New @code_native regression.

[Ismael Venegas Castelló]

Just compiled Julia, tests are running and haven't failed yet or anything, yet I've found another regression related to @code_native, emitting a massive amount of native code:

ismaelvc@server1 ~ % julia-dev

               _

   _       _ _(_)_     |  A fresh approach to technical computing

  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org

   _ _   _| |_  __ _   |  Type "?help" for help.

  | | | | | | |/ _` |  |

  | | |_| | | | (_| |  |  Version 0.5.0-dev+3771 (2016-04-27 16:06 UTC)

 _/ |\__'_|_|_|\__'_|  |  Commit 205d7e3 (0 days old master)

|__/                   |  x86_64-linux-gnu

ERROR: Invalid history file (~/.julia_history) format:

If you have a history file left over from an older version of Julia,

try renaming or deleting it.

Invalid character: '\t' at line 1

 in error(::ASCIIString, ::ASCIIString, ::Vararg{Any}) at ./error.jl:22

 [inlined code] from ./tuple.jl:8

 in hist_from_file(::Base.REPL.REPLHistoryProvider, ::IOStream) at ./REPL.jl:342

 in #setup_interface#26(::Bool, ::Array{Dict{Any,Any},1}, ::Any, ::Base.REPL.LineEditREPL) at ./REPL.jl:749

 [inlined code] from ./REPL.jl:701

 in run_frontend(::Base.REPL.LineEditREPL, ::Base.REPL.REPLBackendRef) at ./REPL.jl:876

 in run_repl(::Base.REPL.LineEditREPL, ::Any) at ./REPL.jl:166

 in _start() at ./client.jl:361

INFO: Disabling history file for this session.

julia> @code_llvm 1 + 1

define i64 @"jlsys_+_32762"(i64, i64) #0 {

top:

  %2 = add i64 %1, %0

  ret i64 %2

}

shell> julia-dev -e "@code_native 1 + 1" > regression.txt

shell> wc -l regression.txt

42850 regression.txt

regression.txt

* https://gist.github.com/Ismael-VC/911156fa3266fe3bed350a1a06fc59a2

I report it here again, because I have no idea what caused this, will open a new issue if asked, thanks!

[Yichao Yu]

https://github.com/JuliaLang/julia/issues/15971

[Ismael Venegas Castelló]

Thanks for pointing it out!

[Scott Jones]

I've been seeing this in a lot of small functions (from Base!), that seems like it may be causing performance regressions compared to v0.4.x (where this doesn't occur):

Source line: 0
 pushq %rbp
 movq %rsp, %rbp

  pushq %r14 

 pushq %rbx
 subq $32, %rsp
 movq %rsi, %r14
 movabsq $jl_get_ptls_states_static, %rax
 callq *%rax
 movq %rax, %rbx
 movq $0, -24(%rbp)
 movq $2, -40(%rbp)
 movq (%rbx), %rax
 movq %rax, -32(%rbp)
 leaq -40(%rbp), %rax
 movq %rax, (%rbx)

It makes the function twice as large as it was in v0.4.x.

I believe this is because of the compiling for thread support by default in v0.5.