[CPU][SME] Add an experimental flag to disable SME tiling (#16211) ATM, when lowering `linalg.matmul` with SME and SSVE enabled: * --iree-llvmcpu-target-cpu-features="+sve,+sme", IREE will always configure tiles sizes to be suitable for SME (i.e. scalable in 2D). Since identical target feature flags are used to enable SSVE, there is no mechanism to make IREE target SSVE instead of SME. Ideally, one would enable SSVE by changing `+sve,+sme` to e.g. `+ssve`. However, LLVM does not support plain `+ssve` as a target feature. Since IREE forwards all CPU target features flags directly to LLVM, this approach is not really feasible (unless IREE were to filter the attributes). Instead, this patch adds a flag, `-iree-experimental-llvmcpu-arm-force-ssve`, that disables "SME tiling". This means that, when using `+sve,+sme`, IREE will: * lower linalg.matmul identically to how lowering for SVE works, * in addition to regular SVE lowering, SME's streaming mode will be enabled. Put differently, this flag will make IREE target _SSVE_ rather than _SME_ when SME+SSVE are enabled
IREE (Intermediate Representation Execution Environment, pronounced as “eerie”) is an MLIR-based end-to-end compiler and runtime that lowers Machine Learning (ML) models to a unified IR that scales up to meet the needs of the datacenter and down to satisfy the constraints and special considerations of mobile and edge deployments.
See our website for project details, user guides, and instructions on building from source.
IREE is still in its early phase. We have settled down on the overarching infrastructure and are actively improving various software components as well as project logistics. It is still quite far from ready for everyday use and is made available without any support at the moment. With that said, we welcome any kind of feedback on any communication channels!
See our website for more information.
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.