OpenJDK-8: Crypto performance is very slow on ARMv7

Alex Kashchenko akashche at redhat.com
Mon Jan 30 11:04:47 UTC 2017


Hi Srinivas,

On 01/30/2017 10:48 AM, Srinivasa Rao Ragolu wrote:
> Hi Andrew,
>
> But hotspot module gives lot of compilation issues . So below statements
> are added to recipe (
> http://git.yoctoproject.org/cgit/cgit.cgi/meta-java/tree/recipes-core/openjdk/openjdk-8-release-102b14.inc
> )
>
> # Enable zero mode for arm based builds, as normal hotspot fails to build
> PACKAGECONFIG_append_arm = " zero"
> Any clue on, how to proceed further? Thanks in advance, Srinivas.

I can suggest to build it first without yocto, using something like this 
(this is for native build with gcc 4 or 5):

bash ./configure \
	--enable-unlimited-crypto=yes \
	--disable-debug-symbols \
	--with-jvm-variants=client \
	--with-debug-level=release \
	--with-boot-jdk=path/to/zero/jdk \
	--with-cacerts-file=path/to/cacerts \
	--with-milestone=[...] \
	--with-update-version=[...] \
	--with-build-number=[...]

And then to proceed with cross-compiling and contributing aarch32 build 
scripts back to yocto.

Or, if your ARM device can run recent Ubuntu or Fedora userspace, you 
can try to use prebuilt aarch32 binaries.

> On Mon, Jan 30, 2017 at 3:45 PM, Andrew Haley <aph at redhat.com> wrote:
>
>> On 30/01/17 07:02, Srinivasa Rao Ragolu wrote:
>>> Any other inputs that I could improve crypto performance?
>>
>> Yes.  Don't use zero.  Zero is slow.
>>
>> Andrew.
>>
>>


-- 
-Alex


More information about the aarch32-port-dev mailing list