GitLab

* Add low-latency kernel usage flag

* Update comments

Allow using MNNVL

* 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

Co-authored-by: Shangyan Zhou <sy.zhou@deepseek.com>

Fix notify_dispatch: using warp 0 to issue send

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

Low-latency P2P code cleanup and bug fixed

* Add 10.0 to TORCH_CUDA_ARCH_LIST
Signed-off-by: Guyue Huang <guyueh@nvidia.com>

* Revert csrc/CMakeLists change; in setup.py make TORCH_CUDA_ARCH_LIST configurable
Signed-off-by: Guyue Huang <guyueh@nvidia.com>

---------
Signed-off-by: Guyue Huang <guyueh@nvidia.com>

Support hidden size 4096

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

Feat: enhance nvidia peer memory detection

Shuffling the starting index of target rank for different ranks and channels

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

Fix DeepEP cannot be used together with code that needs GIL such as Mooncake transfer engine

Add Infrawaves' fork to README.

Support multi-QP for normal kernels

Merge branch 'trmt/internode_multi_qp' of github.com:deepseek-ai/DeepEP into trmt/internode_multi_qp

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+.

Fix typo in nvshmem.patch