adjustments to the latest inlining interface changes

[aviatesk]

Still needs to adjust to JuliaLang/julia#52415.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 62.50000% with 9 lines in your changes are missing coverage. Please review.

Project coverage is 74.95%. Comparing base (7a51e00) to head (a1fac3e).

Files	Patch %	Lines
src/compiler/interpreter.jl	62.50%	9 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1350      +/-   ##
==========================================
- Coverage   75.38%   74.95%   -0.43%     
==========================================
  Files          35       35              
  Lines       10683    10660      -23     
==========================================
- Hits         8053     7990      -63     
- Misses       2630     2670      +40     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[wsmoses]

@vchuravy is this now subsumed by recent 1.11 stuff?

[vchuravy]

No I haven't touched any of that

[wsmoses]

@vchuravy can you take a look at this. I think this is the next pending thing for 1.11 support (with the incoming jll bump:
JuliaPackaging/Yggdrasil#9354

[aviatesk]

I tried updating it, but it seems we might need a fix in Base. Specifically, because of JuliaLang/julia/#52064, the inlining_policy in Base no longer requires mi, so mi isn't being passed to inlining_policy anymore. However, Enzyme's inlining rule still needs it. Since we can't accurately recover mi from info or src, a change to Base might be necessary.

[aviatesk]

When does this error usually occur?

Error During Test at /Users/aviatesk/julia/packages/Enzyme/test/abi.jl:4
  Got exception outside of a @test
  
  No augmented forward pass found for julia.gc_loaded
   at context:   %37 = call {} addrspace(10)* addrspace(13)* @julia.gc_loaded({} addrspace(10)* noundef %17, {} addrspace(10)** noundef %36) #12, !dbg !39
  
  Stacktrace:
   [1] getindex
     @ ./essentials.jl:916
   [2] squareRetArray
     @ ~/julia/packages/Enzyme/test/abi.jl:86
  
  
  Stacktrace:
    [1] getindex
      @ ./essentials.jl:916 [inlined]
    [2] squareRetArray
      @ ~/julia/packages/Enzyme/test/abi.jl:86 [inlined]
    [3] diffejulia_squareRetArray_29137wrap
      @ ~/julia/packages/Enzyme/test/abi.jl:0
    [4] macro expansion
      @ ~/julia/packages/Enzyme/src/compiler.jl:7179 [inlined]
    [5] enzyme_call
      @ ~/julia/packages/Enzyme/src/compiler.jl:6788 [inlined]
    [6] CombinedAdjointThunk
      @ ~/julia/packages/Enzyme/src/compiler.jl:6665 [inlined]
    [7] autodiff
      @ ~/julia/packages/Enzyme/src/Enzyme.jl:320 [inlined]
    [8] autodiff(mode::ReverseMode{false, FFIABI, false, false}, f::var"#squareRetArray#28", ::Type{Const}, args::Duplicated{Vector{Float64}})
      @ Enzyme ~/julia/packages/Enzyme/src/Enzyme.jl:332
    [9] macro expansion
      @ ~/julia/packages/Enzyme/test/abi.jl:91 [inlined]
   [10] macro expansion
      @ ~/julia/julia5/usr/share/julia/stdlib/v1.11/Test/src/Test.jl:1700 [inlined]
   [11] top-level scope
      @ ~/julia/packages/Enzyme/test/abi.jl:6
   [12] include(fname::String)
      @ Main ./sysimg.jl:38
   [13] top-level scope
      @ ~/julia/packages/Enzyme/test/runtests.jl:85
   [14] include(fname::String)
      @ Main ./sysimg.jl:38
   [15] top-level scope
      @ none:6
   [16] eval
      @ ./boot.jl:430 [inlined]
   [17] exec_options(opts::Base.JLOptions)
      @ Base ./client.jl:291
   [18] _start()
      @ Base ./client.jl:526

[wsmoses]

That is fixed by the jll bump -- which presently seems trapped in yggdrassil hell: JuliaPackaging/Yggdrasil#9354

cc @vchuravy @staticfloat @giordano

[wsmoses]

@aviatesk okay a jll should now be available

[aviatesk]

I have switched to an approach using the new CallInfo. This now should work without any changes on the Base side. However, I have noticed the following error occurring in the assertion build (though I am not sure if it is caused by this PR):

Test Summary:    | Pass  Total  Time
Persistent tasks |    1      1  9.0s
Assertion failed: ((IsIntAttr || Attribute::isEnumAttrKind(Kind)) && "Not an enum or int attribute"), function get, file /workspace/srcdir/llvm-project/llvm/lib/IR/Attributes.cpp, line 95.

[18495] signal 6: Abort trap: 6
in expression starting at /Users/aviatesk/julia/packages/Enzyme/test/abi.jl:4
__pthread_kill at /usr/lib/system/libsystem_kernel.dylib (unknown line)
Allocations: 151710873 (Pool: 151707636; Big: 3237); GC: 101

[wsmoses]

That feels unrelated (and likely on me to fix other misc 1.11 stuff)

[wsmoses]

This generally looks good to me, but seems to break 1.10 (which should pass atm — the lower versions have a Yggdrasil issue rn which shouldn’t block this landing)

[wsmoses]

@aviatesk I rebased this [now with all the 1.11 JLL issues remedied, and I must have done something improper as 1.10 fails.

Is it possible to get some help (and sorry about that)

[aviatesk]

Ok, I think the failure has been fixed.