On Wed, 1 Apr 2026 09:16:32 GMT, Andrew Dinn <[email protected]> wrote:
>> Something does indeed. The size of the call is known and fixed, so we need >> to get anything variably sized in first. > > When you say the call size is fixed what exactly do you mean? Does that > include the post call nop? the restore profile code? > > It works hiding the save inside `align_call` (and likewise hiding the restore > in `call` and `ic_call`) for the arches you have currently implemented this > for and it will probably extend ok to riscv. But it doesn't look like it is > as good a fit for s390 where call alignment is handled differently. > > Even if it can be made to work on s390 this still hides an important step > (well, actually both the save and restore) in an innocuous wrapper and that > smells bad to me. Can you not declare generic save and restore methods that > have to be called explicitly at places where a call is (aligned and) planted > -- whether in generic or arch-specific code and then allow them to be > implemented in each arch? Decoupling these operations from the `align_call` > or `call` methods does risk a save and restore being missed but it also > makes it clearer what is happening. Done. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/28541#discussion_r3059405458
