webgpu: fix: conditionally call deprecated GPUAdapter.requestAdapterInfo

[mattvr]

Context

GPUAdapter.requestAdapterInfo is deprecated and non-standard.

It's in use by tfjs-backend-webgpu, but the result is optional, and only used for a single optimization. Yet if called in an environment which doesn't support it, it will throw.

By checking the feature's available before calling it, this unbreaks Deno runtime support for tfjs-backend-webgpu, which doesn't support this feature and throws.

Testing/Verification

yarn test and yarn lint succeed. Verified basic example now works in Deno.

import * as tf from "@tensorflow/tfjs-core";
import "@tensorflow/tfjs-backend-webgpu";

await tf.ready();
tf.randomGamma([2, 2], 1).print();

Related

denoland/deno#22029

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[mattsoulanille]

/gcbrun

[mattsoulanille]

I'm seeing this error in the build:

tfjs-backend-webgpu/src/base.ts:64:38 - error TS2345: Argument of type 'unknown' is not assignable to parameter of type 'GPUAdapterInfo'.
  Type '{}' is missing the following properties from type 'GPUAdapterInfo': __brand, vendor, architecture, device, description

64     return new WebGPUBackend(device, adapterInfo);
                                        ~~~~~~~~~~~

Maybe our WebGPU types from NPM are out of date? Does yarn build work in the tfjs-backend-webgpu directory for you?

[mattvr]

@mattsoulanille Thank you for reviewing, yeah – I'm seeing that too with the previous commit. Oudated WebGPU types was it!

The latest commit fixes the error. It keeps the adapter.requestAdapterInfo() fallback for backwards compatibility, to prevent breaking older environments. This required an any typecast though, let me know if you'd like any changes.

[mattsoulanille]

/gcbrun

[mattsoulanille]

Thanks! The any is probably okay since it gets immediately passed to another function that has a defined type for that argument, so the any type shouldn't leak anywhere.

[mattsoulanille]

Looks like the linter isn't happy with any.

[mattvr]

@mattsoulanille Sorry for the back-and-forth, could've sworn build and test succeeded before I submitted the last two commits. Should be fixed now! 🤞

[mattsoulanille]

No worries! Our build and test scripts are not particularly easy to use (sorry). There's a yarn lint command you can run at the root of the repo to check this.

[mattsoulanille]

/gcbrun

[mattsoulanille]

/gcbrun