Bug 199289 – [WHLSL] 'uint y = x ? 1 : 2' fails to typecheck

• Status: RESOLVED
• Resolution: WONTFIX
• Component: WebGPU

[Myles C. Maxfield]

Right now we will unifyAndConmit both of the ResolvableTypes together, which IIRC will say “hey let’s just resolve these to their preferred type” which is int but the function needs to return uchar so fail!

[Saam Barati]

```
float foo() {
    return true ? 1 : 2
}
```

[Myles C. Maxfield]

This patch should add the following text into the "operator cast" section of the standard library:

operator uchar(bool x) {
    return x ? 1 : 0;
}
operator ushort(bool x) {
    return x ? 1 : 0;
}
operator uint(bool x) {
    return x ? 1 : 0;
}
operator char(bool x) {
    return x ? 1 : 0;
}
operator short(bool x) {
    return x ? 1 : 0;
}
operator half(bool x) {
    return x ? 1 : 0;
}
operator float(bool x) {
    return x ? 1 : 0;
}

[Saam Barati]

Created attachment 373448 [details]
WIP

Trying to fix some final bugs

[Saam Barati]

Created attachment 373449 [details]
WIP

[Saam Barati]

Created attachment 373457 [details]
patch

[Saam Barati]

(In reply to Myles C. Maxfield from comment #2)
> This patch should add the following text into the "operator cast" section of
> the standard library:
> 
> operator uchar(bool x) {
>     return x ? 1 : 0;
> }
> operator ushort(bool x) {
>     return x ? 1 : 0;
> }
> operator uint(bool x) {
>     return x ? 1 : 0;
> }
> operator char(bool x) {
>     return x ? 1 : 0;
> }
> operator short(bool x) {
>     return x ? 1 : 0;
> }
> operator half(bool x) {
>     return x ? 1 : 0;
> }
> operator float(bool x) {
>     return x ? 1 : 0;
> }

I need to rebase and do this, but the patch won't change. So it's still valuable to review in its current form.

[Saam Barati]

Created attachment 373458 [details]
patch

[Robin Morisset]

Comment on attachment 373458 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=373458&action=review

I don't currently understand this code well enough to give it an r+

> Source/WebCore/Modules/webgpu/WHLSL/WHLSLParser.cpp:-368
> -    }

Why change this?

> Source/WebCore/Modules/webgpu/WHLSL/WHLSLPrepare.cpp:63
> +static constexpr bool dumpASTAtEnd = true;

Please remove.

[Saam Barati]

Comment on attachment 373458 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=373458&action=review

>> Source/WebCore/Modules/webgpu/WHLSL/WHLSLParser.cpp:-368
>> -    }
> 
> Why change this?

This is not intentional. Will revert

[Myles C. Maxfield]

WHLSL is no longer relevant.