YJIT - Yet Another Ruby JIT

YJIT is a lightweight, minimalistic Ruby JIT built inside CRuby. It lazily compiles code using a Basic Block Versioning (BBV) architecture. YJIT is currently supported for macOS, Linux and BSD on x86-64 and arm64/aarch64 CPUs. This project is open source and falls under the same license as CRuby.

If you're using YJIT in production, pleaseshare your success stories with us!

If you wish to learn more about the approach taken, here are some conference talks and publications:

• MPLR 2023 talk:Evaluating YJIT’s Performance in a Production Context: A Pragmatic Approach

• RubyKaigi 2023 keynote:Optimizing YJIT’s Performance, from Inception to Production

• RubyKaigi 2023 keynote:Fitting Rust YJIT into CRuby

• RubyKaigi 2022 keynote:Stories from developing YJIT

• RubyKaigi 2022 talk:Building a Lightweight IR and Backend for YJIT

• RubyKaigi 2021 talk:YJIT: Building a New JIT Compiler Inside CRuby

• Blog post:YJIT: Building a New JIT Compiler Inside CRuby

• MPLR 2023 paper:Evaluating YJIT’s Performance in a Production Context: A Pragmatic Approach

• VMIL 2021 paper:YJIT: A Basic Block Versioning JIT Compiler for CRuby

• MoreVMs 2021 talk:YJIT: Building a New JIT Compiler Inside CRuby

• ECOOP 2016 talk:Interprocedural Type Specialization of JavaScript Programs Without Type Analysis

• ECOOP 2016 paper:Interprocedural Type Specialization of JavaScript Programs Without Type Analysis

• ECOOP 2015 talk:Simple and Effective Type Check Removal through Lazy Basic Block Versioning

• ECOOP 2015 paper:Simple and Effective Type Check Removal through Lazy Basic Block Versioning

To cite YJIT in your publications, please cite the MPLR 2023 paper:

@inproceedings{yjit_mplr_2023,author = {Chevalier-Boisvert, Maxime and Kokubun, Takashi and Gibbs, Noah and Wu, Si Xing (Alan) and Patterson, Aaron and Issroff, Jemma},title = {Evaluating YJIT’s Performance in a Production Context: A Pragmatic Approach},year = {2023},isbn = {9798400703805},publisher = {Association for Computing Machinery},address = {New York, NY, USA},url = {https://doi.org/10.1145/3617651.3622982},doi = {10.1145/3617651.3622982},booktitle = {Proceedings of the 20th ACM SIGPLAN International Conference on Managed Programming Languages and Runtimes},pages = {20–33},numpages = {14},keywords = {dynamically typed, optimization, just-in-time, virtual machine, ruby, compiler, bytecode},location = {Cascais, Portugal},series = {MPLR 2023}}

Current Limitations

YJIT may not be suitable for certain applications. It currently only supports macOS, Linux and BSD on x86-64 and arm64/aarch64 CPUs. YJIT will use more memory than the Ruby interpreter because the JIT compiler needs to generate machine code in memory and maintain additional state information. You can change how much executable memory is allocated usingYJIT’s command-line options.

Installation

Requirements

You will need to install:

• All the usual build tools for Ruby. SeeBuilding Ruby

• The Rust compilerrustc

  • The Rust version must be>= 1.58.0.

• Optionally, only if you wish to build in dev/debug mode, Rust’scargo

If you don’t intend on making code changes to YJIT itself, we recommend obtainingrustc through your OS’s package manager since that likely reuses the same vendor which provides the C toolchain.

If you will be changing YJIT’s Rust code, we suggest using thefirst-party installation method for Rust. Rust also provides first classsupport for many source code editors.

Building YJIT

Start by cloning theruby/ruby repository:

git clone https://github.com/ruby/ruby yjitcd yjit

The YJITruby binary can be built with either GCC or Clang. It can be built either in dev (debug) mode or in release mode. For maximum performance, compile YJIT in release mode with GCC. More detailed build instructions are provided in theRuby README.

