YJIT: Replace Array#each only when YJIT is enabled #11955
Conversation
b67e1c3 to
e045b6b
Compare
e045b6b to
192faf1
Compare
2c561a7 to
b66caef
Compare
This comment has been minimized.
This comment has been minimized.
There was a problem hiding this comment.
We should make sure that once the Ruby implementation is subbed in, it's still a basic method, because we have code like the following for example in Enumerable#inject:
if (iter == inject_op_i &&
SYMBOL_P(op) &&
RB_TYPE_P(obj, T_ARRAY) &&
rb_method_basic_definition_p(CLASS_OF(obj), id_each)) {
return ary_inject_op(obj, init, op);
}
There was a problem hiding this comment.
Given that we are likely to port more runtime methods to Ruby (bring back Aaron's ports and add some more?), and that this may increase generated code size, I think it would be good to have a command line flag (or an environment variable?) so we can turn this on or off. This would allow us to test the impact on benchmarks and in production at various stages.
An environment variable might be easier for production?
|
I was so confused by the
Fixed in d06ac2b
Added |
|
Good work Kokubun! 👏 |
This PR resurrects the C implementation of
Array#eachfor the interpreter and lets YJIT use the Ruby one. To minimize incompatibilities between the interpreter and YJIT, it also introducesPrimitive.attr! :c_traceto letArray#eachact like a C method in backtraces.It speeds up the interpreter on lobsters a little. While the rewrite of
Array#eachin Ruby #9533 did not slow downArray#eachon the interpreter with microbenchmarks, it had a small impact on larger benchmarks.