Skip to content

[cleanup/tensor, part 2] Modularized tests for contiguity, retrieval of PyUSMArrayObject* and removed use of a global variable. #931

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Oct 13, 2022
Merged

[cleanup/tensor, part 2] Modularized tests for contiguity, retrieval of PyUSMArrayObject* and removed use of a global variable. #931

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
a496d53
Introduced dpctl::tensor::is_c_contiguous and is_f_contiguous and use…
oleksandr-pavlyk Sep 26, 2022
046d37b
Modularized retrieval of usm_ndarray_ptr
oleksandr-pavlyk Oct 1, 2022
43a6321
Got rid of `static dpctl::tensor::detail::usm_ndarray_types array_typ…
oleksandr-pavlyk Sep 28, 2022
7d2ab88
Changed dpctl/.gitignore to only ignore generated _*.cpp files, not a…
oleksandr-pavlyk Oct 8, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion dpctl/.gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
*.so
*.cpp
_*.cpp
*.cxx
*.c
*.h
Expand Down
48 changes: 28 additions & 20 deletions dpctl/apis/include/dpctl4pybind11.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -439,8 +439,7 @@ class usm_ndarray : public py::object

char *get_data() const
{
PyObject *raw_o = this->ptr();
PyUSMArrayObject *raw_ar = reinterpret_cast<PyUSMArrayObject *>(raw_o);
PyUSMArrayObject *raw_ar = this->usm_array_ptr();

return UsmNDArray_GetData(raw_ar);
}
Expand All @@ -452,16 +451,14 @@ class usm_ndarray : public py::object

int get_ndim() const
{
PyObject *raw_o = this->ptr();
PyUSMArrayObject *raw_ar = reinterpret_cast<PyUSMArrayObject *>(raw_o);
PyUSMArrayObject *raw_ar = this->usm_array_ptr();

return UsmNDArray_GetNDim(raw_ar);
}

const py::ssize_t *get_shape_raw() const
{
PyObject *raw_o = this->ptr();
PyUSMArrayObject *raw_ar = reinterpret_cast<PyUSMArrayObject *>(raw_o);
PyUSMArrayObject *raw_ar = this->usm_array_ptr();

return UsmNDArray_GetShape(raw_ar);
}
Expand All @@ -474,16 +471,14 @@ class usm_ndarray : public py::object

const py::ssize_t *get_strides_raw() const
{
PyObject *raw_o = this->ptr();
PyUSMArrayObject *raw_ar = reinterpret_cast<PyUSMArrayObject *>(raw_o);
PyUSMArrayObject *raw_ar = this->usm_array_ptr();

return UsmNDArray_GetStrides(raw_ar);
}

py::ssize_t get_size() const
{
PyObject *raw_o = this->ptr();
PyUSMArrayObject *raw_ar = reinterpret_cast<PyUSMArrayObject *>(raw_o);
PyUSMArrayObject *raw_ar = this->usm_array_ptr();

int ndim = UsmNDArray_GetNDim(raw_ar);
const py::ssize_t *shape = UsmNDArray_GetShape(raw_ar);
Expand All @@ -499,8 +494,7 @@ class usm_ndarray : public py::object

std::pair<py::ssize_t, py::ssize_t> get_minmax_offsets() const
{
PyObject *raw_o = this->ptr();
PyUSMArrayObject *raw_ar = reinterpret_cast<PyUSMArrayObject *>(raw_o);
PyUSMArrayObject *raw_ar = this->usm_array_ptr();

int nd = UsmNDArray_GetNDim(raw_ar);
const py::ssize_t *shape = UsmNDArray_GetShape(raw_ar);
Expand Down Expand Up @@ -533,36 +527,50 @@ class usm_ndarray : public py::object

sycl::queue get_queue() const
{
PyObject *raw_o = this->ptr();
PyUSMArrayObject *raw_ar = reinterpret_cast<PyUSMArrayObject *>(raw_o);
PyUSMArrayObject *raw_ar = this->usm_array_ptr();

DPCTLSyclQueueRef QRef = UsmNDArray_GetQueueRef(raw_ar);
return *(reinterpret_cast<sycl::queue *>(QRef));
}

int get_typenum() const
{
PyObject *raw_o = this->ptr();
PyUSMArrayObject *raw_ar = reinterpret_cast<PyUSMArrayObject *>(raw_o);
PyUSMArrayObject *raw_ar = this->usm_array_ptr();

return UsmNDArray_GetTypenum(raw_ar);
}

int get_flags() const
{
PyObject *raw_o = this->ptr();
PyUSMArrayObject *raw_ar = reinterpret_cast<PyUSMArrayObject *>(raw_o);
PyUSMArrayObject *raw_ar = this->usm_array_ptr();

