iree
iree copied to clipboard
iree-org
A retargetable MLIR-based machine learning compiler and runtime toolkit.
Lots of this code is legacy and we should standardize on a small number of well supported dialects where possible. It looks like we were/are using this for different lowering...
I've seen multiple people falling down this hole: they run iree-compile on their model, targeting CPU. Then they get performance that is 10x-100x off of any reasonable expectation. Then they...
Progress on https://github.com/iree-org/iree/issues/18561. We have these high level flags (and MLIR attributes) controlling what code the llvm-cpu compiler target generates using LLVM: * `--iree-llvmcpu-target-triple` * e.g. `x86_64-pc-windows-msvc` * See https://github.com/llvm/llvm-project/blob/main/llvm/include/llvm/TargetParser/Triple.h...
As a follow-up to https://github.com/iree-org/iree/pull/17710, this patch renames the compiler target backend from `ROCM` to `HIP` to be consistent with driver. User-facing changes: `IREE_TARGET_BACKEND_ROCM` to `IREE_TARGET_BACKEND_HIP` `iree-hal-target-backends=rocm` to `iree-hal-target-backends=hip`
The prefetch pass assumes that shared memory can be reused in the prologue. This may not be true when nested loops are involved, so we need to explicitly insert a...
### What happened? I have input MLIR that is exported from the test in this [PR](https://github.com/nod-ai/SHARK-Platform/pull/237). It's compilation with assertions on fails with ``` iree-compile: repo/compiler/src/iree/compiler/Dialect/HAL/Analysis/DeviceAnalysis.cpp:180: auto mlir::iree_compiler::IREE::HAL::DeviceAnalysis::gatherDeviceTargets(mlir::Attribute, mlir::Operation *,...
### What happened? ``` latest.mlir:3:10: error: 'func.func' op unhandled function with multiple blocks %0 = flow.dispatch.region -> (tensor) { ^ latest.mlir:2:3: note: called from func.func @simple_test_with_cfg(%arg0: i8) -> tensor {...
In a recent PR, the TopK e2e test fails in CI: https://github.com/iree-org/iree/actions/runs/11107992173/job/30867743807?pr=18634 The following test is what fails: ``` func.func @topk_2d_dim1_inverted_max() { %input_values = util.unfoldable_constant dense : tensor %input_indices =...
This adds an analysis and test pass based on the ValueBounds interface for constructing a set of constraints on the values within a function. The analysis provides a few basic...
