Skip to content

[jit] Let script module buffer attributes can also cast device/type #19700

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

Closed
wanchaol wants to merge 5 commits into from
Closed

[jit] Let script module buffer attributes can also cast device/type #19700

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
9dd12fd
Let script module buffer attributes can also cast device/type
wanchaol Apr 24, 2019
253e2d6
add cpp test
wanchaol Apr 25, 2019
4c5712a
Merge branch 'master' into castfix
wanchaol Apr 25, 2019
3841ebc
address comment
wanchaol Apr 25, 2019
f9bb4c5
minor fix
wanchaol Apr 26, 2019
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
1 change: 1 addition & 0 deletions test/cpp/jit/test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,6 +78,7 @@ namespace jit {
_(ArgumentSpec) \
_(Fusion) \
_(GraphExecutor) \
_(ModuleConversion) \
_(Interp)

#if defined(USE_GTEST)
Expand Down
24 changes: 24 additions & 0 deletions test/cpp/jit/test_misc.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -758,6 +758,30 @@ void testModuleDefine() {
AT_ASSERT(result.toTensor().item<float>() == 6)
}

void testModuleConversion() {
auto m = std::make_shared<script::Module>();
{
// test cuda to cpu for params and buffers
m->register_parameter("foo", torch::ones({}, at::kCUDA), false);
m->register_buffer("bar", torch::ones({}, at::kCUDA));

m->to(at::kCUDA);
m->to(at::kCPU);
AT_ASSERT(m->get_parameter("foo").data().device().is_cpu());
AT_ASSERT(m->get_buffer("bar").data().device().is_cpu());
}
{
// test cpu to cuda for params and buffers
m->register_parameter("foo", torch::ones({}), false);
m->register_buffer("bar", torch::ones({}));

m->to(at::kCUDA);
AT_ASSERT(m->get_parameter("foo").data().device().is_cuda());
AT_ASSERT(m->get_buffer("bar").data().device().is_cuda());
}
}


static int testPassValue = 0;
void fakePass(std::shared_ptr<Graph>& g) {
testPassValue++;
Expand Down
30 changes: 22 additions & 8 deletions torch/csrc/jit/script/module.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -101,25 +101,39 @@ void Module::save(
ExportModule(*this, filename, extra_files);
}

void Module::to_impl(
void module_state_to(
const Slot& s,
const c10::optional<at::Device>& device,
const c10::optional<at::ScalarType>& dtype,
bool non_blocking) {
// First call `to()` on every child module.
for (auto& child : get_modules()) {
child->to_impl(device, dtype, non_blocking);
}
// Then convert every of our parameters.
for (auto& parameter : get_parameters()) {
// Need to access the `at::Tensor` as a `Variable` here.
autograd::Variable variable = parameter.value().toTensor();
autograd::Variable variable = s.value().toTensor();
at::Tensor data = variable.data();
// Use the data's original device or dtype if not supplied here.
auto new_data = data.to(
device.value_or(data.device()),
dtype.value_or(data.scalar_type()),
non_blocking);
variable.set_data(new_data);
}

void Module::to_impl(
const c10::optional<at::Device>& device,
const c10::optional<at::ScalarType>& dtype,
bool non_blocking) {
// First call `to()` on every child module.
for (auto& child : get_modules()) {
child->to_impl(device, dtype, non_blocking);
}
// Then convert every of our parameters.
for (auto& parameter : get_parameters()) {
module_state_to(parameter, device, dtype, non_blocking);
}
// Then convert every tensor attributes (buffers).
for (auto& attr : get_attributes()) {
if (attr.type()->isSubtypeOf(TensorType::get())) {
module_state_to(attr, device, dtype, non_blocking);
}
}
}

Expand Down