Vivian Zhang

> [@MaheshRavishankar](https://github.com/MaheshRavishankar) [@yzhang93](https://github.com/yzhang93) I'm going to set that last IR up in turbine and run some benchmarks to see how it looks. I'm a bit confused about the current status....

A summary of things we have tried so far for this issue. The main issue for non-unit stride convs is that `tensor.insert_slice` with non-unit strides can't be fused into the...

> This issue is fixed now right? The original error is solved as we switched to use multiple dispatches. The remaining issues are just performance wise, and we can open...

For 1x1 filter cases, @zjgarvey proposed a new IR expression which has better performance. ``` // N, Ho, Wo, C, Kh, Kw, F #map = affine_map (d0, d1, d2, d6)>...