From dc3e4f77e330760e6e670726f29056cfdcbdfd36 Mon Sep 17 00:00:00 2001
From: mit kotak <mitak2@illinois.edu>
Date: Tue, 3 Aug 2021 22:09:02 -0500
Subject: [PATCH 01/28] Implemented get_rpow_kernel

---
 pycuda/elementwise.py | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 258bae5c..7b705713 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -665,6 +665,32 @@ def get_pow_array_kernel(dtype_x, dtype_y, dtype_z):
         "pow_method",
     )
 
+@context_dependent_memoize
+def get_rpow_kernel(dtype_x, dtype_y, dtype_z , is_base_array, is_exp_array):
+
+    if is_base_array:
+        x = "x[i]"
+        x_ctype = "%(tp_x)s *x"
+    else:
+        x = "x"
+        x_ctype = "%(tp_x)s x"
+
+    if is_exp_array:
+        y = "y[i]"
+        y_ctype = "%(tp_y)s *y"
+    else:
+        y = "y"
+        y_ctype = "%(tp_y)s y"
+
+
+    return get_elwise_kernel(
+            ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype) % {
+                "tp_x": dtype_to_ctype(dtype_x),
+                "tp_y": dtype_to_ctype(dtype_y),
+                "tp_z": dtype_to_ctype(dtype_z),
+                },
+            "z[i] = %s" % result,
+            name="pow_method")
 
 @context_dependent_memoize
 def get_fmod_kernel():
-- 
GitLab


From caafe5d32f3b9d9658ee31d1c890f888e1fea0ad Mon Sep 17 00:00:00 2001
From: mit kotak <mitak2@illinois.edu>
Date: Tue, 3 Aug 2021 22:09:34 -0500
Subject: [PATCH 02/28] Implemented rpow_scalar and __rpow__

---
 pycuda/gpuarray.py | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index a0bf84c4..b6aa3f8d 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -515,6 +515,12 @@ class GPUArray:
 
         return out
 
+    def _rpow_scalar(result, base, exponent):
+        base = np.array(base)
+        return elementwise.get_rpow_kernel(
+                base.dtype, exponent.dtype, result.dtype,
+                is_base_array=False, is_exp_array=True)
+
     def _new_like_me(self, dtype=None, order="C"):
         strides = None
         if dtype is None:
@@ -839,6 +845,14 @@ class GPUArray:
         """
         return self._pow(other, new=False)
 
+    def __rpow__(self, other):
+        # other must be a scalar
+        common_dtype = _get_common_dtype(self, other)
+        result = self._new_like_me(common_dtype)
+        result.add_event(
+                self._rpow_scalar(result, common_dtype.type(other), self))
+        return result
+
     def reverse(self, stream=None):
         """Return this array in reversed order. The array is treated
         as one-dimensional.
-- 
GitLab


From d651b4f9c5fe4c2144d765c254b2bd7c673b2ffc Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Wed, 4 Aug 2021 22:04:31 +0000
Subject: [PATCH 03/28] Removed blank spaces

---
 pycuda/elementwise.py | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 7b705713..97093cde 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -665,26 +665,23 @@ def get_pow_array_kernel(dtype_x, dtype_y, dtype_z):
         "pow_method",
     )
 
+
 @context_dependent_memoize
-def get_rpow_kernel(dtype_x, dtype_y, dtype_z , is_base_array, is_exp_array):
+def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
     if is_base_array:
-        x = "x[i]"
         x_ctype = "%(tp_x)s *x"
     else:
-        x = "x"
         x_ctype = "%(tp_x)s x"
 
     if is_exp_array:
-        y = "y[i]"
         y_ctype = "%(tp_y)s *y"
     else:
-        y = "y"
         y_ctype = "%(tp_y)s y"
 
 
     return get_elwise_kernel(
-            ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype) % {
+            ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype) %{
                 "tp_x": dtype_to_ctype(dtype_x),
                 "tp_y": dtype_to_ctype(dtype_y),
                 "tp_z": dtype_to_ctype(dtype_z),
-- 
GitLab


From f75ac04085a6a7237226a593f40db5806287bfae Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Sat, 28 Aug 2021 01:42:15 +0000
Subject: [PATCH 04/28] Adjusted spaces in elementwise.py + Added self to
 gpuarray.py

---
 pycuda/elementwise.py | 15 ++++++++-------
 pycuda/gpuarray.py    |  2 +-
 2 files changed, 9 insertions(+), 8 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 97093cde..28269579 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -681,13 +681,14 @@ def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
 
     return get_elwise_kernel(
-            ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype) %{
-                "tp_x": dtype_to_ctype(dtype_x),
-                "tp_y": dtype_to_ctype(dtype_y),
-                "tp_z": dtype_to_ctype(dtype_z),
-                },
-            "z[i] = %s" % result,
-            name="pow_method")
+        ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype) 
+        %{
+            "tp_x": dtype_to_ctype(dtype_x),
+            "tp_y": dtype_to_ctype(dtype_y),
+            "tp_z": dtype_to_ctype(dtype_z),
+        },
+        "z[i] = %s" % result,
+        name="pow_method")
 
