vitalyd at gmail.com
Thu Jan 8 20:42:37 UTC 2015
The java impl you saw would be called by (a) interpreter, (b) if you
explicitly disable intrinsification of this function, or (c) some other JVM
that doesn't intrinsify this method (or any method).
People don't usually disable intrinsics; if they do, it's because they hit
some JIT bug and may disable it.
On Thu, Jan 8, 2015 at 3:34 PM, Zoltan Sziladi <kissziszi at gmail.com> wrote:
> Thanks everyone for all the info.
> So, just to go step by step in understanding this.
> Andrew said HotSpot would ignore my implementation. So why is there an
> implementation of indexOf at all in the JDK, if that's not the code that's
> executed? Is it just a default fallback? When is the indexOf function not
> intrinsified? When do people usually disable intrinsification?
> Sorry if these are newbie questions, I'm new to this part of Java.
> On Tue, Jan 6, 2015 at 1:28 AM, Andrew Haley <aph at redhat.com> wrote:
> > Hi,
> > On 05/01/15 18:59, Zoltan Sziladi wrote:
> > > This discussion was a long time ago, I was just reading through it to
> > check
> > > again what was the last state of the discussion about the
> > > There is one part which I still do not understand, hopefully someone
> > could
> > > shed some light on it. A few emails ago Martin mentioned
> > >
> > > "Hotspot seems to have some intrinsification of String.indexOf, which
> > > confuses me.
> > > Hotspot seems the right place to provide more optimizations for this,
> > since
> > > there has been a fair amount of work creating high-performance
> > > implementations of this idea in C."
> > >
> > > Then Ivan asked what that actually meant, whether hotspot actually
> > replaced
> > > the jdk implementation with a low level optimized C implementation,
> but I
> > > never saw an answer to that.
> > You can have a look at an implementation of
> > in
> > > Can someone please explain this? If we somehow found an algorithm that
> > beat
> > > the naive implementation in the average case, would it be possible to
> > just
> > > implement it in the JDK?
> > No, because HotSpot would ignore it. Any speed improvements have to be
> > done in the architecture-dependent files.
> > Andrew.
More information about the core-libs-dev