Improve throughput for about 20%. Also remove P2P over PCIe which was
left enabled at initial release.
Signed-off-by: Wenkai Du <wenkai.du@amd.com>
[ROCm/rccl commit: f45566a8bd]
Makes a change to allow the user to select between the static CUDA
runtime library (default) and the dynamic CUDA runtime library. Does
this by allowing `CUDARTLIB` to be overridden.
[ROCm/rccl commit: 60a586ded9]
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]
Current SHM object name would only use pidHash and ranks as
identification, which would collide each other when program runs with
multiple communicators. Here we added commId info into pidHash, it makes
'pidHash'es of different communicators keeping in same process will be
distincted with each other.
[ROCm/rccl commit: 161763aab2]
The shared memory filename was only based on the destination. While
this was OK for rings since only one rank would send data to a given
rank, it would crash with trees because they communicate in both
directions.
Co-authored-by: Rong Ou <rong.ou@gmail.com>
[ROCm/rccl commit: 14e0cf644b]
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]
When initNet fails, we should not print the backtrace as it is
supposed to be normal operation (falling back to sockets)
[ROCm/rccl commit: 57368189e1]
Whether the network supported GPU Direct RDMA or not was ignored,
causing sockets to break when cards were local enough that NCCL
tried to use it.
[ROCm/rccl commit: 4b39a4cf91]
This is to handle EINTR/EGAIN properly (issue #137), and also
make the code consistent with the rest.
Unfortunately posix_fallocate and mmap do not follow the classic
return code/errno pattern, so we need to write wrappers around those
functions.
[ROCm/rccl commit: eed8218e17]
SYSCHECKVAL was not retrying when a retry was needed. Since not all
calls are inside a loop, that means we could silently miss an
EINTR/EAGAIN return code.
Also rework the socket connection code and improve error reporting.
[ROCm/rccl commit: 302d538b73]
Dynamically load external network from libnccl-net.so.
Add init function in networks.
Move PCI scoring to net.cu, only ask transport to provide a path.
Simplify CUDA PCI path detection.
Add dummy external network
[ROCm/rccl commit: 0d3a20f96d]
Saad Rahim
saadrahim
Wenkai Du
Yaxun (Sam) Liu
Wenkai Du
Rajat Chopra
gilbertlee-amd
Aaron Enye Shi
Stanley Tsang
Gilbert Lee
David Addison
jakirkham
Gustavo Alvarez
Cao Zongyan
Rong Ou
Sylvain Jeaugey
Christian Sigg
Christian Sigg
Ke Wen
Alex Sergeev