[BYOC] Add GEMM kernel from FasterTransformer as submodule

[masahi]

I extracted fp16 A - int8/4 GEMM kernel from FasterTransformer (see NVIDIA/cutlass#911) to make it easier to build and integrate into TVM. The code has been extracted and cleaned in the repo under tlc-pack and it is being added as a submodule.

A follow-up PR will update the CUTLASS BYOC to support offloading to this kernel. It is going to be useful for weight-quantized LLM inference.

Please review the license stuff etc @tqchen @junrushao @vinx13 @sunggg

[tvm-bot]

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

• cc @billishyahao _{See #10317 for details}

_{Generated by tvm-bot}

[masahi]

hmm the cutlass revision which is submoduled by https://github.com/tlc-pack/cutlass_fpA_intB_gemm is not pulled by the CI apparently. Does anyone know how to tell the CI to do git submodule update --init --recursive?

[yzh119]

LGTM