Skip to content

StaticCompiler upgraded to Julia v1.11 #174

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 7 commits into
base: master
Choose a base branch
from
Open

StaticCompiler upgraded to Julia v1.11 #174

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
62b4be2
update to julia1.11 and GPUCompiler 1.5.2
kylincaster Jun 12, 2025
c5b16a7
update Project.toml
Jun 12, 2025
7af13f2
pass the test for julia 1.8~1.11 in ubuntu
Jun 12, 2025
1eda46a
update the version to 0.7.3
Jun 13, 2025
1c42a57
update for GPUCompiler by removing the deprecated API
Jun 13, 2025
f718e34
Update CI to test on version 1.11
brenhinkeller Jun 14, 2025
d9c3f7b
Update integration test CI to test on version 1.11
brenhinkeller Jun 14, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions .github/workflows/ci-integration.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,9 +18,9 @@ jobs:
fail-fast: false
matrix:
version:
- '1.8'
- '1.9'
- '1.10.0-rc1'
- '1.10'
- '1.11'
os:
- ubuntu-latest
- macOS-latest
Expand Down
2 changes: 1 addition & 1 deletion .github/workflows/ci.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,9 +18,9 @@ jobs:
fail-fast: false
matrix:
version:
- '1.8'
- '1.9'
- '1.10'
- '1.11'
os:
- ubuntu-latest
- macOS-latest
Expand Down
8 changes: 4 additions & 4 deletions Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
name = "StaticCompiler"
uuid = "81625895-6c0f-48fc-b932-11a18313743c"
authors = ["Tom Short and contributors"]
version = "0.7.2"
version = "0.7.3"


[deps]
Expand All @@ -18,11 +18,11 @@ StaticTools = "86c06d3c-3f03-46de-9781-57580aa96d0a"

[compat]
CodeInfoTools = "0.3"
GPUCompiler = "0.21, 0.22, 0.23, 0.24, 0.25, 0.26"
LLVM = "6"
GPUCompiler = "0.21, 0.22, 0.23, 0.24, 0.25, 0.26, 1.5, 1"
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the 1 shouldn't be here, just 1.5

LLVM = "6, 7, 8, 9"
MacroTools = "0.5"
StaticTools = "0.8"
julia = "1.8, 1.9"
julia = "1.8, 1.9, 1.10, 1.11"

[extras]
Formatting = "59287772-0a20-5a39-b81b-1366585eb4c0"
Expand Down
26 changes: 18 additions & 8 deletions src/StaticCompiler.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,5 @@
#__precompile__(false)

module StaticCompiler
using InteractiveUtils
using GPUCompiler: GPUCompiler
Expand All @@ -19,10 +21,10 @@ export static_code_llvm, static_code_typed, static_llvm_module, static_code_nati
export @device_override, @print_and_throw
export StaticTarget

include("quirks.jl")
include("interpreter.jl")
include("target.jl")
include("pointer_warning.jl")
include("quirks.jl")
include("dllexport.jl")

