Hi Matthew, On Sat, Mar 11, 2017 at 06:56:40AM -0800, Matthew Wilcox wrote: > On Mon, Feb 06, 2017 at 12:16:44AM +0900, Minchan Kim wrote: > > +static inline void zram_fill_page(char *ptr, unsigned long len, > > + unsigned long value) > > +{ > > + int i; > > + unsigned long *page = (unsigned long *)ptr; > > + > > + WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long))); > > + > > + if (likely(value == 0)) { > > + memset(ptr, 0, len); > > + } else { > > + for (i = 0; i < len / sizeof(*page); i++) > > + page[i] = value; > > + } > > +} > > I've hacked up memset32/memset64 for both ARM and x86 here: > > http://git.infradead.org/users/willy/linux-dax.git/shortlog/refs/heads/memfill
Thanks for the patch. > > Can you do some performance testing and see if it makes a difference? I tested that zram is *full* with non-zero 100M dedupable data(i.e., it's a ideal case) on x86. With this, I see 7% enhancement. perf stat -r 10 dd if=/dev/zram0 of=/dev/null vanilla: 0.232050465 seconds time elapsed ( +- 0.51% ) memset_l: 0.217219387 seconds time elapsed ( +- 0.07% ) I doubt it makes such benefit in read workload which a small percent non-zero dedup data(e.g., under 3%) but it makes code simple/perform win. Thanks. > > At this point, I'd probably ask for the first 5 patches in that git > branch to be included, and leave out memfill and the shoddy testsuite. > > I haven't actually tested either asm implementation ... only the > C fallback.