RFR: 8335006: C2 SuperWord: add JMH benchmark VectorLoadToStoreForwarding.java

[Aleksey Shipilev]

On Tue, 25 Jun 2024 12:56:16 GMT, Emanuel Peter <epeter at openjdk.org> wrote:

> I'm committing the benchmark mentioned in my blog article:
> https://eme64.github.io/blog/2024/06/24/Auto-Vectorization-and-Store-to-Load-Forwarding.html
> 
> These are the numbers I got on my machine. Read the blog for a full analysis.
> ![image](https://github.com/openjdk/jdk/assets/32593061/28ba310d-1f7e-4c5c-ae0e-1f38e6214603)
> 
> **Quick summary:**
> We see that vectorization is only beneficial for `OFFSET = 0, 1, 2, 4, 8, 16`, at least with the current SuperWord implementation. The regression for `OFFSET = 3 `is especially stark.
> 
> The effect that explains the slowdown of vectorization for the other cases is failed store-to-load-forwarding.
> 
> I ran it like this on my machine:
> `make test TEST="micro:vm.compiler.VectorLoadToStoreForwarding" CONF=linux-x64`

Marked as reviewed by shade (Reviewer).

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

PR Review: https://git.openjdk.org/jdk/pull/19880#pullrequestreview-2138663559