I am wondering if NCCL reduce and allreduce support user defined operations?
I looked at the nccl.h, the ncclRedOp_t is a enum, not a function pointer, so I guess the answer is No?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| reduction operations in cuda | 1 | 1771 | January 22, 2010 | |
| can NCCL be used in distributed environment? across machines. | 0 | 495 | August 10, 2018 | |
| can NCCL be used in distributed environment? across machines. | 0 | 465 | August 10, 2018 | |
| Fast Multi-GPU collectives with NCCL | 14 | 1171 | May 11, 2018 | |
| reduction operation | 8 | 12665 | September 10, 2010 | |
| Reduce kernel in OpenCL | 3 | 12454 | June 25, 2009 | |
| nccl - can we sum up all the values of an array on 1 device GPU to obtain the sum | 1 | 554 | September 4, 2017 | |
| NCCL and D2D data moving across GPU devices | 0 | 1184 | October 28, 2017 | |
| How can I tell whether NCCL is using PCIe or IB network interface while doing AllReduce? | 0 | 774 | March 6, 2020 | |
| Proccess block when call Nccl reduce | 1 | 789 | May 19, 2018 |