[release-1.6] Illegal instruction (simple test case)

[timholy]

I tried uploading an rr bug report but that seems broken (some JSON error), so...

julia> colmajorstrides(sz) = isempty(sz) ? () : (1,cumprod(Int[sz[1:end-1]...])...)::typeof(sz)
colmajorstrides (generic function with 1 method)

julia> colmajorstrides((3,3,3))
Unreachable reached at 0x7f1214e969d8

signal (4): Illegal instruction
in expression starting at REPL[2]:1
colmajorstrides at ./REPL[1]:1
unknown function (ip: 0x7f1214e96a50)
_jl_invoke at /home/tim/src/julia-master/src/gf.c:2224 [inlined]
jl_apply_generic at /home/tim/src/julia-master/src/gf.c:2406
jl_apply at /home/tim/src/julia-master/src/julia.h:1701 [inlined]
do_call at /home/tim/src/julia-master/src/interpreter.c:115
eval_value at /home/tim/src/julia-master/src/interpreter.c:204
eval_stmt_value at /home/tim/src/julia-master/src/interpreter.c:155 [inlined]
eval_body at /home/tim/src/julia-master/src/interpreter.c:557
jl_interpret_toplevel_thunk at /home/tim/src/julia-master/src/interpreter.c:669
jl_toplevel_eval_flex at /home/tim/src/julia-master/src/toplevel.c:879
jl_toplevel_eval_flex at /home/tim/src/julia-master/src/toplevel.c:827
jl_toplevel_eval_flex at /home/tim/src/julia-master/src/toplevel.c:827
jl_toplevel_eval_in at /home/tim/src/julia-master/src/toplevel.c:931
eval at ./boot.jl:360
_jl_invoke at /home/tim/src/julia-master/src/gf.c:2224 [inlined]
jl_apply_generic at /home/tim/src/julia-master/src/gf.c:2406
eval_user_input at /home/tim/src/julia-master/usr/share/julia/stdlib/v1.6/REPL/src/REPL.jl:139
repl_backend_loop at /home/tim/src/julia-master/usr/share/julia/stdlib/v1.6/REPL/src/REPL.jl:200
start_repl_backend at /home/tim/src/julia-master/usr/share/julia/stdlib/v1.6/REPL/src/REPL.jl:185
#run_repl#42 at /home/tim/src/julia-master/usr/share/julia/stdlib/v1.6/REPL/src/REPL.jl:317
run_repl at /home/tim/src/julia-master/usr/share/julia/stdlib/v1.6/REPL/src/REPL.jl:305
_jl_invoke at /home/tim/src/julia-master/src/gf.c:2224 [inlined]
jl_apply_generic at /home/tim/src/julia-master/src/gf.c:2406
#872 at ./client.jl:387
jfptr_YY.872_44432 at /home/tim/src/julia-master/usr/lib/julia/sys.so (unknown line)
_jl_invoke at /home/tim/src/julia-master/src/gf.c:2224 [inlined]
jl_apply_generic at /home/tim/src/julia-master/src/gf.c:2406
jl_apply at /home/tim/src/julia-master/src/julia.h:1701 [inlined]
do_apply at /home/tim/src/julia-master/src/builtins.c:672
jl_f__apply_latest at /home/tim/src/julia-master/src/builtins.c:722
#invokelatest#2 at ./essentials.jl:707 [inlined]
invokelatest at ./essentials.jl:706 [inlined]
run_main_repl at ./client.jl:372
exec_options at ./client.jl:302
_start at ./client.jl:485
jfptr__start_43916 at /home/tim/src/julia-master/usr/lib/julia/sys.so (unknown line)
_jl_invoke at /home/tim/src/julia-master/src/gf.c:2224 [inlined]
jl_apply_generic at /home/tim/src/julia-master/src/gf.c:2406
jl_apply at /home/tim/src/julia-master/src/julia.h:1701 [inlined]
true_main at /home/tim/src/julia-master/src/jlapi.c:569
repl_entrypoint at /home/tim/src/julia-master/src/jlapi.c:688
main at /home/tim/src/julia-master/cli/loader_exe.c:51
__libc_start_main at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
_start at /home/tim/src/julia-master/usr/bin/julia (unknown line)
Allocations: 630837 (Pool: 630467; Big: 370); GC: 1

Without the ::typeof(sz) there is no crash. Conversely, colmajorstrides(sz::Dims{N}) where N = (1,cumprod(Int[sz[1:end-1]...])...)::Dims{N} does crash.

julia> versioninfo()
Julia Version 1.6.0-DEV.1780
Commit de69b02a48* (2020-12-19 19:43 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: Intel(R) Core(TM) i7-8850H CPU @ 2.60GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-11.0.0 (ORCJIT, skylake)
Environment:
  JULIA_CPU_THREADS = 4

[Keno]

I can't reproduce this - an rr trace would be helpful.

[martinholters]

I can reproduce on 1.6.0-beta1.1, but seems fixed on master (aka 1.7.0-DEV.367). In particular, on 1.6:

julia> @code_typed colmajorstrides((3,3,3))
CodeInfo(
# ...
14 ─ %27 = Core._apply_iterate(Base.iterate, Core.tuple, (1,), %24)::Core.PartialStruct(Tuple{Int64, Vararg{Int64, N} where N}, Any[Core.Const(1), Vararg{Int64, N} where N])
│          Core.typeassert(%27, Tuple{Int64, Int64, Int64})::Union{}
│          π (%27, Union{})
└───       unreachable
) => Union{}

While on master:

julia> @code_typed colmajorstrides((3,3,3))
CodeInfo(
# ...
26 ─ %46 = Core._apply_iterate(Base.iterate, Core.tuple, (1,), %43)::Core.PartialStruct(Tuple{Int64, Vararg{Int64}}, Any[Core.Const(1), Vararg{Int64}])
│          Core.typeassert(%46, Tuple{Int64, Int64, Int64})::Core.PartialStruct(Tuple{Int64, Vararg{Int64}}, Any[Core.Const(1), Vararg{Int64}])
│    %48 = π (%46, Core.PartialStruct(Tuple{Int64, Vararg{Int64}}, Any[Core.Const(1), Vararg{Int64}]))
└───       return %48
) => Tuple{Int64, Vararg{Int64}}

Might be worth bisecting to identify the fix and backport it. Tangentially, note that the inference result on master is also not as precise as it could be. And indeed, on 1.5:

julia> @code_typed colmajorstrides((3,3,3))
CodeInfo(
# ...
14 ─ %27 = Core._apply_iterate(Base.iterate, Core.tuple, (1,), %24)::Core.Compiler.PartialStruct(Tuple{Int64,Vararg{Int64,N} where N}, Any[Core.Compiler.Const(1, false), Vararg{Int64,N} where N])
│          Core.typeassert(%27, Tuple{Int64,Int64,Int64})::Tuple{Int64,Int64,Int64}
│    %29 = π (%27, Tuple{Int64,Int64,Int64})
└───       return %29
) => Tuple{Int64,Int64,Int64}

Here, we loose the Const(1), but get the more precise return type in the end.

[martinholters]

Bisected the regression to 8a0e0c8 and the fix to 80ace52. Unfortunately, the latter is probably not eligible for backport, so I guess we need to come up with a specific fix for 1.6.

[martinholters]

An adapted excerpt from #38136 is on the backports branch now, so this should be fixed in 1.6.0-rc1.

[timholy]

Thanks folks. @Keno, sorry I missed your reply. As I mentioned in the OP, Julia's rr submission is broken for me.