# Configure in release mode for maximum performance, build and install./autogen.sh./configure --enable-yjit --prefix=$HOME/.rubies/ruby-yjit --disable-install-docmake -j && make install

or

# Configure in lower-performance dev (debug) mode for development, build and install./autogen.sh./configure --enable-yjit=dev --prefix=$HOME/.rubies/ruby-yjit --disable-install-docmake -j && make install

Dev mode includes extended YJIT statistics, but can be slow. For only statistics you can configure in stats mode:

# Configure in extended-stats mode without slow runtime checks, build and install./autogen.sh./configure --enable-yjit=stats --prefix=$HOME/.rubies/ruby-yjit --disable-install-docmake -j && make install

On macOS, you may need to specify where to find some libraries:

# Install dependenciesbrew install openssl libyaml# Configure in dev (debug) mode for development, build and install./autogen.sh./configure --enable-yjit=dev --prefix=$HOME/.rubies/ruby-yjit --disable-install-doc --with-opt-dir="$(brew --prefix openssl):$(brew --prefix readline):$(brew --prefix libyaml)"make -j && make install

Typically configure will choose the default C compiler. To specify the C compiler, use

# Choosing a specific c compilerexport CC=/path/to/my/chosen/c/compiler

before running./configure.

You can test that YJIT works correctly by running:

# Quick tests found in /bootstraptestmake btest# Complete set of testsmake -j test-all

Usage

Examples

Once YJIT is built, you can either use./miniruby from within your build directory, or switch to the YJIT version ofruby by using thechruby tool:

chruby ruby-yjitruby myscript.rb

You can dump statistics about compilation and execution by running YJIT with the--yjit-stats command-line option:

./miniruby --yjit-stats myscript.rb

You can see what YJIT has compiled by running YJIT with the--yjit-log command-line option:

./miniruby --yjit-log myscript.rb

The machine code generated for a given method can be printed by addingputs RubyVM::YJIT.disasm(method(:method_name)) to a Ruby script. Note that no code will be generated if the method is not compiled.

Command-Line Options

YJIT supports all command-line options supported by upstream CRuby, but also adds a few YJIT-specific options:

• --yjit: enable YJIT (disabled by default)

• --yjit-mem-size=N: soft limit on YJIT memory usage in MiB (default: 128). Tries to limitcode_region_size + yjit_alloc_size

• --yjit-exec-mem-size=N: hard limit on executable memory block in MiB. Limitscode_region_size

• --yjit-call-threshold=N: number of calls after which YJIT begins to compile a function. It defaults to 30, and it’s then increased to 120 when the number of ISEQs in the process reaches 40,000.

• --yjit-cold-threshold=N: number of global calls after which an ISEQ is considered cold and not compiled, lower values mean less code is compiled (default 200K)

• --yjit-stats: print statistics after the execution of a program (incurs a run-time cost)

• --yjit-stats=quiet: gather statistics while running a program but don’t print them. Stats are accessible throughRubyVM::YJIT.runtime_stats. (incurs a run-time cost)

• --yjit-log[=file|dir]: log all compilation events to the specified file or directory. If no name is supplied, the last 1024 log entries will be printed to stderr when the application exits.

• --yjit-log=quiet: gather a circular buffer of recent YJIT compilations. The compilation log entries are accessible throughRubyVM::YJIT.log and old entries will be discarded if the buffer is not drained quickly. (incurs a run-time cost)

• --yjit-disable: disable YJIT despite other--yjit* flags for lazily enabling it withRubyVM::YJIT.enable

• --yjit-code-gc: enable codeGC (disabled by default as of Ruby 3.3). It will cause all machine code to be discarded when the executable memory size limit is hit, meaning JIT compilation will then start over. This can allow you to use a lower executable memory size limit, but may cause a slight drop in performance when the limit is hit.

• --yjit-perf: enable frame pointers and profiling with theperf tool

• --yjit-trace-exits: produce aMarshal dump of backtraces from all exits. Automatically enables--yjit-stats

• --yjit-trace-exits=COUNTER: produce aMarshal dump of backtraces from a counted exit or a fallback. Automatically enables--yjit-stats

• --yjit-trace-exits-sample-rate=N: trace exit locations only every Nth occurrence. Automatically enables--yjit-trace-exits

Note that there is also an environment variableRUBY_YJIT_ENABLE which can be used to enable YJIT. This can be useful for some deployment scripts where specifying an extra command-line option to Ruby is not practical.

You can also enable YJIT at run-time usingRubyVM::YJIT.enable. This can allow you to enable YJIT after your application is done booting, which makes it possible to avoid compiling any initialization code.

