OpenJDK7 port to the little endian PowerPC64

[Lindenmaier, Goetz]

Hi Alexander,

I had a look at the hotspot change.

For jdk7, we don't need a bug id, as this repo is only managed by 
the ppc-aix-port team.
But I assume you also want to get these changes into jdk8/9,
for what we do need one.  I'll generate them once we have
a change ready to push to 8/9.

I would propose to split the hotspot change into the fixes for the
new abi and those for the endianess. The abi changes are almost ppc-only 
and thus easier to get reviewed. I prepared a change including only
that:
http://cr.openjdk.java.net/~goetz/webrevs/abi_elfv2/

I fixed some minor issues:
 - some alignments
 - avoid ! in #ifs
 - There was a stray '}' in macroAssembler.inline.hpp breaking our build.

In elfFile.cpp, you should please use PPC64 for the test.  Else you 
might break the PPC32 build. I fixed that, too.

I removed the PrintStubCode sections. I think this should be fixed 
in a general way working for all platforms. 

Please have a look at this change and test it.  . Feel free to further
change it.  If the change is ok for you I'll push it right away


For the remaining changes to hotspot it would be great if we could
avoid all the duplication of files. They are empty, anyways.
Adding all the #includes in shared code will raise discussions
in the openJDK community.  I think you just need to change arch_model and 
os_arch_model in platform_ppc64le.

I would prefer if you use VM_LITTLE_ENDIAN to distinguish in C-code, e.g.
in bytes_ppc.hpp. You could also name bytes_linux_ppc64le.inline.hpp 
without 'le' and guard the content by VM_LITTLE_ENDIAN.

Did you build and use SA? Does it work? So far, we never tried to.
This also might raise discussions in the openJDK community, as
you define it for ppc32, too.

Your change in os_linux.cpp to arch_array looks wrong. As you use 
the same code entry for both ppc64, in the loop below, running_arch_code 
will always point to your new entry.

Further I think, the CPU architecture string should be 

#elif defined(PPC64)
#ifdef VM_LITTLE_ENDIAN
static char cpu_arch[] = "ppc64le";
#else
static char cpu_arch[] = "ppc64";
#endif


I removed the warning in adlc by downporting a fix from jdk9.

Best regards,
  Goetz.

-----Original Message-----
From: Alexander Smundak [mailto:asmundak at google.com] 
Sent: Samstag, 15. Februar 2014 00:38
To: Lindenmaier, Goetz
Cc: ppc-aix-port-dev at openjdk.java.net
Subject: Re: OpenJDK7 port to the little endian PowerPC64

I haven't done this before, and I don't have an account on openjdk, so
I had to ask my coworker to upload the webrevs. The one for the
hotspot is at http://cr.openjdk.java.net/~rasbold/ppc64le/hotspot/webrev.00,
and the one for jdk is at
http://cr.openjdk.java.net/~rasbold/ppc64le/jdk/webrev.00
The change for the jdk/ is small, but the change for the hotspot/ is
fairly large and I am willing to break it into pieces if you say so.
Oh, and can you create a bug ID for this?

Alexander

On Fri, Feb 14, 2014 at 12:04 AM, Lindenmaier, Goetz
<goetz.lindenmaier at sap.com> wrote:
> Hi Alexander,
>
> Nice to hear you are using the port!
>
> Yes, we can commit these changes to jdk7u.  Can you supply
> two webrevs, for hotspot and jdk?
>
> Best regards,
>   Goetz.
>
> -----Original Message-----
> From: ppc-aix-port-dev-bounces at openjdk.java.net [mailto:ppc-aix-port-dev-bounces at openjdk.java.net] On Behalf Of Alexander Smundak
> Sent: Donnerstag, 13. Februar 2014 20:07
> To: ppc-aix-port-dev at openjdk.java.net
> Subject: OpenJDK7 port to the little endian PowerPC64
>
> I am working on porting OpenJDK7 to the little endian PowerPC64
> running Linux (little endian support is part of the OpenPower
> initiative).
> The port is based on OpenJDK7 for PowerPC in the
> http://hg.openjdk.java.net/ppc-aix-port/jdk7u. It is functioning: it
> can build itself, and all but a few Jtreg tests in jdk/test succeed.
> The port was initially targeting the "old" ABI (that is, the same ABI
> that big endian PowerPC64 is using), and is now targeting v2 ABI (the
> function descriptors are dropped and two words are dropped from the
> base frame).
> Would it be possible to add this port to the ppc-aix-port/jdk7u
> repository? The changes are moderate, 'hg diff' for the hotspot ~2300
> lines, plus ~400 lines for jdk et al.
>
> Alexander Smundak