Correcting alignment

theano/sandbox/cuda/basic_ops.py

@@ -2516,11 +2516,11 @@ class GpuAdvancedIncSubtensor1(tensor.AdvancedIncSubtensor1, GpuOp):
      		unsigned int num_threads_per_block = std::min(numcolsX, (unsigned int)NUM_VECTOR_OP_THREADS_PER_BLOCK);
      		unsigned int num_blocks = std::min(size ,(unsigned int)NUM_VECTOR_OP_BLOCKS);
 
-    		dim3 n_blocks(num_blocks);
+     		dim3 n_blocks(num_blocks);
      		dim3 n_threads(num_threads_per_block);
      		long *d_indices_arr = NULL;
 
-		PyArrayObject *cpu_indices_arr = PyArray_GETCONTIGUOUS(indices_arr);
+    		PyArrayObject *cpu_indices_arr = PyArray_GETCONTIGUOUS(indices_arr);
 
     		d_indices_arr = (long *)device_malloc(PyArray_NBYTES(cpu_indices_arr));
      		assert(d_indices_arr);
@@ -2530,7 +2530,7 @@ class GpuAdvancedIncSubtensor1(tensor.AdvancedIncSubtensor1, GpuOp):
                                              PyArray_NBYTES(cpu_indices_arr),
                                              cudaMemcpyHostToDevice);
 
-		assert(err == cudaSuccess);
+    		assert(err == cudaSuccess);
 
      		k_vector_add_fast<<<n_blocks, n_threads>>>(shapeX[0],
                                                            shapeX[1],

theano/sandbox/cuda/cuda_ndarray.cu

@@ -1776,7 +1776,6 @@ CudaNdarray_inplace_elemwise(PyObject* py_self, PyObject * py_other, operator_t
     return 0;
 }
 
-
 /*
  * We need this inplace Add to support IncSubTensor
  * It returns py_self on success with an additional reference. Else NULL.

theano/sandbox/cuda/cuda_ndarray.cuh

@@ -490,7 +490,6 @@ DllExport PyObject * CudaNdarray_View(const CudaNdarray * self);
 DllExport PyObject * CudaNdarray_inplace_add(PyObject* py_self, PyObject * py_other);
 DllExport PyObject * CudaNdarray_Subscript(PyObject * py_self, PyObject * key);
 DllExport int CudaNdarray_inplace_elemwise(PyObject* py_self, PyObject * py_other, operator_t fct_nb);
-DllExport void CudaNdarray_vector_add_fast(CudaNdarray* py_self, CudaNdarray *py_other, PyArrayObject *indices_arr);
 
 // Ensures that *arr is a pointer to a contiguous ndarray of the specified
 // dimensions.

theano/sandbox/cuda/opt.py

@@ -1353,6 +1353,7 @@ def local_gpualloc_memset_0(node):
             new_out = GpuAlloc(memset_0=True)(*node.inputs)
             return [new_out]
 
+
 def safe_to_gpu(x):
     if (isinstance(x.type, tensor.TensorType) and
         x.type.dtype == 'float32'):