fix_name(f::Function) = fix_name(string(nameof(f)))
Expand Down Expand Up @@ -450,9 +452,9 @@ function static_llvm_module(f, tt, name=fix_name(f); demangle=true, target::Stat
if !demangle
name = "julia_"*name
end
job, kwargs = static_job(f, tt; name, target, kwargs...)
job, kwargs = static_job(f, tt; name, target, strip=true, only_entry=false, validate=false, libraries=false, kwargs...)
m = GPUCompiler.JuliaContext() do context
m, _ = GPUCompiler.codegen(:llvm, job; strip=true, only_entry=false, validate=false, libraries=false)
m, _ = GPUCompiler.compile(:llvm, job; kwargs...)
locate_pointers_and_runtime_calls(m)
m
end
Expand All @@ -467,17 +469,17 @@ function static_llvm_module(funcs::Union{Array,Tuple}; demangle=true, target::St
if !demangle
name_f = "julia_"*name_f
end
job, kwargs = static_job(f, tt; name = name_f, target, kwargs...)
mod,_ = GPUCompiler.codegen(:llvm, job; strip=true, only_entry=false, validate=false, libraries=false)
job, kwargs = static_job(f, tt; name = name_f, target, strip=true, only_entry=false, validate=false, libraries=false, kwargs...)
mod,_ = GPUCompiler.compile(:llvm, job; kwargs...)
if length(funcs) > 1
for func in funcs[2:end]
f,tt = func
name_f = fix_name(f)
if !demangle
name_f = "julia_"*name_f
end
job, kwargs = static_job(f, tt; name = name_f, target, kwargs...)
tmod,_ = GPUCompiler.codegen(:llvm, job; strip=true, only_entry=false, validate=false, libraries=false)
job, kwargs = static_job(f, tt; name = name_f, target, strip=true, only_entry=false, validate=false, libraries=false, kwargs...)
tmod,_ = GPUCompiler.compile(:llvm, job; kwargs...)
link!(mod,tmod)
end
end
Expand Down Expand Up @@ -587,8 +589,16 @@ function generate_obj(funcs::Union{Array,Tuple}, path::String = tempname(), file
obj_path = joinpath(path, "$filenamebase.o")
obj = GPUCompiler.JuliaContext() do ctx
fakejob, _ = static_job(f, tt; target, kwargs...)
@static if VERSION < v"1.9"
obj, _ = GPUCompiler.emit_asm(fakejob, mod; strip=strip_asm, validate=false, format=LLVM.API.LLVMObjectFile)
obj
else
@static if pkgversion(GPUCompiler) < v"1.3.0"
obj, _ = GPUCompiler.emit_asm(fakejob, mod; strip=strip_asm, validate=false, format=LLVM.API.LLVMObjectFile)
else
obj, _ = GPUCompiler.emit_asm(fakejob, mod, LLVM.API.LLVMObjectFile)
end
end
obj
Comment on lines +592 to +601
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It seems that the new way to handle kwargs like strip and validate is GPUCompierConfig: JuliaGPU/GPUCompiler.jl#668

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Indeed, GPUCompiler has changed significantly between versions 0.x and 1.x.
Supporting both versions requires considerable effort, and dropping support for older versions would make maintenance much smoother.

end
open(obj_path, "w") do io
write(io, obj)
Expand Down
96 changes: 53 additions & 43 deletions src/interpreter.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,61 +1,68 @@
## interpreter

using Core.Compiler:
AbstractInterpreter, InferenceResult, InferenceParams, InferenceState, MethodInstance, OptimizationParams, WorldView, get_world_counter
AbstractInterpreter, InferenceResult, InferenceParams, InferenceState, MethodInstance, OptimizationParams, WorldView
using GPUCompiler:
@safe_debug, AbstractCompilerParams, CodeCache, CompilerJob, methodinstance
@safe_debug, AbstractCompilerParams, CompilerJob, methodinstance, CodeInstance, inference_params, optimization_params, get_inference_world
Comment on lines 5 to +6
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think it'd be better to import some of this stuff from Core.Compiler where it's defined rather than `GPUCompiler.

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Unfortunately, Core.Compiler is an internal API and not guaranteed to be stable or long-term supported.
Since all the compilation-related data originates from GPUCompiler, I believe it's better to import the necessary functionality from GPUCompiler rather than relying directly on Core.Compiler.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Those functions and types are not part of the public interface of GPUCompiler either. By importing them from GPUCompiler, you're relying on GPUCompiler internals and Core.Compiler internals. By importing those names out of GPUCompiler, you make it more vulnerable to instability, not less.

using CodeInfoTools
using CodeInfoTools: resolve


const HAS_INTEGRATED_CACHE = GPUCompiler.HAS_INTEGRATED_CACHE
@static if HAS_INTEGRATED_CACHE
const CodeCache = Nothing

else
using GPUCompiler: CodeCache
end

# https://github.com/JuliaGPU/GPUCompiler.jl/src/jlgen.jl8#L322
# as from struct GPUInterpreter <: CC.AbstractInterpreter
struct StaticInterpreter <: AbstractInterpreter
global_cache::CodeCache
# The world age we're working inside of
world::UInt
method_table::Union{Nothing,Core.MethodTable}

@static if HAS_INTEGRATED_CACHE
token::Any
else
code_cache::CodeCache # global cache
end

# Cache of inference results for this particular interpreter
local_cache::Vector{InferenceResult}
# The world age we're working inside of
world::UInt

