PrecompileTools
Overview
PrecompileTools
is designed to help reduce delay on first usage of Julia code. It can force precompilation of specific workloads; particularly with Julia 1.9 and higher, the precompiled code is automatically saved to disk, so that it doesn't need to be compiled freshly in each Julia session. You can use PrecompileTools
as a package developer, to reduce the latency experienced by users of your package for "typical" workloads; you can also use PrecompileTools
as a user, creating custom "Startup" package(s) that precompile workloads important for your work.
The main tool in PrecompileTools
is a macro, @compile_workload
, which precompiles all the code needed to execute the workload. It also includes a second macro, @setup_workload
, which can be used to "mark" a block of code as being relevant only for precompilation but which does not itself force compilation of @setup_workload
code. (@setup_workload
is typically used to generate test data using functions that you don't need to precompile in your package.) Finally, PrecompileTools
includes @recompile_invalidations
to mitigate the undesirable consequences of invalidations. These different tools are demonstrated below.
The latency reductions from PrecompileTools are maximally effective for Julia versions 1.9 and higher, and intermediate for Julia 1.8. Julia versions 1.7 and earlier may see some limited benefit as well, but have also occasionally been found to suffer from precompilation-induced runtime performance regressions. If you wish, you can disable precompilation on older Julia versions by wrapping precompilation statements (see below) with if Base.VERSION >= v"1.8" ... end
. On older Julia versions, you may wish to test packages for performance regressions when introducing precompilation directives.
Tutorial: forcing precompilation with workloads
No matter whether you're a package developer or a user looking to make your own workloads start faster, the basic workflow of PrecompileTools
is the same. Here's an illustration of how you might use @compile_workload
and @setup_workload
:
module MyPackage
using PrecompileTools: @setup_workload, @compile_workload # this is a small dependency
struct MyType
x::Int
end
struct OtherType
str::String
end
@setup_workload begin
# Putting some things in `@setup_workload` instead of `@compile_workload` can reduce the size of the
# precompile file and potentially make loading faster.
list = [OtherType("hello"), OtherType("world!")]
@compile_workload begin
# all calls in this block will be precompiled, regardless of whether
# they belong to your package or not (on Julia 1.8 and higher)
d = Dict(MyType(1) => list)
x = get(d, MyType(2), nothing)
last(d[MyType(1)])
end
end
end
When you build MyPackage
, it will precompile the following, including all their callees:
Pair(::MyPackage.MyType, ::Vector{MyPackage.OtherType})
Dict(::Pair{MyPackage.MyType, Vector{MyPackage.OtherType}})
get(::Dict{MyPackage.MyType, Vector{MyPackage.OtherType}}, ::MyPackage.MyType, ::Nothing)
getindex(::Dict{MyPackage.MyType, Vector{MyPackage.OtherType}}, ::MyPackage.MyType)
last(::Vector{MyPackage.OtherType})
In this case, the "top level" calls were fully inferrable, so there are no entries on this list that were called by runtime dispatch. Thus, here you could have gotten the same result with manual precompile
directives. The key advantage of @compile_workload
is that it works even if the functions you're calling have runtime dispatch.
Once you set up a block using PrecompileTools
, try your package and see if it reduces the time to first execution, using the same workload you put inside the @compile_workload
block.
If you're happy with the results, you're done! If you want deeper verification of whether it worked as expected, or if you suspect problems, the SnoopCompile package provides diagnostic tools. Potential sources of trouble include invalidation (diagnosed with SnoopCompileCore.@snoopr
and related tools) and omission of intended calls from inside the @compile_workload
block (diagnosed with SnoopCompileCore.@snoopi_deep
and related tools).
@compile_workload
works by monitoring type-inference. If the code was already inferred prior to @compile_workload
(e.g., from prior usage), you might omit any external methods that were called via runtime dispatch.
You can use multiple @compile_workload
blocks if you need to interleave @setup_workload
code with code that you want precompiled. You can use @snoopi_deep
to check for any (re)inference when you use the code in your package. To fix any specific problems, you can combine @compile_workload
with manual precompile
directives.
Tutorial: local "Startup" packages
Users who want to precompile workloads that have not been precompiled by the packages they use can follow the recipe above, creating custom "Startup" packages for each project. Imagine that you have three different kinds of analyses you do: you could have a folder
MyData/
Project1/
Project2/
Project3/
From each one of those Project
folders you could do the following:
(@v1.9) pkg> activate .
Activating new project at `/tmp/Project1`
(Project1) pkg> generate Startup
Generating project Startup:
Startup/Project.toml
Startup/src/Startup.jl
(Project1) pkg> dev ./Startup
Resolving package versions...
Updating `/tmp/Project1/Project.toml`
[e9c42744] + Startup v0.1.0 `Startup`
Updating `/tmp/Project1/Manifest.toml`
[e9c42744] + Startup v0.1.0 `Startup`
(Project1) pkg> activate Startup/
Activating project at `/tmp/Project1/Startup`
(Startup) pkg> add PrecompileTools LotsOfPackages...
In the last step, you add PrecompileTools
and all the package you'll need for your work on Project1
as dependencies of Startup
. Then edit the Startup/src/Startup.jl
file to look similar to the tutorial previous section, e.g.,
module Startup
using LotsOfPackages...
using PrecompileTools
@compile_workload begin
# inside here, put a "toy example" of everything you want to be fast
end
end
Then when you're ready to start work, from the Project1
environment just say using Startup
. All the packages will be loaded, together with their precompiled code.
