RFR: 8032218: Emit single post-constructor barrier for chain of superclass constructors

[Joshua Cao]

On Fri, 19 Apr 2024 22:31:10 GMT, Joshua Cao <duke at openjdk.org> wrote:

> Opening this PR on top of https://github.com/openjdk/jdk/pull/18505. This PR is only valid if we agree it is sufficient to use `StoreStore` barriers at the end of constructors instead of `Release` barriers.
> 
> Currently on master, [C2 emits a Release barrier for each constructor call](https://github.com/openjdk/jdk/blob/32946e1882e9b22c983cbba3c6bda3cc7295946a/src/hotspot/share/opto/parse1.cpp#L1019) in a chain of superclass constructor calls. After https://github.com/openjdk/jdk/pull/18505 is merged, it is the same except that the barrier is a `StoreStore`. It is unnecessary. We only need to emit a single barrier for each object allocation / each pair of `Allocation/InitializeNode`. 
> 
> [Macro expansion emits a trailing StoreStore after an InitializeNode](https://github.com/openjdk/jdk/blob/32946e1882e9b22c983cbba3c6bda3cc7295946a/src/hotspot/share/opto/macro.cpp#L1610-L1628). This `StoreStore` is sufficient as the post-constructor barrier. From the [InitializeNode definition](https://github.com/openjdk/jdk/blob/32946e1882e9b22c983cbba3c6bda3cc7295946a/src/hotspot/share/opto/memnode.cpp#L3639-L3642):
> 
>> // An InitializeNode collects and isolates object initialization after
> // an AllocateNode and before the next possible safepoint.  As a
> // memory barrier (MemBarNode), it keeps critical stores from drifting
> // down past any safepoint or any publication of the allocation.
> 
> All the writes that occur in the constructor will come before the `InitializeNode/StoreStore`. This PR removes the emitting of `StoreStore` barriers in `Parse::do_exits()`. This PR subsumes https://github.com/openjdk/jdk/pull/18505.
> 
> Passes hotspot tier1 locally on x86 linux machine. New tests make sure that there is a single `StoreStore` for chained constructors.

> Doesn't this only work if the allocation and call to ctor are compiled together? Where is the StoreStore added if we compile a method by itself? The allocation could be in the interpreter but the when calling we call compiled code.

Yes, I think you're right. We should probably still emit the barrier when it is the outermost method. Unfortunately, I don't think we can directly IR test on a constructor.

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

PR Comment: https://git.openjdk.org/jdk/pull/18870#issuecomment-2071044716