Han-Chung Wang
Han-Chung Wang
Okay, I have verified that e2e is working with the new changes. I still don't get a good name for the attribute interface, perhaps I'll just call it EncodingAttrInterface for...
I think we can name it to `EncodingLayoutAttrInterface`. All the methods are about layouts, e.g., storage size calculation, materialized layout shape, generating operations for device layout, etc. I also want...
> @hanhanW Are there any aarch64 IREE benchmarks now? ([benchmarks:android-cpu](https://github.com/iree-org/iree/labels/benchmarks%3Aandroid-cpu) seems to no longer function) @ScottTodd knows the details. I think they are off for a while; I don't know...
I took a look this morning, but I did not get time to write down my observation. I'll do it soon
Here is the IR before vectorization. It is very similar to what @pashu123 and I saw in broadcast + mmt4d fusion. The dequant op is not fused into the reduction...
I think there are numeric issues because you initialize the `acc` to zeros every time. That's why I'm saying that we should only fuse the input operands when tiling the...
@MaheshRavishankar @IanWood1 why do we have a tensor.expand_shape in between? I thought that the reshape ops become `flow.reshape` ops and we don't fuse them into dispatches? If it is expected,...
My understanding is that all the compute ops in the ssa-chain should implement TilingInterface. Otherwise, we don't have much things to do in codegen. So (1), (2) and (3) are...
This is done by recent CPU work. I can compile the program without errors, closing the issue.
(CC myself for notification)