vkd3d-proton upstream (17-05-2023)

Just using the test suite, test_advanced_cbv_layout is failing on Intel.
It seems the failing is in d3d12_device_validate_shader_meta on this line :

    if ((meta->flags & VKD3D_SHADER_META_FLAG_USES_FP64) && !device->d3d12_caps.options.DoublePrecisionFloatShaderOps)
    {
        WARN("Attempting to use FP64 operations in shader %016"PRIx64", but this is not supported.", meta->hash);
        return false;
    }

Now the shader being compiled doesn't seem to have any FP64 values/types :

cbuffer Cbuf : register(b0)
{
    uint64_t4 values_root[8];
};

cbuffer Cbuf : register(b0, space1)
{
    uint64_t4 values_table[8];
};

RWStructuredBuffer<uint> RWBuf : register(u0);

uint pack4(uint4 v)
{
    return v.x | (v.y << 8) | (v.z << 16) | (v.w << 24);
}

[numthreads(8, 1, 1)]
void main(uint thr : SV_DispatchThreadID)
{
    uint64_t4 v = values_root[thr] + values_table[thr];
    uint4 lo = uint4(v);
    uint4 hi = uint4(v >> 32);
    RWBuf[2 * thr + 0] = pack4(lo);
    RWBuf[2 * thr + 1] = pack4(hi);
}

Is the dxil/dxc compiler introducing fp64 operations?

System information

• GPU: Intel DG2 A750
• Driver: Upstream Mesa/Anv
• Wine version: N/A
• VKD3D-Proton version: ff665a37352a477a780619e196706576484220bd