Re: [PATCH] perf bench: Add benchmark of find_next_bit

[Arnaldo Carvalho de Melo]

Em Wed, Jul 29, 2020 at 04:59:18PM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Tue, Jul 28, 2020 at 08:51:52AM -0300, Arnaldo Carvalho de Melo escreveu:
> > Em Fri, Jul 24, 2020 at 12:19:59AM -0700, Ian Rogers escreveu:
> > > for_each_set_bit, or similar functions like for_each_cpu, may be hot
> > > within the kernel. If many bits were set then one could imagine on
> > > Intel a "bt" instruction with every bit may be faster than the function
> > > call and word length find_next_bit logic. Add a benchmark to measure
> > > this.
> > 
> > Thanks, applied.

> > > This benchmark on AMD rome and Intel skylakex shows "bt" is not a good
> > > option except for very small bitmaps.
> 
> > > +++ b/tools/perf/bench/find-bit-bench.c
> 
> > > +#if defined(__i386__) || defined(__x86_64__)
> > > +static bool asm_test_bit(long nr, const unsigned long *addr)
> > > +{
> > > +	bool oldbit;
> > > +
> > > +	asm volatile("bt %2,%1"
> > > +		     : "=@ccc" (oldbit)
> > > +		     : "m" (*(unsigned long *)addr), "Ir" (nr) : "memory");
> > > +
> > > +	return oldbit;
> 
> Some old clang versions are not liking this:

Failed with:

  clang version 3.8.0 (tags/RELEASE_380/final)
  clang version 3.8.1 (tags/RELEASE_381/final)
  clang version 4.0.0 (tags/RELEASE_400/final)
  Alpine clang version 8.0.0 (tags/RELEASE_800/final) (based on LLVM 8.0.0)
  Alpine clang version 5.0.0 (tags/RELEASE_500/final) (based on LLVM 5.0.0)
  Alpine clang version 5.0.1 (tags/RELEASE_501/final) (based on LLVM 5.0.1)
  Alpine clang version 5.0.1 (tags/RELEASE_502/final) (based on LLVM 5.0.1)

Worked with:

  Alpine clang version 9.0.0 (https://git.alpinelinux.org/aports f7f0d2c2b8bcd6a5843401a9a702029556492689) (based on LLVM 9.0.0)
  Alpine clang version 10.0.0 (https://gitlab.alpinelinux.org/alpine/aports.git 7445adce501f8473efdb93b17b5eaf2f1445ed4c)
  Alpine clang version 10.0.0 (git://git.alpinelinux.org/aports 7445adce501f8473efdb93b17b5eaf2f1445ed4c)


Also failed for;

# grep FAIL dm.log/summary  | grep -v alpine
 alt:p8: FAIL
   clang version 3.8.0 (tags/RELEASE_380/final)
 alt:p9: FAIL
   clang version 7.0.1
 amazonlinux:1: FAIL
   clang version 3.6.2 (tags/RELEASE_362/final)
 amazonlinux:2: FAIL
   clang version 7.0.1 (Amazon Linux 2 7.0.1-1.amzn2.0.2)
#