From 0f45153177aa7ffab10acbdc07ec2be341c8e07e Mon Sep 17 00:00:00 2001
From: lhausamm <loic_hausammann@hotmail.com>
Date: Thu, 26 Oct 2017 09:11:46 +0200
Subject: [PATCH 1/2] Add ipow

---
 pycuda/gpuarray.py | 44 ++++++++++++++++++++++++++++++++++----------
 1 file changed, 34 insertions(+), 10 deletions(-)

diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index 3f6fd32d..99b63969 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -605,14 +605,10 @@ class GPUArray(object):
 
         return result
 
-    def __pow__(self, other):
-        """pow function::
-
-           example:
-                   array = pow(array)
-                   array = pow(array,4)
-                   array = pow(array,array)
-
+    def _pow(self, other, new):
+        """
+        Do the pow operator.
+        with new, the user can choose between ipow or just pow
         """
 
         if isinstance(other, GPUArray):
@@ -622,7 +618,10 @@ class GPUArray(object):
 
             assert self.shape == other.shape
 
-            result = self._new_like_me(_get_common_dtype(self, other))
+            if new:
+                result = self._new_like_me(_get_common_dtype(self, other))
+            else:
+                result = self
 
             func = elementwise.get_pow_array_kernel(
                     self.dtype, other.dtype, result.dtype)
@@ -637,7 +636,10 @@ class GPUArray(object):
                 raise RuntimeError("only contiguous arrays may "
                         "be used as arguments to this operation")
 
-            result = self._new_like_me()
+            if new:
+                result = self._new_like_me()
+            else:
+                result = self
             func = elementwise.get_pow_kernel(self.dtype)
             func.prepared_async_call(self._grid, self._block, None,
                     other, self.gpudata, result.gpudata,
@@ -645,6 +647,28 @@ class GPUArray(object):
 
             return result
 
+    def __pow__(self, other):
+        """pow function::
+
+           example:
+                   array = pow(array)
+                   array = pow(array,4)
+                   array = pow(array,array)
+
+        """
+        return self._pow(other,new=True)
+
+    def __ipow__(self, other):
+        """ipow function::
+
+           example:
+                   array **= 4
+                   array **= array
+
+        """
+        return self._pow(other,new=False)
+        
+
     def reverse(self, stream=None):
         """Return this array in reversed order. The array is treated
         as one-dimensional.
-- 
GitLab


From e8816d87f5512cacf44f7659c719eaff4c28f554 Mon Sep 17 00:00:00 2001
From: lhausamm <loic_hausammann@hotmail.com>
Date: Fri, 27 Oct 2017 15:27:21 +0200
Subject: [PATCH 2/2] Add test for ipow

---
 test/test_gpuarray.py | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/test/test_gpuarray.py b/test/test_gpuarray.py
index 0a3218e1..103bba3d 100644
--- a/test/test_gpuarray.py
+++ b/test/test_gpuarray.py
@@ -36,6 +36,10 @@ class TestGPUArray:
         result = (a_gpu**a_gpu).get()
         assert (np.abs(pow(a, a) - result) < 1e-3).all()
 
+        a_gpu **= a_gpu
+        a_gpu = a_gpu.get()
+        assert (np.abs(pow(a, a) - a_gpu) < 1e-3).all()
+
     @mark_cuda_test
     def test_pow_number(self):
         a = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]).astype(np.float32)
@@ -44,6 +48,10 @@ class TestGPUArray:
         result = pow(a_gpu, 2).get()
         assert (np.abs(a**2 - result) < 1e-3).all()
 
+        a_gpu **= 2
+        a_gpu = a_gpu.get()
+        assert (np.abs(a**2 - a_gpu) < 1e-3).all()
+
     @mark_cuda_test
     def test_numpy_integer_shape(self):
         gpuarray.empty(np.int32(17), np.float32)
-- 
GitLab