RFR: Load coops base shift from AOTRuntimeConstants in AOT code [v4]

[Andrew Dinn]

On Wed, 25 Sep 2024 13:58:15 GMT, Andrew Dinn <adinn at openjdk.org> wrote:

>> This PR modifies AOT compiled method code to load compressed oops base and shift constants via the AOTRuntimeConstants area rather than encode them as immediates. It also unlatches the currently forced setting of UseCompatibleCompressedOops, allowing the heap to be allocated wherever it will fit.
>
> Andrew Dinn has updated the pull request incrementally with one additional commit since the last revision:
> 
>   fix oop decode not to side-effect flags

I debugged the deopt and found that the interpreter frame was restored at the ifeq (bci 109) as you might expect but that the restored top of stack was `0xdeaddeaf00000001`. This value ought to have been 0 since we were seeing a failed instanceof.

Digging deeper I found that the problem was actually in the generated method code. The problem was actually an aarch64 compiler/macro assembler issue because C2 had interleaved an oop decode between the ifeq test (cmp) and subsequent deopt branch (beq). This was down to a detail of how I implemented the decode for aot code.

My implementation of `decode_heap_oop_for_aot` followed `encode_heap_oop_for_aot` by using `cmp` + `csel`. However, I failed to notice that `encode_heap_oop` can only use `cmp` + `csel` safely because C2 instruction `encodeHeapOop` declares `effect(KILL cr)`. I could have fixed this by adding a kill effect to `decodeheapOop` but instead I thought it safer to modify `decode_heap_oop_for_aot` to follow `decode_heap_oop` and employ a local `cbnz` branch. This has fixed the problem.

I'll try to get some performance figures now that it is working.

-------------

PR Comment: https://git.openjdk.org/leyden/pull/20#issuecomment-2374302414