Scott Todd

The IR right before the error, as well as the full error message for `error: failed to legalize operation 'arith.extui' that was explicitly marked illegal` would be useful in triaging...

Since WebGPU uses the SPIR-V backend in generating WGSL, this is also affecting the WebGPU compilation target: `iree-compile --iree-hal-target-backends=webgpu --iree-input-type=mhlo tests/e2e/xla_ops/fft.mlir -o /dev/null` ```mlir // -----// IR Dump Before ConvertToSPIRV...

@GMNGeoffrey do you have more plans for this? The git trailers and current filtering both seem to be working pretty well now.

I wonder if this reproduces on the latest Tracy / newer phones. This issue is quite old 🤔

This still feels like we're trying to fit a square peg into a round hole by tightly coupling build/test CI with benchmarking CI. > Our P0 is to support manual...

Migrating more build/test jobs to the pkgci style and using whatever magic Stella set up with local git mirrors ([discord discussion here](https://discord.com/channels/689900678990135345/1169307503633383664/1178905294500274206), [workflow file here](https://github.com/nod-ai/iree-amd-aie/blob/main/.github/workflows/ci.yml)) should both help with the...

Going to call this obsolete, as the TensorFlow export APIs have changed a few times and the TensorFlow frontend is a lower priority.

This may be a bit easier to land now that https://github.com/openxla/iree/pull/12645 is merged.

I've been looking at this issue (similar but not identical dispatches) as the number of dispatches correlates directly with compile time and some one our CI workflows are bottlenecked on...

(After more debugging) If you can narrow down a test case and expand the CTS that would be great. What's in there today was written after a few reads through...