[QST] Do batched gemm/gemv support multiple alpha and beta?

[SubjectNoi]

Given a batched gemm/gemv with say 32 batch, does cutlass support assigning 32 different alpha and beta to 32 batched gemm/gemv? For now I only see the usage of sharing one {alpha, beta} among 32 batch.

[hwu36]

gemv code is here https://github.com/NVIDIA/cutlass/blob/main/include/cutlass/gemm/kernel/gemv.h very simple. easy to hack. (@NVJiangShao)

you can use newly released evt to support vector alpha/beta. check https://github.com/NVIDIA/cutlass/blob/main/examples/47_ampere_gemm_universal_streamk/ampere_gemm_universal_streamk_broadcast.cu. you need #1120 to run it correctly (@apuaaChen). the alternative way is to use the old way like https://github.com/NVIDIA/cutlass/blob/main/test/unit/gemm/device/gemm_f16t_f16n_f16t_tensor_op_f16_broadcast_sm80.cu

[mnicely]

@SubjectNoi is your question resolved?

[mnicely]

Closing due to inactivity. Feel free to reopen if needed.