On 28 May 2014 19:35, Regan, Brian (EPC COE) wrote: > Agree result is the same - just in a non-compliant-ey type of way. (opcodes are just my means of looking under the compiler hood to understand the compiler behaviour.) But it's not non-compliant! The compiler knows, without a shadow of a doubt, that the constant 200. fits in a single-precision float, so it could either convert 'a' into a double-precision value and then compare that to 200., or it could just compare 'a' directly to 200.f Knowing that the result will be identical, presumably it chooses the more efficient option, which is what most people want their compiler to do. That's still a perfectly valid representation of your source code. The standard specifies behaviour in terms of the observable side effects of the abstract machine on the execution environment not in terms of which instructions or registers are used. The observable behaviour of GCC's output is exactly the same as your supposed "compliant" version. "In the abstract machine, all expressions are evaluated as specified by the semantics. An actual implementation need not evaluate part of an expression if it can deduce that its value is not used and that no needed side effects are produced (including any caused by calling a function or accessing a volatile object)." Converting to double to perform the comparison is "part of an expression" that need not be evaluated because the compiler can deduce that it's entirely pointless.
