Debug Travis macOS

[maleadt]

No description provided.

[maleadt]

The failing test:

@testset "loops" begin
    program = """
    def fib(x) {
        var a = 1, b = 1
        for i = 3, i < x, 1.0 {
            var c = a + b
            a = b
            b = c
        }
        b
    }

    def entry() {
        fib(10)
    }
    """

    LLVM.Context() do ctx
        m = Kaleidoscope.generate_IR(program, ctx)
        Kaleidoscope.optimize!(m)
        mktemp() do path, io
            Kaleidoscope.write_objectfile(m, path)
        end
        println(m)
        v = Kaleidoscope.run(m, "entry")
        @test v == 55.0
    end
end

Generated module looks sane:

; ModuleID = 'KaleidoscopeModule'
source_filename = "KaleidoscopeModule"
target datalayout = "e-m:o-p:32:32-f64:32:64-f80:128-n8:16:32-S128"
define double @fib(double %x) {
entry:
  br label %loop
loop:                                             ; preds = %loop, %entry
  %i6 = phi double [ %nextvar, %loop ], [ 3.000000e+00, %entry ]
  %b3 = phi double [ %addtmp, %loop ], [ 1.000000e+00, %entry ]
  %a2 = phi double [ %b3, %loop ], [ 1.000000e+00, %entry ]
  %addtmp = fadd double %a2, %b3
  %cmptmp = fcmp olt double %i6, %x
  %nextvar = fadd double %i6, 1.000000e+00
  br i1 %cmptmp, label %loop, label %afterloop
afterloop:                                        ; preds = %loop
  ret double %addtmp
}
define double @entry() {
entry:
  %calltmp = call double @fib(double 1.000000e+01)
  ret double %calltmp
}
; Function Attrs: nounwind
declare void @llvm.stackprotector(i8*, i8**) #0
attributes #0 = { nounwind }

[maleadt]

Huh, can reproduce with binary releases of Julia.

[maleadt]

So the difference between a locally-built Julia/LLVM and the binary one is the datalayout...

• binary Julia: e-m:o-p:32:32-f64:32:64-f80:128-n8:16:32-S128
• built Julia: e-m:o-i64:64-f80:128-n8:16:32:64-S128

Not sure how that differs between different LLVM builds.

[maleadt]

According to https://github.com/ghc/ghc/blob/master/llvm-targets these are respectively the 32 and 64 bit macOS target...

,("i386-apple-darwin", ("e-m:o-p:32:32-f64:32:64-f80:128-n8:16:32-S128", "yonah", ""))
,("x86_64-apple-darwin", ("e-m:o-i64:64-f80:128-n8:16:32:64-S128", "core2", ""))

[maleadt]

https://github.com/dslab-epfl/asap/blob/5e69c08f0c2973c17bc9c120445067ff14dde31a/lib/Target/X86/X86TargetMachine.cpp#L60-L64

[maleadt]

Triple is wrong.

Binary:

$ julia -e "using InteractiveUtils; versioninfo(); println(unsafe_string(ccall(:LLVMGetDefaultTargetTriple, Cstring, ())))"
Julia Version 1.0.2
Commit d789231e99 (2018-11-08 20:11 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin14.5.0)
  CPU: Intel(R) Core(TM) i5-4308U CPU @ 2.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.0 (ORCJIT, haswell)
Environment:
  JULIA_DEPOT_PATH = /Users/julia/tbesard/depot
  JULIA_PKG_DEVDIR = /Users/julia/tbesard/dev

i386-apple-darwin15.6.0

Built:

$ ~/Julia/julia-1.0/julia -e "using InteractiveUtils; versioninfo(); println(unsafe_string(ccall(:LLVMGetDefaultTargetTriple, Cstring, ())))"
Julia Version 1.0.3-pre.0
Commit 4f6c8c8 (2018-11-09 04:07 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin15.6.0)
  CPU: Intel(R) Core(TM) i5-4308U CPU @ 2.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.0 (ORCJIT, haswell)
Environment:
  JULIA_DEPOT_PATH = /Users/julia/tbesard/depot
  JULIA_PKG_DEVDIR = /Users/julia/tbesard/dev

x86_64-apple-darwin15.6.0

getDefaultTargetTriple() - Return the default target triple the compiler has been configured to produce code for.

Well, it's not wrong. So we can't rely on the default triple being well suited for the local host.

[maleadt]

On the other hand: https://llvm.org/docs/tutorial/LangImpl08.html

LLVM provides sys::getDefaultTargetTriple, which returns the target triple of the current machine.

It's set by CMake (LLVM_DEFAULT_TARGET_TRIPLE) though so that can't work.