You can verify that YJIT is enabled usingRubyVM::YJIT.enabled? or by checking thatruby --yjit -v includes the string+YJIT:

ruby --yjit -vruby 3.3.0dev (2023-01-31T15:11:10Z master 2a0bf269c9) +YJIT dev [x86_64-darwin22]ruby --yjit -e "p RubyVM::YJIT.enabled?"trueruby -e "RubyVM::YJIT.enable; p RubyVM::YJIT.enabled?"true

Benchmarking

We have collected a set of benchmarks and implemented a simple benchmarking harness in theyjit-bench repository. This benchmarking harness is designed to disable CPU frequency scaling, set process affinity and disable address space randomization so that the variance between benchmarking runs will be as small as possible.

Performance Tips for Production Deployments

While YJIT options default to what we think would work well for most workloads, they might not necessarily be the best configuration for your application. This section covers tips on improving YJIT performance in case YJIT does not speed up your application in production.

Increasing –yjit-mem-size

The--yjit-mem-size value can be used to set the maximum amount of memory that YJIT is allowed to use. This corresponds to the total ofRubyVM::YJIT.runtime_stats[:code_region_size] andRubyVM::YJIT.runtime_stats[:yjit_alloc_size] Increasing the--yjit-mem-size value means more code can be optimized by YJIT, at the cost of more memory usage.

If you start Ruby with--yjit-stats, e.g. using an environment variableRUBYOPT=--yjit-stats,RubyVM::YJIT.runtime_stats[:ratio_in_yjit] shows the percentage of total YARV instructions executed by YJIT as opposed to the CRuby interpreter. Ideally,ratio_in_yjit should be as large as 99%, and increasing--yjit-mem-size often helps improvingratio_in_yjit.

Running workers as long as possible

It’s helpful to call the same code as many times as possible before a process restarts. If a process is killed too frequently, the time taken for compiling methods may outweigh the speedup obtained by compiling them.

You should monitor the number of requests each process has served. If you’re periodically killing worker processes, e.g. withunicorn-worker-killer orpuma_worker_killer, you may want to reduce the killing frequency or increase the limit.

Reducing YJIT Memory Usage

YJIT allocates memory for JIT code and metadata. Enabling YJIT generally results in more memory usage. This section goes over tips on minimizing YJIT memory usage in case it uses more than your capacity.

Decreasing –yjit-mem-size

YJIT uses memory for compiled code and metadata. You can change the maximum amount of memory that YJIT can use by specifying a different--yjit-mem-size command-line option. The default value is currently128. When changing this value, you may want to monitorRubyVM::YJIT.runtime_stats[:ratio_in_yjit] as explained above.

Enabling YJIT lazily

If you enable YJIT by--yjit options orRUBY_YJIT_ENABLE=1, YJIT may compile code that is used only during the application boot.RubyVM::YJIT.enable allows you to enable YJIT from Ruby code, and you can call this after your application is initialized, e.g. on Unicorn’safter_fork hook. If you use any YJIT options (--yjit-*), YJIT will start at boot by default, but--yjit-disable allows you to start Ruby with the YJIT-disabled mode while passing YJIT tuning options.

Code Optimization Tips

This section contains tips on writing Ruby code that will run as fast as possible on YJIT. Some of this advice is based on current limitations of YJIT, while other advice is broadly applicable. It probably won’t be practical to apply these tips everywhere in your codebase. You should ideally start by profiling your application using a tool such asstackprof so that you can determine which methods make up most of the execution time. You can then refactor the specific methods that make up the largest fractions of the execution time. We do not recommend modifying your entire codebase based on the current limitations of YJIT.

• Avoid usingOpenStruct

• Avoid redefining basic integer operations (i.e. +, -, <, >, etc.)

• Avoid redefining the meaning ofnil, equality, etc.

• Avoid allocating objects in the hot parts of your code

• Minimize layers of indirection

• Avoid writing wrapper classes if you can (e.g. a class that only wraps a Ruby hash)

• Avoid methods that just call another method

• Ruby method calls are costly. Avoid things such as methods that only return a value from a hash

• Try to write code so that the same variables and method arguments always have the same type

• Avoid usingTracePoint as it can cause YJIT to deoptimize code

• Avoid usingbinding as it can cause YJIT to deoptimize code

You can also use the--yjit-stats command-line option to see which bytecodes cause YJIT to exit, and refactor your code to avoid using these instructions in the hottest methods of your code.

