Author: Romain Guillebert <romain...@gmail.com>
Branch: numpypy-nditer
Changeset: r64813:8c3a4fc396d3
Date: 2013-06-06 17:21 +0200
http://bitbucket.org/pypy/pypy/changeset/8c3a4fc396d3/
Log: Refactor the way nditer iterates
diff --git a/pypy/module/micronumpy/interp_nditer.py
b/pypy/module/micronumpy/interp_nditer.py
--- a/pypy/module/micronumpy/interp_nditer.py
+++ b/pypy/module/micronumpy/interp_nditer.py
@@ -5,10 +5,41 @@
from pypy.module.micronumpy.base import W_NDimArray, convert_to_array
from pypy.module.micronumpy.strides import (calculate_broadcast_strides,
shape_agreement_multiple)
-from pypy.module.micronumpy.iter import MultiDimViewIterator
+from pypy.module.micronumpy.iter import MultiDimViewIterator, SliceIterator
from pypy.module.micronumpy import support
from pypy.module.micronumpy.arrayimpl.concrete import SliceArray
+class AbstractIterator(object):
+ def done(self):
+ raise NotImplementedError("Abstract Class")
+
+ def next(self):
+ raise NotImplementedError("Abstract Class")
+
+ def getitem(self, array):
+ raise NotImplementedError("Abstract Class")
+
+class IteratorMixin(object):
+ _mixin_ = True
+ def __init__(self, it, op_flags):
+ self.it = it
+ self.op_flags = op_flags
+
+ def done(self):
+ return self.it.done()
+
+ def next(self):
+ self.it.next()
+
+ def getitem(self, space, array):
+ return self.op_flags.get_it_item(space, array, self.it)
+
+class BoxIterator(IteratorMixin):
+ pass
+
+class SliceIterator(IteratorMixin):
+ pass
+
def parse_op_arg(space, name, w_op_flags, n, parse_one_arg):
ret = []
if space.is_w(w_op_flags, space.w_None):
@@ -53,6 +84,13 @@
#it.dtype.setitem(res, 0, it.getitem())
return W_NDimArray(res)
+def get_readonly_slice(space, array, it):
+ #XXX Not readonly
+ return W_NDimArray(it.getslice())
+
+def get_readwrite_slice(space, array, it):
+ return W_NDimArray(it.getslice())
+
def parse_op_flag(space, lst):
op_flag = OpFlag()
for w_item in lst:
@@ -191,11 +229,11 @@
self.iters=[]
self.shape = iter_shape = shape_agreement_multiple(space, self.seq)
if self.external_loop:
- xxx find longest contiguous shape
+ #XXX find longest contiguous shape
iter_shape = iter_shape[1:]
for i in range(len(self.seq)):
- self.iters.append(get_iter(space, self.order,
- self.seq[i].implementation, iter_shape))
+ self.iters.append(BoxIterator(get_iter(space, self.order,
+ self.seq[i].implementation, iter_shape),
self.op_flags[i]))
def descr_iter(self, space):
return space.wrap(self)
@@ -220,8 +258,7 @@
raise OperationError(space.w_StopIteration, space.w_None)
res = []
for i in range(len(self.iters)):
- res.append(self.op_flags[i].get_it_item(space, self.seq[i],
- self.iters[i]))
+ res.append(self.iters[i].getitem(space, self.seq[i]))
self.iters[i].next()
if len(res) <2:
return res[0]
diff --git a/pypy/module/micronumpy/iter.py b/pypy/module/micronumpy/iter.py
--- a/pypy/module/micronumpy/iter.py
+++ b/pypy/module/micronumpy/iter.py
@@ -32,13 +32,13 @@
shape dimension
which is back 25 and forward 1,
which is x.strides[1] * (x.shape[1] - 1) + x.strides[0]
-so if we precalculate the overflow backstride as
+so if we precalculate the overflow backstride as
[x.strides[i] * (x.shape[i] - 1) for i in range(len(x.shape))]
we can go faster.
All the calculations happen in next()
next_skip_x() tries to do the iteration for a number of steps at once,
-but then we cannot gaurentee that we only overflow one single shape
+but then we cannot gaurentee that we only overflow one single shape
dimension, perhaps we could overflow times in one big step.
"""
@@ -266,6 +266,30 @@
def reset(self):
self.offset %= self.size
+class SliceIterator(object):
+ def __init__(self, arr, stride, backstride, shape, dtype=None):
+ self.step = 0
+ self.arr = arr
+ self.stride = stride
+ self.backstride = backstride
+ self.shape = shape
+ if dtype is None:
+ dtype = arr.implementation.dtype
+ self.dtype = dtype
+ self._done = False
+
+ def done():
+ return self._done
+
+ def next():
+ self.step += self.arr.implementation.dtype.get_size()
+ if self.step == self.backstride - self.implementation.dtype.get_size():
+ self._done = True
+
+ def getslice(self):
+ from pypy.module.micronumpy.arrayimpl.concrete import SliceArray
+ return SliceArray(self.step, [self.stride], [self.backstride],
self.shape, self.arr.implementation, self.arr, self.dtype)
+
class AxisIterator(base.BaseArrayIterator):
def __init__(self, array, shape, dim, cumultative):
self.shape = shape
@@ -288,7 +312,7 @@
self.dim = dim
self.array = array
self.dtype = array.dtype
-
+
def setitem(self, elem):
self.dtype.setitem(self.array, self.offset, elem)
_______________________________________________
pypy-commit mailing list
pypy-commit@python.org
http://mail.python.org/mailman/listinfo/pypy-commit
