RFR: 8287788: Implement a better allocator for downcalls [v9]

Maurizio Cimadamore mcimadamore at openjdk.org
Thu Jan 23 12:43:58 UTC 2025


On Wed, 22 Jan 2025 12:38:25 GMT, Matthias Ernst <duke at openjdk.org> wrote:

>> Right. Using SlicingAllocator now.
>> 
>> Which brings up one question about alignment of the stack frames. When the linker asks for a buffer, it also has an alignment requirement. Do we guarantee anything about the alignment of Arena.ofConfined().allocate() ? Does the linker overallocate already?
>> 
>> To accomodate for this, I added a frame alignment parameter as well => the stack itself is well-served by using a SlicingAllocator. For this, I added methods to query and reset the position, as well as whether it could satisfy a certain allocation request.
>
> And yes, this requires that a client operate in stack order. I added checks for this.

> To accomodate for this, I added a frame alignment parameter as well => the stack itself is well-served by using a SlicingAllocator. For this, I added methods to query and reset the position, as well as whether it could satisfy a certain allocation request.

I like what you did w.r.t. alignment - I think the current revision seems to now handle that correctly

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

PR Review Comment: https://git.openjdk.org/jdk/pull/23142#discussion_r1926916051


More information about the core-libs-dev mailing list