Misc #16258: [PATCH] Combine call info and cache to speed up method invocation

• Status: Closed
• Assignee: ko1 (Koichi Sasada)

[alanwu (Alan Wu)]

Proposed change: https://github.com/ruby/ruby/pull/2564

To perform a regular method call, the VM needs two structs, rb_call_info
and rb_call_cache. At the moment, we allocate these two structures in
separate buffers. In the worst case, the CPU needs to read 4 cache lines to
complete a method call. Putting the two structures together reduces the
maximum number of cache line reads to 2.

Combining the structures also saves 8 bytes per call site as the current
layout uses separate pointers for the call info and the call cache. This
change saves about 2 MiB on Discourse.

The Optcarrot benchmark receives a performance improvement from this patch. I
collected the following results using make install binaries compiled with
-DRUBY_NDEBUG, with a sample size of 50 for each category:

	master-a5245c	after patch	speed-up ratio
plain	42.39	50.17	18.35%
jit	71.72	72.73	1.41%

These are medium FPS from the benchmark output. For raw benchmark results and
basic stats, see
https://gist.github.com/XrXr/ce5cb7cf2c3c4d29e58c919fa5c86b33. I took these
results with a i7-8750H CPU @ 2.20GHz on a 2018 MacBook Pro. I also ran the
benchmark with a AMD 2400G running Arch Linux and observed a 3% improvement
without the jit.

Complications

• A new instruction attribute comptime_sp_inc is introduced to calculate
  SP increase at compile time without using call caches. At compile time, a
  TS_CALLDATA operand points to a call info struct, but at runtime, the
  same operand points to a call data struct. Instruction that explicitly
  define sp_inc also need to define comptime_sp_inc.
• MJIT code for copying call cache becomes slightly more complicated.
• This changes the bytecode format, which might break existing tools.

I think this patch offers a good general performance boost for a manageable amount
of code change.

[ko1 (Koichi Sasada)]

• Assignee set to ko1 (Koichi Sasada)

Thank you for your patch.

Conclusion: OK.

Points:

• Current implementation separates ci and cc because of CoW friendliness (ci is immutable data and cc is mutable data). However, there are no measurements how it affect on CoW friendliness. Bcause ci is immutable data, we can pre-compile these data and it will improve startup time. However, there are no implementation of it.
• For Guild, I will rewrite inline cache (cc) because of atomicity. However, Ruby 2.7 doesn't have this change. For Ruby 2.7 only this patch is accepted.

[alanwu (Alan Wu)]

• Status changed from Open to Closed

Applied in changeset git|89e7997622038f82115f34dbb4ea382e02bed163.

---

Combine call info and cache to speed up method invocation

To perform a regular method call, the VM needs two structs,
rb_call_info and rb_call_cache. At the moment, we allocate these two
structures in separate buffers. In the worst case, the CPU needs to read
4 cache lines to complete a method call. Putting the two structures
together reduces the maximum number of cache line reads to 2.

Combining the structures also saves 8 bytes per call site as the current
layout uses separate two pointers for the call info and the call cache.
This saves about 2 MiB on Discourse.

This change improves the Optcarrot benchmark at least 3%. For more
details, see attached bugs.ruby-lang.org ticket.

Complications:

• A new instruction attribute comptime_sp_inc is introduced to
  calculate SP increase at compile time without using call caches. At
  compile time, a TS_CALLDATA operand points to a call info struct, but
  at runtime, the same operand points to a call data struct. Instruction
  that explicitly define sp_inc also need to define comptime_sp_inc.
• MJIT code for copying call cache becomes slightly more complicated.
• This changes the bytecode format, which might break existing tools.

[Misc #16258]