Testing BSplineKit asserts in LLVM with Attribute after last parameter!

[KristofferC]

PkgEval:

(jl_XXRoRG) pkg> test BSplineKit
     Testing BSplineKit
      Status `/tmp/jl_UPK8kk/Project.toml`
...
Precompiling project...
  8 dependencies successfully precompiled in 3 seconds. 28 already precompiled.
     Testing Running tests...
Test Summary:   | Pass  Total  Time
DifferentialOps |   17     17  0.2s
Test Summary: | Pass  Total  Time
Knots         |   86     86  0.0s
Attribute after last parameter!
  %645 = call token (...) @llvm.julia.gc_preserve_begin(), !dbg !492
Attribute after last parameter!
  %646 = call token (...) @llvm.julia.gc_preserve_begin(), !dbg !494
Attribute after last parameter!
  %647 = call token (...) @llvm.julia.gc_preserve_begin(), !dbg !495
julia: /home/kc/julia/src/llvm-alloc-opt.cpp:1187: bool {anonymous}::AllocOpt::runOnFunction(llvm::Function&, llvm::function_ref<llvm::DominatorTree&()>): Assertion `!verifyFunction(F, &errs())' failed.

[151419] signal (6): Aborted
in expression starting at /home/kc/.julia/packages/BSplineKit/ltMBY/test/bsplines.jl:136
pthread_kill at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
raise at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
abort at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
unknown function (ip: 0x7fc5fe42871a)
__assert_fail at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
runOnFunction at /home/kc/julia/src/llvm-alloc-opt.cpp:1187 [inlined]
runOnFunction at /home/kc/julia/src/llvm-alloc-opt.cpp:1179
runOnFunction at /home/kc/julia/src/llvm-alloc-opt.cpp:1201
_ZN4llvm13FPPassManager13runOnFunctionERNS_8FunctionE at /home/kc/julia/usr/bin/../lib/libLLVM-14jl.so (unknown line)
_ZN4llvm13FPPassManager11runOnModuleERNS_6ModuleE at /home/kc/julia/usr/bin/../lib/libLLVM-14jl.so (unknown line)
_ZN4llvm6legacy15PassManagerImpl3runERNS_6ModuleE at /home/kc/julia/usr/bin/../lib/libLLVM-14jl.so (unknown line)
operator() at /home/kc/julia/src/jitlayers.cpp:1109 [inlined]
withModuleDo<(anonymous namespace)::OptimizerT::operator()(llvm::orc::ThreadSafeModule, llvm::orc::MaterializationResponsibility&)::<lambda(llvm::Module&)> > at /home/kc/julia/usr/include/llvm/ExecutionEngine/Orc/ThreadSafeModule.h:136 [inlined]
operator() at /home/kc/julia/src/jitlayers.cpp:1081 [inlined]
CallImpl<(anonymous namespace)::OptimizerT> at /home/kc/julia/usr/include/llvm/ADT/FunctionExtras.h:222
`` 

[apaz-cli]

@KristofferC I cannot seem to reproduce this on my machine, all the tests pass and I can't get the assertion to fail. Do you have any guidance on this?

[gbaraldi]

Are you testing with both FORCE_ASSERTIONS=1 and LLVM_ASSERTIONS=1?

[apaz-cli]

@gbaraldi I am, yes.

[gbaraldi]

It might be machine difference, or maybe something we did fixed it :)

[apaz-cli]

I should mention that I tried running the tests on antarctic also, and they all passed there as well. And yeah, maybe a re-run would be useful.

[gbaraldi]

If you have the time, get like a 15 day old master and see if it fails.

[vtjnash]

It failed for me, so maybe this is helpful:

julia> versioninfo()
Julia Version 1.9.0-DEV.1706
Commit 3661a08750 (2022-11-01 18:31 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin21.6.0)
  CPU: 8 × Intel(R) Core(TM) i7-8559U CPU @ 2.70GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, skylake)
  Threads: 1 on 8 virtual cores

[gbaraldi]

I also can reproduce on

julia> versioninfo()
Julia Version 1.9.0-DEV.1742
Commit d0ba259f63 (2022-11-03 16:22 UTC)
Platform Info:
  OS: macOS (arm64-apple-darwin22.1.0)
  CPU: 8 × Apple M1
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, apple-m1)
  Threads: 4 on 4 virtual cores
Environment:
  JULIA_NUM_PRECOMPILE_TASKS = 4
  JULIA_NUM_THREADS = 4

[gbaraldi]

@apaz-cli Did you do make cleanall? I think the LLVM toolchain doesn't get redownloaded if you don't do make cleanall.

[apaz-cli]

I did make cleanall, then make debug, didn't work. Re-cloned the repo, copied my Make.user, did make debug, still didn't work. I also tried a ~15 day old julia (faa49a0ef62190e38f67f5b83cf00ad88fb997a5 because I already had it checked out), and couldn't get it to work there either.

[gbaraldi]

What's your machine? It's quite interesting that you can't reproduce it, because it's there in two different OSs and architectures

[apaz-cli]

I'm using Debian sid on an HP Elitebook 745. I've got an AMD Ryzen 5 PRO 2500U as my processor. But I also can't seem to reproduce it on antarctic either.

[gbaraldi]

I can only reproduce this from inside the test harness .

[apaz-cli]

I just reproduced it locally. I rebuilt from the same 15 day old master again, and I guess I must have missed something.

Now though, I'm running into the same problem as with #47247, minus the sycall_traced() issue.

I figured out how to demangle the function and make the call inside the assert

(rr) define srch
>c
>p !(bool)_ZN4llvm14verifyFunctionERKNS_8FunctionEPNS_11raw_ostreamE(F, null)
>end

And then I proceeded to srch through the entire trace. It never prints false.

[apaz-cli]

It turns out that I was on the wrong process the whole time. Makes sense.