 @context_dependent_memoize
 def get_fmod_kernel():
diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index b6aa3f8d..a85367ff 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -515,7 +515,7 @@ class GPUArray:
 
         return out
 
-    def _rpow_scalar(result, base, exponent):
+    def _rpow_scalar(self, result, base, exponent):
         base = np.array(base)
         return elementwise.get_rpow_kernel(
                 base.dtype, exponent.dtype, result.dtype,
-- 
GitLab


From b4a8989cbd3f7ace09e4960640ce82a3625923e8 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Sat, 28 Aug 2021 01:48:06 +0000
Subject: [PATCH 05/28] replaced result with func

---
 pycuda/elementwise.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 28269579..74233f8e 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -681,15 +681,16 @@ def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
 
     return get_elwise_kernel(
-        ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype) 
+        ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype)
         %{
             "tp_x": dtype_to_ctype(dtype_x),
             "tp_y": dtype_to_ctype(dtype_y),
             "tp_z": dtype_to_ctype(dtype_z),
         },
-        "z[i] = %s" % result,
+        "z[i] = %s" % func,
         name="pow_method")
 
+
 @context_dependent_memoize
 def get_fmod_kernel():
     return get_elwise_kernel(
-- 
GitLab


From 9fb83772d1badc397d5c1fe024fc1c55a35443d1 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Thu, 16 Sep 2021 01:43:56 +0000
Subject: [PATCH 06/28] Defined func in elementwise.get_rpow_kernel

---
 pycuda/elementwise.py | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 74233f8e..c7cfecf2 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -671,13 +671,17 @@ def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
     if is_base_array:
         x_ctype = "%(tp_x)s *x"
+        func    = pow(x[i],y)
     else:
         x_ctype = "%(tp_x)s x"
+        func    = pow(x,y)
 
     if is_exp_array:
         y_ctype = "%(tp_y)s *y"
+        func    = pow(x,y[i])
     else:
         y_ctype = "%(tp_y)s y"
+        func    = pow(x,y)
 
 
     return get_elwise_kernel(
-- 
GitLab


From c84e42781f1b5ac0120ac9c07707a0a24917c15f Mon Sep 17 00:00:00 2001
From: mit kotak <mitak2@illinois.edu>
Date: Wed, 29 Sep 2021 20:47:24 -0500
Subject: [PATCH 07/28] Fixed typos

---
 pycuda/elementwise.py | 12 ++++++------
 pycuda/gpuarray.py    |  7 +------
 2 files changed, 7 insertions(+), 12 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index c7cfecf2..7e4aef78 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -669,19 +669,19 @@ def get_pow_array_kernel(dtype_x, dtype_y, dtype_z):
 @context_dependent_memoize
 def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
-    if is_base_array:
+    if is_base_array and not is_exp_array:
         x_ctype = "%(tp_x)s *x"
-        func    = pow(x[i],y)
+        func    = "pow(x[i],y)"
     else:
         x_ctype = "%(tp_x)s x"
-        func    = pow(x,y)
+        func    = "pow(x,y)"
 
-    if is_exp_array:
+    if is_exp_array and not is_exp_array:
         y_ctype = "%(tp_y)s *y"
-        func    = pow(x,y[i])
+        func    = "pow(x,y[i])"
     else:
         y_ctype = "%(tp_y)s y"
-        func    = pow(x,y)
+        func    = "pow(x,y)"
 
 
     return get_elwise_kernel(
diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index a85367ff..905b496a 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -515,12 +515,6 @@ class GPUArray:
 
         return out
 
-    def _rpow_scalar(self, result, base, exponent):
-        base = np.array(base)
-        return elementwise.get_rpow_kernel(
-                base.dtype, exponent.dtype, result.dtype,
-                is_base_array=False, is_exp_array=True)
-
     def _new_like_me(self, dtype=None, order="C"):
         strides = None
         if dtype is None:
@@ -847,6 +841,7 @@ class GPUArray:
 
     def __rpow__(self, other):
         # other must be a scalar
+        assert np.isscalar(other)
         common_dtype = _get_common_dtype(self, other)
         result = self._new_like_me(common_dtype)
         result.add_event(
-- 
GitLab


From 27864e0face645b8c8465af97ba3c7a0504069ba Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Thu, 30 Sep 2021 01:53:16 +0000
Subject: [PATCH 08/28] Removed rpow_scalar + Fixed typos

---
 pycuda/elementwise.py | 12 ++++++------
 pycuda/gpuarray.py    |  7 +------
 2 files changed, 7 insertions(+), 12 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index c7cfecf2..9781a1ed 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -669,19 +669,19 @@ def get_pow_array_kernel(dtype_x, dtype_y, dtype_z):
 @context_dependent_memoize
 def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
-    if is_base_array:
+    if is_base_array and not_exp_array:
         x_ctype = "%(tp_x)s *x"
-        func    = pow(x[i],y)
+        func    = "pow(x[i],y)"
     else:
         x_ctype = "%(tp_x)s x"
-        func    = pow(x,y)
+        func    = "pow(x,y)"
 
