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

2.8.3-1

Browse Source 
Optimization for Tree allreduce on A100.
Improve aggregation performance.
Use shared buffers for inter-node send/recv.
Add NVTX profiling hooks.
Accelerate alltoall connections by merging communication for all
channels.
Add support for one hop communication through NVLink, for faster
send/recv communication on cubemesh topologies like DGX-1.
Improve alltoall scheduling to better balance intra/inter node
communication.
Increase send/recv parallelism by 8x, each warp sending or
receiving to a different peer.
Net: move to v4.
Net: make flush operation asynchronous to accelerate alltoall.
Net: define maximum number of requests.
Fix hang when using LL128 protocol after 2^31 steps.
Fix #379 : topology injection failing when using less GPUs than
described in the XML.
Fix #394 : protocol mismatch causing hangs or crashes when using
one GPU per node.


[ROCm/rccl commit: 920dbe5b35]
This commit is contained in:
Sylvain Jeaugey
2020-09-04 14:35:05 -07:00
parent 0277094dd2
commit a8908b34ee
90 changed files with 11172 additions and 3209 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/include/nvmlwrap.h
+1 -15
View File
@@ -1,5 +1,5 @@
/*************************************************************************
* Copyright (c) 2015-2019, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2015-2020, NVIDIA CORPORATION. All rights reserved.
*
* See LICENSE.txt for license information
************************************************************************/
@@ -45,14 +45,6 @@ static ncclResult_t wrapNvmlDeviceGetIndex(nvmlDevice_t device, unsigned* index)
NVMLCHECK(nvmlDeviceGetIndex(device, index));
return ncclSuccess;
}
static ncclResult_t wrapNvmlDeviceGetHandleByIndex(unsigned int index, nvmlDevice_t *device) {
NVMLCHECK(nvmlDeviceGetHandleByIndex(index,device));
return ncclSuccess;
}
static ncclResult_t wrapNvmlDeviceGetHandleByPciInfo(nvmlDevice_t device, nvmlPciInfo_t* pci) {
NVMLCHECK(nvmlDeviceGetPciInfo(device, pci));
return ncclSuccess;
}
static ncclResult_t wrapNvmlDeviceGetNvLinkState(nvmlDevice_t device, unsigned int link, nvmlEnableState_t *isActive) {
NVMLCHECK(nvmlDeviceGetNvLinkState(device, link, isActive));
return ncclSuccess;
@@ -66,10 +58,6 @@ 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;
}
static ncclResult_t wrapNvmlDeviceGetCudaComputeCapability(nvmlDevice_t device, int* major, int* minor) {
NVMLCHECK(nvmlDeviceGetCudaComputeCapability(device, major, minor));
return ncclSuccess;
@@ -150,12 +138,10 @@ ncclResult_t wrapNvmlShutdown(void);
ncclResult_t wrapNvmlDeviceGetHandleByPciBusId(const char* pciBusId, nvmlDevice_t* device);
ncclResult_t wrapNvmlDeviceGetIndex(nvmlDevice_t device, unsigned* index);
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);
ncclResult_t wrapNvmlDeviceGetCudaComputeCapability(nvmlDevice_t device, int* major, int* minor);
#endif // NVML_DIRECT