1

There is mоvq and mоvd, but mоvb and mоvw aren't exist. Why? Don't we need to mоve bytes and words?

prytkov
  • 11
  • 1
  • 2
    MMX is (was) for moving a lot of memory at a time efficiently. Why would you want to move a single byte through those registers? If this is just a random question looking for orthagonal code, great, but are you attempting to do something with these byte moves? – Michael Dorgan Apr 14 '15 at 21:23
  • I wish I had `movw` to an `xmm` register, for LUT lookups. (for par2 GF16 multiply). I've been doing `movd` and then `pinsrw`, but if I pack the LUT of 16bit values into 16bits per element, the `movd`s can cross cachelines. (The old non-simd code used a zero-padded LUT, which I might keep since it only takes 2kiB of L1 cache. https://github.com/pcordes/par2-asm-experiments/blob/master/asm-pinsrw.s) – Peter Cordes Jul 08 '15 at 23:48

0 Answers0