Other Statistics

If you runruby with--yjit-stats, YJIT will track and return performance statistics inRubyVM::YJIT.runtime_stats.

$RUBYOPT="--yjit-stats"irbirb(main):001:0>RubyVM::YJIT.runtime_stats=>{:inline_code_size=>340745,:outlined_code_size=>297664,:all_stats=>true,:yjit_insns_count=>1547816,:send_callsite_not_simple=>7267,:send_kw_splat=>7,:send_ivar_set_method=>72,...

Some of the counters include:

• :yjit_insns_count - how many Ruby bytecode instructions have been executed

• :binding_allocations - number of bindings allocated

• :binding_set - number of variables set via a binding

• :code_gc_count - number of garbage collections of compiled code since process start

• :vm_insns_count - number of instructions executed by the Ruby interpreter

• :compiled_iseq_count - number of bytecode sequences compiled

• :inline_code_size - size in bytes of compiled YJIT blocks

• :outline_code_size - size in bytes of YJIT error-handling compiled code

• :side_exit_count - number of side exits taken at runtime

• :total_exit_count - number of exits, including side exits, taken at runtime

• :avg_len_in_yjit - avg. number of instructions in compiled blocks before exiting to interpreter

Counters starting with “exit_” show reasons for YJIT code taking a side exit (return to the interpreter.)

Performance counter names are not guaranteed to remain the same between Ruby versions. If you’re curious what each counter means, it’s usually best to search the source code for it — but it may change in a later Ruby version.

The printed text after a--yjit-stats run includes other information that may be named differently than the information inRubyVM::YJIT.runtime_stats.

Contributing

We welcome open source contributions. You should feel free to open new issues to report bugs or just to ask questions. Suggestions on how to make this readme file more helpful for new contributors are most welcome.

Bug fixes and bug reports are very valuable to us. If you find a bug in YJIT, it’s very possible be that nobody has reported it before, or that we don’t have a good reproduction for it, so please open an issue and provide as much information as you can about your configuration and a description of how you encountered the problem. List the commands you used to run YJIT so that we can easily reproduce the issue on our end and investigate it. If you are able to produce a small program reproducing the error to help us track it down, that is very much appreciated as well.

If you would like to contribute a large patch to YJIT, we suggest opening an issue or a discussion on theShopify/ruby repository so that we can have an active discussion. A common problem is that sometimes people submit large pull requests to open source projects without prior communication, and we have to reject them because the work they implemented does not fit within the design of the project. We want to save you time and frustration, so please reach out so we can have a productive discussion as to how you can contribute patches we will want to merge into YJIT.

Source Code Organization

The YJIT source code is divided between:

• yjit.c: code YJIT uses to interface with the rest of CRuby

• yjit.h: C definitions YJIT exposes to the rest of the CRuby

• yjit.rb:YJIT Ruby module that is exposed to Ruby

• yjit/src/asm/*: in-memory assembler we use to generate machine code

• yjit/src/codegen.rs: logic for translating Ruby bytecode to machine code

• yjit/src/core.rb: basic block versioning logic, core structure of YJIT

• yjit/src/stats.rs: gathering of run-time statistics

• yjit/src/options.rs: handling of command-line options

• yjit/src/cruby.rs: C bindings manually exposed to the Rust codebase

• yjit/bindgen/src/main.rs: C bindings exposed to the Rust codebase through bindgen

The core of CRuby’s interpreter logic is found in:

• insns.def: defines Ruby’s bytecode instructions (gets compiled intovm.inc)

• vm_insnshelper.c: logic used by Ruby’s bytecode instructions

• vm_exec.c: Ruby interpreter loop

Generating C bindings with bindgen

In order to expose C functions to the Rust codebase, you will need to generate C bindings:

CC=clang ./configure --enable-yjit=devmake -j yjit-bindgen

This uses the bindgen tools to generate/updateyjit/src/cruby_bindings.inc.rs based on the bindings listed inyjit/bindgen/src/main.rs. Avoid manually editing this file as it could be automatically regenerated at a later time. If you need to manually add C bindings, add them toyjit/cruby.rs instead.