-    if is_exp_array:
+    if is_exp_array and not_exp_array:
         y_ctype = "%(tp_y)s *y"
-        func    = pow(x,y[i])
+        func    = "pow(x,y[i])"
     else:
         y_ctype = "%(tp_y)s y"
-        func    = pow(x,y)
+        func    = "pow(x,y)"
 
 
     return get_elwise_kernel(
diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index a85367ff..905b496a 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -515,12 +515,6 @@ class GPUArray:
 
         return out
 
-    def _rpow_scalar(self, result, base, exponent):
-        base = np.array(base)
-        return elementwise.get_rpow_kernel(
-                base.dtype, exponent.dtype, result.dtype,
-                is_base_array=False, is_exp_array=True)
-
     def _new_like_me(self, dtype=None, order="C"):
         strides = None
         if dtype is None:
@@ -847,6 +841,7 @@ class GPUArray:
 
     def __rpow__(self, other):
         # other must be a scalar
+        assert np.isscalar(other)
         common_dtype = _get_common_dtype(self, other)
         result = self._new_like_me(common_dtype)
         result.add_event(
-- 
GitLab


From 6829adbe18b8429df3a6ff44b989a447d0c63d89 Mon Sep 17 00:00:00 2001
From: mit kotak <mitak2@illinois.edu>
Date: Fri, 8 Oct 2021 20:09:00 -0500
Subject: [PATCH 09/28] Removed _get_scalar_func_loopy_program

---
 pycuda/elementwise.py | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 7e4aef78..b2e50ab6 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -672,17 +672,15 @@ def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
     if is_base_array and not is_exp_array:
         x_ctype = "%(tp_x)s *x"
         func    = "pow(x[i],y)"
-    else:
+    elif not is_base_array and is_exp_array:
         x_ctype = "%(tp_x)s x"
-        func    = "pow(x,y)"
+        func    = "pow(x,y[i])"
 
-    if is_exp_array and not is_exp_array:
+    elif is_exp_array and is_exp_array:
         y_ctype = "%(tp_y)s *y"
-        func    = "pow(x,y[i])"
-    else:
-        y_ctype = "%(tp_y)s y"
         func    = "pow(x,y)"
-
+    else:
+        raise Assertion
 
     return get_elwise_kernel(
         ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype)
-- 
GitLab


From ab3101592e637e8783a1265127183364c3943db6 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Fri, 22 Oct 2021 04:53:27 +0000
Subject: [PATCH 10/28] Corrected if condition logic in rpow_scalar

---
 pycuda/elementwise.py | 15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 9781a1ed..65a57d5d 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -669,20 +669,19 @@ def get_pow_array_kernel(dtype_x, dtype_y, dtype_z):
 @context_dependent_memoize
 def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
-    if is_base_array and not_exp_array:
+    #XOR check that raises an Assertion when both are scalar
+    raise AssertionError if not (is_exp_array or is_base_array)
+    if is_base_array and not is_exp_array:
         x_ctype = "%(tp_x)s *x"
         func    = "pow(x[i],y)"
-    else:
-        x_ctype = "%(tp_x)s x"
-        func    = "pow(x,y)"
 
-    if is_exp_array and not_exp_array:
+    if not is_base_array and is_exp_array:
         y_ctype = "%(tp_y)s *y"
         func    = "pow(x,y[i])"
-    else:
-        y_ctype = "%(tp_y)s y"
-        func    = "pow(x,y)"
 
+    if is_base_array and exp_array:
+        x_ctype = "%(tp_x)s x"
+        func    = "pow(x,y)"
 
     return get_elwise_kernel(
         ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype)
-- 
GitLab


From 34af6ea30dd243d0f8353971ef4ca2a6d51b8bea Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Fri, 22 Oct 2021 05:07:15 +0000
Subject: [PATCH 11/28] test for rpow_array

---
 test/test_gpuarray.py | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/test/test_gpuarray.py b/test/test_gpuarray.py
index dbf4b2f7..f9411d17 100644
--- a/test/test_gpuarray.py
+++ b/test/test_gpuarray.py
@@ -39,6 +39,31 @@ class TestGPUArray:
         a_gpu **= 2
         a_gpu = a_gpu.get()
         assert (np.abs(a ** 2 - a_gpu) < 1e-3).all()
+    @mark_cuda_test
+
+    def test_rpow_array(self):
+        scalar = 2
+        a = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]).astype(np.float32)
+        a_gpu = gpuarray.to_gpu(a)
+
+        result = rpow(a_gpu, scalar).get()
+        assert (np.abs(a ** scalar - result) < 1e-3).all()
+
+        a_gpu **= scalar
+        a_gpu = a_gpu.get()
+        assert (np.abs(a ** scalar - a_gpu) < 1e-3).all()
+
+        result = rpow(scalar, a_gpu).get()
+        assert (np.abs(scalar ** a - result) < 1e-3).all()
+
+        a_gpu = a_gpu.get()
+        assert (np.abs(scalar ** a - scalar ** a_gpu) < 1e-3).all()
+
+        result = rpow(a_gpu, a_gpu).get()
+        assert (np.abs(a ** a - result) < 1e-3).all()
+
+        a_gpu = a_gpu.get()
+        assert (np.abs(a ** a - a_gpu ** a_gpu) < 1e-3).all()
 
