New attribute trace_equivalent

[shyouhei]

Some instructions are spacial cases for another ones. These instructions need not preserve their trace counterparts. By reducing those unnecessary trace instructions we can strip binary size of vm_exec_core from 25,759 bytes to 24,924 bytes on my machine.

Yes, this changeset slows traces down a bit. But is that a problem?

[shyouhei]

This is a part of #2100.

[ko1]

These instructions need not preserve their trace counterparts

could you explain more with examples?

[shyouhei]

These instructions need not preserve their trace counterparts

could you explain more with examples?

For instance opt_plus has trace_opt_plus now. But trace_opt_send_without_block works 100% well when tracing opt_plus. So we are deleting trace_opt_plus and use trace_opt_send_without_block instead.

[k0kubun]

I don't have a strong opinion on which we should prioritize, a binary size or performance, but as long as we're making some decision on trade-off, please share the benchmark results just in case.

---

Possibly we may want to reuse the attr for

ruby/tool/ruby_vm/views/mjit_compile.inc.erb

Lines 29 to 32 in 14dd8d6

	% send_compatible_opt_insns = RubyVM::BareInstructions.to_a.select do |insn|
	% insn.name.start_with?('opt_') && opt_send_without_block.opes == insn.opes &&
	% insn.expr.expr.lines.any? { |l| l.match(/\A\s+CALL_SIMPLE_METHOD\(\);\s+\z/) }
	% end.map(&:name)

. If the information is not exactly the same as what you want to achieve, never mind.

[shyouhei]

@k0kubun I don't know any benchmark instance that focuses on tracing performance. Do you have one? This changeset obviously does not affect normal runs.

[ko1]

For instance opt_plus has trace_opt_plus now. But trace_opt_send_without_block works 100% well when tracing opt_plus. So we are deleting trace_opt_plus and use trace_opt_send_without_block instead.

How to replace opt_plus to trace_opt_send_without_block ? Or make trace_opt_plus insn and it goto trace_opt_send_without_block impl immediately?

@k0kubun I don't know any benchmark instance that focuses on tracing performance. Do you have one? This changeset obviously does not affect normal runs.

Now, if we replace trace_xxx insn, then we don't restore to xxx insn even if trace is disabled. It can cause performance downgrade.

[shyouhei]

How to replace opt_plus to trace_opt_send_without_block ? Or make trace_opt_plus insn and it goto trace_opt_send_without_block impl immediately?

Look at the patch :) That's the same as before. We only change how encoded_insn_data is constructed.

Now, if we replace trace_xxx insn, then we don't restore to xxx insn even if trace is disabled. It can cause performance downgrade.

Yes. That's what I refer in "is that a problem?" in the original comment. Can we measure the magnitude somehow?

[shyouhei]

#1970 can be something related to this.