BF16 Batched GEMM Functions Available?

[zhongyilin926]

Hi everyone, I'm benchmarking BF16 GEMM functions of BLIS, and so far I only see a series of BF16 GEMM function which are aocl_gemm_xx. I was wondering if there are any BF16 batched GEMM functions available?

[BhaskarNallani]

Hi @zhongyilin926 ,

Currently we only support two bf16 GEMM API's aocl_gemm_bf16bf16f32o[ f32 | bf16 ] ( ) with a support of fusing multiple post-ops along with their combinations.

We have plans to support batch GEMM for all low precision data types in upcoming releases.

Thank you.

[moderato]

Hi @zhongyilin926 ,

Currently we only support two bf16 GEMM API's aocl_gemm_bf16bf16f32o[ f32 | bf16 ] ( ) with a support of fusing multiple post-ops along with their combinations.

We have plans to support batch GEMM for all low precision data types in upcoming releases.

Thank you.

Hi @BhaskarNallani , thank you for the reply. I was wondering if you could be a bit more specific? Are they gonna be supported right in the next release? And if so is there an ETA for the next release? Thank you!

[BhaskarNallani]

@zhongyilin926 ,
We added support for batch gemm now for all data types in dev branch for now and will be released as part of AOCL5.1 in near future. Please check latest code from dev branch ( https://github.com/amd/blis/tree/dev ) where we update it bi-weekly.