     @mark_cuda_test
     def test_numpy_integer_shape(self):
-- 
GitLab


From bf005c183e0eac0924ccbda6a48fadc92c7bfaf4 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Fri, 22 Oct 2021 05:13:24 +0000
Subject: [PATCH 12/28] Corrected assert error typo

---
 pycuda/elementwise.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 65a57d5d..cefa0c7d 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -670,7 +670,7 @@ def get_pow_array_kernel(dtype_x, dtype_y, dtype_z):
 def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
     #XOR check that raises an Assertion when both are scalar
-    raise AssertionError if not (is_exp_array or is_base_array)
+    assert if not (is_exp_array or is_base_array)
     if is_base_array and not is_exp_array:
         x_ctype = "%(tp_x)s *x"
         func    = "pow(x[i],y)"
-- 
GitLab


From 9b6e730334c5b00656764f85ddcb1b90234bde31 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Fri, 22 Oct 2021 05:22:39 +0000
Subject: [PATCH 13/28] Removed awkard mark_cuda space

---
 test/test_gpuarray.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/test/test_gpuarray.py b/test/test_gpuarray.py
index f9411d17..b119e66e 100644
--- a/test/test_gpuarray.py
+++ b/test/test_gpuarray.py
@@ -39,8 +39,8 @@ class TestGPUArray:
         a_gpu **= 2
         a_gpu = a_gpu.get()
         assert (np.abs(a ** 2 - a_gpu) < 1e-3).all()
-    @mark_cuda_test
 
+    @mark_cuda_test
     def test_rpow_array(self):
         scalar = 2
         a = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]).astype(np.float32)
-- 
GitLab


From 15536d63e0547f9dae7276229ddbc432df930a98 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Fri, 22 Oct 2021 05:32:33 +0000
Subject: [PATCH 14/28] Changed format for Assertion Error

---
 pycuda/elementwise.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index cefa0c7d..d650597b 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -670,7 +670,9 @@ def get_pow_array_kernel(dtype_x, dtype_y, dtype_z):
 def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
     #XOR check that raises an Assertion when both are scalar
-    assert if not (is_exp_array or is_base_array)
+    if not (is_exp_array or is_base_array):
+        raise AssertionError
+
     if is_base_array and not is_exp_array:
         x_ctype = "%(tp_x)s *x"
         func    = "pow(x[i],y)"
-- 
GitLab


From 7b9a5a33fbc3aa020b49828714bb92d847ba2126 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Fri, 22 Oct 2021 06:30:34 +0000
Subject: [PATCH 15/28] rpow_kernel separate conditions for all cases

---
 pycuda/gpuarray.py | 34 +++++++++++++++++++++++++++++++---
 1 file changed, 31 insertions(+), 3 deletions(-)

diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index 905b496a..5b9e0c34 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -840,14 +840,42 @@ class GPUArray:
         return self._pow(other, new=False)
 
     def __rpow__(self, other):
-        # other must be a scalar
-        assert np.isscalar(other)
         common_dtype = _get_common_dtype(self, other)
         result = self._new_like_me(common_dtype)
+
+        if np.isscalar(other):
+            base = np.array(common_dtype.type(other))
+            result.add_event(
+                    elementwise.get_rpow_kernel(
+                    base.dtype, self.dtype, result.dtype,
+                    is_base_array=False, is_exp_array=True)
+            )
+            return result
+
+        if np.isscalar(self):
+            exponent = np.array(common_dtype.type(self))
+            result.add_event(
+                    elementwise.get_rpow_kernel(
+                    base.dtype, self.dtype, result.dtype,
+                    is_base_array=True, is_exp_array=False)
+            )
+            return result
+
         result.add_event(
-                self._rpow_scalar(result, common_dtype.type(other), self))
+                    elementwise.get_rpow_kernel(
+                    base.dtype, self.dtype, result.dtype,
+                    is_base_array=True, is_exp_array=True)
+            )
         return result
 
+
+    def _rpow_scalar(self, result, base, exponent):
+        base = np.array(base)
+        return elementwise.get_rpow_kernel(
+                base.dtype, exponent.dtype, result.dtype,
+                is_base_array=False, is_exp_array=True)
+
+
     def reverse(self, stream=None):
         """Return this array in reversed order. The array is treated
         as one-dimensional.
-- 
GitLab


