OpenJDK-8: Crypto performance is very slow on ARMv7

Srinivasa Rao Ragolu sragolu at mvista.com
Fri Jan 27 13:09:23 UTC 2017


Hi Edward,

Thanks for the precious input. Let me try out source from aarch32-port and
test.

I will update on results once I finish my testing.

Thanks in advance,
Srinivas.

On Fri, Jan 27, 2017 at 5:14 PM, Edward Nevill <edward.nevill at gmail.com>
wrote:

> Hi Srinivas,
>
> On Fri, 2017-01-27 at 15:48 +0530, Srinivasa Rao Ragolu wrote:
> > Hi All,
> >
> > I have ported openjdk-8-72b05(from Yocto) for armv7 . I am sharing
> --------------------------^^^^^
>
> I am not familiar with this, but my guess is that you are building a Zero
> image. IE. You are building the reference interpreter implementation.
>
> I repeated your tests with a C1 build form armv7
>
> With C1 I get
>
> real    0m10.745s
> user    0m10.925s
> sys     0m0.235s
>
> With -Xint I get
>
> real    3m15.089s
> user    3m14.550s
> sys     0m0.280s
>
> Which is the same ballpark that you are getting.
>
> My guess is that yocto is picking up the sources from
>
> http://hg.openjdk.java.net/jdk8u/jdk8u/
>
> which does not have support for the C1 compiler.
>
> To build a version with the C1 compiler you need to use the source at
>
> http://hg.openjdk.java.net/aarch32-port/jdk8u/
>
> I have put a prebuilt image at
>
> http://camswl.com/images/jdk8-arm32-release.tgz
>
> which may work on your system (or may not depending on your shared libs)
> if you wish to give it a try.
>
> All the best,
> Ed.
>
>
> > code via link, which I have run to see the crypto performance and found
> it
> > is very slow.
> >
> > Code Location:
> > https://drive.google.com/open?id=0BxeL64n1TNdJMlItSWpld1lIdjg
> >
>
>


More information about the aarch32-port-dev mailing list