[CIR] [CodeGen] Handling case not in switch scope directly

[ChuanqiXu9]

Reproducer:

int func(int v) {
    switch(v) {
        case 1:
            return 43;

        default:
            if (v > 5) {
                case 2:
                    return 35;

                return 47;
            }

            return  99;
    }
}

Reduced from Spec2017. https://godbolt.org/z/zrqxsM9K4

Now it will crash at

clangir/clang/lib/CIR/CodeGen/CIRGenFunction.h

Line 2205 in 24b2455

assert(isSwitch() && "expected switch scope");

due to we're trying to create a switch region but not in swtich scope. The direct issue of layering issue is relatively easy to fix.

What confuses me is what is the expected CIR generated for the above code. A direct idea may be:

"cir.switch"(%value) (
case (equal, 1) {
    codes for case 1
},
case (default) {
   // codes for default case
   "cir.scope"() {
        // codes for if
        if (v > 5)
            cir.br ^bb2
        
        return 47
   }
},
case (equal, 2) {
    // codes fro case2
    ^bb2:
        return 35
})

I am hesitating since in

clangir/clang/lib/CIR/CodeGen/CIRGenFunction.h

Line 2171 in 24b2455

// have their own scopes but are distinct regions nonetheless.

I saw comments like the regions of cases are distinct regions, so I am afraid the above codes may break assumptions of isolations of regions (not sure if it is true)

[Lancern]

The current design of cir.switch operation simply cannot handle cases like this, as it requires each case to be isolated from other cases. In #522 @wenpen has proposed a new design that could help resolve this problem. Also it contains some further discussions.

[ChuanqiXu9]

Thanks. Then this one is duplicated