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

Fully remove barrier FIFO designs (#200)

Browse Source 
* Fully remove FIFO slots

* Fully remove FIFO buffers

* Minor fix styles

* Fix some typos

* Bugs fixed

* Cleanup `ibgda_poll_cq`
This commit is contained in:
Chenggang Zhao
2025-06-10 16:23:20 +08:00
committed by GitHub
parent a16af40531
commit 8da2d7b38d
10 changed files with 121 additions and 181 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
csrc/deep_ep.hpp
View File

@@ -51,10 +51,9 @@ private:
// After IPC/NVSHMEM synchronization, this flag will be true
bool available = false;
// Task fifo
int head = 0;
int* task_fifo_ptrs[NUM_MAX_NVL_PEERS] = {nullptr};
int** task_fifo_ptrs_gpu = nullptr;
// Barrier signals
int* barrier_signal_ptrs[NUM_MAX_NVL_PEERS] = {nullptr};
int** barrier_signal_ptrs_gpu = nullptr;
// Workspace
void* workspace = nullptr;
@@ -75,9 +74,6 @@ private:
volatile int* low_latency_usage_flag = nullptr;
int* low_latency_usage_flag_mapped = nullptr;
private:
void move_fifo_slots(int num_slots = 1);
public:
Buffer(int rank, int num_ranks, int64_t num_nvl_bytes, int64_t num_rdma_bytes, bool low_latency_mode);