addressed @notoraptor comments

theano/gpuarray/linalg.py

@@ -8,9 +8,6 @@ from theano import Op
 
 from theano.gpuarray import basic_ops, GpuArrayType
 
-from pygpu.gpuarray import GpuKernel, GpuArray
-from string import Template
-
 import numpy as np
 from numpy.linalg.linalg import LinAlgError
 

theano/gpuarray/opt.py

@@ -1975,7 +1975,7 @@ def local_gpu_solve(op, context_name, inputs, outputs):
 def local_gpu_cholesky(op, context_name, inputs, outputs):
     if not cusolver_available:
         return
-    return GpuCholesky()
+    return GpuCholesky(lower=op.lower, inplace=op.destructive)
 
 # Do not register in fast_run or fast_compile.
 # It will be added to fast_run if the GPU is enabled.

theano/gpuarray/tests/test_linalg.py

@@ -191,4 +191,3 @@ class TestGpuCholesky(unittest.TestCase):
         A_val = -M_val.dot(M_val.T)
         fn = self.get_gpu_cholesky_func(True, False)
         self.assertRaises(LinAlgError, fn, A_val)
-

theano/gpuarray/tests/test_opt.py

@@ -573,7 +573,7 @@ def test_local_lift_solve():
     A = tensor.fmatrix()
     b = tensor.fmatrix()
     o = slinalg.solve(A, b)
-    f_cpu = theano.function([A, b], o)
+    f_cpu = theano.function([A, b], o, mode_without_gpu)
     f_gpu = theano.function([A, b], o, mode=mode_with_gpu)
     assert not any(isinstance(n.op, slinalg.Solve)
                    for n in f_gpu.maker.fgraph.apply_nodes)
@@ -589,7 +589,7 @@ def test_local_lift_cholesky():
         raise SkipTest('No cuSolver')
     A = tensor.fmatrix()
     o = slinalg.cholesky(A)
-    f_cpu = theano.function([A], o)
+    f_cpu = theano.function([A], o, mode=mode_without_gpu)
     f_gpu = theano.function([A], o, mode=mode_with_gpu)
     assert not any(isinstance(n.op, slinalg.Cholesky)
                    for n in f_gpu.maker.fgraph.apply_nodes)