Hello all,
Thanks for this great package which ease so much the use of cuda !
I think I found out a small bug in the current git version of the code.
If you run the following code in ipython :
>>> from pycuda import autoinit
>>> from pycuda import driver, compiler, gpuarray, tools
>>> a = gpuarray.ones(16, dtype=float32)
>>> a += 1
you get :
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
/home/data/projets/pycuda_mult/<ipython console> in <module>()
/usr/lib/python2.6/site-packages/pycuda-0.94rc-py2.6-linux-x86_64.egg/pycuda/gpuarray.py
in __iadd__(self, other)
258
259 def __iadd__(self, other):
--> 260 return self._axpbyz(1, other, 1, self)
261
262 def __isub__(self, other):
/usr/lib/python2.6/site-packages/pycuda-0.94rc-py2.6-linux-x86_64.egg/pycuda/gpuarray.py
in _axpbyz(self, selffac, other, otherfac, out, add_timer, stream)
140 """Compute ``out = selffac * self + otherfac*other``,
141 where `other` is a vector.."""
--> 142 assert self.shape == other.shape
143
144 func = elementwise.get_axpbyz_kernel(self.dtype,
other.dtype, out.dtype)
AttributeError: 'int' object has no attribute 'shape'
The following patch should solve this issue. It adds also a ones function as
it exist in numpy and define __sub__ using __add__ to avoid code
duplication.
Tell me what you think about those changes. Also I do not know if patches
should be submitted to the mailing list but could not find another way.
Cheers,
Nicolas
path follows:
diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index a0a1da1..b685e52 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -235,17 +235,7 @@ class GPUArray(object):
def __sub__(self, other):
"""Substract an array from an array or a scalar from an array."""
-
- if isinstance(other, GPUArray):
- result = self._new_like_me(_get_common_dtype(self, other))
- return self._axpbyz(1, other, -1, result)
- else:
- if other == 0:
- return self
- else:
- # create a new array for the result
- result = self._new_like_me()
- return self._axpbz(1, -other, result)
+ return self.__add__(-other)
def __rsub__(self,other):
"""Substracts an array by a scalar or an array::
@@ -257,10 +247,18 @@ class GPUArray(object):
return self._axpbz(-1, other, result)
def __iadd__(self, other):
- return self._axpbyz(1, other, 1, self)
+ if isinstance(other, GPUArray):
+ # add another vector
+ return self._axpbyz(1, other, 1, self)
+ else:
+ # add a scalar
+ if other == 0:
+ return self
+ else:
+ return self._axpbz(1, other, self)
def __isub__(self, other):
- return self._axpbyz(1, other, -1, self)
+ return self.__iadd__(-other)
def __neg__(self):
result = self._new_like_me()
@@ -631,6 +629,13 @@ def zeros(shape, dtype, allocator=drv.mem_alloc):
result.fill(0)
return result
+def ones(shape, dtype, allocator=drv.mem_alloc):
+ """Returns an array of the given shape and dtype filled with 1's."""
+
+ result = GPUArray(shape, dtype, allocator)
+ result.fill(1)
+ return result
+
def empty_like(other_ary):
result = GPUArray(
other_ary.shape, other_ary.dtype, other_ary.allocator)
_______________________________________________
PyCUDA mailing list
[email protected]
http://lists.tiker.net/listinfo/pycuda
