Unary operators don't produce overflow flags for integer operations

[andykaylor]

Classic clang codegen sets the nsw flag for the add and sub operations produced for inc, dec, and minus operators on signed integer operations. When lowering through CIR, this flag is not set.

For example:

int f(int b) {
  return ++b;
}

Classic codegen produces:

%inc = add nsw i32 %0, 1

CIR codegen produces:

%5 = add i32 %4, 1

It's not clear to me whether the CIR should also have an attribute for this.

[andykaylor]

A comment in the code for inc/dec handling says that it is relying on the module flag for overflow behavior and that it will be handled during lowering, but it is not handled during lowering.

[bcardosolopes]

(cc @Lancern)

[Lancern]

Looking into this.

[andykaylor]

@Lancern It looks like it would be easy enough to add the flag during lowering as current comments say we should, but I think maybe it would be better to add an attribute to the UnaryOp (like what BinOp has) rather than rely on a module-wide flag.

[Lancern]

I think maybe it would be better to add an attribute to the UnaryOp (like what BinOp has) rather than rely on a module-wide flag.

Agree. I'll draft a patch for this later.

[Lancern]

Opened #1485 to resolve this issue.