[PATCH 0/6] x86: a few more optimizations

[Cui, Lili]

> On 17.06.2024 10:09, Cui, Lili wrote:
> >> APX {nf} insn forms present a number of interesting optimization
> >> opportunities, often mostly for size. There are a few more that I'm
> >> aware of, but where I'm less convinced that input code would really
> >> ever be written like
> >> this:
> >>
> >> 1) Quite a few operations could be converted to plain MOV. For
> >> example
> >>
> >> 	{nf} xor %cl, %cl
> >> 	{nf} sub %cl, %cl
> >> 	{nf} and $0, %cl
> >>
> >> can all be replaced by the much shorter
> >>
> >> 	mov	$0, %cl
> >>
> >> 2) Certain forms of IMUL{,ZU} with a power-of-2 immediate could be
> >> converted to SHL. This could be beneficial even when size doesn't
> >> shrink, for SHL still having better latency/throughput.
> >>
> >> Thoughts?
> >
> > These two ideas are so good that several people think that GCC should
> support these two conversions. Would you consider implementing these
> conversions in gcc?
> 
> Multiplication to shift conversion is already done by the compiler, afaik. As to
> 1), are you suggesting the compiler may emit such code? Imo it shouldn't.

Perhaps, but we can still add a pattern for it.

Lili.

> Plus, as indicated in the reply to Haochen, I view optimizations in gas as
> primarily targeting hand-written code anyway. Compilers shouldn't rely on the
> assembler doing further transformations.
>