On 07/12/2010 10:30 AM, Ian Lance Taylor wrote: > Andrew Haley <aph@xxxxxxxxxx> writes: > >> But you can trivially do the job with an inline asm in a header file >> in almost all cases, so I don't really understand why you want to >> submit to the pain of defining a new intrinsic. > > I would say that the advantage to an intrinsic over an asm is that the > instruction is exposed to the scheduler. On some processors good > instruction scheduling is essential for good performance. And on some > special purpose processors intrinsic functions (or asm statements) are > needed for a lot of useful code. Well, yeah, maybe. But given the work involved in defining a new builtin, I want to be sure the OP has carefully considered the possibility of using an inline asm. Andrew.
