String.indexOf optimization

[Vitaly Davidovich]

I'm fairly certain that nobody would object if you send a patch with
reproducible benchmark code demonstrating the gain.  I think this mostly
comes down to cost/benefit of doing this work, and the benefit is
marginalized out of the gate by virtue of the intrinsic.  You'd probably
also have to demonstrate no loss in performance for the common case of a
short string.

I don't have an answer to your test suite question, unfortunately.
Moreover, any such benchmarks would probably be measuring peak
performance,  which means the JIT intrinsic code.

Sent from my phone
On Jan 8, 2015 6:05 PM, "Zoltan Sziladi" <kissziszi at gmail.com> wrote:

> Thanks for the info.
> So that basically means we have 2 implementations of indexOf currently,
> one is in HotSpot, the other is in the JDK itself, which rarely gets
> executed. Aside from this later fact, isn't it still worth improving the
> JDK implementation if it is possible? I know that the intrinsified method
> gets executed most of the time, but still, if we can improve the JDK
> implementation also, why not? I don't know much about other JVMs but maybe
> a few don't intrinsify it?
>
> Is there any existing test suite which is considered conclusive enough
> that if an implementation beats the naive algorithm in those testcases then
> it could be considered as a replacement in the JDK?
>
> Regards,
> Zoltan
>
>
> On Thu, Jan 8, 2015 at 12:42 PM, Vitaly Davidovich <vitalyd at gmail.com>
> wrote:
>
>> 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:
>>
>>> Hi,
>>>
>>> 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.
>>>
>>> Regards,
>>> Zoltan
>>>
>>> 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
>>> String.indexOf.
>>> > > 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
>>> low-level
>>> > > 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
>>> MacroAssembler::string_indexof
>>> > in
>>> >
>>> >
>>> >
>>> http://hg.openjdk.java.net/jdk9/jdk9/hotspot/file/b6b89b8f8531/src/cpu/x86/vm/macroAssembler_x86.cpp
>>> >
>>> > > 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.
>>> >
>>> >
>>>
>>
>>
>