From a8b99bf98c7e9bbe6e15657d96d56f40b1068a81 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Fri, 29 Oct 2021 23:38:36 +0000
Subject: [PATCH 16/28] Resolved typo + optimization issues in pycuda/
 elementiwise.py

---
 pycuda/elementwise.py | 19 +++++++++----------
 1 file changed, 9 insertions(+), 10 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index d650597b..bec7d688 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -669,21 +669,20 @@ def get_pow_array_kernel(dtype_x, dtype_y, dtype_z):
 @context_dependent_memoize
 def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
-    #XOR check that raises an Assertion when both are scalar
-    if not (is_exp_array or is_base_array):
-        raise AssertionError
-
-    if is_base_array and not is_exp_array:
-        x_ctype = "%(tp_x)s *x"
-        func    = "pow(x[i],y)"
-
     if not is_base_array and is_exp_array:
         y_ctype = "%(tp_y)s *y"
         func    = "pow(x,y[i])"
 
-    if is_base_array and exp_array:
+    elif is_base_array and exp_array:
         x_ctype = "%(tp_x)s x"
-        func    = "pow(x,y)"
+        func    = "pow(x[i],y[i])"
+
+    elif is_base_array and not is_exp_array:
+        x_ctype = "%(tp_x)s *x"
+        func    = "pow(x[i],y)"
+
+    else:
+        raise AssertionError
 
     return get_elwise_kernel(
         ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype)
-- 
GitLab


From e3efef690e9965060b2f7e3aae79d5029e024478 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Sat, 30 Oct 2021 00:33:51 +0000
Subject: [PATCH 17/28] Removed if conditions in __rpow__

---
 pycuda/gpuarray.py | 30 ++----------------------------
 1 file changed, 2 insertions(+), 28 deletions(-)

diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index 5b9e0c34..ce93fc46 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -842,40 +842,14 @@ class GPUArray:
     def __rpow__(self, other):
         common_dtype = _get_common_dtype(self, other)
         result = self._new_like_me(common_dtype)
-
-        if np.isscalar(other):
-            base = np.array(common_dtype.type(other))
-            result.add_event(
-                    elementwise.get_rpow_kernel(
-                    base.dtype, self.dtype, result.dtype,
-                    is_base_array=False, is_exp_array=True)
-            )
-            return result
-
-        if np.isscalar(self):
-            exponent = np.array(common_dtype.type(self))
-            result.add_event(
-                    elementwise.get_rpow_kernel(
-                    base.dtype, self.dtype, result.dtype,
-                    is_base_array=True, is_exp_array=False)
-            )
-            return result
-
+        base = np.array(common_dtype.type(other))
         result.add_event(
                     elementwise.get_rpow_kernel(
                     base.dtype, self.dtype, result.dtype,
-                    is_base_array=True, is_exp_array=True)
+                    is_base_array= not np.isscalar(self), is_exp_array= not np.isscalar(other))
             )
         return result
 
-
-    def _rpow_scalar(self, result, base, exponent):
-        base = np.array(base)
-        return elementwise.get_rpow_kernel(
-                base.dtype, exponent.dtype, result.dtype,
-                is_base_array=False, is_exp_array=True)
-
-
     def reverse(self, stream=None):
         """Return this array in reversed order. The array is treated
         as one-dimensional.
-- 
GitLab


From d1c545b65f19972e4f7703bbb9641b84932f623d Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Sat, 30 Oct 2021 00:49:12 +0000
Subject: [PATCH 18/28] Fixed test_pow_array to call __rpow__

---
 test/test_gpuarray.py | 18 ++++--------------
 1 file changed, 4 insertions(+), 14 deletions(-)

diff --git a/test/test_gpuarray.py b/test/test_gpuarray.py
index b119e66e..12cc0743 100644
--- a/test/test_gpuarray.py
+++ b/test/test_gpuarray.py
@@ -46,24 +46,14 @@ class TestGPUArray:
         a = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]).astype(np.float32)
         a_gpu = gpuarray.to_gpu(a)
 
-        result = rpow(a_gpu, scalar).get()
-        assert (np.abs(a ** scalar - result) < 1e-3).all()
-
-        a_gpu **= scalar
-        a_gpu = a_gpu.get()
-        assert (np.abs(a ** scalar - a_gpu) < 1e-3).all()
-
-        result = rpow(scalar, a_gpu).get()
+        result = (scalar ** a_gpu).get()
         assert (np.abs(scalar ** a - result) < 1e-3).all()
 
-        a_gpu = a_gpu.get()
-        assert (np.abs(scalar ** a - scalar ** a_gpu) < 1e-3).all()
-
-        result = rpow(a_gpu, a_gpu).get()
+        result = (a_gpu ** a_gpu).get()
         assert (np.abs(a ** a - result) < 1e-3).all()
 
-        a_gpu = a_gpu.get()
-        assert (np.abs(a ** a - a_gpu ** a_gpu) < 1e-3).all()
+        result = (a_gpu ** scalar).get()
+        assert (np.abs(a ** scalar - result) < 1e-3).all()
 