return UsmNDArray_GetFlags(raw_ar);
}

int get_elemsize() const
{
PyObject *raw_o = this->ptr();
PyUSMArrayObject *raw_ar = reinterpret_cast<PyUSMArrayObject *>(raw_o);
PyUSMArrayObject *raw_ar = this->usm_array_ptr();

return UsmNDArray_GetElementSize(raw_ar);
}

bool is_c_contiguous() const
{
int flags = this->get_flags();
return static_cast<bool>(flags & USM_ARRAY_C_CONTIGUOUS);
}

bool is_f_contiguous() const
{
int flags = this->get_flags();
return static_cast<bool>(flags & USM_ARRAY_F_CONTIGUOUS);
}

private:
PyUSMArrayObject *usm_array_ptr() const
{
return reinterpret_cast<PyUSMArrayObject *>(m_ptr);
}
};

} // end namespace tensor
Expand Down
92 changes: 40 additions & 52 deletions dpctl/tensor/libtensor/source/tensor_py.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,8 +42,6 @@

namespace py = pybind11;

static dpctl::tensor::detail::usm_ndarray_types array_types;

namespace
{

Expand Down Expand Up @@ -301,6 +299,7 @@ copy_usm_ndarray_into_usm_ndarray(dpctl::tensor::usm_ndarray src,
int src_typenum = src.get_typenum();
int dst_typenum = dst.get_typenum();

auto array_types = dpctl::tensor::detail::usm_ndarray_types::get();
int src_type_id = array_types.typenum_to_lookup_id(src_typenum);
int dst_type_id = array_types.typenum_to_lookup_id(dst_typenum);

Expand All @@ -322,15 +321,16 @@ copy_usm_ndarray_into_usm_ndarray(dpctl::tensor::usm_ndarray src,
throw py::value_error("Arrays index overlapping segments of memory");
}

int src_flags = src.get_flags();
int dst_flags = dst.get_flags();
bool is_src_c_contig = src.is_c_contiguous();
bool is_src_f_contig = src.is_f_contiguous();

bool is_dst_c_contig = dst.is_c_contiguous();
bool is_dst_f_contig = dst.is_f_contiguous();

// check for applicability of special cases:
// (same type && (both C-contiguous || both F-contiguous)
bool both_c_contig = ((src_flags & USM_ARRAY_C_CONTIGUOUS) &&
(dst_flags & USM_ARRAY_C_CONTIGUOUS));
bool both_f_contig = ((src_flags & USM_ARRAY_F_CONTIGUOUS) &&
(dst_flags & USM_ARRAY_F_CONTIGUOUS));
bool both_c_contig = (is_src_c_contig && is_dst_c_contig);
bool both_f_contig = (is_src_f_contig && is_dst_f_contig);
if (both_c_contig || both_f_contig) {
if (src_type_id == dst_type_id) {

Expand Down Expand Up @@ -360,12 +360,6 @@ copy_usm_ndarray_into_usm_ndarray(dpctl::tensor::usm_ndarray src,
int nd = src_nd;
const py::ssize_t *shape = src_shape;

bool is_src_c_contig = ((src_flags & USM_ARRAY_C_CONTIGUOUS) != 0);
bool is_src_f_contig = ((src_flags & USM_ARRAY_F_CONTIGUOUS) != 0);

bool is_dst_c_contig = ((dst_flags & USM_ARRAY_C_CONTIGUOUS) != 0);
bool is_dst_f_contig = ((dst_flags & USM_ARRAY_F_CONTIGUOUS) != 0);

constexpr py::ssize_t src_itemsize = 1; // in elements
constexpr py::ssize_t dst_itemsize = 1; // in elements

Expand Down Expand Up @@ -550,6 +544,7 @@ copy_usm_ndarray_for_reshape(dpctl::tensor::usm_ndarray src,
const py::ssize_t *src_shape = src.get_shape_raw();
const py::ssize_t *dst_shape = dst.get_shape_raw();

auto array_types = dpctl::tensor::detail::usm_ndarray_types::get();
int type_id = array_types.typenum_to_lookup_id(src_typenum);

auto fn = copy_for_reshape_generic_dispatch_vector[type_id];
Expand All @@ -576,14 +571,13 @@ copy_usm_ndarray_for_reshape(dpctl::tensor::usm_ndarray src,

const py::ssize_t *src_strides = src.get_strides_raw();
if (src_strides == nullptr) {
int src_flags = src.get_flags();
if (src_flags & USM_ARRAY_C_CONTIGUOUS) {
if (src.is_c_contiguous()) {
const auto &src_contig_strides =
c_contiguous_strides(src_nd, src_shape);
std::copy(src_contig_strides.begin(), src_contig_strides.end(),
packed_host_shapes_strides_shp->begin() + src_nd);
}
else if (src_flags & USM_ARRAY_F_CONTIGUOUS) {
else if (src.is_f_contiguous()) {
const auto &src_contig_strides =
f_contiguous_strides(src_nd, src_shape);
std::copy(src_contig_strides.begin(), src_contig_strides.end(),
Expand All @@ -602,15 +596,14 @@ copy_usm_ndarray_for_reshape(dpctl::tensor::usm_ndarray src,

