[ruby-core:100610] [Ruby master Feature#17288] Optimize __send__ call with a literal method name

[eregontp@gmail.com]

Issue #17288 has been updated by Eregon (Benoit Daloze).


I think `obj.send(:foo)` should be optimized too, to not create a an arbitrary performance difference between `send` and `__send__`.
`send` is also used ~15x used more frequently than `__send__` in gems, so that would be even more valuable.

Of course, optimizing `send` means checking that `obj.method(:send)` corresponds to `Kernel#send` (somewhat similar to `+`).

@matz suggested an operator for `__send__` in https://twitter.com/yukihiro_matz/status/1320496276476063744, that might be a cleaner solution.
OTOH, it would not benefit existing code and not be adopted for a few years.

FWIW in TruffleRuby I've been thinking to optimize all 3 sends (`__send__,` send, public_send), since they are hidden from backtraces, etc, and optimizing them makes it simpler to remove them from backtraces, etc. It would also reduce the overhead in interpreter for `send/__send__(:private_method)` calls (in JITed code, those have no overhead). I would still check the correct method in all cases, changing semantics should IMHO be the last resort.

----------------------------------------
Feature #17288: Optimize __send__ call with a literal method name
https://bugs.ruby-lang.org/issues/17288#change-88241

* Author: mrkn (Kenta Murata)
* Status: Open
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
----------------------------------------
I made a patch to optimize a `__send__` call with a literal method name.  This optimization replaces a `__send__` method call with a `send` instruction.  The patch is available in [this pull-request](https://github.com/ruby/ruby/pull/3707).

By this change, the redefined `__send__` method is no longer called when it is called by a literal method name.  I guess it is no problem because the following warning message is displayed for a long time.

    $ ruby -e 'def __send__; end'
    -e:1: warning: redefining `__send__' may cause serious problems

This change makes the optimized case x5~x6 faster.  The benchmark result is below:

```
$ make benchmark COMPARE_RUBY="../../ruby/build-o3/ruby" ITEM=vm_send.yml
(snip)
# Iteration per second (i/s)

|             |compare-ruby|built-ruby|
|:------------|-----------:|---------:|
|vm_send      |     18.536M|  113.778M|
|             |           -|     6.14x|
|vm_send_var  |     18.085M|   16.595M|
|             |       1.09x|         -|
```



-- 
https://bugs.ruby-lang.org/