     @mark_cuda_test
     def test_numpy_integer_shape(self):
-- 
GitLab


From 6ee4f8197aaefa152cb2f380c2132d4a64deaed8 Mon Sep 17 00:00:00 2001
From: mit kotak <mitak2@illinois.edu>
Date: Fri, 29 Oct 2021 20:19:46 -0500
Subject: [PATCH 19/28] Fixed Flake8 issues

---
 pycuda/elementwise.py | 13 ++++++-------
 pycuda/gpuarray.py    |  3 ++-
 2 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index b2e50ab6..caad3560 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -671,20 +671,19 @@ def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
 
     if is_base_array and not is_exp_array:
         x_ctype = "%(tp_x)s *x"
-        func    = "pow(x[i],y)"
+        func = "pow(x[i],y)"
     elif not is_base_array and is_exp_array:
         x_ctype = "%(tp_x)s x"
-        func    = "pow(x,y[i])"
-
-    elif is_exp_array and is_exp_array:
+        func = "pow(x,y[i])"
+    elif is_base_array and is_exp_array:
         y_ctype = "%(tp_y)s *y"
-        func    = "pow(x,y)"
+        func = "pow(x,y)"
     else:
-        raise Assertion
+        raise AssertionError
 
     return get_elwise_kernel(
         ("%(tp_z)s *z, " + x_ctype + ", "+y_ctype)
-        %{
+        % {
             "tp_x": dtype_to_ctype(dtype_x),
             "tp_y": dtype_to_ctype(dtype_y),
             "tp_z": dtype_to_ctype(dtype_z),
diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index 905b496a..a3ff90b4 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -845,7 +845,8 @@ class GPUArray:
         common_dtype = _get_common_dtype(self, other)
         result = self._new_like_me(common_dtype)
         result.add_event(
-                self._rpow_scalar(result, common_dtype.type(other), self))
+                self._rpow_scalar(result, common_dtype.type(other), self)
+                )
         return result
 
     def reverse(self, stream=None):
-- 
GitLab


From 8968000caf313256298992eda413e10680704833 Mon Sep 17 00:00:00 2001
From: Kaushik Kulkarni <kaushikcfd@gmail.com>
Date: Fri, 29 Oct 2021 21:39:56 -0500
Subject: [PATCH 20/28] collective hacking during meeting

---
 pycuda/elementwise.py |  6 ++++++
 pycuda/gpuarray.py    | 15 +++++++++------
 2 files changed, 15 insertions(+), 6 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 764182c0..2357b132 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -668,17 +668,23 @@ def get_pow_array_kernel(dtype_x, dtype_y, dtype_z):
 
 @context_dependent_memoize
 def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
+    """
+    Returns the kernel for the operation: ``z = x ** y``
+    """
 
     if not is_base_array and is_exp_array:
+        x_ctype = "%(tp_x)s x"
         y_ctype = "%(tp_y)s *y"
         func = "pow(x,y[i])"
 
     elif is_base_array and is_exp_array:
         x_ctype = "%(tp_x)s x"
+        y_ctype = "%(tp_y)s *y"
         func = "pow(x[i],y[i])"
 
     elif is_base_array and not is_exp_array:
         x_ctype = "%(tp_x)s *x"
+        y_ctype = "%(tp_y)s y"
         func = "pow(x[i],y)"
 
     else:
diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index c17eadb7..94fdc4f3 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -842,12 +842,15 @@ class GPUArray:
     def __rpow__(self, other):
         common_dtype = _get_common_dtype(self, other)
         result = self._new_like_me(common_dtype)
-        base = np.array(common_dtype.type(other))
-        result.add_event(
-                elementwise.get_rpow_kernel(
-                    base.dtype, self.dtype, result.dtype,
-                    is_base_array=not np.isscalar(self), is_exp_array=not np.isscalar(other))
-                )
+        # FIXME:  We are assuming that 'base' is always scalar. Get rid of that
+        # assumptions.
+        base = common_dtype.type(other)
+        func = elementwise.get_rpow_kernel(
+            base.dtype, self.dtype, result.dtype,
+            is_base_array=not np.isscalar(other), is_exp_array=True)
+        func.prepared_async_call(self._grid, self._block, None,
+                                 result.gpudata, base, self.gpudata,
+                                 self.mem_size)
         return result
 
     def reverse(self, stream=None):
-- 
GitLab


From 9f4b97ba889fbb25d0b113b766401e0398dfe5e0 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Sat, 6 Nov 2021 12:30:25 +0000
Subject: [PATCH 21/28] Removed basearray is scalar assumption

