amd64: make memmove and memcpy less slow with mov

[mjg]

Summary

The macro is in the current form in order to accomodate switching copyin/copyout to use it. Note the resulting code is not big for the other routines as they don't have to handle overlap. It provides a win of similar nature to the one seen with memset.

[kib]

Can you show me one other use of the macro ? In particular, I am curious about other begin/end definitions.

[mjg]

I did not finish copyin/copyout yet. The macro has 2 exit points for non-overlapping case. The idea is to use it inside existing COPYIN/COPYOUT macros with something of this sort:

.macro COPYOUT_BEGIN
.endm
.macro COPYOUT_END
        SMAP_ENABLE \smap
        xorl %eax,%eax
        movq    %rax,PCB_ONFAULT(%r11)
        POP_FRAME_POINTER
.endm

For initial commit I can drop the two args. Will make it easier to import to libc.

They can be added later.

[kib]

I do not care much about these two single-use arguments. I am more worried that to understand the flow it is easier to read disassembly than source code.