Version 0.7.0-rc1.0, permutedims crash with Illegal inttoptr

[PaulSoderlind]

This crashes Version 0.7.0-rc1.0, Win64

x = rand(4,3,2)
y = permutedims(x,[3,1,2])

[fredrikekre]

Same on Linux; full stacktrace:

julia> y = permutedims(x,[3,1,2])
Illegal inttoptr
	  %49 = ptrtoint double addrspace(13)* %48 to i64
Illegal inttoptr
	  %104 = inttoptr i64 %103 to i8 addrspace(13)*

signal (6): Aborted
in expression starting at no file:0
raise at /build/glibc-Cl5G7W/glibc-2.23/signal/../sysdeps/unix/sysv/linux/raise.c:54
abort at /build/glibc-Cl5G7W/glibc-2.23/stdlib/abort.c:89
runOnFunction at /home/fredrik/julia-master/src/llvm-gc-invariant-verifier.cpp:178
_ZN4llvm13FPPassManager13runOnFunctionERNS_8FunctionE at /home/fredrik/julia-master/usr/bin/../lib/libLLVM-6.0.so (unknown line)
_ZN4llvm13FPPassManager11runOnModuleERNS_6ModuleE at /home/fredrik/julia-master/usr/bin/../lib/libLLVM-6.0.so (unknown line)
_ZN4llvm6legacy15PassManagerImpl3runERNS_6ModuleE at /home/fredrik/julia-master/usr/bin/../lib/libLLVM-6.0.so (unknown line)
operator() at /home/fredrik/julia-master/src/jitlayers.cpp:481
addModule at /home/fredrik/julia-master/usr/include/llvm/ExecutionEngine/Orc/IRCompileLayer.h:57 [inlined]
addModule at /home/fredrik/julia-master/src/jitlayers.cpp:608
jl_add_to_ee at /home/fredrik/julia-master/src/jitlayers.cpp:855 [inlined]
jl_finalize_function at /home/fredrik/julia-master/src/jitlayers.cpp:863
getAddressForFunction at /home/fredrik/julia-master/src/codegen.cpp:1300
jl_generate_fptr at /home/fredrik/julia-master/src/codegen.cpp:1411
jl_compile_method_internal at /home/fredrik/julia-master/src/gf.c:1804 [inlined]
jl_fptr_trampoline at /home/fredrik/julia-master/src/gf.c:1811
jl_apply_generic at /home/fredrik/julia-master/src/gf.c:2165
do_call at /home/fredrik/julia-master/src/interpreter.c:324
eval_value at /home/fredrik/julia-master/src/interpreter.c:428
eval_stmt_value at /home/fredrik/julia-master/src/interpreter.c:363 [inlined]
eval_body at /home/fredrik/julia-master/src/interpreter.c:682
jl_interpret_toplevel_thunk_callback at /home/fredrik/julia-master/src/interpreter.c:799
unknown function (ip: 0xfffffffffffffffe)
unknown function (ip: 0x7f31983d850f)
unknown function (ip: (nil))
jl_interpret_toplevel_thunk at /home/fredrik/julia-master/src/interpreter.c:808
jl_toplevel_eval_flex at /home/fredrik/julia-master/src/toplevel.c:831
jl_toplevel_eval_in at /home/fredrik/julia-master/src/builtins.c:633
eval at ./boot.jl:319
jl_apply_generic at /home/fredrik/julia-master/src/gf.c:2165
eval_user_input at /home/fredrik/julia-master/usr/share/julia/stdlib/v0.7/REPL/src/REPL.jl:85
macro expansion at /home/fredrik/julia-master/usr/share/julia/stdlib/v0.7/REPL/src/REPL.jl:117 [inlined]
#28 at ./task.jl:262
jl_fptr_trampoline at /home/fredrik/julia-master/src/gf.c:1812
jl_apply_generic at /home/fredrik/julia-master/src/gf.c:2165
jl_apply at /home/fredrik/julia-master/src/julia.h:1538 [inlined]
start_task at /home/fredrik/julia-master/src/task.c:268
unknown function (ip: 0xffffffffffffffff)
Allocations: 5796103 (Pool: 5794648; Big: 1455); GC: 12
Aborted (core dumped)

Looks related to Keno's recent LLVM fixes?

Edit:

julia> versioninfo()
Julia Version 0.7.0-rc1.4
Commit 0ef8826 (2018-08-01 00:52 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: Intel(R) Core(TM) i5-7600K CPU @ 3.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.0 (ORCJIT, skylake)

[fredrikekre]

Probably dup of #28379?

[dlfivefifty]

If it is a dup, better to close #28379 as this is a much better MWE.

[gdkrmr]

I cannot reproduce this on 0.7.0-rc1.5, Ubuntu 16.04

[dlfivefifty]

I get it on rc1.4:

               _
   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: https://docs.julialang.org
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.7.0-rc1.4 (2018-08-01 00:52 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit 0ef882679f (0 days old master)
|__/                   |  x86_64-apple-darwin17.7.0

julia> x = rand(4,3,2)
4×3×2 Array{Float64,3}:
[:, :, 1] =
 0.547093  0.681695  0.259469 
 0.450266  0.939295  0.590317 
 0.925171  0.85919   0.548892 
 0.182937  0.288661  0.0223785

[:, :, 2] =
 0.151585    0.0777777  0.167315
 0.333083    0.595943   0.748375
 0.00221635  0.153361   0.20937 
 0.613113    0.0917188  0.383782

julia> y = permutedims(x,[3,1,2])
Illegal inttoptr
	  %49 = ptrtoint double addrspace(13)* %48 to i64
Illegal inttoptr
	  %104 = inttoptr i64 %103 to i8 addrspace(13)*

signal (6): Abort trap: 6
in expression starting at no file:0
__pthread_kill at /usr/lib/system/libsystem_kernel.dylib (unknown line)
Allocations: 5956874 (Pool: 5955405; Big: 1469); GC: 12
Abort trap: 6

[timholy]

Just FYI, I don't on what is basically the same (it has one more commit, #28386):

julia> versioninfo()
Julia Version 0.7.0-rc1.4
Commit 0ef882679f* (2018-08-01 00:52 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: Intel(R) Core(TM) i7-5500U CPU @ 2.40GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.0 (ORCJIT, broadwell)
Environment:
  JULIAFUNCDIR = /home/tim/juliafunc
  JULIA_CPU_THREADS = 2

[fredrikekre]

For those of you who do not get this failure, have you rebuilt llvm after Keno's patches? (i.e. make -C deps distclean-llvm)

[gdkrmr]

llvm was rebuilt today but I did not have to make -C deps distclean-llvm

[mgr327]

I do not get this failure on two different machines, both 0.7.0-rc1.5 on Fedora. I did make -C deps distclean-llvm one one machine (yesterday, for rc1.4 I believe) but not today on the other machine. Today's build 'recognized' by itself that there are patches to llvm that need to be applied...

versioninfo()
Julia Version 0.7.0-rc1.5
Commit 25a8c95 (2018-08-01 11:41 UTC)
Platform Info:
OS: Linux (x86_64-redhat-linux)
CPU: Intel(R) Xeon(R) CPU X5650 @ 2.67GHz
WORD_SIZE: 64
LIBM: libopenlibm
LLVM: libLLVM-6.0.0 (ORCJIT, westmere)

[vchuravy]

That sounds like #28377 fixed this occurrence. Recent LLVM patches have not caused a recompilation of LLVM due to a sticky patch at the end, now they should retrigger a compilation when they get applied.