RFR: 8291752: AArch64: Remove check_emit_size parameter from trampoline_call [v3]
Andrew Haley
aph at openjdk.org
Wed Aug 10 09:51:38 UTC 2022
On Mon, 8 Aug 2022 15:25:55 GMT, Evgeny Astigeevich <duke at openjdk.org> wrote:
>> `MacroAssembler::trampoline_call` can skip generation of a trampoline stub if C2 phase output is in the checking emitted code size mode. If it is not C2 compilation, `trampoline_call` always generates a trampoline stub.
>>
>> `gen_continuation_enter` as a part of C2 compilation needed to disable the functionality above. Now `trampoline_call` has the `check_emit_size` which allows to disable the functionality completely. Its default value is `true` which means the functionality is enabled.
>>
>> There are problems:
>> - C2 specific code which must be properly guarded if it is not C2 compilation. The code has the risk of undefined behaviour (UB). It is done with `is_c2_compile`. [JDK-8291654](https://bugs.openjdk.org/browse/JDK-8291654) is an example of changes caused UB.
>> - The default value of `check_emit_size`. The current value `true` allows C1 to access the C2 specific code. Setting the default value to `false` will require to review and to update every existing use of `trampoline_call`. It will complicate uses of `trampoline_call`: each time a decision needs to be done - use or not use the default.
>>
>> Summary of changes:
>> - Have a virtual `in_scratch_emit_size` returning `false` by default. It is privately overridden in `C2_MacroAssembler`. The overridden version returns `true` if C2 phase output is in the checking emitted code size mode.
>> - `check_emit_size` is removed because it is not needed any more.
>>
>> Tested with fastdebug/release builds:
>> - `gtest`: Passed.
>> - `tier1`...`tier4`: Passed.
>
> Evgeny Astigeevich has updated the pull request incrementally with one additional commit since the last revision:
>
> Add virtual in_scratch_emit_size instead of emit_trampoline_stub
Marked as reviewed by aph (Reviewer).
-------------
PR: https://git.openjdk.org/jdk/pull/9782
More information about the hotspot-compiler-dev
mailing list