iree
iree copied to clipboard
iree-org
A retargetable MLIR-based machine learning compiler and runtime toolkit.
I tried running an example similar to the checked in test [here](https://github.com/iree-org/iree/blob/main/tools/test/iree-run-module-multi.mlir) with the exception that I would like to run across GPU and CPU. Here is the sample mlir...
I ran into this input IR for the `iree-stream-schedule-execution` pass that results in an after-pass verification failure ``` error: partition set out of order; value captured declared as escaping below...
### What happened? On compiling a model with int8 quantization one of the dispatches fails to compile with the following error: ``` error: One or more operations with large vector...
### What happened? Converting the following models https://github.com/iree-gd/iree.zoo/blob/main/.github/workflows/multiple-models.yml#L10 (31 models so far) from **kaggle** with version `20240812.983` results in (run here https://github.com/iree-gd/iree.zoo/actions/runs/10377948022) a success rate of 14/31. Logs are in...
According to the C++ standard, the main function shall not be declared with a linkage-specification. For upcomming [Clang] Strengthen checks for `main` to meet `[basic.start.main]p3`’s requirements (#101853) which will lead...
Hi, I have some question about the numStages in `compiler/src/iree/compiler/Codegen/Common/GPU/GPUPipelining.cpp` As explaiend in this pr https://github.com/iree-org/iree/pull/12603 , we assume that numStages >= 3. But when we schedule instructions here, we...
At least command buffer fill fails if `iree_hal_buffer_ref_t` has a length of `IREE_WHOLE_BUFFER`. It could be made to work but the whole buffer mechanism is sloppy and it's not worth...
Running SDXL int8 with aggressive fusion enabled produces different results from running without aggressive fusion enabled. ### Repro Instructions ### 1. Checkout https://github.com/iree-org/iree/tree/shared/sdxl_quantized in IREE 2. Clone https://github.com/nod-ai/sdxl-scripts and `cd...
Some of the models (like SDXL) from torch seem to contain these types of patterns ``` %collapsed = tensor.collapse_shape %input [[0], [1, 2], [3]] : tensor into tensor %expanded =...