If desired, the Reexport package can be used to ensure these packages are also exported by Startup
.
Tutorial: "healing" invalidations
Julia sometimes invalidates previously compiled code (see Why does Julia invalidate code?). PrecompileTools provides a mechanism to recompile the invalidated code so that you get the full benefits of precompilation. This capability can be used in "Startup" packages (like the one described above), as well as by package developers.
Excepting piracy (which is heavily discouraged), type-stable (i.e., well-inferred) code cannot be invalidated. If invalidations are a problem, an even better option than "healing" the invalidations is improving the inferrability of the "victim": not only will you prevent invalidations, you may get faster performance and slimmer binaries. Packages that can help identify inference problems and invalidations include SnoopCompile, JET, and Cthulhu.
The basic usage is simple: wrap expressions that might invalidate with @recompile_invalidations
. Invalidation can be triggered by defining new methods of external functions, including during package loading. Using the "Startup" package above, you might wrap the using
statements:
module Startup
using PrecompileTools
@recompile_invalidations begin
using LotsOfPackages...
end
# Maybe a @compile_workload here?
end
Note that recompiling invalidations can be useful even if you don't add any additional workloads.
Alternatively, if you're a package developer worried about "collateral damage" you may cause by extending functions owned by Base or other package (i.e., those that require import
or module-scoping when defining the method), you can wrap those method definitions:
module MyContainers
using AnotherPackage
using PrecompileTools
struct Container
list::Vector{Any}
end
# This is a function created by this package, so it doesn't need to be wrapped
make_container() = Container([])
@recompile_invalidations begin
# Only those methods extending Base or other packages need to go here
Base.push!(obj::Container, x) = ...
function AnotherPackage.foo(obj::Container)
⋮
end
end
end
You can have more than one @recompile_invalidations
block in a module. For example, you might use one to wrap your using
s, and a second to wrap your method extensions.
Package developers should be aware of the tradeoffs in using @recompile_invalidations
to wrap method extensions:
- the benefit is that you might deliver a better out-of-the-box experience for your users, without them needing to customize anything
- the downside is that it will increase the precompilation time for your package. Worse, what can be invalidated once can sometimes be invalidated again by a later package, and if that happens the time spent recompiling is wasted.
Using @recompile_invalidations
in a "Startup" package is, in a sense, safer because it waits for all the code to be loaded before recompiling anything. On the other hand, this requires users to implement their own customizations.
Package developers are encouraged to try to fix "known" invalidations rather than relying reflexively on @recompile_invalidations
.
When you can't run a workload
There are cases where you might want to precompile code but cannot safely execute that code: for example, you may need to connect to a database, or perhaps this is a plotting package but you may be currently on a headless server lacking a display, etc. In that case, your best option is to fall back on Julia's own precompile
function. However, as explained in How PrecompileTools works, there are some differences between precompile
and @compile_workload
; most likely, you may need multiple precompile
directives. Analysis with SnoopCompile may be required to obtain the results you want; in particular, combining @snoopi_deep
and parcel
will allow you to generate a set of precompile
directives that can be include
d in your module definition.
Be aware that precompile
directives are more specific to the Julia version, CPU (integer width), and OS than running a workload.
Troubleshooting
Ensure your workload "works" (runs without error) when copy/pasted into the REPL. If it produces an error only when placed inside @precompile_workload
, check whether your workload runs when wrapped in a
let
# workload goes here
end
block.
Package developers: reducing the cost of precompilation during development
If you're frequently modifying one or more packages, you may not want to spend the extra time precompiling the full set of workloads that you've chosen to make fast for your "shipped" releases. One can locally reduce the cost of precompilation for selected packages using the Preferences.jl
-based mechanism and the "precompile_workload"
key: from within your development environment, use
using MyPackage, Preferences
set_preferences!(MyPackage, "precompile_workload" => false; force=true)
This will write the following to LocalPreferences.toml alongside your active environment Project.toml
[MyPackage]
precompile_workload = false
After restarting julia, the @compile_workload
and @setup_workload
workloads will be disabled (locally) for MyPackage
. You can also specify additional packages (e.g., dependencies of MyPackage
) if you're co-developing a suite of packages. Simply run set_preferences!
for the additional packages, or edit LocalPreferences.toml directly.
Changing precompile_workload
will result in a one-time recompilation of all packages that depend on the package(s) from the current environment. Package developers may wish to set this preference locally within the "main" package's environment; precompilation will be skipped while you're actively developing the project, but not if you use the package from an external environment. This will also keep the precompile_workload
setting independent and avoid needless recompilation of large environments.
Finally, it is possible to fully disable PrecompileTools.jl for all packages with
using PrecompileTools, Preferences
set_preferences!(PrecompileTools, "precompile_workloads" => false; force=true)
This can be helpful to reduce the system image size generated when using PackageCompiler.jl by only compiling calls made in a precompilation script.
Seeing what got precompiled
If you want to see the list of calls that will be precompiled, navigate to the MyPackage
folder and use
julia> using PrecompileTools
julia> PrecompileTools.verbose[] = true # runs the block even if you're not precompiling, and print precompiled calls
julia> include("src/MyPackage.jl");
This will only show the direct- or runtime-dispatched method instances that got precompiled (omitting their inferrable callees). For a more comprehensive list of all items stored in the compile_workload file, see PkgCacheInspector.