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

ScatterUpdate CPU shape_infer by entryFirstPassthrough (#13950)

Browse Source 
* ScatterUpdate shape_infer tests

* Add ScatterUpdate to entryFirstPassthrough

* Headers cleanup

* Style update

* namespace update
This commit is contained in:
Katarzyna Mitrus
2022-12-16 11:28:16 +01:00
committed by GitHub
parent 4bc80ca446
commit e0c21ce302
2 changed files with 136 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
src/plugins/intel_cpu/src/utils/shape_inference/shape_inference.cpp
View File

@@ -420,7 +420,8 @@ std::shared_ptr<IShapeInferCommon> make_shape_inference(const std::shared_ptr<ng
ov::is_type<ov::opset1::HardSigmoid>(op) || ov::is_type<ov::opset1::Selu>(op) ||
ov::is_type<ov::opset1::PRelu>(op) || ov::is_type<ov::opset3::CumSum>(op) ||
ov::is_type<ov::opset1::BatchNormInference>(op) || ov::is_type<ov::opset5::BatchNormInference>(op) ||
ov::is_type<ov::opset4::Swish>(op) || ov::is_type<ov::opset1::NormalizeL2>(op)) {
ov::is_type<ov::opset4::Swish>(op) || ov::is_type<ov::opset1::NormalizeL2>(op) ||
ov::is_type<ov::opset3::ScatterUpdate>(op)) {
return std::make_shared<entryFirstPassthrough>(op);
} else if (ov::is_type<ov::op::util::BinaryElementwiseArithmetic>(op) ||
ov::is_type<ov::op::util::BinaryElementwiseComparison>(op) ||

134
src/plugins/intel_cpu/tests/unit/shape_inference_test/scatter_update_shape_inference_test.cpp Normal file
View File

