Support for julia 1.12 #181

VarLad · 2025-11-28T06:40:23Z

Based on #174
~~Currently filing this as a draft because Bumper.jl integration seems to be broken.~~

VarLad · 2025-11-28T06:57:58Z

Okay, never mind, turns out I needed to add @inbounds to the Bumper.jl example...

using Bumper, StaticTools, StaticCompiler

function times_table(argc::Int, argv::Ptr{Ptr{UInt8}})
           argc == 3 || return printf(c"Incorrect number of command-line arguments\n")
           rows = argparse(Int64, argv, 2)            # First command-line argument
           cols = argparse(Int64, argv, 3)            # Second command-line argument

           buf = MallocSlabBuffer()
           @no_escape buf begin
               M = @alloc(Int, rows, cols)
               @inbounds for i=1:rows
                   @inbounds for j=1:cols
                       M[i,j] = i*j
                   end
               end
               printf(M)
           end
           free(buf)
end
       
filepath = compile_executable(times_table, (Int64, Ptr{Ptr{UInt8}}), "./")

VarLad · 2025-11-28T07:16:22Z

This likely has similar issues to:
#174 (comment)

MasonProtter

Thanks for trying to revive this @VarLad. I mostly have the same comment as I did for the previous one, so I'll just copy-paste it:

I think it probably doesn't make sense to try and have so many if version(...) checks in here, and we should just increase our version support bounds.

If people aren't able to upgrade to newer versions of julia or GPUCompiler, they will get stuck with older versions of StaticCompiler which is fine.

Also, the pkgversion function was added in v1.9, so using it will break the v1.8 tests.

Since v1.10 is the current julia LTS, I say we drop support for v1.8 and 1.9, and drop support for GPUCompiler < 1.5

That should make this simpler and hopefully more robust too.

Except I'd recommend now that we also drop v1.10 support and move onto v1.11/v1.12 support since there's already a released version that works on v1.10 anyways, and this isn't adding new features that people on 1.10 would want (if anything, it seems that some features have regressed, since it seems you changed some of the tests to remove the possibility of them emitting boundserrors)

src/StaticCompiler.jl

src/interpreter.jl

src/quirks.jl

test/runtests.jl

MasonProtter · 2025-11-28T11:46:23Z

test/scripts/loopvec_matrix.jl

 @inline function mul!(C::MallocArray, A::MallocArray, B::MallocArray)
-    @turbo for n ∈ indices((C,B), 2), m ∈ indices((C,A), 1)
+    #@turbo for n ∈ indices((C,B), 2), m ∈ indices((C,A), 1)
+    @turbo for n ∈ indices(C, 2), m ∈ indices(C, 1)
        Cmn = zero(eltype(C))
-        for k ∈ indices((A,B), (2,1))
+        # for k ∈ indices((A,B), (2,1))
+        for k ∈ indices(A, 2)
            Cmn += A[m,k] * B[k,n]


Why is this removed? Was it to avoid boundserrors causing it to fail to compile? If so, we should see if there's a device_override we could do to help the compiler not get tripped up here.

I've reverted this, but both now and on the previous PR, loop_vector related stuff fail to compile

test/scripts/loopvec_matrix.jl

test/scripts/loopvec_matrix_stack.jl

MasonProtter · 2025-11-28T11:48:28Z

test/scripts/loopvec_matrix_stack.jl

 @inline function mul!(C::StackArray, A::StackArray, B::StackArray)
-    @turbo for n ∈ indices((C,B), 2), m ∈ indices((C,A), 1)
+    # error since Julia v1.11
+    #@turbo for n ∈ indices((C,B), 2), m ∈ indices((C,A), 1) 
+    @turbo for n ∈ indices(C, 2), m ∈ indices(C, 1)
        Cmn = zero(eltype(C))
-        for k ∈ indices((A,B), (2,1))
+        # for k ∈ indices((A,B), (2,1))
+        for k ∈ indices(A, 2)
            Cmn += A[m,k] * B[k,n]


same as above

MasonProtter · 2025-11-28T11:48:57Z

test/testcore.jl

    #Compile dylib
-    name = repr(fib)
+    name = string(nameof(fib)) # repr(fib)


did repr stop working?

This is what I see if I use repr:

Testing Running tests... Standalone Dylibs: Error During Test at ~/.julia/dev/StaticCompiler/test/testcore.jl:5 Got exception outside of a @test UndefVarError: `name` not defined in `Main` Suggestion: check for spelling errors or missing imports. Hint: a global variable of this name also exists in LLVM.

src/quirks.jl

MasonProtter · 2025-11-28T12:01:36Z

In the future, if you're reviving old PRs, please make sure you address the review feedback that's already been given on the older version of the PR before re-requesting that someone review it. Because of the amount of time that had past, I didn't immediately recognize that I had already made the exact same review comments back in June, so I spent some time basically just rediscovering the same complaints I already had back then.