---
 pycuda/gpuarray.py | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index 94fdc4f3..dcc63c7b 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -842,12 +842,13 @@ class GPUArray:
     def __rpow__(self, other):
         common_dtype = _get_common_dtype(self, other)
         result = self._new_like_me(common_dtype)
-        # FIXME:  We are assuming that 'base' is always scalar. Get rid of that
-        # assumptions.
-        base = common_dtype.type(other)
+        if np.isscalar(other):
+            base = common_dtype.type(other)
+        else:
+            base = common_dtype.type(self)
         func = elementwise.get_rpow_kernel(
             base.dtype, self.dtype, result.dtype,
-            is_base_array=not np.isscalar(other), is_exp_array=True)
+            is_base_array=not np.isscalar(other), is_exp_array=not np.isscalar(self))
         func.prepared_async_call(self._grid, self._block, None,
                                  result.gpudata, base, self.gpudata,
                                  self.mem_size)
-- 
GitLab


From cbb8007596533c7336ed23f01be44278514b1460 Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Sat, 6 Nov 2021 12:41:20 +0000
Subject: [PATCH 22/28] Tried fixing a ** a issue

---
 pycuda/gpuarray.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index dcc63c7b..dbcea0bc 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -842,10 +842,10 @@ class GPUArray:
     def __rpow__(self, other):
         common_dtype = _get_common_dtype(self, other)
         result = self._new_like_me(common_dtype)
-        if np.isscalar(other):
-            base = common_dtype.type(other)
-        else:
+        if not np.isscalar(other):
             base = common_dtype.type(self)
+        else:
+            base = common_dtype.type(other)
         func = elementwise.get_rpow_kernel(
             base.dtype, self.dtype, result.dtype,
             is_base_array=not np.isscalar(other), is_exp_array=not np.isscalar(self))
-- 
GitLab


From 5ccd0dd44e891cfcd9d205d7c09ab1332d51d026 Mon Sep 17 00:00:00 2001
From: Kaushik Kulkarni <kaushikcfd@gmail.com>
Date: Sat, 6 Nov 2021 09:36:06 -0500
Subject: [PATCH 23/28] some more collective hacking during the meeting

---
 pycuda/elementwise.py |  3 ++-
 pycuda/gpuarray.py    | 10 ++++++++--
 2 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 2357b132..18672826 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -671,6 +671,7 @@ def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
     """
     Returns the kernel for the operation: ``z = x ** y``
     """
+    # FIXME: This needs to be more precise in its invocation of pow/powf.
 
     if not is_base_array and is_exp_array:
         x_ctype = "%(tp_x)s x"
@@ -678,7 +679,7 @@ def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
         func = "pow(x,y[i])"
 
     elif is_base_array and is_exp_array:
-        x_ctype = "%(tp_x)s x"
+        x_ctype = "%(tp_x)s *x"
         y_ctype = "%(tp_y)s *y"
         func = "pow(x[i],y[i])"
 
diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index dbcea0bc..5e7a5607 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -842,15 +842,21 @@ class GPUArray:
     def __rpow__(self, other):
         common_dtype = _get_common_dtype(self, other)
         result = self._new_like_me(common_dtype)
+
         if not np.isscalar(other):
-            base = common_dtype.type(self)
+            # Base is a gpuarray => do not cast.
+            base = other
         else:
             base = common_dtype.type(other)
+
         func = elementwise.get_rpow_kernel(
             base.dtype, self.dtype, result.dtype,
             is_base_array=not np.isscalar(other), is_exp_array=not np.isscalar(self))
+        # Evaluates z = x ** y
         func.prepared_async_call(self._grid, self._block, None,
-                                 result.gpudata, base, self.gpudata,
+                                 result.gpudata,  # z
+                                 base if np.isscalar(base) else base.gpudata,  # x
+                                 self.gpudata,  # y
                                  self.mem_size)
         return result
 
-- 
GitLab


From dae3cb1f56a237cb24bdfa750ca5bec7e21a5922 Mon Sep 17 00:00:00 2001
From: mit kotak <mitak2@illinois.edu>
Date: Fri, 12 Nov 2021 18:42:49 -0600
Subject: [PATCH 24/28] Increased precision using pow/powf

---
 pycuda/elementwise.py | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/pycuda/elementwise.py b/pycuda/elementwise.py
index 18672826..607e9f9c 100644
--- a/pycuda/elementwise.py
+++ b/pycuda/elementwise.py
@@ -671,22 +671,25 @@ def get_rpow_kernel(dtype_x, dtype_y, dtype_z, is_base_array, is_exp_array):
     """
     Returns the kernel for the operation: ``z = x ** y``
     """
-    # FIXME: This needs to be more precise in its invocation of pow/powf.
+    if np.float64 in [dtype_x, dtype_y]:
+        func = "pow"
+    else:
+        func = "powf"
 
     if not is_base_array and is_exp_array:
         x_ctype = "%(tp_x)s x"
         y_ctype = "%(tp_y)s *y"
-        func = "pow(x,y[i])"
+        func = "%s(x,y[i])" % func
 
     elif is_base_array and is_exp_array:
         x_ctype = "%(tp_x)s *x"
         y_ctype = "%(tp_y)s *y"
-        func = "pow(x[i],y[i])"
+        func = "%s(x[i],y[i])" % func
 
     elif is_base_array and not is_exp_array:
         x_ctype = "%(tp_x)s *x"
         y_ctype = "%(tp_y)s y"
-        func = "pow(x[i],y)"
+        func = "%s(x[i],y)" % func
 
     else:
         raise AssertionError
-- 
GitLab


From 932b4bc0cb1d0e20b9832bfead1a1a060d912f95 Mon Sep 17 00:00:00 2001
From: mit kotak <mitak2@illinois.edu>
Date: Fri, 12 Nov 2021 19:43:25 -0600
Subject: [PATCH 25/28] Modified test_rpow float32 -> float64

