RFR: 8352696: JFR: assert(false): EA: missing memory path [v2]

[Aleksey Shipilev]

On Mon, 24 Mar 2025 12:22:10 GMT, Markus Grönlund <mgronlun at openjdk.org> wrote:

>> Greetings,
>> 
>> In debug builds, we can sometimes assert because C2's Escape Analysis does not recognize a pattern where one input of memory Phi node is a MergeMem node, and another is a RAW store.
>> 
>> This pattern is created by the jdk.jfr.internal.JVM.commit() intrinsic, which is inlined because of inlining JFR events, for example jdk.VirtualThreadStart.
>> 
>> As a result, EA complains about a strange memory graph. 
>> 
>> Testing: jdk_jfr
>> 
>> Thanks
>> Markus
>
> Markus Grönlund has updated the pull request incrementally with two additional commits since the last revision:
> 
>  - merged requires
>  - indentation

src/hotspot/share/opto/library_call.cpp line 3179:

> 3177:   store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, MemNode::release);
> 3178: #else
> 3179:   store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, MemNode::release);

BTW, you can write this is less duplication, like:


store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);


(I wish there was a `X`-style macro like `MakeConX`, but I don't think there is one for `BasicType`)

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

PR Review Comment: https://git.openjdk.org/jdk/pull/24192#discussion_r2010087366