source: codes/icosagcm/devel/Python/dev/numba.py @ 985

from __future__ import absolute_import     
from __future__ import print_function

import numpy as np
import numba
from numba import int32, int64, float64

class NumbaData(object):
    """A base class to extract data from derived class instances and use it as argument to @jit functions. Derived classes must set the 'signature' attribute."""
    def data(self):
        """Returns a jitclass instance containing attributes copied from self, using self.signature which is of the form type,names,type,names ... where names is a string 'attr1 attr2 attr3' containing space-separated names of attributes of self. Those attributes are declared to numba with the type preceding them. The result of data() can be used as argument to a @jit function."""
        cls = self.__class__.__name__
        spec = []
        for name in dir(self):
            attr = getattr(self,name)
            if not name.startswith('__') and not callable(attr):
                tp = None
                if isinstance(attr, int): tp=int64
                if isinstance(attr, float): tp=float64
                if isinstance(attr, np.ndarray):
                    dtype=attr.dtype
                    if dtype == np.int32 : dtype=int32
                    elif dtype == np.float64 : dtype=float64
                    else: dtype=None
                    if dtype is None: print('Unknown dtype ', attr.dtype)
                        
                    if   len(attr.shape)==1 : tp = dtype[:]
                    elif len(attr.shape)==2 : tp = dtype[:,:]
                    elif len(attr.shape)==3 : tp = dtype[:,:,:]
                    else: print('%s.%s is a numpy array with unsupported rank >3'%(cls,name))
                        
                if tp is None:
                    print('Type of attribute %s.%s is not recognized'%(cls,name), type(attr))
                else:
                    spec.append( (name, tp) )
        
        @numba.jitclass(spec)
        class JitClass(object):
            def __init__(self): pass

        data=JitClass()
        for name,thetype in spec: 
            print( 'Making %s.%s available @jit functions : '%(self.__class__.__name__, name), type(getattr(self,name)))
            setattr(data, name, getattr(self,name))
        return data

jit=numba.jit(nopython=True, nogil=True, error_model='numpy', fastmath=True)