iree
iree copied to clipboard
iree-org
A retargetable MLIR-based machine learning compiler and runtime toolkit.
### What happened? for the given IR ```mlir #map = affine_map (d0, d1, d2, d3)> #map1 = affine_map (0, d1, d2, d3)> module { ml_program.global private mutable @global_seed(dense : tensor)...
Splitting https://github.com/iree-org/iree/pull/18519 into four patches. This is the last patch, dependent on #18784, #18800, and #18822, which adds the configurations required by the VectorDistribute pipeline for the reduction ops. A...
Moving the test, renaming, and cleaning. We are moving the test from `compiler/plugins/target/ROCM/test/` to `samples/compiler_plugins/rocm/` to make it more in line with the organization of the project. This also changes...
To enable faster flash attention, we'd like to be able to force different vector widths => we'd like different contraction to potentially have different intrinsics. This PR introduces a way...
This replaces some constants what were hardcoded in GPUMaterializeEncoding.cpp by actual GPU target parameters. The logic in `getSwizzle` was doing wonky things with its own local `const int targetPreferredLoadBitWidth =...
bump llvm-project to https://github.com/llvm/llvm-project/commit/864902e9b4d8bc6d3f0852d5c475e3dc97dd8335
Our StableHLO build has incurred some maintenance costs due to the complexities of how source dependencies on Bazel and CMake projects work. See for example https://github.com/iree-org/iree/issues/18785 and https://github.com/openxla/stablehlo/issues/2255. We currently...
