YJIT: Implement splat for cfuncs. Split exit cases
This also implements a new check for ruby2keywords as the last
argument of a splat. This does mean that we generate more code, but in
actual benchmarks where we gained speed from this (binarytrees) I
don't see any significant slow down. I did have to struggle here with
the register allocator to find code that didn't allocate too many
registers. It's a bit hard when everything is implicit. But I think I
got to the minimal amount of copying and stuff given our current
allocation strategy.
This reverts commit 650a20a3e1.
Now that 3.2.0 is released, let's disable flaky tests. Koichi said he'll
rework Ractor implementation for this, and it has not been done yet.
I noticed this while running test_yjit with --mjit-call-threshold=1,
which redefines `Integer#<`. When Ruby is monkey-patched,
MJIT itself could be broken.
Similarly, Ruby scripts could break MJIT in many different ways. I
prepared the same set of hooks as YJIT so that we could possibly
override it and disable it on those moments. Every constant under
RubyVM::MJIT is private and thus it's an unsupported behavior though.
This would be still flaky, but we want to make sure there's no
YJIT-specific issue when Ruby 3.2 is released. We might skip it again
after the release.
* YJIT: Make case-when optimization respect === redefinition
Even when a fixnum key is in the dispatch hash, if there is a case such
that its basic operations for === is redefined, we need to fall back to
checking each case like the interpreter. Semantically we're always
checking each case by calling === in order, it's just that this is not
observable when basic operations are intact.
When all the keys are fixnums, though, we can do the optimization we're
doing right now. Check for this condition.
* Update yjit/src/cruby_bindings.inc.rs
Co-authored-by: Takashi Kokubun <takashikkbn@gmail.com>
Co-authored-by: Takashi Kokubun <takashikkbn@gmail.com>
Previously we essentially never freed block even after invalidation.
Their reference count never reached zero for a couple of reasons:
1. `Branch::block` formed a cycle with the block holding the branch
2. Strong count on a branch that has ever contained a stub never
reached 0 because we increment the `.clone()` call for
`BranchRef::into_raw()` didn't have a matching decrement.
It's not safe to immediately deallocate blocks during
invalidation since `branch_stub_hit()` can end up
running with a branch pointer from an invalidated branch.
To plug the leaks, we wait until code GC or global invalidation and
deallocate the blocks for iseqs that are definitely not running.
@casperisfine reporting a bug in this gist https://gist.github.com/casperisfine/d59e297fba38eb3905a3d7152b9e9350
After investigating I found it was caused by a combination of send and a c_func that we have overwritten in the JIT. For send calls, we need to do some stack manipulation before making the call. Because of the way exits works, we need to do that stack manipulation at the last possible moment. In this case, we weren't doing that stack manipulation at all. Unfortunately, with how the code is structured there isn't a great place to do that stack manipulation for our overridden C funcs.
Each overridden C func can return a boolean stating that it shouldn't be used. We would need to do the stack manipulation after all of those checks are done. We could pass a lambda(?) or separate out the logic for "can I run this override" from "now generate the code for it". Since we are coming up on a release, I went with the path of least resistence and just decided to not use these overrides if we are in a send call.
We definitely should revist this in the future.
Follow-up for 2b8191bdad. Since that
commit, we stopped doing code invalidation the second time the call and
return events are enabled. We need to do it every time these events are
enabled because we might have generated code while these events are
disabled.
Also rename locals and edit comments to make it more clear that the iseq
rewrite code path only happens the first time a particular iseq trace
event is enabled.
The internal location in ractor.rb is not usefull at all.
```
$ ruby -e 'Ractor.new {}'
<internal:ractor>:267: warning: Ractor is experimental, ...
```
* YJIT: fix a parameter name
* YJIT: add support for calling bmethods
This commit adds support for the VM_METHOD_TYPE_BMETHOD method type in
YJIT. You can get these type of methods from facilities like
Kernel#define_singleton_method and Module#define_method.
Even though the body of these methods are blocks, the parameter setup
for them is exactly the same as VM_METHOD_TYPE_ISEQ, so we can reuse
the same logic in gen_send_iseq(). You can see this from how
vm_call_bmethod() eventually calls setup_parameters_complex() with
arg_setup_method.
Bmethods do need their frame environment to be setup differently. We
handle this by allowing callers of gen_send_iseq() to control the iseq,
the frame flag, and the prev_ep. The `prev_ep` goes into the same
location as the block handler would go into in an iseq method frame.