const py::ssize_t *dst_strides = dst.get_strides_raw();
if (dst_strides == nullptr) {
int dst_flags = dst.get_flags();
if (dst_flags & USM_ARRAY_C_CONTIGUOUS) {
if (dst.is_c_contiguous()) {
const auto &dst_contig_strides =
c_contiguous_strides(dst_nd, dst_shape);
std::copy(dst_contig_strides.begin(), dst_contig_strides.end(),
packed_host_shapes_strides_shp->begin() + 2 * src_nd +
dst_nd);
}
else if (dst_flags & USM_ARRAY_F_CONTIGUOUS) {
else if (dst.is_f_contiguous()) {
const auto &dst_contig_strides =
f_contiguous_strides(dst_nd, dst_shape);
std::copy(dst_contig_strides.begin(), dst_contig_strides.end(),
Expand Down Expand Up @@ -736,6 +729,7 @@ void copy_numpy_ndarray_into_usm_ndarray(
py::detail::array_descriptor_proxy(npy_src.dtype().ptr())->type_num;
int dst_typenum = dst.get_typenum();

auto array_types = dpctl::tensor::detail::usm_ndarray_types::get();
int src_type_id = array_types.typenum_to_lookup_id(src_typenum);
int dst_type_id = array_types.typenum_to_lookup_id(dst_typenum);

Expand All @@ -744,14 +738,13 @@ void copy_numpy_ndarray_into_usm_ndarray(
char *dst_data = dst.get_data();

int src_flags = npy_src.flags();
int dst_flags = dst.get_flags();

// check for applicability of special cases:
// (same type && (both C-contiguous || both F-contiguous)
bool both_c_contig = ((src_flags & py::array::c_style) &&
(dst_flags & USM_ARRAY_C_CONTIGUOUS));
bool both_f_contig = ((src_flags & py::array::f_style) &&
(dst_flags & USM_ARRAY_F_CONTIGUOUS));
bool both_c_contig =
((src_flags & py::array::c_style) && dst.is_c_contiguous());
bool both_f_contig =
((src_flags & py::array::f_style) && dst.is_f_contiguous());
if (both_c_contig || both_f_contig) {
if (src_type_id == dst_type_id) {
int src_elem_size = npy_src.itemsize();
Expand Down Expand Up @@ -791,8 +784,8 @@ void copy_numpy_ndarray_into_usm_ndarray(
bool is_src_c_contig = ((src_flags & py::array::c_style) != 0);
bool is_src_f_contig = ((src_flags & py::array::f_style) != 0);

bool is_dst_c_contig = ((dst_flags & USM_ARRAY_C_CONTIGUOUS) != 0);
bool is_dst_f_contig = ((dst_flags & USM_ARRAY_F_CONTIGUOUS) != 0);
bool is_dst_c_contig = dst.is_c_contiguous();
bool is_dst_f_contig = dst.is_f_contiguous();

// all args except itemsizes and is_?_contig bools can be modified by
// reference
Expand Down Expand Up @@ -906,18 +899,18 @@ usm_ndarray_linear_sequence_step(py::object start,
"usm_ndarray_linspace: Expecting 1D array to populate");
}

int flags = dst.get_flags();
if (!(flags & USM_ARRAY_C_CONTIGUOUS)) {
if (!dst.is_c_contiguous()) {
throw py::value_error(
"usm_ndarray_linspace: Non-contiguous arrays are not supported");
}

sycl::queue dst_q = dst.get_queue();
if (dst_q != exec_q && dst_q.get_context() != exec_q.get_context()) {
if (!dpctl::utils::queues_are_compatible(exec_q, {dst_q})) {
throw py::value_error(
"Execution queue context is not the same as allocation context");
"Execution queue is not compatible with the allocation queue");
}

auto array_types = dpctl::tensor::detail::usm_ndarray_types::get();
int dst_typenum = dst.get_typenum();
int dst_typeid = array_types.typenum_to_lookup_id(dst_typenum);

Expand Down Expand Up @@ -955,18 +948,18 @@ usm_ndarray_linear_sequence_affine(py::object start,
"usm_ndarray_linspace: Expecting 1D array to populate");
}

int flags = dst.get_flags();
if (!(flags & USM_ARRAY_C_CONTIGUOUS)) {
if (!dst.is_c_contiguous()) {
throw py::value_error(
"usm_ndarray_linspace: Non-contiguous arrays are not supported");
}

sycl::queue dst_q = dst.get_queue();
if (dst_q != exec_q && dst_q.get_context() != exec_q.get_context()) {
if (!dpctl::utils::queues_are_compatible(exec_q, {dst_q})) {
throw py::value_error(
"Execution queue context is not the same as allocation context");
}

auto array_types = dpctl::tensor::detail::usm_ndarray_types::get();
int dst_typenum = dst.get_typenum();
int dst_typeid = array_types.typenum_to_lookup_id(dst_typenum);

