Closed #4052.
--
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/dmlc/tvm/issues/4052#event-2753595541
I have chatted with @minminsun and his team these days. Just as then mentioned
https://github.com/dmlc/tvm/issues/4105#issuecomment-542032766. We can have
different frontends but only one backend. In my previous implement, users can
only use fragments with 16x16x16 shape and row-major layout. To
@soiferj Thank you for such a helpful comment. I have just made the extension
into the schedule for BatchMatMul. You can check the schedule in my fork repo:
https://github.com/Hzfengsy/tvm/blob/master/tests/python/unittest/test_schedule_tensor_core.py#L101
--
You are receiving this because you
Would it be easy to extend your gemm schedule into a schedule for BatchMatMul?
That would help round out the TensorCore story for matrix multiplication.
--
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/dmlc
@Hzfengsy Sure, we will show the code as well as a sample schedule very soon.
It's being under internal review now. As you will see, the schedule for
TensorCore CodeGen looks no different than a normal matmul schedule for GPU.
Everything is done in IR passes including matrix_a/matrix_b/accumulat
@yangjunpro Really happy to see another solution for TensorCore.
You are right! I just extend tvm intrinsic to support it. It does cause
programmers who write the schedule some trouble. It is not easy to write a
high-performance schedule.
I'm really curious about how to use IR passes to recogn
@tmoreau89 Exactly! For now, we use the NCHWnc layout, the same layout with VTA.
--
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/dmlc/tvm/issues/4052#issuecomment-537816661
cc @Laurawly
--
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/dmlc/tvm/issues/4052#issuecomment-537808438
Nice to see other folks working on adding TensorCore support into TVM, we have
also been working on enhancing TVM to incorporate TensorCore schedule support.
If my understanding is correct, @Hzfengsy your solution is based on extending
TVM's intrinsic while our solution put most of the complexit
Very welcome work @Hzfengsy !
--
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/dmlc/tvm/issues/4052#issuecomment-537791018
Tensor Core is a defining feature of the NVIDIA new Volta and Turing GPU
Architecture, which gives a massive boost for matrix multiplication and
convolution. Tensor Cores enable us to use mixed-precision to achieve higher
throughput without sacrificing accuracy.
## Tensor Core Overview
Each Ten
11 matches
Mail list logo
