-
### What happened?
Some changes during dispatch creation (encountered here https://github.com/iree-org/iree/pull/18341) causes f16 SDXL to produce incorrect results under aggressive fusion. This on…
-
This collects the tasks to enable direct pack/unpack vectorization by default.
### Tasks
- [x] Land https://github.com/iree-org/iree/pull/16692
- [x] Finish vector input sizes inference from IR f…
-
For this, we need to lower it from multiple models and multiple frameworks and see "what they do". At least two potential options: `exp(arg)` and `exp(arg-mean)`, might be more. We need to pick one, p…
-
The VectorConvertToLLVM pass is not converting the %85 = vector.extract %75[0, %83] : i1 from vector
to LLVM dialect. It appers the conversion code converts only 1-d types.
```
module {
func.fu…
-
It would be useful to explicitly block optimizations in the conformance suite, so that constant folding doesn't fold the entire suite away. IREE [has](https://github.com/openxla/iree/blob/96b61ba6fe20…
-
This is an umbrella issue for implementing a tuning infrastructure. By tuning we mean a type of Profile Guided Optimization flow where we compile a program/model with extra instrumentation and use the…
kuhar updated
2 months ago
-
I'm working on testing some models using `functorch` along with `torch-mlir` and IREE. I don't see an analog of jax's `tree_map`. Is this something it makes sense for `functorch` to implement, or is…
-
There seems to be a requirement in transform dialect lowering strategies to set the pass pipeline in translation_info to None https://github.com/iree-org/iree/blob/main/compiler/src/iree/compiler/Code…
-
### Request description
Based on a discussion in the IREE discord server.
This would help StableHLO interoperate more seamlessly in the MLIR ecosystem. Need to investigate how to express targetVer…
-
Several language models (and efficientnets) fail during runtime complaining of invalid PTX JIT compilation:
```
E RuntimeError: Error registering modules: c/runtime/src/iree/hal/drivers/cuda/nat…