# Parameters for inference and optimization
inf_params::InferenceParams
opt_params::OptimizationParams

function StaticInterpreter(cache::CodeCache, mt::Union{Nothing,Core.MethodTable}, world::UInt, ip::InferenceParams, op::OptimizationParams)
# token_or_cache = token::Any, code_cache::CodeCache
function StaticInterpreter(world::UInt, mt::Union{Nothing,Core.MethodTable}, token_or_cache, ip::InferenceParams, op::OptimizationParams)
@assert world <= Base.get_world_counter()

return new(
cache,
mt,

# Initially empty cache
Vector{InferenceResult}(),

# world age counter
world,

# parameters for inference and optimization
ip,
op
)
# mt = get_method_table_view(world, mt)
local_cache = Vector{Core.Compiler.InferenceResult}() # Initially empty cache
return new(world, mt, token_or_cache, local_cache, ip, op)
end
end


Core.Compiler.InferenceParams(interp::StaticInterpreter) = interp.inf_params
Core.Compiler.OptimizationParams(interp::StaticInterpreter) = interp.opt_params
Core.Compiler.get_world_counter(interp::StaticInterpreter) = interp.world
# Core.Compiler.get_world_counter(interp::StaticInterpreter) = interp.world
GPUCompiler.get_inference_world(interp::StaticInterpreter) = interp.world
Core.Compiler.get_inference_cache(interp::StaticInterpreter) = interp.local_cache
Core.Compiler.code_cache(interp::StaticInterpreter) = WorldView(interp.global_cache, interp.world)
@static if HAS_INTEGRATED_CACHE
Core.Compiler.cache_owner(interp::StaticInterpreter) = interp.token
else
Core.Compiler.code_cache(interp::StaticInterpreter) = WorldView(interp.code_cache, interp.world)
end

# No need to do any locking since we're not putting our results into the runtime cache
Core.Compiler.lock_mi_inference(interp::StaticInterpreter, mi::MethodInstance) = nothing
Core.Compiler.unlock_mi_inference(interp::StaticInterpreter, mi::MethodInstance) = nothing

function Core.Compiler.add_remark!(interp::StaticInterpreter, sv::InferenceState, msg)
@safe_debug "Inference remark during static compilation of $(sv.linfo): $msg"
@safe_debug "Inference remark during static compilation of $(sv.linfo): $msg"
end


#####
##### Pre-inference
#####
Expand All @@ -77,16 +84,19 @@ function custom_pass!(interp::StaticInterpreter, result::InferenceResult, mi::Co
end

function Core.Compiler.InferenceState(result::InferenceResult, cache::Symbol, interp::StaticInterpreter)
world = get_world_counter(interp)
src = @static if VERSION >= v"1.10.0-DEV.873"
world = get_inference_world(interp)
src = @static if VERSION >= v"1.10.0-DEV.873"
Core.Compiler.retrieve_code_info(result.linfo, world)
else
Core.Compiler.retrieve_code_info(result.linfo)
Core.Compiler.retrieve_code_info(result.linfo)
end
mi = result.linfo
src = custom_pass!(interp, result, mi, src)
src === nothing && return nothing
Core.Compiler.validate_code_in_debug_mode(result.linfo, src, "lowered")
src === nothing && return @static if VERSION < v"1.11"
Core.Compiler.validate_code_in_debug_mode(result.linfo, src, "lowered")
else
Core.Compiler.maybe_validate_code(result.linfo, src, "lowered")
end
return InferenceState(result, src, cache, interp)
end

Expand All @@ -95,7 +105,6 @@ Core.Compiler.may_compress(interp::StaticInterpreter) = true
Core.Compiler.may_discard_trees(interp::StaticInterpreter) = true
Core.Compiler.verbose_stmt_info(interp::StaticInterpreter) = false


if isdefined(Base.Experimental, Symbol("@overlay"))
using Core.Compiler: OverlayMethodTable
if v"1.8-beta2" <= VERSION < v"1.9-" || VERSION >= v"1.9.0-DEV.120"
Expand All @@ -112,13 +121,13 @@ end

# semi-concrete interepretation is broken with overlays (JuliaLang/julia#47349)
@static if VERSION >= v"1.9.0-DEV.1248"
function Core.Compiler.concrete_eval_eligible(interp::StaticInterpreter,
@nospecialize(f), result::Core.Compiler.MethodCallResult, arginfo::Core.Compiler.ArgInfo)
ret = @invoke Core.Compiler.concrete_eval_eligible(interp::AbstractInterpreter,
f::Any, result::Core.Compiler.MethodCallResult, arginfo::Core.Compiler.ArgInfo)
ret === false && return nothing
return ret
end
function Core.Compiler.concrete_eval_eligible(interp::StaticInterpreter,
@nospecialize(f), result::Core.Compiler.MethodCallResult, arginfo::Core.Compiler.ArgInfo)
ret = @invoke Core.Compiler.concrete_eval_eligible(interp::AbstractInterpreter,
f::Any, result::Core.Compiler.MethodCallResult, arginfo::Core.Compiler.ArgInfo)
ret === false && return nothing
return ret
end
end

