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

[LPT] tests extending

Browse Source
This commit is contained in:
Edward Shogulin
2020-06-01 10:37:39 +03:00
committed by Alexander Peskov
parent 1d51d2185a
commit 4d01adbe01
3 changed files with 37 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
inference-engine/tests/functional/plugin/cpu/shared_tests_instances/low_precision_transformations/fully_connected_transformation.cpp
View File

@@ -22,7 +22,7 @@ const std::vector<InferenceEngine::SizeVector> dimensions = {
const std::vector<LayerTransformation::Params> trasformationParamValues = {
LayerTestsUtils::LayerTransformationParamsFactory::createParams(),
LayerTestsUtils::LayerTransformationParamsFactory::createParamsI8I8(),
// LayerTestsUtils::LayerTransformationParamsFactory::createParamsI8I8(),
LayerTestsUtils::LayerTransformationParamsFactory::createParamsU8I8()
};

10
inference-engine/tests/functional/plugin/shared/src/low_precision_transformations/fully_connected_transformation.cpp
View File

@@ -88,7 +88,15 @@ void FullyConnectedTransformation::validate() {
EXPECT_TRUE(outputLayer != nullptr);
EXPECT_EQ("ScaleShift", outputLayer->type);
checkParentPrecision(outputLayer, false);
const InferenceEngine::CNNLayerPtr layer = InferenceEngine::details::CNNNetworkHelper::getParent(*outputLayer);
if (params.updatePrecisions) {
checkPrecisions(
*layer,
{ { InferenceEngine::Precision::U8 }, { InferenceEngine::Precision::I8 }, { netPrecision } },
{ getDeviceInternalPrecision(netPrecision) });
} else {
checkPrecisions(*layer, netPrecision);
}
}
IE_SUPPRESS_DEPRECATED_END

41
inference-engine/tests/functional/plugin/shared/src/low_precision_transformations/gemm_transformation.cpp
View File

@@ -39,27 +39,30 @@ void GemmTransformation::SetUp() {
std::tie(netPrecision, inputShape, targetDevice, params) = this->GetParam();
auto ngPrecision = FuncTestUtils::PrecisionUtils::convertIE2nGraphPrc(netPrecision);
const auto paramNode1 = std::make_shared<ngraph::opset1::Parameter>(ngPrecision, ngraph::Shape(inputShape));
const auto fakeQuantizeOnAcitvations = ngraph::builder::makeFakeQuantize(
paramNode1, ngPrecision, 256ul, { 1ul },
{ 0.f }, { 255.f / 4.f }, { 0.f }, { 255.f / 4.f });
fakeQuantizeOnAcitvations->set_friendly_name("fakeQuantizeOnAcitvations");
const float low = params.precisionsOnActivations[0] == InferenceEngine::Precision::U8 ? 0.f : -128.f;
const float high = params.precisionsOnActivations[0] == InferenceEngine::Precision::U8 ? 255.f : 127.f;
const auto paramNode2 = std::make_shared<ngraph::opset1::Parameter>(ngPrecision, ngraph::Shape(inputShape));
const auto fakeQuantizeOnWeights = ngraph::builder::makeFakeQuantize(
paramNode2, ngPrecision, 256ul, { 1ul },
{ -128.f / 8.f }, { 127.f / 8.f }, { -128.f / 8.f }, { 127.f / 8.f });
fakeQuantizeOnWeights->set_friendly_name("fakeQuantizeOnWeights");
const auto input1 = std::make_shared<ngraph::opset1::Parameter>(ngPrecision, ngraph::Shape(inputShape));
const auto fakeQuantize1 = ngraph::builder::makeFakeQuantize(
input1, ngPrecision, 256ul, { 1ul },
{ low / 4.f }, { high / 4.f }, { low / 4.f }, { high / 4.f });
fakeQuantize1->set_friendly_name("fakeQuantize1");
const auto input2 = std::make_shared<ngraph::opset1::Parameter>(ngPrecision, ngraph::Shape(inputShape));
const auto fakeQuantize2 = ngraph::builder::makeFakeQuantize(
input2, ngPrecision, 256ul, { 1ul },
{ low / 8.f }, { high / 8.f }, { low / 8.f }, { high / 8.f });
fakeQuantize2->set_friendly_name("fakeQuantize2");
const auto matMul = std::make_shared<ngraph::opset1::MatMul>(
fakeQuantizeOnAcitvations->output(0),
fakeQuantizeOnWeights->output(0),
fakeQuantize1->output(0),
fakeQuantize2->output(0),
false,
false);
matMul->set_friendly_name("matMul");
ngraph::ResultVector results {std::make_shared<ngraph::opset1::Result>(matMul)};
function = std::make_shared<ngraph::Function>(results, ngraph::ParameterVector { paramNode1, paramNode2 }, "GemmTransformation");
function = std::make_shared<ngraph::Function>(results, ngraph::ParameterVector { input1, input2 }, "GemmTransformation");
// TODO: move to some another place
validate();
@@ -83,7 +86,17 @@ void GemmTransformation::validate() {
EXPECT_TRUE(outputLayer != nullptr);
EXPECT_EQ("ScaleShift", outputLayer->type);
checkParentPrecision(outputLayer, false);
const InferenceEngine::CNNLayerPtr layer = InferenceEngine::details::CNNNetworkHelper::getParent(*outputLayer);
for (const InferenceEngine::DataWeakPtr insDataWeak : layer->insData) {
const InferenceEngine::DataPtr insData = insDataWeak.lock();
EXPECT_TRUE(insData != nullptr) << "insert data is nullable";
const InferenceEngine::Precision precision = insData->getTensorDesc().getPrecision();
const std::unordered_set<uint8_t> expectedPrecisions = params.updatePrecisions ?
std::unordered_set<uint8_t>({ params.precisionsOnActivations[0] }) :
std::unordered_set<uint8_t>({ InferenceEngine::Precision::FP16, InferenceEngine::Precision::FP32 });
EXPECT_TRUE(expectedPrecisions.find(precision) != expectedPrecisions.end()) <<
" actual precision is " << precision;
}
}
IE_SUPPRESS_DEPRECATED_END