mirror of
https://github.com/deepseek-ai/DeepGEMM
synced 2025-05-05 08:44:22 +00:00
Minor update
This commit is contained in:
parent
d7c068d467
commit
24517316af
@ -276,7 +276,7 @@ fp8_gemm_kernel(__nv_bfloat16* gmem_d, float* scales_b, int* grouped_layout,
|
||||
|
||||
// Empty barrier arrival
|
||||
auto empty_barrier_arrive = [&](int s) {
|
||||
if (kNumTMAMulticast == 1) {
|
||||
if constexpr (kNumTMAMulticast == 1) {
|
||||
lane_idx == 0 ? empty_barriers[s]->arrive() : void();
|
||||
} else {
|
||||
auto target_cta = scheduler.is_peer_cta_alive ? lane_idx : cute::block_rank_in_cluster();
|
||||
|
Loading…
Reference in New Issue
Block a user