non device files from *.cu to *.cc
[ROCm/rccl commit: f40ce73e89]
52 строки
1.4 KiB
C++
52 строки
1.4 KiB
C++
/*************************************************************************
|
|
* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
|
|
*
|
|
* See LICENSE.txt for license information
|
|
************************************************************************/
|
|
|
|
#ifndef NCCL_ALLOC_H_
|
|
#define NCCL_ALLOC_H_
|
|
|
|
#include "nccl.h"
|
|
#include "checks.h"
|
|
#include <sys/mman.h>
|
|
|
|
static inline ncclResult_t ncclCudaHostAlloc(void** ptr, void** devPtr, size_t size) {
|
|
CUDACHECK(cudaHostAlloc(ptr, size, cudaHostAllocMapped));
|
|
memset(*ptr, 0, size);
|
|
*devPtr = *ptr;
|
|
return ncclSuccess;
|
|
}
|
|
|
|
static inline ncclResult_t ncclCudaHostFree(void* ptr) {
|
|
CUDACHECK(cudaFreeHost(ptr));
|
|
return ncclSuccess;
|
|
}
|
|
|
|
template <typename T>
|
|
static ncclResult_t ncclCalloc(T** ptr, size_t nelem) {
|
|
void* p = malloc(nelem*sizeof(T));
|
|
if (p == NULL) {
|
|
WARN("Failed to malloc %ld bytes", nelem*sizeof(T));
|
|
return ncclSystemError;
|
|
}
|
|
memset(p, 0, nelem*sizeof(T));
|
|
*ptr = (T*)p;
|
|
return ncclSuccess;
|
|
}
|
|
|
|
template <typename T>
|
|
static ncclResult_t ncclCudaCalloc(T** ptr, size_t nelem) {
|
|
CUDACHECK(cudaMalloc(ptr, nelem*sizeof(T)));
|
|
CUDACHECK(cudaMemset(*ptr, 0, nelem*sizeof(T)));
|
|
return ncclSuccess;
|
|
}
|
|
|
|
template <typename T>
|
|
static ncclResult_t ncclCudaMemcpy(T* dst, T* src, size_t nelem) {
|
|
CUDACHECK(cudaMemcpy(dst, src, nelem*sizeof(T), cudaMemcpyDefault));
|
|
return ncclSuccess;
|
|
}
|
|
|
|
#endif
|