commit | 187aa7f13238958bb866422dfd42524f7932d3aa | [log] [tgz] |
---|---|---|
author | Ben Vanik <benvanik@google.com> | Fri Jun 11 17:09:22 2021 -0700 |
committer | GitHub <noreply@github.com> | Fri Jun 11 17:09:22 2021 -0700 |
tree | b101cfe6edb61471f6d6522c9e8767f40718b93a | |
parent | 7c611b95c5318c8f12abdfc2d84d35495375eaa3 [diff] |
Flip flags to enable IREE native ABI. (#6155) * Fixing rounding in VM float op folders to match C semantics. * Supporting tensor->buffer conversion in !iree.list. * Compute stream buffer sizes based on the tensor shape we carry around. We know the full tensor type and can avoid the buffer view length query. * Use !hal.buffer_view at the TF ABI boundary instead of !hal.buffer. This lets us carry across the shape and data type information as is needed to match the dynamic nature of the ABI (lists of lists of type-erased tensors, etc). * Update the custom_modules_test to use buffer views for I/O. (this should be updated to use the new runtime API at some point) * Flip flags to enable IREE native ABI. * TF passes changed to enable native ABI/disable SIP. * HLO pass added to enable native ABI. * IREEVM flags flipped. * Make the TensorCastOp conversion create a new BufferView when casting from a tensor. * It is not a legal optimization in general to just return the original buffer view, as a previous step in the conversion may have rerouted an update which changes the metadata of the backing buffer. The tensor cast has the correct shape that should be applied and we must create the new BufferView with it. * Fix python tests and add assert for npe benchmark var case. * Disable 'enable_benchmark' flag in compiler test (#6196). * Make strings and tensorlist dialects implement the inliner interface. * Was missing for no good reason and blocks inlining of the new ABI constructs. * Disabling dynamic_compare_and_select test. It does a funny shape.shape_of canonicalization dance that has no quick fix. Improvements to dynamic shapes that avoid this issue will make this better without any brittle hacks. Co-authored-by: Stella Laurenzo <stellaraccident@gmail.com>
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!
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.