Improve handling of endianness

[simonameng]

• Add native function isBigEndian in JITHelpers.java following the implementation of Java_sun_misc_Unsafe_isBigEndian0.

• Replace IS_PLATFORM_LITTLE_ENDIAN with the new implemented function isBigEndian in JITHelpers.java and remove IS_PLATFORM_LITTLE_ENDIAN relevant code.

• Use putByteInArrayByIndex function to put the zero byte into the right index into char[], instead of checking the byteOrder and then doing &= 0x00FF or &= 0xFF00

• Inline and fold away java/nio/ByteOrder.nativeOrder and java/nio/Bits.byteOrder. The reason I inline and fold away java/nio/Bits.byteOrder is that JDK8 calls Bits.byteOrder in ByteOrder.nativeOrder instead of ByteOrder.NATIVE_ORDER, which is a static field in JDK11.

Closes: #4741

[fjeremic]

Changes look good! We'll need to carry out performance experiments once we get an approval from someone on the optimizer side.

[andrewcraik]

I would appreciate a justification for the change to JITHelpers - adding the recognized methods so that we can try to inline them and do good things for the java/nio package seems reasonable, but the changes to JITHelpers seem like a step backwards since we cannot constant fold and add a complex dependency from JITHelpers which it has never previously had.

[fjeremic]

Jenkins test sanity all jdk8,jdk11

[pshipton]

@fjeremic pls don't run all for more than one jdk version at a time. I believe your request caused jenkins to restart, and we lost all the running builds.

[fjeremic]

@fjeremic pls don't run all for more than one jdk version at a time. I believe your request caused jenkins to restart, and we lost all the running builds.

Sorry about that. Will not do this in the future. Can we disable this combination on the infrastructure side to prevent people from launching the builds? Is there an issue tracking a fix? IIRC this used to work in the past.

[fjeremic]

Jenkins test sanity all jdk8

[pshipton]

Can we disable this combination on the infrastructure side?

@AdamBrousseau ?

Is there an issue tracking a fix?

https://bugs.eclipse.org/bugs/show_bug.cgi?id=553326

[pshipton]

IIRC this used to work in the past.

https://openj9.slack.com/archives/C8PQL5N65/p1575512682081300

[AdamBrousseau]

We've disabled keywords all for versions and all for testing. That doesn't stop a user from spelling them all out. For platforms, all maps to a (large) subset of platforms (found here). One caveat is this only maps in a PR build right now, so personal builds (internally) which launch all platforms actually get all platforms, which is a different problem/discussion. I don't want to hardcode logic for "if you launch all, limit to 1 version", since we can run 4 versions simultaneously internally just fine.

[fjeremic]

Ok, I see so this is an external machine connection flakiness issue. I agree with not hardcoding. I'll keep an eye if I see people launching too many builds to inform them of this restriction. Thanks!

Jenkins test sanity all jdk11

[fjeremic]

@andrewcraik this change is ready. I'll wait for your approval before merging.

[andrewcraik]

Can we please get an updated PR message? The commits need a squash too

[fjeremic]

Revoking approval to prevent myself from accidentally merging. @simonameng let's address the previous comment and we should be ready here.

[fjeremic]

Force push link says there is nothing different between this version and the last, sans the commit message fixup and squashing. For sanity I'll just kick off a single build, although it is not warranted given previous testing.

Jenkins test sanity xlinux jdk8,jdk11