Using rel32 addressing when calling cfuncs, instead of r64

[64kramsystem]

The current Runtime#call_cfunc uses 64-bit addressing:

@fisk.mov(@fisk.rax, @fisk.uimm(func_loc))
  .call(@fisk.rax)

This clobbers RAX, and makes it impossible to perform some calls, e.g. libc's printf, which uses RAX to count the floating point parameters (ie. xmm registers used).

I'm not familiar with the subject, but this seems to be required by the assumption of a large/huge memory model.

In ISEQCompiler though, a 32-bits relative address is used for jumps (see operations using @fisk.absolute), which assumes a small memory model.

Are there cases where 64-bit addressing is used for cfuncs? If there isn't, call_cfunc could be simplified. I've actually tested on a branch if the change works, and it does, however, I don't know if there are other situations where this wouldn't work.

[tenderlove]

AFAIK there aren't any cases where we actually need 64 bit addressing, though I haven't confirmed. I think if the tests pass with your PR we should just merge it? I don't really see any downsides 😄

[64kramsystem]

Hi there!

My reasoning is that the allocated buffers may reside anywhere in the address space, so that, if a buffer is located too far (GBs) from the function behind called, 32-bit relative addressing would fail.

However, I have no experience in operating systems, so this is 100% speculation 😬

I'll follow up with my latest work packaged in a PR, including this change.