diff --git a/include/cutlass/gemm/collective/sm90_mma_tma_gmma_ss_warpspecialized_fp8_blockwise_scaling.hpp b/include/cutlass/gemm/collective/sm90_mma_tma_gmma_ss_warpspecialized_fp8_blockwise_scaling.hpp index a931e988..5b8f1059 100644 --- a/include/cutlass/gemm/collective/sm90_mma_tma_gmma_ss_warpspecialized_fp8_blockwise_scaling.hpp +++ b/include/cutlass/gemm/collective/sm90_mma_tma_gmma_ss_warpspecialized_fp8_blockwise_scaling.hpp @@ -1001,7 +1001,7 @@ struct CollectiveMma< // Advance smem_pipe_read and smem_pipe_release ++smem_pipe_release; } - { + if (k_tile_count == 1) { pipeline.consumer_wait(smem_pipe_read, barrier_token); //