How to lower to matrix instructions?

[xqdan]

Hi, all,

How to lower tensor to matrix instructions, like 16*16 matrix/tensor multiply or add, rather than two nests loops with operations in tvm?

Thanks!

[tqchen]

this is a planned feature that is yet to be officially announced so please stay tuned

[derisavi]

I'm interested in this too. Do you have any rough timeline on developing this feature or the required infrastructure?

[tqchen]

All the elements are in, it is mainly effort of documentation and testing

[xqdan]

that's great!

[tqchen]

close this for now and will update when more documents get in