VarLad · 2025-11-28T14:31:51Z

@MasonProtter apologies for wasting time, I forgot about the reviews in last PR. I'll be careful about it in future PRs. 🥲

I've addressed most of the comments. I've removed explicit support for anything pre-1.11 from the codebase.
I've reverted the loopvec based tests (they were failing before, and are still failing)

printdlm function call fails test as well with:

/usr/bin/ld: ./times_table.o: in function `times_table':
start:(.text+0x330): undefined reference to `ijl_error'
clang: error: linker command failed with exit code 1 (use -v to see invocation)

VarLad · 2025-11-28T14:34:28Z

There are a lot of warnings about:

WARNING: llvmcall with integer pointers is deprecated.
Use actual pointers instead, replacing i32 or i64 with i8* or ptr

throughout the codebase 🤔

brenhinkeller · 2025-11-28T16:28:36Z

Thanks for taking this on!

brenhinkeller · 2025-11-28T16:30:18Z

There are a lot of warnings about:
WARNING: llvmcall with integer pointers is deprecated.
Use actual pointers instead, replacing i32 or i64 with i8* or ptr
throughout the codebase 🤔

Yeah, this is probably due to the switch to opaque pointers in LLVM, and will require a bunch of changes in StaticTools, but I think they should be relatively straightforward/repetitive

MasonProtter · 2025-11-29T10:21:59Z

@MasonProtter apologies for wasting time, I forgot about the reviews in last PR. I'll be careful about it in future PRs. 🥲

No problem, and sorry if I came off as grumpy, I was just explaining some best practices for submitting PRs like this.

I've reverted the loopvec based tests (they were failing before, and are still failing)

Looking into this a bit more, it seems that there's been a change to how the errors from indices are handled so that they don't throw a bounderror, that's unfortunate. Here's what happens:

julia> indices(([1, 2], [3, 4, 5]), (1, 1))
ERROR: 2 and 3 are not equal
Stacktrace:
 [1] error(s::String)
   @ Base ./error.jl:44
 [2] _static_promote
   @ ~/.julia/packages/Static/d7YOk/src/Static.jl:332 [inlined]
 [3] static_promote
   @ ~/.julia/packages/Static/d7YOk/src/Static.jl:330 [inlined]
 [4] static_promote
   @ ~/.julia/packages/Static/d7YOk/src/Static.jl:363 [inlined]
 [5] static_promote
   @ ~/.julia/packages/Static/d7YOk/src/Static.jl:371 [inlined]
 [6] macro expansion
   @ ~/.julia/packages/Static/d7YOk/src/Static.jl:751 [inlined]
 [7] reduce_tup
   @ ~/.julia/packages/Static/d7YOk/src/Static.jl:751 [inlined]
 [8] indices(x::Tuple{Vector{Int64}, Vector{Int64}}, dim::Tuple{Int64, Int64})
   @ StaticArrayInterface ~/.julia/packages/StaticArrayInterface/lkDPR/src/ranges.jl:60
 [9] top-level scope
   @ REPL[3]:1

and the relevant method of _static_promote is here: https://github.com/SciML/Static.jl/blob/master/src/Static.jl#L331C1-L334C4

I would still like some boundschecking in the tests (just to make sure we still eliminate boundserrors) so could you maybe just put

@boundscheck checkbounds(...)

calls into one of the loop vectorization tests? e.g. something like

Base.@propagate_inbounds function mul!(C::StackArray, A::StackArray, B::StackArray)
    @boundscheck begin
        checkbounds(axes(C, 2), axes(B, 2))
        checkbounds(axes(B, 2), axes(C, 2))
        checkbounds(axes(C, 1), axes(A, 1))
        checkbounds(axes(A, 1), axes(C, 1))
        checkbounds(axes(A, 2), axes(B, 1))
        checkbounds(axes(B, 1), axes(A, 2))
    end
    @turbo for n ∈ axes(C, 2), m ∈ axes(C, 1)
        Cmn = zero(eltype(C))
        for k ∈ axes(A, 2)
            Cmn += A[m,k] * B[k,n]
        end
        C[m,n] = Cmn
    end
    return C
end

VarLad · 2025-11-30T16:59:39Z

@MasonProtter I think even simple loopvec is broke now on StaticCompiler.jl.
I am getting a lot of undefined reference error in this basic example:

using StaticCompiler, StaticTools, LoopVectorization

Base.@propagate_inbounds function loopvec_product(argc::Int, argv::Ptr{Ptr{UInt8}})
    argc == 3 || return printf(stderrp(), c"Incorrect number of command-line arguments\n")
    rows = argparse(Int64, argv, 2)            # First command-line argument
    cols = argparse(Int64, argv, 3)            # Second command-line argument

    s = 0
    @turbo for i=1:rows
        for j=1:cols
           s += i*j
        end
    end
    printf(s)
end

# Attempt to compile
path = compile_executable(loopvec_product, (Int64, Ptr{Ptr{UInt8}}), "./")

(this compiles fine if I remove @turbo)

MasonProtter · 2025-11-30T18:10:05Z

Unfortunate. I guess lets just mark those all inside @test_broken blocks for now then

VarLad · 2025-12-01T03:54:11Z

There's the issue of printdlm erroring in tests, which I have commented out.
Likely related to brenhinkeller/StaticTools.jl#72 and #181 (comment)
I'll take a look at the switch to opaque pointers for StaticTools (and re-enabling printdlm tests here) when I get more time.