Co-authored-by: John Hawthorn <john@hawthorn.email>
Co-authored-by: John Hawthorn <john@hawthorn.email>
Object Shapes is used for accessing instance variables and representing the
"frozenness" of objects. Object instances have a "shape" and the shape
represents some attributes of the object (currently which instance variables are
set and the "frozenness"). Shapes form a tree data structure, and when a new
instance variable is set on an object, that object "transitions" to a new shape
in the shape tree. Each shape has an ID that is used for caching. The shape
structure is independent of class, so objects of different types can have the
same shape.
For example:
```ruby
class Foo
def initialize
# Starts with shape id 0
@a = 1 # transitions to shape id 1
@b = 1 # transitions to shape id 2
end
end
class Bar
def initialize
# Starts with shape id 0
@a = 1 # transitions to shape id 1
@b = 1 # transitions to shape id 2
end
end
foo = Foo.new # `foo` has shape id 2
bar = Bar.new # `bar` has shape id 2
```
Both `foo` and `bar` instances have the same shape because they both set
instance variables of the same name in the same order.
This technique can help to improve inline cache hits as well as generate more
efficient machine code in JIT compilers.
This commit also adds some methods for debugging shapes on objects. See
`RubyVM::Shape` for more details.
For more context on Object Shapes, see [Feature: #18776]
Co-Authored-By: Aaron Patterson <tenderlove@ruby-lang.org>
Co-Authored-By: Eileen M. Uchitelle <eileencodes@gmail.com>
Co-Authored-By: John Hawthorn <john@hawthorn.email>
Object Shapes is used for accessing instance variables and representing the
"frozenness" of objects. Object instances have a "shape" and the shape
represents some attributes of the object (currently which instance variables are
set and the "frozenness"). Shapes form a tree data structure, and when a new
instance variable is set on an object, that object "transitions" to a new shape
in the shape tree. Each shape has an ID that is used for caching. The shape
structure is independent of class, so objects of different types can have the
same shape.
For example:
```ruby
class Foo
def initialize
# Starts with shape id 0
@a = 1 # transitions to shape id 1
@b = 1 # transitions to shape id 2
end
end
class Bar
def initialize
# Starts with shape id 0
@a = 1 # transitions to shape id 1
@b = 1 # transitions to shape id 2
end
end
foo = Foo.new # `foo` has shape id 2
bar = Bar.new # `bar` has shape id 2
```
Both `foo` and `bar` instances have the same shape because they both set
instance variables of the same name in the same order.
This technique can help to improve inline cache hits as well as generate more
efficient machine code in JIT compilers.
This commit also adds some methods for debugging shapes on objects. See
`RubyVM::Shape` for more details.
For more context on Object Shapes, see [Feature: #18776]
Co-Authored-By: Aaron Patterson <tenderlove@ruby-lang.org>
Co-Authored-By: Eileen M. Uchitelle <eileencodes@gmail.com>
Co-Authored-By: John Hawthorn <john@hawthorn.email>
* Create code generation func
* Make rb_vm_concat_array available to use in Rust
* Map opcode to code gen func
* Implement code gen for concatarray
* Add test for concatarray
* Use new asm backend
* Add comment to C func wrapper
`YJIT.simulate_oom!` used to leave one byte of space in the code block,
so our test didn't expose a problem with asserting that the write
position is in bounds in `CodeBlock::set_pos`. We do the following when
patching code:
1. save current write position
2. seek to middle of the code block and patch
3. restore old write position
The bounds check fails on (3) when the code block is already filled up.
Leaving one byte of space also meant that when we write that byte, we
need to fill the entire code region with trapping instruction in
`VirtualMem`, which made the OOM tests unnecessarily slow.
Remove the incorrect bounds check and stop leaving space in the code
block when simulating OOM.
* Port gen_send_iseq to the new backend IR
* Replace occurrences of 8 by SIZEOF_VALUE
Co-authored-by: Maxime Chevalier-Boisvert <maximechevalierb@gmail.com>
Co-authored-by: Maxime Chevalier-Boisvert <maxime.chevalierboisvert@shopify.com>
* Port gen_send_cfunc to the new backend
* Remove an obsoleted test
* Add more cfunc tests
* Use csel_e instead and more into()
Co-authored-by: Maxime Chevalier-Boisvert <maximechevalierb@gmail.com>
* Add a missing lea for build_kwargs
* Split cfunc test cases
Co-authored-by: Maxime Chevalier-Boisvert <maximechevalierb@gmail.com>
Teach getblockparamproxy to handle the no-block case without exiting
Co-authored-by: John Hawthorn <john@hawthorn.email>
Co-authored-by: John Hawthorn <john@hawthorn.email>
The test in [1] was removed because it stopped working when we limited
the power of Kernel#binding in [2]. However, the underlying issue could
still be reproduced using blocks. Add back a regression test.
I tested the test by commenting out the fix from [1].
[1]: 54c91042ed
[2]: 343ea9967e
