[hlsl-out] Stores of vec2-based matrix inside struct tries to access raw matrix directly

[cwfitzgerald]

WGSL

struct Ah {
    inner: mat2x2<f32>,
};
@group(0) @binding(0)
var<storage, write> ah: Ah;

@compute @workgroup_size(1)
fn cs_main() {
    ah = Ah(mat2x2<f32>(vec2<f32>(0.0), vec2<f32>(1.0)));
}

HLSL

struct Ah {
    float2 inner_0; float2 inner_1;
};

RWByteAddressBuffer ah : register(u0);

Ah ConstructAh(float2x2 arg0) {
    Ah ret = (Ah)0;
    ret.inner_0 = arg0[0];
    ret.inner_1 = arg0[1];
    return ret;
}

[numthreads(1, 1, 1)]
void cs_main()
{
    {
        Ah _value2 = ConstructAh(float2x2((0.0).xx, (1.0).xx));
        {
            // ERROR: Inner doesn't exist
            float2x2 _value3 = _value2.inner;
            ah.Store2(0+0, asuint(_value3[0]));
            ah.Store2(0+8, asuint(_value3[1]));
        }
    }
    return;
}

[ErichDonGubler]

Firefox bug tracking this: https://bugzilla.mozilla.org/show_bug.cgi?id=1859304

[jimblandy]

Since this causes HLSL compilation errors, it should be pretty easy to spot when we start running into it on the CTS. (If the symptom was just corrupted data, that would be harder to debug.)

In that light, Mozilla is going to wait until we run into this in our CTS runs to prioritize it.

Once that happens, this looks like a good first bug: easily reproduced, the HLSL backend isn't lovely but it's manageable, snapshot testing should suffice.