Skip to content

perf: optimize Python bindings array conversion performance (~27% speedup) #1311

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

Open
benksih-code wants to merge 1 commit into
base: main
Choose a base branch
from
Open

perf: optimize Python bindings array conversion performance (~27% speedup) #1311

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
22 changes: 19 additions & 3 deletions compilers/concrete-compiler/compiler/lib/Bindings/Python/CompilerAPIModule.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -202,15 +202,31 @@ getPythonTypeTransformer(const Message<concreteprotocol::GateInfo> &info) {

template <typename T> Tensor<T> arrayToTensor(pybind11::array &input) {
auto data_ptr = (const T *)input.data();
std::vector<T> data = std::vector(data_ptr, data_ptr + input.size());
// Optimized: copy directly from numpy buffer to vector without intermediate default initialization
std::vector<T> data(data_ptr, data_ptr + input.size());
auto dims = std::vector<size_t>(input.ndim(), 0);
for (ssize_t i = 0; i < input.ndim(); i++) {
dims[i] = input.shape(i);
}
return std::move(Tensor<T>(std::move(data), std::move(dims)));
return Tensor<T>(std::move(data), std::move(dims));
}

template <typename T> pybind11::array tensorToArray(Tensor<T> input) {
template <typename T> pybind11::array tensorToArray(Tensor<T>&& input) {
auto dims = input.dimensions;
// Optimized: move vector data to heap, transfer ownership to numpy array without copy
auto* values_ptr = new std::vector<T>(std::move(input.values));
return pybind11::array(
pybind11::dtype::of<T>(),
dims,
values_ptr->data(),
pybind11::capsule(values_ptr, [](void *ptr) {
delete static_cast<std::vector<T>*>(ptr);
})
);
}

// Lvalue fallback for backwards compatibility
template <typename T> pybind11::array tensorToArray(const Tensor<T>& input) {
return pybind11::array(pybind11::array::ShapeContainer(input.dimensions),
input.values.data());
}
Expand Down