Example crashes on Mac OS X

[numba]

Example crashes on Mac OS X

[aterrel]

I'm getting the same error on examples/numexpr_compare.py but examples/example.py works. @bayerj do you get this error for examples/example.py

[bayerj]

This is my output for example.py, seems valid:

$ py example.py 
; ModuleID = 'sinc_mod'

define double @sinc(double %x) {
Entry:
  %0 = fcmp oeq double %x, 0.000000e+00
  br i1 %0, label %CONT_9, label %IF_FALSE_9

CONT_9:                                           ; preds = %Entry
  ret double 1.000000e+00

IF_FALSE_9:                                       ; preds = %Entry
  %1 = fmul double %x, 0x400921FB54442D18
  %2 = call double @llvm.sin.f64(double %1)
  %3 = fmul double 0x400921FB54442D18, %x
  %4 = fdiv double %2, %3
  ret double %4

RETURN_37:                                        ; No predecessors!
  ret double 0.000000e+00
}

declare double @llvm.sin.f64(double) nounwind readonly

0.059888 1.000000 ** 
; ModuleID = 'sinc_mod'

define double @sinc(double %x) {
Entry:
  %0 = fcmp oeq double %x, 0.000000e+00
  br i1 %0, label %CONT_9, label %IF_FALSE_9

CONT_9:                                           ; preds = %Entry
  ret double 1.000000e+00

IF_FALSE_9:                                       ; preds = %Entry
  %1 = fmul double %x, 0x400921FB54442D18
  %2 = call double @llvm.sin.f64(double %1)
  %3 = fmul double 0x400921FB54442D18, %x
  %4 = fdiv double %2, %3
  ret double %4

RETURN_37:                                        ; No predecessors!
  ret double 0.000000e+00
}

declare double @llvm.sin.f64(double) nounwind readonly

0.059888 1.000000 ** 
[  3.89817183e-17   3.96971926e-02   6.64556414e-02   6.80812328e-02
   4.15063827e-02  -5.12469250e-03  -5.46515558e-02  -8.68921740e-02
  -8.64379535e-02  -4.93794333e-02   1.37553357e-02   8.03754744e-02
   1.23178051e-01   1.20247416e-01   6.44510924e-02  -3.13845935e-02
  -1.35838264e-01  -2.07001488e-01  -2.05197221e-01  -1.06420158e-01
   8.79177183e-02   3.48410566e-01   6.23566834e-01   8.52825194e-01
   9.82960198e-01   9.82960198e-01   8.52825194e-01   6.23566834e-01
   3.48410566e-01   8.79177183e-02  -1.06420158e-01  -2.05197221e-01
  -2.07001488e-01  -1.35838264e-01  -3.13845935e-02   6.44510924e-02
   1.20247416e-01   1.23178051e-01   8.03754744e-02   1.37553357e-02
  -4.93794333e-02  -8.64379535e-02  -8.68921740e-02  -5.46515558e-02
  -5.12469250e-03   4.15063827e-02   6.80812328e-02   6.64556414e-02
   3.96971926e-02   3.89817183e-17]

[jriehl]

This is most likely an issue with LLVM and llvm-py. I saw this exact problem using LLVM version 3.2svn (from the head of the development repository), and it was caused by the intrinsic codes in LLVM mismatching with those in llvm-py's _intrinsic_ids.py. Fixed by regenerating these codes in llvm-py.

You can verify this for yourself by doing a comparison of the output of ".../llvm-py/tools/intrgen.py llvm-config --includedir/llvm/Intrisics.gen" with the codes in ".../llvm-py/llvm/_intrinsic_ids.py". Note that you might have multiple installations of LLVM, so check the output of llvm-config --version against the LLVM version you built.

(Sorry about the crappy misuse of the markdown, hopefully this is parse-able.)