@@ -0,0 +1,134 @@
// Copyright (C) 2018-2022 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include <gtest/gtest.h>
#include <openvino/op/ops.hpp>
#include <openvino/op/parameter.hpp>
#include <utils/shape_inference/shape_inference.hpp>
#include <utils/shape_inference/static_shape.hpp>
using namespace ov;
using namespace ov::intel_cpu;
TEST(StaticShapeInferenceTest, ScatterUpdate_3D_axis_1) {
auto data_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1, -1});
auto indices_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1});
auto updates_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1, -1, -1});
auto axis_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1});
auto scatter_update = std::make_shared<op::v3::ScatterUpdate>(data_param, indices_param, updates_param, axis_param);
int32_t axis_val[] = {1};
std::map<size_t, std::shared_ptr<ngraph::runtime::HostTensor>> constant_data;
constant_data[3] = std::make_shared<ngraph::runtime::HostTensor>(element::Type_t::i32, Shape{1}, axis_val);
std::vector<StaticShape> input_shapes = {StaticShape{2, 3, 4},
StaticShape{2, 1},
StaticShape{2, 2, 1, 4},
StaticShape{1}},
output_shapes = {StaticShape{}};
shape_inference(scatter_update.get(), input_shapes, output_shapes, constant_data);
EXPECT_EQ(output_shapes[0], StaticShape({2, 3, 4}));
}
TEST(StaticShapeInferenceTest, ScatterUpdate_4D_axis_2) {
auto data_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1, -1, -1});
auto indices_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1});
auto updates_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1, -1, -1, -1});
auto axis_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1});
auto scatter_update = std::make_shared<op::v3::ScatterUpdate>(data_param, indices_param, updates_param, axis_param);
int32_t axis_val[] = {2};
std::map<size_t, std::shared_ptr<ngraph::runtime::HostTensor>> constant_data;
constant_data[3] = std::make_shared<ngraph::runtime::HostTensor>(element::Type_t::i32, Shape{1}, axis_val);
std::vector<StaticShape> input_shapes = {StaticShape{1000, 256, 10, 15},
StaticShape{125, 20},
StaticShape{1000, 125, 20, 10, 15},
StaticShape{1}},
output_shapes = {StaticShape{}};
shape_inference(scatter_update.get(), input_shapes, output_shapes, constant_data);
EXPECT_EQ(output_shapes[0], StaticShape({1000, 256, 10, 15}));
}
TEST(StaticShapeInferenceTest, ScatterUpdate_4D_incompatible_axis) {
auto data_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1, -1, -1});
auto indices_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1});
auto updates_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1, -1, -1, -1});
auto axis_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1});
auto scatter_update = std::make_shared<op::v3::ScatterUpdate>(data_param, indices_param, updates_param, axis_param);
int32_t axis_val[] = {1};
std::map<size_t, std::shared_ptr<ngraph::runtime::HostTensor>> constant_data;
constant_data[3] = std::make_shared<ngraph::runtime::HostTensor>(element::Type_t::i32, Shape{1}, axis_val);
std::vector<StaticShape> input_shapes = {StaticShape{1000, 256, 10, 15},
StaticShape{125, 20},
StaticShape{1000, 125, 20, 10, 15},
StaticShape{1}},
output_shapes = {StaticShape{}};
shape_inference(scatter_update.get(), input_shapes, output_shapes, constant_data);
EXPECT_EQ(output_shapes[0], StaticShape({1000, 256, 10, 15}));
}
TEST(StaticShapeInferenceTest, ScatterUpdate_axis_as_const) {
auto data_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1, -1, -1});
auto indices_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1});
auto updates_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape{-1, -1, -1, -1, -1});
auto axis_const = std::make_shared<ov::op::v0::Constant>(element::i32, Shape{1}, std::vector<int32_t>{1});
auto scatter_update = std::make_shared<op::v3::ScatterUpdate>(data_param, indices_param, updates_param, axis_const);
std::vector<StaticShape> input_shapes = {StaticShape{1000, 256, 10, 15},
StaticShape{125, 20},
StaticShape{1000, 125, 20, 10, 15},
StaticShape{1}},
output_shapes = {StaticShape{}};
shape_inference(scatter_update.get(), input_shapes, output_shapes);
EXPECT_EQ(output_shapes[0], StaticShape({1000, 256, 10, 15}));
}
TEST(StaticShapeInferenceTest, ScatterUpdate_dynamic_rank) {
auto data_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape::dynamic());
auto indices_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape::dynamic());
auto updates_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape::dynamic());
auto axis_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape::dynamic());
auto scatter_update = std::make_shared<op::v3::ScatterUpdate>(data_param, indices_param, updates_param, axis_param);
int32_t axis_val[] = {1};
std::map<size_t, std::shared_ptr<ngraph::runtime::HostTensor>> constant_data;
constant_data[3] = std::make_shared<ngraph::runtime::HostTensor>(element::Type_t::i32, Shape{1}, axis_val);
std::vector<StaticShape> input_shapes = {StaticShape{1000, 256, 10, 15},
StaticShape{125, 20},
StaticShape{1000, 125, 20, 10, 15},
StaticShape{1}},
output_shapes = {StaticShape{}};
shape_inference(scatter_update.get(), input_shapes, output_shapes, constant_data);
EXPECT_EQ(output_shapes[0], StaticShape({1000, 256, 10, 15}));
}
TEST(StaticShapeInferenceTest, ScatterUpdate_params_dynamic_rank_incorrect_updates_shape) {
auto data_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape::dynamic());
auto indices_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape::dynamic());
auto updates_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape::dynamic());
auto axis_param = std::make_shared<ov::op::v0::Parameter>(element::i32, PartialShape::dynamic());
auto scatter_update = std::make_shared<op::v3::ScatterUpdate>(data_param, indices_param, updates_param, axis_param);
int32_t axis_val[] = {1};
std::map<size_t, std::shared_ptr<ngraph::runtime::HostTensor>> constant_data;
constant_data[3] = std::make_shared<ngraph::runtime::HostTensor>(element::Type_t::i32, Shape{1}, axis_val);
// Incorrect rank of the third input shape
std::vector<StaticShape> input_shapes = {StaticShape{1000, 256, 10, 15},
StaticShape{125, 20, 1, 1, 1},
StaticShape{1000, 125, 20, 10},
StaticShape{1}},
output_shapes = {StaticShape{}};
// ScatterUpdate shape_inference is implemented by usage of entryFirstPassthrough, no additional checks
shape_inference(scatter_update.get(), input_shapes, output_shapes, constant_data);
EXPECT_EQ(output_shapes[0], StaticShape({1000, 256, 10, 15}));
}