Cross compiling openjdk9, problem with HotSpot assembly code

[Bruno Kremel]

Hi Alex,

Just a quick question. Does the toolchain have to be hardfloat?
Currently I use arm-linux-gnueabi (not hf, it's linaro toolchain).

Thanks
Bruno

Alex Kashchenko <akashche at redhat.com> writes:

> Hi Bruno,
>
> On 03/29/2017 08:42 PM, Bruno Kremel wrote:
>> Hi All,
>>
>> I am trying to compile Openjdk 9 (HG tag changeset 162) and I do
>> following:
>> ./configure --with-jvm-interpreter=cpp --with-jvm-variants=zero \
>>
>> [...]
>>
>> It seems that it tries to compile some x86_64 assembly code which does
>> not make sense when clearly configure states that I have the ARM target.
>>
>> The other thing I don't understand why does it even compile hotspot
>> code? I have asked for zero interpreter AND openjdk only, I actually
>> don't want to use HotSpot I intend to use JamVM with OpenJDK classpath.
>>
>> Do you have any suggestions.
>
> jdk9 build system tries to build a native "build jdk" (to not be
> confused with "boot jdk") first and then uses it during the cross-build.
> If you cross compiler has correct prefix (arm-linux-gnueabihf,
> aarch64-linux-gnu etc) and default compiler is a native one - that
> should work out of the box. Alternatively you can build native x86_64
> jdk9 from the same hg revision manually and specify it with
> "--with-build-jdk" switch.
>
> Also AFAIK Zero port is currently not functional in jdk9, though you can
> build a full ARM32 VM (server/client/minimal) instead using
> "--openjdk-target=arm-linux-gnueabihf" and
> "--with-abi-profile=arm-vfp-hflt" switches.