Currently only workgroup_id and num_workgroups built-in values are considered uniform according to spec.

Treating subgroup_size as uniform may help writing flexible kernel that fit to different actual subgroup size, and allow using functions like workgroupBarrier and subgroupBroadcast (if requiring that the subgroup operations only be used in uniform control flow) under condition of subgroup size, like

    if (subgroupSize == 16) {
        ...
        workgroupBarrier();
        let a0 = subgroupBroadcast(a[0], 0u);
        ...
    } else {
        // Fall back path
        ...
    }

The Vulkan spec, in SubgroupSize of 15.9. Built-In Variables section, required that

The value may vary across a single draw call, and for fragment shaders may vary across a single primitive. In compute dispatches, SubgroupSize must be uniform with command scope.

I don't find explicit requirement in MSL/HLSL spec, but I think they should work in the same way for compute kernel.