| commit | fb9e1b68f73ace6ae799ed27ad06a1288deccdb6 | [log] [tgz] |
|---|---|---|
| author | Ben Vanik <ben.vanik@gmail.com> | Tue Sep 19 17:57:16 2023 -0700 |
| committer | GitHub <noreply@github.com> | Tue Sep 19 20:57:16 2023 -0400 |
| tree | c096453667d1a887d522423e76b923d1d32e62a7 | |
| parent | 0f4dd7350fe4e6c44149afbb3572e04b6572b298 [diff] |
Adding `--iree-vulkan-experimental-indirect-bindings=true` flag. (#14977) This makes all descriptor set layouts have the new `Indirect` bit set and plumbs it all the way through to the runtime `IREE_HAL_DESCRIPTOR_SET_LAYOUT_FLAG_INDIRECT` bit. SPIR-V codegen can inspect the pipeline layout attr of exports to discover which descriptor sets are indirect and lower via `VK_KHR_buffer_device_address` and for the runtime to specially handle the indirect descriptor sets by producing device address buffers. The flag is currently experimental as interop with non-indirect dispatches (custom/produced by other higher layers like IREE input dialects/plugins) and multi-versioning (producing both direct and indirect) are TBD. It should be sufficient for users targeting specific Vulkan devices where they know the support is present, though. Note that while this is just the plumbing for the flag and the IR/runtime bits nothing is either lowering differently or setting up the appropriate runtime structures but it should allow codegen to start experimenting with alternative lowerings. Progress on #13945.
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.