Fix#45 : All_gather and reduce_scatter algorithm bandwidth was
computed as time/count*(nranks-1) which is not consistent with the
way we compute it for other collectives.
This change makes algbw higher; busbw is unchanged.
These APIs launche RCCL kernel implementation by default. If environmental
variable RCCL_ALLTOALL_KERNEL_DISABLE=1, then the APIs use wrapper around
ncclSend and ncclRecv.
In some cases, the MPI library is not in $(MPI_HOME)/lib but
in $(MPI_HOME)/lib64. For example, on RedHat like Linux system
(CentOS, Amazon Linux), and MPI is installed by yum or rpm.
Under such circumstance, the current make file will cause failure.
This patch address this issue by adding -L$(MPI_HOME)/lib64 to
NVLDFLAGS in src/Makefile.
Signed-off-by: Wei Zhang <wzam@amazon.com>
* Adding initial unit test and Jenkins code.
Fixing scope of unit tests
Adding unit tests and files for CI
Fixing Jenkinsfile
* Removing typos from Jenkinsfile
* Making some fixes to the Jenkins file; temporarily disabling MPI
* Making corrections to Jenkinsfile
* Correcting dockerNodes entry in Jenkinsfile
* Fixed Jenkinsfile for CI
* Correcting Jenkinsfile for CI
* Updating README to include instructions on how to run unit tests.
Major rework to merge most of the changes from the NCCL internal
tests into the public ones
Added "-m <agg_iters>" operation aggregation option.
Data integrity checking is now much more performant at scale.
Startup times at scale are improved.
Test latency units are now displayed in usec.
Wenkai Du
Wenkai Du
Luke Yeager
Sylvain Jeaugey
Luke Yeager
Sourav Chakraborty
Wei Zhang
Gilbert Lee
Stanley Tsang
David Addison