Re: [PATCH] phiopt: Optimize (x <=> y) cmp z [PR94589]

[Marc Glisse <marc.glisse@inria.fr>]

On Thu, 6 May 2021, Jakub Jelinek via Gcc-patches wrote:

> Though, (x&1) == x is equivalent to both (x&~1)==0 and to x < 2U
> and from the latter two it isn't obvious which one is better/more canonical.
> On aarch64 I don't see differences in number of insns nor in their size:
>  10:	13001c00 	sxtb	w0, w0
>  14:	721f781f 	tst	w0, #0xfffffffe
>  18:	1a9f17e0 	cset	w0, eq  // eq = none
>  1c:	d65f03c0 	ret
> vs.
>  20:	12001c00 	and	w0, w0, #0xff
>  24:	7100041f 	cmp	w0, #0x1
>  28:	1a9f87e0 	cset	w0, ls  // ls = plast
>  2c:	d65f03c0 	ret
> On x86_64 same number of insns, but the comparison is shorter (note, the
> spaceship result is a struct with signed char based enum):
>  10:	31 c0                	xor    %eax,%eax
>  12:	81 e7 fe 00 00 00    	and    $0xfe,%edi
>  18:	0f 94 c0             	sete   %al
>  1b:	c3                   	retq
>  1c:	0f 1f 40 00          	nopl   0x0(%rax)
> vs.
>  20:	31 c0                	xor    %eax,%eax
>  22:	40 80 ff 01          	cmp    $0x1,%dil
>  26:	0f 96 c0             	setbe  %al
>  29:	c3                   	retq
> Generally, I'd think that the comparison should be better because it
> will be most common in user code that way and VRP will be able to do the
> best thing for it.

We can probably do it in 2 steps, first something like

(for cmp (eq ne)
  (simplify
   (cmp (bit_and:c @0 @1) @0)
   (cmp (@0 (bit_not! @1)) { build_zero_cst (TREE_TYPE (@0)); })))

to get rid of the double use, and then simplify X&C==0 to X<=~C if C is a 
mask 111...000 (I thought we already had a function to detect such masks, 
or the 000...111, but I can't find them anymore).

I agree that the comparison seems preferable, although if X is signed, the 
way GIMPLE represents types will add an inconvenient cast. And I think VRP 
already manages to use the bit test to derive a range.

-- 
Marc Glisse