Python and Numpy for Xeon Phi

Makefile

PY_VER = 2.7.3
SRC = $(shell readlink -f python)
MIC_PY_HOME = $(SRC)/_install
MIC_PY_PATH = $(MIC_PY_HOME)/lib/python2.7
CTYPES = $(SRC)/Modules/_ctypes

all: $(MIC_PY_PATH)/_ctypes

install:
	@echo "To install, copy $(MIC_PY_HOME) where it can be accessed from the MIC card."
	$(error)

python.tgz:
	wget http://www.python.org/ftp/python/$(PY_VER)/Python-$(PY_VER).tgz -O $@

$(SRC): python.tgz
	mkdir -p $@
	tar -xf $< -C $@ --strip-components=1

$(SRC)/hostpython: | $(SRC)
	# TODO: fail if patched
	cd $< && ./configure && make
	cd $< && mv python hostpython && mv Parser/pgen Parser/hostpgen && make distclean

xcompile.patch:
	wget http://randomsplat.com/wp-content/uploads/2012/10/Python-$(PY_VER)-xcompile.patch -O $@

$(SRC)/patched: xcompile.patch $(SRC)/hostpython
	cd $(SRC) && patch -p1 < ../$<
	touch $@

$(SRC)/Makefile: $(SRC)/patched $(CTYPES)/.git
	cd $(SRC) && ./configure CC="icc -mmic" CXX="icpc -mmic" --host=x86_64 --without-gcc

$(CTYPES)/.git: | $(SRC)
	mv $(CTYPES) $(CTYPES).orig
	git clone https://github.com/bpartridge/xeon_phi_ctypes.git $(CTYPES)

$(MIC_PY_PATH)/_ctypes: $(SRC)/Makefile $(SRC)/patched $(CTYPES)/.git
	sed -e "s/-OPT:Olimit=0//g" -i.backup $(SRC)/Makefile
	cd $(SRC) && make HOSTPYTHON=./hostpython HOSTPGEN=./Parser/hostpgen \
		CROSS_COMPILE=k1om- CROSS_COMPILE_TARGET=yes HOSTARCH=x86_64 BUILDARCH=x86_64-linux-gnu \
		EXTRA_CFLAGS="-fp-model precise -shared -fPIC" LDFLAGS=""
	mkdir -p $(MIC_PY_HOME)
	cd $(SRC) && make install HOSTPYTHON=./hostpython CROSS_COMPILE_TARGET=yes prefix=$(MIC_PY_HOME)

micclean: miccleanobjs
	rm $(SRC)/Makefile

miccleanobjs:
	rm -rvf $(SRC)/**/*.o
	rm -rf $(SRC)/build
	rm -rf $(MIC_PY_HOME)

numpy.tgz:
	wget http://downloads.sourceforge.net/project/numpy/NumPy/1.8.0/numpy-1.8.0.tar.gz -O $@

numpy: numpy.tgz
	mkdir $@
	tar -xf $< -C $@ --strip-components=1

numpyclean: numpy
	rm -rf numpy/build
	rm -rvf numpy/**/*.o

numpyxc: numpy | $(MIC_PY_HOME)
	# build_clib forces config.h to be created; it needs to be modified before extensions are built
	cd numpy && PYTHONXCPREFIX=$(MIC_PY_HOME) python numpyxc.py build_clib
	sed '/INTRIN/ d' -i numpy/build/*/numpy/core/include/numpy/config.h
	cd numpy && PYTHONXCPREFIX=$(MIC_PY_HOME) python numpyxc.py build

.PHONY: all micclean miccleanobjs numpyxc

numpyxc.py

import os

xcprefix = os.environ.get('PYTHONXCPREFIX', None)
if not xcprefix:
  raise Exception("Must specify PYTHONXCPREFIX as an argument to make this work")

from setup import setup_package

from numpy.distutils import fcompiler, ccompiler
from numpy.distutils.intelccompiler import IntelCCompiler

class IntelMICCompiler(IntelCCompiler):
  """ A modified Intel compiler compatible with an gcc built Python."""
  compiler_type = 'intelmic'
  cc_exe = 'icc'

  def __init__ (self, verbose=0, dry_run=0, force=0):
    IntelCCompiler.__init__ (self, verbose, dry_run, force)
    compiler = "icc -mmic -fPIC -I{0}/include -L{0}/lib -L{0}/lib/python2.7".format(xcprefix)
    self.set_executables(compiler=compiler + ' -shared',
                         compiler_so=compiler,
                         compiler_cxx=compiler,
                         linker_exe=compiler,
                         linker_so=compiler + ' -shared')

# Export it into a module that is __import__ed by np.distutils.ccompiler
import sys, imp
imc = imp.new_module('numpy.distutils.intelmiccompiler')
sys.modules['numpy.distutils.intelmiccompiler'] = imc
imc.IntelMICCompiler = IntelMICCompiler

# print ccompiler.compiler_class # {name: (module_name, module_member_name, desc), ...}
ccompiler.compiler_class['intelmic'] = ('intelmiccompiler', 'IntelMICCompiler', 'Intel Compiler for Xeon Phi')

fcompiler.load_all_fcompiler_classes()
# print fcompiler.fcompiler_class # {name: (name, klass, desc), ...}

# The following taken from https://bitbucket.org/lambacck/distutilscross
# import os
# from distutils import sysconfig
# xcprefix = os.environ.get('PYTHONXCPREFIX', None)
# if xcprefix:
#   _get_python_lib = sysconfig.get_python_lib
#   def get_python_lib(plat_specific=0, standard_lib=0, prefix=None):
#     return _get_python_lib(plat_specific, standard_lib, xcprefix)
#   sysconfig.get_python_lib = get_python_lib

#   _get_python_inc = sysconfig.get_python_inc
#   def get_python_inc(plat_specific=0, prefix=None):
#     return _get_python_inc(plat_specific, xcprefix)
#   sysconfig.get_python_inc = get_python_inc

#   sysconfig.PREFIX = xcprefix
#   sysconfig.EXEC_PREFIX = xcprefix
#   # now force a reload from the given prefix
#   sysconfig._config_vars = None
#   print sysconfig.get_config_var("LDSHARED")

# Disable SSE intrinsics since they don't seem to work

# import numpy.core.setup as core_setup
# setup_common.OPTIONAL_HEADERS = []
# setup_common.OPTIONAL_INTRINSICS = [] # TODO: reintroduce other builtins

setup_package()