Fix device_pool refactor

[greggman]

The old version made one device with no descriptor. If that fail then it stopped creating any devices period.

The previous refactor was bad in that if the any device failed it would just stop making devices.

This one is simplified. It just used the pool as a pool. There is no special handling for failed devices.

Note: I could check if no description is passed in and it fails then assume all other requests will fail. Should I add that in or leave it as is?

---

Requirements for PR author:

• All missing test coverage is tracked with "TODO" or .unimplemented().
• New helpers are /** documented */ and new helper files are found in helper_index.txt.
• Test behaves as expected in a WebGPU implementation. (If not passing, explain above.)
• Test have be tested with compatibility mode validation enabled and behave as expected. (If not passing, explain above.)

Requirements for reviewer sign-off:

• Tests are properly located in the test tree.
• Test descriptions allow a reader to "read only the test plans and evaluate coverage completeness", and accurately reflect the test code.
• Tests provide complete coverage (including validation control cases). Missing coverage MUST be covered by TODOs.
• Helpers and types promote readability and maintainability.

When landing this PR, be sure to make any necessary issue status updates.

[greggman]

note: I tested by pasting this at top of device_pool.ts

GPU.prototype.requestAdapter = (function (origFn) {
  return async function (this: GPU, desc) {
    const adapter = await origFn.call(this, desc);
    if (adapter) {
      Object.defineProperty(adapter, 'features', {
        value: new Set([...adapter.features].filter(v => v !== 'shader-f16')),
      });
    }
    return adapter;
  };
})(GPU.prototype.requestAdapter);

GPUAdapter.prototype.requestDevice = (function (origFn) {
  return async function (this: GPUAdapter, desc) {
    if (desc && desc.requiredFeatures) {
      if ([...desc.requiredFeatures].includes('shader-f16')) {
        throw new Error('f16 unsupported');
      }
    }
    return await origFn.call(this, desc);
  };
})(GPUAdapter.prototype.requestDevice);

I then ran http://localhost:8080/standalone/?q=webgpu:shader,execution,expression,binary,f16_addition:*

Both before the bad change (works) and after the bad change (failed) and with this fix (works)

[ben-clayton]

If you have dawn node set up, and you have a machine without f16 support, please can you see if the following correctly shows SKIPs?

tools/run run-cts 'webgpu:shader,execution,expression,binary,f16_subtraction:*'

[greggman]

I don't have a machine without f16 support. But the code I pasted in above repos the bug and after the fix correctly skips.

[greggman]

oh, maybe I can force swiftshader, assuming it doesn't support f16

[ben-clayton]

Yeah, SwiftShader should SKIP.

[ben-clayton]

Confirmed that this fixes the dawn/node issues by checking out your branch 👍🏼

[ben-clayton]

LGTM, but would also be good to get @kainino0x 's thumbs up

[greggman]

I couldn't figure out how to run swiftshader on dawn.node on Mac. OTOH I can run the web version with the fallback adapter which uses swiftshader (need --enable-unsafe-webgpu)

[ben-clayton]

Build swiftshader, and set VK_ICD_FILENAMES to the vk_swiftshader_icd.json file in the output directory. Example:

VK_ICD_FILENAMES=path_to_vk_swiftshader_icd.json tools/run run-cts 'webgpu:shader,execution,expression,binary,f16_subtraction:*'

[kainino0x]

This seems fine. Honestly not sure why I added all that complex logic to begin with. I think all it does is make tests fail faster when webgpu isn't working. Hopefully I'm not forgetting something.