mirror of
https://github.com/deepseek-ai/DeepEP
synced 2025-06-26 18:28:11 +00:00
Fully remove barrier FIFO designs (#200)
* 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:
@@ -12,13 +12,13 @@ namespace deep_ep {
|
||||
namespace intranode {
|
||||
|
||||
template<int kNumRanks>
|
||||
__global__ void barrier(int** task_fifo_ptrs, int head, int rank) {
|
||||
barrier_device<kNumRanks>(task_fifo_ptrs, head, rank);
|
||||
__global__ void barrier(int** barrier_signal_ptrs, int rank) {
|
||||
barrier_block<kNumRanks>(barrier_signal_ptrs, rank);
|
||||
}
|
||||
|
||||
void barrier(int** task_fifo_ptrs, int head, int rank, int num_ranks, cudaStream_t stream) {
|
||||
void barrier(int** barrier_signal_ptrs, int rank, int num_ranks, cudaStream_t stream) {
|
||||
#define BARRIER_LAUNCH_CASE(ranks) \
|
||||
LAUNCH_KERNEL(&cfg, barrier<ranks>, task_fifo_ptrs, head, rank); \
|
||||
LAUNCH_KERNEL(&cfg, barrier<ranks>, barrier_signal_ptrs, rank); \
|
||||
break
|
||||
|
||||
SETUP_LAUNCH_CONFIG(1, 32, stream);
|
||||
|
||||
Reference in New Issue
Block a user