DeepSeek/DeepEP
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepEP synced 2025-06-26 18:28:11 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Use IBGDA only (#177)

Browse Source
This commit is contained in:
Shangyan Zhou
2025-05-28 16:40:14 +08:00
committed by GitHub
parent aae9fa9a6d
commit 9fe9021f29
3 changed files with 77 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
csrc/kernels/runtime.cu
View File

@@ -58,12 +58,6 @@ int init(const std::vector<uint8_t> &root_unique_id_val, int rank, int num_ranks
EP_HOST_ASSERT(cpu_rdma_team != NVSHMEM_TEAM_INVALID);
}
// TODO: we still use `nvshmem_barrier` under IBRC mode, which should be switch to IBGDA mode later
nvshmemi_device_host_state_t* dev_state_ptr = nullptr;
CUDA_CHECK(cudaGetSymbolAddress(reinterpret_cast<void**>(&dev_state_ptr), nvshmemi_device_state_d));
bool ibgda_is_initialized = false;
CUDA_CHECK(cudaMemcpy(&dev_state_ptr->ibgda_is_initialized, &ibgda_is_initialized, sizeof(bool), cudaMemcpyHostToDevice));
nvshmem_barrier_all();
return nvshmem_my_pe();
}