Expand Down Expand Up @@ -1010,23 +1003,20 @@ usm_ndarray_full(py::object py_value,
return std::make_pair(sycl::event(), sycl::event());
}

int dst_flags = dst.get_flags();

sycl::queue dst_q = dst.get_queue();
if (dst_q != exec_q && dst_q.get_context() != exec_q.get_context()) {
if (!dpctl::utils::queues_are_compatible(exec_q, {dst_q})) {
throw py::value_error(
"Execution queue context is not the same as allocation context");
"Execution queue is not compatible with the allocation queue");
}

auto array_types = dpctl::tensor::detail::usm_ndarray_types::get();
int dst_typenum = dst.get_typenum();
int dst_typeid = array_types.typenum_to_lookup_id(dst_typenum);

char *dst_data = dst.get_data();
sycl::event full_event;

if (dst_nelems == 1 || (dst_flags & USM_ARRAY_C_CONTIGUOUS) ||
(dst_flags & USM_ARRAY_F_CONTIGUOUS))
{
if (dst_nelems == 1 || dst.is_c_contiguous() || dst.is_f_contiguous()) {
auto fn = full_contig_dispatch_vector[dst_typeid];

sycl::event full_contig_event =
Expand Down Expand Up @@ -1068,6 +1058,7 @@ eye(py::ssize_t k,
"allocation queue");
}

auto array_types = dpctl::tensor::detail::usm_ndarray_types::get();
int dst_typenum = dst.get_typenum();
int dst_typeid = array_types.typenum_to_lookup_id(dst_typenum);

Expand All @@ -1079,8 +1070,8 @@ eye(py::ssize_t k,
return std::make_pair(sycl::event{}, sycl::event{});
}

bool is_dst_c_contig = ((dst.get_flags() & USM_ARRAY_C_CONTIGUOUS) != 0);
bool is_dst_f_contig = ((dst.get_flags() & USM_ARRAY_F_CONTIGUOUS) != 0);
bool is_dst_c_contig = dst.is_c_contiguous();
bool is_dst_f_contig = dst.is_f_contiguous();
if (!is_dst_c_contig && !is_dst_f_contig) {
throw py::value_error("USM array is not contiguous");
}
Expand Down Expand Up @@ -1182,6 +1173,8 @@ tri(sycl::queue &exec_q,
throw py::value_error("Arrays index overlapping segments of memory");
}

auto array_types = dpctl::tensor::detail::usm_ndarray_types::get();

int src_typenum = src.get_typenum();
int dst_typenum = dst.get_typenum();
int src_typeid = array_types.typenum_to_lookup_id(src_typenum);
Expand All @@ -1203,9 +1196,8 @@ tri(sycl::queue &exec_q,
using shT = std::vector<py::ssize_t>;
shT src_strides(src_nd);

int src_flags = src.get_flags();
bool is_src_c_contig = ((src_flags & USM_ARRAY_C_CONTIGUOUS) != 0);
bool is_src_f_contig = ((src_flags & USM_ARRAY_F_CONTIGUOUS) != 0);
bool is_src_c_contig = src.is_c_contiguous();
bool is_src_f_contig = src.is_f_contiguous();

const py::ssize_t *src_strides_raw = src.get_strides_raw();
if (src_strides_raw == nullptr) {
Expand All @@ -1227,9 +1219,8 @@ tri(sycl::queue &exec_q,

shT dst_strides(src_nd);

int dst_flags = dst.get_flags();
bool is_dst_c_contig = ((dst_flags & USM_ARRAY_C_CONTIGUOUS) != 0);
bool is_dst_f_contig = ((dst_flags & USM_ARRAY_F_CONTIGUOUS) != 0);
bool is_dst_c_contig = dst.is_c_contiguous();
bool is_dst_f_contig = dst.is_f_contiguous();

const py::ssize_t *dst_strides_raw = dst.get_strides_raw();
if (dst_strides_raw == nullptr) {
Expand Down Expand Up @@ -1457,9 +1448,6 @@ PYBIND11_MODULE(_tensor_impl, m)
init_copy_for_reshape_dispatch_vector();
import_dpctl();

// populate types constants for type dispatching functions
array_types = dpctl::tensor::detail::usm_ndarray_types::get();

m.def(
"_contract_iter", &contract_iter,
"Simplifies iteration of array of given shape & stride. Returns "
Expand Down