On 04/09/2014 22:13, John Rose wrote:
We have standard OpenJDK source locations for platform-specific code, as
described in http://openjdk.java.net/jeps/201 .
:
But, there is no place at present for (a) ISA-dependent source files, or (b)
ISA-and-platform-dependent source files.
To date then we've only had a very small amount of ISA specific code in
the jdk repo and just not enough to be important in the source code
organization. Aside from the 2D code then there are a few places using
ifdef to deal with alignment and endianness on specific architectures
but it's rare.
If there is a significant amount of ISA specific code coming in then the
layout should support it. Introducing siblings to share and $OS make
sense to me. A slight variation is 'share'. $ISA to make it consistent
with the share tree (but perhaps the 'cpu.' was chosen to be consistent
with the hotspot repo?).
An important topic for the layout discussion is that it should be easy
for the build to determine what needs to be compiled. Erik and Magnus
are the folks to comment on topic. With the new layout then we hope to
get to the point soon where the $OS-specific sources are moved out of
the unix tree to the appropriate place in the $OS tree (this has been a
historical source of excludes and complications in the build). I'm just
wondering if introducing ISA specific directories has any issues - I
assume the starting assumption for the build is that all sources
matching the target $OS and $ISA are compiled, non-matching directories
are ignored. In the example that you discussed with Alex then I assume
that unix.sparcv9 works well when the target build is solaris-sparcv9 or
linux-sparcv9. There might be some VIS version specific instructions but
I would think this goes along with version detection at runtime and that
the code would still be placed in the unix.sparcv9 tree.
-Alan
