IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[GPU] Updated to get lockable memories for device inputs in prepare_input() (#14532)

Browse Source 
* updated to get ockable memory for device inputs

* gpuFuncTest to set f64 scalar tensor

* aligned type names

* updated to get lockable memory for some specific precisions
This commit is contained in:
Eddy Kim
2022-12-09 22:00:32 +09:00
committed by GitHub
parent 67bbc7361c
commit 083a17e89e
2 changed files with 30 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/plugins/intel_gpu/src/plugin/infer_request.cpp
View File

@@ -918,7 +918,10 @@ void InferRequest::prepare_input(const cldnn::primitive_id& inputName, Blob::Ptr
_deviceInputs[inputName] = reinterpret_device_blob(_deviceInputs[inputName], inputBlob->getTensorDesc());
}
} else if (input_layout.is_static() && !is_dev_input && can_use_usm) {
allocate_dev_mem_if_needed(_deviceInputs, inputBlob, inputName, input_layout);
bool need_lockable_mem = false;
if (prec == Precision::I16 || prec == Precision::U16 || prec == Precision::FP64 || prec == Precision::U64 || prec == Precision::U32)
need_lockable_mem = true;
allocate_dev_mem_if_needed(_deviceInputs, inputBlob, inputName, input_layout, need_lockable_mem);
}
OPENVINO_ASSERT(_deviceInputs.find(inputName) != _deviceInputs.end(), "[GPU] Couldn't find device blob allocated for ", inputName, " input");
auto reqBlob = _deviceInputs.at(inputName)->as<gpu::ClBlob>();

26
src/tests/functional/plugin/gpu/behavior/infer_request.cpp
View File

@@ -137,3 +137,29 @@ TEST(TensorTest, smoke_canSetShapeForPreallocatedTensor) {
ASSERT_NO_THROW(output_tensor.set_shape({1, 10, 10, 10}));
ASSERT_NO_THROW(output_tensor.set_shape({2, 10, 20, 20}));
}
TEST(TensorTest, smoke_canSetScalarTensor) {
std::vector<std::vector<size_t>> scalar_shape = {{}};
auto params = ngraph::builder::makeParams(ngraph::element::f64, scalar_shape);
params.front()->set_friendly_name("Scalar_1");
params.front()->output(0).get_tensor().set_names({"scalar1"});
std::vector<size_t> const_shape = {1};
auto const1 = ngraph::opset1::Constant::create(ngraph::element::i64, ngraph::Shape{1}, const_shape);
const1->set_friendly_name("Const_1");
const1->output(0).get_tensor().set_names({"const1"});
const1->fill_data(ov::element::i64, 0);
auto unsqueeze1 = std::make_shared<ngraph::opset1::Unsqueeze>(params.front(), const1);
ngraph::ResultVector results{std::make_shared<ngraph::opset1::Result>(unsqueeze1)};
std::shared_ptr<ngraph::Function> fnPtr = std::make_shared<ngraph::Function>(results, params);
auto ie = ov::Core();
auto compiled_model = ie.compile_model(fnPtr, CommonTestUtils::DEVICE_GPU);
auto request = compiled_model.create_infer_request();
double real_data = 1.0;
ov::Tensor input_data(ngraph::element::f64, {}, &real_data);
request.set_tensor("scalar1", input_data);
ASSERT_NO_THROW(request.infer());
}