def test_cusparseScsrmm2_notranspose():
A = np.random.laplace(size=(5, 3)).astype(np.float32)
A[A<0.1] = 0
A = sparse.csr_matrix(A, dtype=np.float32)
B = np.random.normal(size=(3, 6)).astype(np.float32, order="f")
C = np.ones((A.shape[0], B.shape[1]), dtype=np.float32)
X_exp = (A*B) + 0.5*C
a_data = gpu.to_gpu(A.data)
a_indptr = gpu.to_gpu(A.indptr)
a_indices = gpu.to_gpu(A.indices)
b = gpu.to_gpu(B)
h = cusparse.cusparseCreate()
descrA = cusparse.cusparseCreateMatDescr()
c = gpu.empty((C.shape[1], C.shape[0]), dtype=A.dtype)
c.fill(1.0)
cusparse.cusparseScsrmm2(h, cusparse.CUSPARSE_OPERATION_NON_TRANSPOSE,
cusparse.CUSPARSE_OPERATION_NON_TRANSPOSE,
c.shape[1], c.shape[0], b.shape[0], A.nnz, 1.0,
descrA, a_data.gpudata, a_indptr.gpudata, a_indices.gpudata,
b.gpudata, b.shape[0], 0.5, c.gpudata, c.shape[1])
assert_allclose(c.get().T, X_exp, rtol=1e-4)
def test_cusparseScsr2dense():
A = np.random.laplace(size=(3, 5)).astype(np.float32)
A[A<0.1] = 0
A = sparse.csr_matrix(A, dtype=np.float32)
A.sort_indices()
a_data = gpu.to_gpu(A.data)
a_indptr = gpu.to_gpu(A.indptr)
a_indices = gpu.to_gpu(A.indices)
out = gpu.empty((A.shape[0], A.shape[1]), dtype=A.dtype, order="F")
h = cusparse.cusparseCreate()
descrA = cusparse.cusparseCreateMatDescr()
cusparse.cusparseScsr2dense(h, A.shape[0], A.shape[1],
descrA, a_data.gpudata, a_indptr.gpudata, a_indices.gpudata,
out.gpudata, out.shape[0])
assert_allclose(out.get(), A.A, rtol=1e-4)
def __init__(self, array, dtype=None, allocator=mem_alloc, stream=None):
self.dtype = array.dtype if dtype is None else dtype
self.nnz = array.nnz
self.shape = array.shape
if self.nnz == 0: # let's not waste time
return
if not sparse.isspmatrix_csr(array):
array = sparse.csr_matrix(array, dtype=self.dtype)
if not array.has_sorted_indices:
array = array.sorted_indices()
if stream is not None:
self.data = gpuarray.to_gpu_async(array.data.astype(dtype=self.dtype), allocator=allocator, stream=stream)
self.indptr = gpuarray.to_gpu_async(array.indptr, allocator=allocator, stream=stream)
self.indices = gpuarray.to_gpu_async(array.indices, allocator=allocator, stream=stream)
else:
self.data = gpuarray.to_gpu(array.data.astype(dtype=self.dtype), allocator=allocator)
self.indptr = gpuarray.to_gpu(array.indptr, allocator=allocator)
self.indices = gpuarray.to_gpu(array.indices, allocator=allocator)
self.descr = cusparse.cusparseCreateMatDescr()
def __init__(self, array, dtype=None, allocator=mem_alloc, stream=None):
self.dtype = array.dtype if dtype is None else dtype
self.nnz = array.nnz
self.shape = array.shape
if self.nnz == 0: # let's not waste time
return
if not sparse.isspmatrix_csr(array):
array = sparse.csr_matrix(array, dtype=self.dtype)
if not array.has_sorted_indices:
array = array.sorted_indices()
if stream is not None:
self.data = gpuarray.to_gpu_async(array.data.astype(dtype=self.dtype), allocator=allocator, stream=stream)
self.indptr = gpuarray.to_gpu_async(array.indptr, allocator=allocator, stream=stream)
self.indices = gpuarray.to_gpu_async(array.indices, allocator=allocator, stream=stream)
else:
self.data = gpuarray.to_gpu(array.data.astype(dtype=self.dtype), allocator=allocator)
self.indptr = gpuarray.to_gpu(array.indptr, allocator=allocator)
self.indices = gpuarray.to_gpu(array.indices, allocator=allocator)
self.descr = cusparse.cusparseCreateMatDescr()
