Evaluate destination subtrees early when deferring destination eval

[rmnattas]

In x86 ArrayStoreCHK, with OffHeap the destination tree evaluation is deferred but its subtrees still need to be evaluated early especially if subtrees are complex and might introduce more instructions and virtual-registers.

WIP: Testing

[rmnattas]

@zl-wang @0xdaryl

[rmnattas]

Ready for merge

[zl-wang]

Jenkins test sanity xlinux jdk8,jdk23

[zl-wang]

@rmnattas you might need to take a look at the failures.

[rmnattas]

testJITServer_2 #19084 (dup of #14594)

jdk_security4_0 jdk_security4_1, connection issue @AdamBrousseau?, machine looks online now https://openj9-jenkins.osuosl.org/computer/ubu24%2Dx86%2D1/

[2025-02-26T22:17:49.267Z] STDERR:
[2025-02-26T22:17:49.267Z] java.net.UnknownHostException: ubu24-x86-1: ubu24-x86-1: Name or service not known
[2025-02-26T22:17:49.267Z] 	at java.base/java.net.InetAddress.getLocalHost(InetAddress.java:1941)
[2025-02-26T22:17:49.267Z] 	at NoAddresses.main(NoAddresses.java:51)
[2025-02-26T22:17:49.267Z] 	at java.base/jdk.internal.reflect.DirectMethodHandleAccessor.invoke(DirectMethodHandleAccessor.java:103)
...
[2025-02-26T22:17:49.267Z] Caused by: java.net.UnknownHostException: ubu24-x86-1: Name or service not known
[2025-02-26T22:17:49.267Z] 	at java.base/java.net.Inet6AddressImpl.lookupAllHostAddr(Native Method)
[2025-02-26T22:17:49.267Z] 	at java.base/java.net.Inet6AddressImpl.lookupAllHostAddr(Inet6AddressImpl.java:52)

[AdamBrousseau]

@rmnattas please open an infra issue for that. Could be a config problem related to hostname.

[zl-wang]

Not sure what range would that be, if for example offset <= (MAX_INT32 << MAX_SHIFT) is the appropriate check for that?

in x86-64 addressing modes, the scale value can be 1,2,4,or8 (i.e. index not shifted or 1/2/3 left shifted ... naturally aligned with the typical array element access as well). the index is expected to be in a register as usual (otherwise, it should be constant-folded already) ... that component is represented in our IL as lshl.

[rmnattas]

PR ready for merge. Extra condition and assert is sufficient to catch unexpected cases and after discussions trying to tailor for the different addressing modes would be too complex given that 1) this is for off-heap only and 2) its a temporary fix until commoning dataAddr is enabled.

[zl-wang]

Jenkins test sanity xlinux jdk8,jdk21