GitLab

* Add automatic warp count control for low-latency dispatch

* Add automatic warp count control for low-latency combine

* More assertions

let the sender SM use the channel_id, and the receiver SM use channel_id + num_channels

* Update README

* Update `setup.py`

* Fix headers

* Add `DISABLE_NVSHMEM` for APIs

* Fix launch

* Fix TMA settings

* Fix TMA usages

* Fix dlink

* Separate layout kernels

* Update version

* Add `is_sm90_compiled`

* Fix tests

* Add NVLink connection checks

* Update README

* Fix tests

* Add some comments

* Minor fix

* Minor fix

* Fix bugs

* Fully remove FIFO slots

* Fully remove FIFO buffers

* Minor fix styles

* Fix some typos

* Bugs fixed

* Cleanup `ibgda_poll_cq`

* Add low-latency kernel usage flag

* Update comments

* Update CMake files

* Use TMA instead of LD/ST for intranode dispatch

* Use TMA instead of LD/ST for intranode combine

* Adjust configs

* Test default configs as well

* More warps for combine

* Add inter-thread fence

* Enable more warps

* Do not use TMA for senders

* Update configs

* Remove useless wait

Signed-off-by: wzc.wuzhicheng <wzc.wuzhicheng@linux.alibaba.com>

Co-authored-by: zhyncs <me@zhyncs.com>
Co-authored-by: yinfan98 <1106310035@qq.com>

To mitigate incast congestion, shuffle the starting index of target rank for different ranks and channels
Signed-off-by: wangfakang <fakangwang@gmail.com>

In the Internode Normal Kernel, when using nvshmem ibrc for RDMA data transmission, a single QP is used for data transfer between two GPUs, which limits kernel performance in network card dual-port and RoCE network scenarios.

In our optimized Internode Normal Kernel, we implemented multiple QPs for data transmission between two GPUs, setting a different QP for each channel. Additionally, we modified the transmission method from IBRC to IBGAD.

Through these optimizations, the Internode Normal Kernel achieves optimal performance in both H800 and H20 environments, with RDMA transmission performance nearly reaching the physical network performance limit. Using the current default statistical method, in 4-node H800 and H20 environments, RDMA performance can reach 60GB/s+.

For the SMs which calculate metadata in notify_dispatch, each warp in the SM is used to calculate the metadata of one channel. The default configuration is 8 warps for 10 channels, which needs two rounds of loop. Maybe the number of warps can be configured to the number of the channels so that one loop is enough.