Hi Animesh, The problem is that I need padding added in the middle of TIR on my (transformed) data tensor.
I.e., something like ``` A1 = im2col(A) A2 = pad(A1) C_padded = te.compute([M,N], lambda i, j : sum(A2[i,k]*B[k,j], k) C = unpad(C)+requantization ``` Then I tile on `C` and tensorize on the inner tile (which is where the problem started). Note that I cannot fuse the requantization to the main computation because of the `unpad` Also, it would be nice to not pad `A` at all, but to work on a solution that can automatically detect the borders and invoke different kind of tensorizations (if provided) or use scalar computation for the borders (if multiple `tensorizations` are not provided). In this way I don't need unpadding and the computation could become: ``` A1 = im2col(A) C = te.compute([M,N], lambda i, j : sum(A1[i,k]*B[k,j], k)) + requantization #tensorization handles everything automatically ``` What do you think? --- [Visit Topic](https://discuss.tvm.ai/t/loop-partitioning-padding-and-tensorization/7753/3) to respond. You are receiving this because you enabled mailing list mode. To unsubscribe from these emails, [click here](https://discuss.tvm.ai/email/unsubscribe/c5fcc72c3588f10fc2346acc6ba2cb5e8a6a6064681dc5d80f5891be005c2c10).
