Specialization constant conversion to float is disallowed by VkGLSL spec, but is used in an example.

[NicolBolas]

The GL_KHR_vulkan_glsl specification states that you cannot do a specialization constant conversion from an integer type to a floating-point type. It states this by omission, as the "Specialization Constant Operations" section lists the allowable operations on a specialization constant (that still yields a specialization constant). And while integer-to-integer conversions are allowed, conversions between integers and floats are disallowed.

However, the spec also has this example:

    layout(constant_id = 18) const int scX = 1;
    layout(constant_id = 19) const int scZ = 1;
    const vec3 scVec = vec3(scX, 1, scZ);  // partially specialized vector

This performs implicit conversions from scX and scZ to floating-point types, which is not permitted. As such, this should use ivec3 instead of vec3.