Aleksey Bragin wrote:
Why wrap into #ifdefs? I thought it's supposed to be runtime, not compile time.
Because CPU_TRANSMETA, CPU_CENTAUR and CPU_RISE aren't defined yet, and the identification routine returns zero instead when those vendors are detected. As soon as each CPU vendor id is defined, its respective initialization code will be enabled. I don't know what initialization code is required, other than detecting/enabling CMPXCHG8B, and I don't have any CPUs of said vendors to test it anyway, so I won't enable support for them myself