Hello.
I’m not familiar with C/C++ language specification and Clang implementation, so please let me ask a question.
It is said that the compiler assumes the behavior of integer overflow is undefined and add nsw to arithmetic operations when -fno-wrapv is enabled. The description of the option -fwrapv is the following:
Treat signed integer overflow as two’s complement
This doesn’t mention what are the target operations, and I’m not sure whether nsw is added to all operations even if some of them are created by the compiler in private.
Actually, I’m thinking of introducing this option into Flang. Fortran standard also prohibited integer overflow but internal operations such as address calculations doesn’t seem to be considered. If the option is related only to the language specification, I should create another option to add nsw to address calculations.