struct StaticCompilerParams <: AbstractCompilerParams
Expand All @@ -127,8 +136,9 @@ struct StaticCompilerParams <: AbstractCompilerParams
cache::CodeCache
end

function StaticCompilerParams(; opt = false,
optlevel = Base.JLOptions().opt_level,
cache = CodeCache())
function StaticCompilerParams(; opt=false,
optlevel=Base.JLOptions().opt_level,
cache=CodeCache()
)
return StaticCompilerParams(opt, optlevel, cache)
end
3 changes: 2 additions & 1 deletion src/pointer_warning.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,8 +30,9 @@ function locate_pointers_and_runtime_calls(mod)
end
end
if warned
lines = split(string(func),"\n")
@warn("LLVM function generated warnings due to raw pointers embedded in the code. This will likely cause errors or undefined behaviour.",
func = func)
func = join(lines[1:min(20, end)], "\n")) # just print the first 20 lines
end
end
end
Expand Down
44 changes: 39 additions & 5 deletions src/quirks.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,41 @@
libcexit(x::Int32) = @symbolcall exit(x::Int32)::Nothing
@static if isdefined(Base.Experimental, Symbol("@overlay"))
Base.Experimental.@MethodTable(method_table)
Base.Experimental.@MethodTable(empty_table)
else
const method_table = nothing
end

"""
```julia
@device_override old_bad_method(arg1::Type1, arg2::Type2) = new_good_method(arg1, arg2)
```
Override a non-static-compilable method (e.g. `old_bad_method(::Type1, ::Type2)`)
with a more compileable replacement.
### Examples
```
@device_override @noinline Core.throw_inexacterror(f::Symbol, ::Type{T}, val) where {T} =
@print_and_throw c"Inexact conversion"
```
"""
macro device_override(ex)
ex = macroexpand(__module__, ex)
if Meta.isexpr(ex, :call)
@show ex = eval(ex)
error()
end
code = quote
$Base.Experimental.@overlay($StaticCompiler.method_table, $ex)
end
return esc(code)
end

macro print_and_throw(err)
quote
println($err)
printf($err)
libcexit(Int32(1))
end
end
libcexit(x::Int32) = @symbolcall exit(x::Int32)::Nothing

# math.jl
@device_override @noinline Base.Math.throw_complex_domainerror(f::Symbol, x) =
Expand Down Expand Up @@ -37,9 +68,12 @@ end
@device_override @noinline Core.throw_inexacterror(f::Symbol, ::Type{T}, val) where {T} =
@print_and_throw c"Inexact conversion"

# abstractarray.jl
@device_override @noinline Base.throw_boundserror(A, I) =
@print_and_throw c"Out-of-bounds array access"
# abstractarray.jl
# Base.throw_boundserror is removed since v1.11
if VERSION < v"1.11"
@device_override @noinline Base.throw_boundserror(A, I) =
@print_and_throw c"Out-of-bounds array access"
end

Comment on lines -40 to +76
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Base.throw_boundserror definitely was not removed in v1.11.

Copy link
Author

@kylincaster kylincaster Jun 15, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In my tests, Bumper v0.6.0 fails to work with Julia v1.11 due to its attempt to override Base.throw_boundserror.

However, when switching to Bumper v0.7.0, everything works smoothly, as this version uses a different backend (UnsafeArrays).

Although the throw_boundserror function still exists in Julia v1.11, its definition has been moved from abstractarray.jl to essential.jl, which might indicate that it has been made non-overridable.

Copy link
Collaborator

@MasonProtter MasonProtter Jun 15, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Weird. Let's just drop Bumper 0.6 support then and put a compat bound with Bumper v0.7 as the minimum.

Please restore the device overrride for the boundserror in v1.11+, we need that.

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Unfortunately, Bumper v0.6 is required by StaticTools.jl. However, I believe that updating Bumper in StaticTools will not affect its functionality. Currently, I’m helping the developers with some new functionality for StaticTools.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The version of StaticTools that we registered yesterday (0.8.11) should support Bumper v0.7 since the PR right before yours bumped the compat bounds on Bumper

# trig.jl
@device_override @noinline Base.Math.sincos_domain_error(x) =
Expand Down
Loading
Loading