-
Recently I'm trying to run the model compiled by onnx-mlir on GPU.
Here is my idea about the compilation flow:
affine/scf/std Dialect -> gpu Dialect ->NVVM IR -> LLVM IR -> ptx assemble.
I have no …
-
### What happened?
`iree-compile` llvm-cpu` backend with `riscv32` cpu and `soft_fp` failed to lower `softmax` op
```
iree-compile --output-format=vm-bytecode --mlir-print-op-on-diagnostic=false …
-
It looks like these flags are enabled by default, which is extremely dangerous given the consequent correctness issues and unexpected behavior to the user. Not sure why we have them enable by default …
-
### What happened?
This program [in-place-sharded-tensor.zip](https://github.com/user-attachments/files/17403601/in-place-sharded-tensor.zip) fails to do a proper in-place update.
The program is equ…
-
To reproduce, `cmake -DIREE_TARGET_BACKEND_DEFAULTS=OFF`, clear cached CMake variables controlling backends, and rebuild and test.
These tests are currently failing: (and that's with #14035 fixing…
-
We've reached the point where the time it takes to run the IREE compiler and the memory that the compiler uses are both significant pain points for certain input programs and compilation modes. (Not t…
-
### What happened?
1. I used the Torch-MLIR tool to convert the Alexnet model from the TorchVision library to linalg expression, i get **alexnet.mlir**, python script as followed
```
import torch…
-
Seen in Windows CI logs:
https://github.com/openxla/iree/actions/runs/5027770230/jobs/9017669780#step:8:528
```
-- Performing Test IREE_UK_BUILD_X86_64_AVX2_FMA
-- Performing Test IREE_UK_BUIL…
-
-
### What happened?
We have an experimental version of an IREE compiler plugin that lowers GEMMs and some other ops into our high-perf implementations via library calls. Running `iree-compile` to co…