Add complex erf,erfc,erfcx,erfi,dawson routines from Faddeeva package. * libinterp/corefcn/mappers.cc: Add erfi and dawson mapper functions, and add complex-argument test cases for erf, erfc, erfcx, erfi, and dawson. * libinterp/octave-value/ov-base.cc, libinterp/octave-value/ov-base.h: Add erfi and dawson mapper functions. * libinterp/octave-value/ov-complex.cc, libinterp/octave-value/ov-cx-mat.cc, libinterp/octave-value/ov-cx-sparse.cc, libinterp/octave-value/ov-float.cc, libinterp/octave-value/ov-flt-complex.cc, libinterp/octave-value/ov-flt-cx-mat.cc, libinterp/octave-value/ov-flt-re-mat.cc, libinterp/octave-value/ov-re-mat.c, libinterp/octave-value/ov-re-sparse.cc, libinterp/octave-value/ov-scalar.cc, libinterp/octave-value/ov.h: Support erf, erfc, erfcx, erfi, and dawson mapper functions for real and complex matrices and scalars. * liboctave/cruft/Faddeeva/Faddeeva.cc, liboctave/cruft/Faddeeva/Faddeeva.hh: liboctave/cruft/Faddeeva/module.mk, liboctave/cruft/Makefile.am: Add Faddeeva package (from http://ab-initio.mit.edu/Faddeeva) to libcruft, to provide the various complex-argument error functions. * liboctave/numeric/lo-specfun.cc, liboctave/numeric/lo-specfun.h: Add complex-argument erf, erfc, erfcx, erfi, and dawson functions to liboctave API. Delete previous real-argument erfcx implementation in favor of Faddeeva::erfcx (which seems to be slightly faster in gcc/x86-64 benchmarks, with similar accuracy). * doc/interpreter/arith.txi: Include erfi and dawson documentation.

--- a/liboctave/numeric/module.mk
+++ b/liboctave/numeric/module.mk
@@ -179,6 +179,7 @@
 numeric_libnumeric_la_CPPFLAGS = \
   $(liboctave_la_CPPFLAGS) \
   $(ARPACK_CPPFLAGS) \
+  -I$(srcdir)/cruft/Faddeeva \
   $(FFTW_XCPPFLAGS) \
   $(SPARSE_XCPPFLAGS)