[LLVMdev] Swapping function arguments in libcalls

[Borja Ferrer]

Hello, I don't know if this has ever been discussed or mentioned, it's
something simple so it could have been talked before, forgive me in that
case. The idea is to basically commute/swap function arguments just like in
instructions if it's beneficial, mainly for libcalls. Consider this example:

typedef float t;
t foo(t a, t b)
{
  return a*b;
}

this gets compiled into (msp430 asm, arm does the same):
    call    #__mulsf3
    ret

but if we instead do "return b*a;" we get:
    push.w    r11
    push.w    r10
    mov.w    r14, r11
    mov.w    r15, r10
    mov.w    r13, r15
    mov.w    r12, r14
    mov.w    r10, r13
    mov.w    r11, r12
    call    #__mulsf3
    pop.w    r10
    pop.w    r11
    ret

Notice the arguments are getting swapped for a commutable operation were
there order shouldn't matter. Just an idea to push the optimizations
further.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20110330/b17fafe9/attachment.html>