Coding & Debugging Protips

There are multiple test suites:

• make btest (see/bootstraptest)

• make test-all

• make test-spec

• make check runs all of the above

• make yjit-check runs quick checks to see that YJIT is working correctly

The tests can be run in parallel like this:

make -j test-all RUN_OPTS="--yjit-call-threshold=1"

Or single-threaded like this, to more easily identify which specific test is failing:

make test-all TESTOPTS=--verbose RUN_OPTS="--yjit-call-threshold=1"

To run a single test file withtest-all:

make test-all TESTS='test/-ext-/marshal/test_usrmarshal.rb' RUNRUBYOPT=--debugger=lldb RUN_OPTS="--yjit-call-threshold=1"

It’s also possible to filter tests by name to run a single test:

make test-all TESTS='-n /test_float_plus/' RUN_OPTS="--yjit-call-threshold=1"

You can also run one specific test inbtest:

make btest BTESTS=bootstraptest/test_ractor.rb RUN_OPTS="--yjit-call-threshold=1"

There are shortcuts to run/debug your own test/repro intest.rb:

make run  # runs ./miniruby test.rbmake lldb # launches ./miniruby test.rb in lldb

You can use the Intel syntax for disassembly in LLDB, keeping it consistent with YJIT’s disassembly:

echo "settings set target.x86-disassembly-flavor intel" >> ~/.lldbinit

Running x86 YJIT on Apple’s Rosetta

For development purposes, it is possible to run x86 YJIT on an Apple M1 via Rosetta. You can find basic instructions below, but there are a few caveats listed further down.

First, install Rosetta:

$ softwareupdate --install-rosetta

Now any command can be run with Rosetta via thearch command line tool.

Then you can start your shell in an x86 environment:

$ arch -x86_64 zsh

You can double check your current architecture via thearch command:

$ arch -x86_64 zsh$ archi386

You may need to set the default target forrustc to x86-64, e.g.

$ rustup default stable-x86_64-apple-darwin

While in your i386 shell, install Cargo and Homebrew, then hack away!

Rosetta Caveats

1. You must install a version of Homebrew for each architecture

2. Cargo will install in $HOME/.cargo by default, and I don’t know a good way to change architectures after install

If you use Fish shell you canread this link for information on making the dev environment easier.

Profiling with Linux perf

--yjit-perf allows you to profile JIT-ed methods along with other native functions using Linux perf. When you run Ruby withperf record, perf looks up/tmp/perf-{pid}.map to resolve symbols in JIT code, and this option lets YJIT write method symbols into that file as well as enabling frame pointers.

Call graph

Here’s an example way to use this option withFirefox Profiler (See also:Profiling with Linux perf):

# Compile the interpreter with frame pointers enabled./configure --enable-yjit --prefix=$HOME/.rubies/ruby-yjit --disable-install-doc cflags=-fno-omit-frame-pointermake -j && make install# [Optional] Allow running perf without sudoecho 0 | sudo tee /proc/sys/kernel/kptr_restrictecho -1 | sudo tee /proc/sys/kernel/perf_event_paranoid# Profile Ruby with --yjit-perfcd ../yjit-benchPERF="record --call-graph fp" ruby --yjit-perf -Iharness-perf benchmarks/liquid-render/benchmark.rb# View results on Firefox Profiler https://profiler.firefox.com.# Create /tmp/test.perf as below and upload it using "Load a profile from file".perf script --fields +pid > /tmp/test.perf

YJIT codegen

You can also profile the number of cycles consumed by code generated by each YJIT function.

# Install perfapt-get install linux-tools-common linux-tools-generic linux-tools-`uname -r`# [Optional] Allow running perf without sudoecho 0 | sudo tee /proc/sys/kernel/kptr_restrictecho -1 | sudo tee /proc/sys/kernel/perf_event_paranoid# Profile Ruby with --yjit-perf=codegencd ../yjit-benchPERF=record ruby --yjit-perf=codegen -Iharness-perf benchmarks/lobsters/benchmark.rb# Aggregate resultsperf script > /tmp/perf.txt../ruby/misc/yjit_perf.py /tmp/perf.txt

Building perf with Python support

The above instructions work fine for most people, but you could also use a handyperf script -s interface if you build perf from source.

# Build perf from source for Python supportsudo apt-get install libpython3-dev python3-pip flex libtraceevent-dev \  libelf-dev libunwind-dev libaudit-dev libslang2-dev libdw-devgit clone --depth=1 https://github.com/torvalds/linuxcd linux/tools/perfmakemake install# Aggregate resultsperf script -s ../ruby/misc/yjit_perf.py