Optimize set.pop() to advance a pointer instead of indexing.

[rhettinger]

Gives approx 20% speed-up using clang depending on the number of elements in the set (the less dense the set, the more the speed-up).

Uses the same entry++ logic used elsewhere in the setobject.c code.

[rhettinger]

FWIW, here is disassembly of the inner-loop which is now very tight:

LBB24_5:                                ## =>This Inner Loop Header: Depth=1
	movq	%rsi, %rax
	addq	$16, %rax
	movq	%r10, %rsi
	cmpq	%rcx, %rax
	jbe	LBB24_6
## %bb.7:                               ##   in Loop: Header=BB24_5 Depth=1
	movq	(%rsi), %rax
	testq	%rax, %rax
	je	LBB24_5
LBB24_8:                                ##   in Loop: Header=BB24_5 Depth=1
	cmpq	%r9, %rax
	je	LBB24_5

The set_pop() function entry and exit code is also tighter (formerly it saved and restored three registers, and now it skips that work).

[serhiy-storchaka]

Could you please provide microbenchmarks?

In most cases the loop ends after 1 or 2 iterations. Additional operations before and after the loop can eat the benefit of the optimization of short loop.

[serhiy-storchaka]

Is + 1 missed?

[rhettinger]

Yes. Thanks for noticing.

[rhettinger]

I can't run new benchmarks right now (my build has been broken for a couple of days since the extensive include file changes went in). The benchmark looked like this:

$ python -m timeit -r11 -s 's=set(range(10_000))' -s 's_pop=s.pop' 'while s: s_pop()'

[miss-islington]

@rhettinger: Status check is done, and it's a success ✅ .