On 3/08/2018 8:23 AM, Andrew Haley wrote:
On 08/02/2018 10:44 PM, Magnus Ihse Bursie wrote:
2 aug. 2018 kl. 14:07 skrev Andrew Haley <a...@redhat.com>:
On 08/02/2018 07:35 AM, David Holmes wrote:
In theory yes - in practice I don't know if anyone has tried it. How
would you do a native build using the ARM64 sources rather than the
aarch64 sources?
It's fine. I used:
sh ./configure --with-native-debug-symbols=internal
--disable-warnings-as-errors --disable-hotspot-gtest --enable-dtrace=no
--with-jtreg=/home/aph/jtreg
--with-boot-jdk=/local/jdk10-pristine/build/linux-aarch64-normal-server-release/images/jdk/
--enable-precompiled-headers --with-debug-level=release
--with-jvm-features=-aot,-jvmci
... but the important part is to disable aot and jvmci.
I think what David meant was that it's unclear if it's possible to build the
ARM64 port natively, i.e. using --with-cpu-port=arm64, instead of the default
--with-cpu-port=aarch64.
Sorry, I typo'd the configuration line.
In fact, --with-cpu-port=arm64 doesn't work at all because
/local/jdk-jdk11/src/hotspot/cpu/arm/gc/shared/barrierSetAssembler_arm.cpp:70:30:
error: ‘src’ was not declared in this scope
__ encode_heap_oop(src);
and this fails regardless of cross compilation. So arm(64) does not matter:
it's obsolete and does not build.
Broken by:
changeset: 49950:7b916885654d
user: shade
date: Wed May 02 19:26:42 2018 +0200
summary: 8201786: Modularize interpreter GC barriers: leftovers for
ARM32
Bob flagged the port for removal but not sure what the state of that is.
David
-----
