| commit | 339d9d7acce9e127169fe2eea9d0d7a8176defdf | [log] [tgz] |
|---|---|---|
| author | Scott Todd <scotttodd@google.com> | Mon Oct 03 10:21:31 2022 -0700 |
| committer | GitHub <noreply@github.com> | Mon Oct 03 17:21:31 2022 +0000 |
| tree | 171f235ef9e02c6b1e5fd25d1500505b46ad7654 | |
| parent | 939212280e6fd41c45079dc1ff4a1d627e52810a [diff] |
Update WebGPU target using latest Tint code. (#10134)
Tested on a few programs, seems to still work: simple programs (no push
constants) compile to WGSL that looks reasonable and complex programs
(push constants) fail with errors like this:
```
Tint reported 1 error(s) for a SPIR-V program, see diagnostics:
error: unknown SPIR-V storage class: 9SPIR-V pointer type with ID 7 has invalid storage class 9
D:\dev\projects\iree/../iree-tmp/webgpu/unidirectional_lstm_webgpu_2022_08_18/\module__main_dispatch_0.mlir:2:2: error: failed to compile SPIR-V to WGSL. Consider inspecting the shader program using -iree-hal-dump-executable-intermediates.
hal.executable.variant public @webgpu_wgsl_fb, target = <"webgpu", "webgpu-wgsl-fb", {spv.target_env = #spv.target_env<#spv.vce<v1.0, [Shader], [SPV_KHR_storage_buffer_storage_class]>, #spv.resource_limits<>>}> {
^
D:\dev\projects\iree/../iree-tmp/webgpu/unidirectional_lstm_webgpu_2022_08_18/\module__main_dispatch_0.mlir:2:2: note: see current operation: "hal.executable.variant"() ({
...
```
I want to route those diagnostics to MLIR and add line breaks at some
point (that will really help output legibility when the compiler runs
multithreaded)
---
This also adds a few compiler lit tests ~and enables building and
running those tests on our CI (`build_all` and `test_all`)~. As
compatibility issues are fixed, some of those lit tests will be removed
or adjusted. At that point I plan to enable the rest of our test suite,
similar to how Wasm is tested:
https://github.com/iree-org/iree/blob/2b226445bec8c8e28e863a467fc9e9db70a53296/tests/e2e/xla_ops/BUILD#L393-L402IREE (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.