---
 test/test_gpuarray.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/test/test_gpuarray.py b/test/test_gpuarray.py
index 12cc0743..12ed9f83 100644
--- a/test/test_gpuarray.py
+++ b/test/test_gpuarray.py
@@ -43,7 +43,7 @@ class TestGPUArray:
     @mark_cuda_test
     def test_rpow_array(self):
         scalar = 2
-        a = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]).astype(np.float32)
+        a = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]).astype(np.float64)
         a_gpu = gpuarray.to_gpu(a)
 
         result = (scalar ** a_gpu).get()
-- 
GitLab


From c52bddfaa0957cb263d3bfe3866ab4b760d4a131 Mon Sep 17 00:00:00 2001
From: mit kotak <mitak2@illinois.edu>
Date: Mon, 22 Nov 2021 19:48:31 -0600
Subject: [PATCH 26/28] Added assert np.testing.assert_allclose()

---
 test/test_gpuarray.py | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/test/test_gpuarray.py b/test/test_gpuarray.py
index 12ed9f83..7ecd01f6 100644
--- a/test/test_gpuarray.py
+++ b/test/test_gpuarray.py
@@ -42,18 +42,18 @@ class TestGPUArray:
 
     @mark_cuda_test
     def test_rpow_array(self):
-        scalar = 2
-        a = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]).astype(np.float64)
+        scalar = np.random.rand()
+        a = abs(np.random.rand(10).astype(np.float64))
         a_gpu = gpuarray.to_gpu(a)
 
         result = (scalar ** a_gpu).get()
-        assert (np.abs(scalar ** a - result) < 1e-3).all()
+        assert np.testing.assert_allclose(scalar ** a, result)
 
         result = (a_gpu ** a_gpu).get()
-        assert (np.abs(a ** a - result) < 1e-3).all()
+        assert np.testing.assert_allclose(a ** a, result)
 
         result = (a_gpu ** scalar).get()
-        assert (np.abs(a ** scalar - result) < 1e-3).all()
+        assert np.testing.assert_allclose(a ** scalar, result)
 
     @mark_cuda_test
     def test_numpy_integer_shape(self):
-- 
GitLab


From 978c30e8accf5f6f60eff17d401399fce5f52a2e Mon Sep 17 00:00:00 2001
From: Mit Kotak <mitak2@illinois.edu>
Date: Tue, 23 Nov 2021 02:04:00 +0000
Subject: [PATCH 27/28] Removed AssertionError: assert None

---
 test/test_gpuarray.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/test/test_gpuarray.py b/test/test_gpuarray.py
index 7ecd01f6..51e118dc 100644
--- a/test/test_gpuarray.py
+++ b/test/test_gpuarray.py
@@ -47,13 +47,13 @@ class TestGPUArray:
         a_gpu = gpuarray.to_gpu(a)
 
         result = (scalar ** a_gpu).get()
-        assert np.testing.assert_allclose(scalar ** a, result)
+        np.testing.assert_allclose(scalar ** a, result)
 
         result = (a_gpu ** a_gpu).get()
-        assert np.testing.assert_allclose(a ** a, result)
+        np.testing.assert_allclose(a ** a, result)
 
         result = (a_gpu ** scalar).get()
-        assert np.testing.assert_allclose(a ** scalar, result)
+        np.testing.assert_allclose(a ** scalar, result)
 
     @mark_cuda_test
     def test_numpy_integer_shape(self):
-- 
GitLab


From 6f6bd9b606f540200047356cd6cccedd77fd72e2 Mon Sep 17 00:00:00 2001
From: Kaushik Kulkarni <kgk2@illinois.edu>
Date: Tue, 23 Nov 2021 03:11:32 +0000
Subject: [PATCH 28/28] removes unnecessary type cast

---
 test/test_gpuarray.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/test/test_gpuarray.py b/test/test_gpuarray.py
index 51e118dc..5234d8a2 100644
--- a/test/test_gpuarray.py
+++ b/test/test_gpuarray.py
@@ -43,7 +43,7 @@ class TestGPUArray:
     @mark_cuda_test
     def test_rpow_array(self):
         scalar = np.random.rand()
-        a = abs(np.random.rand(10).astype(np.float64))
+        a = abs(np.random.rand(10))
         a_gpu = gpuarray.to_gpu(a)
 
         result = (scalar ** a_gpu).get()
-- 
GitLab