AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

2.4.2-1

Browse Source 
Add tree algorithms for allreduce to improve performance at scale.
Add ncclCommAbort() and ncclCommGetAsyncError() to properly handle
network errors and be permit recover.
Detect initial CPU affinity and no longer escape it.


[ROCm/rccl commit: 1450d42675]
This commit is contained in:
Sylvain Jeaugey
2018-12-13 15:56:12 -08:00
parent 3a7974d4c4
commit ffc55560a1
66 changed files with 3746 additions and 3251 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/include/nvmlwrap.h
+7 -11
View File
@@ -7,7 +7,7 @@
#ifndef NCCL_NVMLWRAP_H_
#define NCCL_NVMLWRAP_H_
#include "core.h"
#include "nccl.h"
//#define NVML_DIRECT 1
#ifdef NVML_DIRECT
@@ -32,14 +32,6 @@ static ncclResult_t wrapNvmlDeviceGetIndex(nvmlDevice_t device, unsigned* index)
NVMLCHECK(nvmlDeviceGetIndex(device, index));
return ncclSuccess;
}
static ncclResult_t wrapNvmlDeviceSetCpuAffinity(nvmlDevice_t device) {
NVMLCHECK(nvmlDeviceSetCpuAffinity(device));
return ncclSuccess;
}
static ncclResult_t wrapNvmlDeviceClearCpuAffinity(nvmlDevice_t device) {
NVMLCHECK(nvmlDeviceClearCpuAffinity(device));
return ncclSuccess;
}
static ncclResult_t wrapNvmlDeviceGetHandleByIndex(unsigned int index, nvmlDevice_t *device) {
NVMLCHECK(nvmlDeviceGetHandleByIndex(index,device));
return ncclSuccess;
@@ -61,6 +53,10 @@ static ncclResult_t wrapNvmlDeviceGetNvLinkCapability(nvmlDevice_t device, unsig
NVMLCHECK(nvmlDeviceGetNvLinkCapability(device, link, capability, capResult));
return ncclSuccess;
}
static ncclResult_t wrapNvmlDeviceGetMinorNumber(nvmlDevice_t device, unsigned int* minorNumber) {
NVMLCHECK(nvmlDeviceGetMinorNumber(device, minorNumber));
return ncclSuccess;
}
#else
// Dynamically handle dependencies on NVML
@@ -136,14 +132,14 @@ ncclResult_t wrapNvmlInit(void);
ncclResult_t wrapNvmlShutdown(void);
ncclResult_t wrapNvmlDeviceGetHandleByPciBusId(const char* pciBusId, nvmlDevice_t* device);
ncclResult_t wrapNvmlDeviceGetIndex(nvmlDevice_t device, unsigned* index);
ncclResult_t wrapNvmlDeviceSetCpuAffinity(nvmlDevice_t device);
ncclResult_t wrapNvmlDeviceClearCpuAffinity(nvmlDevice_t device);
ncclResult_t wrapNvmlDeviceGetHandleByIndex(unsigned int index, nvmlDevice_t *device);
ncclResult_t wrapNvmlDeviceGetPciInfo(nvmlDevice_t device, nvmlPciInfo_t* pci);
ncclResult_t wrapNvmlDeviceGetNvLinkState(nvmlDevice_t device, unsigned int link, nvmlEnableState_t *isActive);
ncclResult_t wrapNvmlDeviceGetNvLinkRemotePciInfo(nvmlDevice_t device, unsigned int link, nvmlPciInfo_t *pci);
ncclResult_t wrapNvmlDeviceGetNvLinkCapability(nvmlDevice_t device, unsigned int link,
nvmlNvLinkCapability_t capability, unsigned int *capResult);
ncclResult_t wrapNvmlDeviceGetMinorNumber(nvmlDevice_t device, unsigned int* minorNumber);
#endif // NVML_DIRECT
#endif // End include guard