mengbingrock edited a comment on pull request #8605:
URL: https://github.com/apache/tvm/pull/8605#issuecomment-1039109093
Hi @Tantalus13A98B5F
Your work is very helpful!
I'm evaluating the performance of sparse kernels. I noticed that PR
currently only supports padding = 1 case if I'm right. I've been struggling to
add padding = 0 support.
I've modified the conv defined in `spconv2d_3x3_nhwc(cfg, data, wdat, wind,
wptr, layout="NHWC")` and I think it could produce the correct result, or at
least its dimension is correct. But it reports dimension error else where.
I will appreciate it if any guidance could be provided.
Thank you
The
```
sparse_func, params = relay.data_dep_optimization.bsr_conv2d.convert2( func,
{}, (2, 1), 0.3, "NHWC", 3)
```
the sparse_func returned has input/output shape
```
fn (%data: Tensor[(1, 118, 118, 128), float32]) -> Tensor[(1, 118, 118,
128), float32]
```
but it should be (1, 118, 118, 128)->(1,116,116,128) for padding = 0,
kernel_size = 3 case.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
