This fixes a bug in cusolver solve op when m > k

theano/gpuarray/linalg.py

@@ -7,7 +7,7 @@ from theano import Op
 from theano.gpuarray import basic_ops, GpuArrayType
 
 try:
-    from pygpu import gpuarray
+    import pygpu
 except ImportError:
     pass
 
@@ -109,12 +109,12 @@ class GpuCusolverSolve(Op):
                 raise ValueError('A and b must be aligned.')
 
             lda = max(1, n)
-            ldb = max(1, k, m)
+            ldb = max(1, k)
 
             # We copy A and b as cusolver operates inplace
-            b = gpuarray.array(b, copy=True, order='F')
+            b = pygpu.array(b, copy=True, order='F')
             if not self.inplace:
-                A = gpuarray.array(A, copy=True)
+                A = pygpu.array(A, copy=True)
             A_ptr = A.gpudata
             b_ptr = b.gpudata
 
@@ -129,18 +129,18 @@ class GpuCusolverSolve(Op):
 
             if (thunk.workspace is None or
                     thunk.workspace.size != workspace_size):
-                thunk.workspace = gpuarray.zeros((workspace_size,),
+                thunk.workspace = pygpu.zeros(workspace_size,
                                               dtype='float32',
                                               context=context)
 
             if thunk.pivots is None or thunk.pivots.size != min(n, n):
-                thunk.pivots = gpuarray.zeros((min(n, n),),
-                                              dtype='float32',
+                thunk.pivots = pygpu.zeros(n,
+                                           dtype='int32',
                                            context=context)
 
             if thunk.dev_info is None:
-                thunk.dev_info = gpuarray.zeros((1,),
-                                                dtype='float32',
+                thunk.dev_info = pygpu.zeros((1,),
+                                             dtype='int32',
                                              context=context)
 
             workspace_ptr = thunk.workspace.gpudata