MasonProtter

Almost there! Thanks for sticking with it, just a few nitpicks then we can merge it.

src/StaticCompiler.jl

src/interpreter.jl

src/quirks.jl

src/target.jl

test/testintegration.jl

src/interpreter.jl

VarLad · 2025-12-01T14:15:58Z

This is quite weird, StaticCompiler both builds and all tests (pass + broken) work for me locally 🤔

MasonProtter · 2025-12-01T14:31:43Z

Are you running the integration tests locally?

VarLad · 2025-12-01T14:42:23Z

Yeah, I just ran it again, I got:

Test Summary:                     | Pass  Broken  Total     Time
Standalone Executable Integration |   18      15     33  4m11.0s
     Testing StaticCompiler tests passed

I can try commenting out printdlm again, thats the only thing that should've changed here.

VarLad · 2025-12-01T14:52:50Z

I have commented out printdlm again (and have put a comment about when to re-enable it)
Lets see if that appeases the CI gods....

brenhinkeller · 2025-12-01T16:03:59Z

Closes #166

brenhinkeller · 2025-12-01T16:17:58Z

I'd say some amount of breakage in integration tests is acceptable given those may need to be fixed on the StaticTools side

MasonProtter · 2025-12-01T17:21:58Z

Nice!

kylincaster and others added 6 commits June 12, 2025 16:16

update to julia1.11 and GPUCompiler 1.5.2

62b4be2

update Project.toml

c5b16a7

pass the test for julia 1.8~1.11 in ubuntu

7af13f2

update the version to 0.7.3

1eda46a

update for GPUCompiler by removing the deprecated API

1c42a57

initial 1.12 support

66fe017

VarLad marked this pull request as ready for review November 28, 2025 07:08

MasonProtter requested changes Nov 28, 2025

View reviewed changes

MasonProtter reviewed Nov 28, 2025

View reviewed changes

src/quirks.jl Outdated Show resolved Hide resolved

VarLad added 2 commits November 28, 2025 23:16

address reviews

4d22563

re-enable Bumper tests, comment printdlm

bdadd58

revert printf

a549bad

use 1.11 and 1.12 in ci

b02c8fc

mark loopvec tests as broken

f9ffc04

VarLad requested a review from MasonProtter December 1, 2025 11:16

MasonProtter reviewed Dec 1, 2025

View reviewed changes

address comments

c5be3c5

VarLad requested a review from MasonProtter December 1, 2025 14:23

comment printdlm again

81d3d0e

MasonProtter merged commit 7e36691 into tshort:master Dec 1, 2025
11 of 16 checks passed

Support for julia 1.12 #181

Support for julia 1.12 #181

Conversation

VarLad commented Nov 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

VarLad commented Nov 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

VarLad commented Nov 28, 2025

Uh oh!

MasonProtter left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

MasonProtter Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

VarLad Nov 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

MasonProtter Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

MasonProtter Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

VarLad Nov 28, 2025

Choose a reason for hiding this comment

Uh oh!

MasonProtter Nov 29, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

MasonProtter commented Nov 28, 2025

Uh oh!

VarLad commented Nov 28, 2025

Uh oh!

VarLad commented Nov 28, 2025

Uh oh!

brenhinkeller commented Nov 28, 2025

Uh oh!

brenhinkeller commented Nov 28, 2025

Uh oh!

MasonProtter commented Nov 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

VarLad commented Nov 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

MasonProtter commented Nov 30, 2025

Uh oh!

VarLad commented Dec 1, 2025

Uh oh!

MasonProtter left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

VarLad commented Dec 1, 2025

Uh oh!

MasonProtter commented Dec 1, 2025

Uh oh!

VarLad commented Dec 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

VarLad commented Nov 28, 2025 •

edited

Loading

VarLad commented Nov 28, 2025 •

edited

Loading

MasonProtter left a comment •

edited

Loading

VarLad Nov 28, 2025 •

edited

Loading

MasonProtter commented Nov 29, 2025 •

edited

Loading

VarLad commented Nov 30, 2025 •

edited

Loading

VarLad commented Dec 1, 2025 •

edited

Loading

VarLad commented Dec 1, 2025 •

edited

Loading