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

[IE CLDNN] Fix gather dimensions calculation (#960)

Browse Source
This commit is contained in:
Konrad Dobros
2020-06-17 23:31:17 +02:00
committed by GitHub
parent aeb70036d7
commit cf60baf2f0
2 changed files with 190 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/gather/gather_kernel_ref.cpp vendored
View File

@@ -88,7 +88,7 @@ static std::string GetDictionaryIndexOrder(const gather_params& params, size_t a
const std::string zeroVal = "0";
size_t dictionary_dims_num = GetNonEmptyDimsNumber(params.inputs[0]);
size_t indices_dims_num = GetNonEmptyDimsNumber(params.inputs[1]);
size_t indices_dims_num = GetNonEmptyDimsNumber(params.output) - dictionary_dims_num + 1;
// Shift indices of Gather dictionary input related to output dims
for (size_t i = axis + 1; i < dictionary_dims_num; i++)

190
inference-engine/thirdparty/clDNN/tests/test_cases/gather_gpu_test.cpp vendored
View File

@@ -1,4 +1,4 @@
// Copyright (c) 2019 Intel Corporation
// Copyright (c) 2019-2020 Intel Corporation
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
@@ -750,3 +750,191 @@ TEST(gather_gpu_int32, d22_axisY) {
EXPECT_EQ(expected_results[i], output_ptr[i]);
}
}
TEST(gather_gpu_fp32, d41_axisB) {
// Dictionary : 2x2x3x1
// Indexes : 4x1x1x1
// Axis : 0
// Output : 4x1x2x3
// Input values in fp32, indices in i32
// Indexes:
// 0, 1, 1, 0
//
// Dictionary:
// 1, 2, 3, 4, 5, 6,
// 7, 8, 9, 10, 11, 12
//
// Output:
// 1, 2, 3, 4, 5, 6,
// 7, 8, 9, 10, 11, 12
// 7, 8, 9, 10, 11, 12
// 1, 2, 3, 4, 5, 6,
engine engine;
auto input1 = memory::allocate(engine, { data_types::f32, format::bfyx, { 2, 2, 1, 3 } }); // Dictionary
auto input2 = memory::allocate(engine, { data_types::i32, format::bfyx, { 4, 1, 1, 1 } }); // Indexes
auto axis = cldnn::gather::gather_axis::along_b;
set_values(input1, {
1.f, 2.f, 3.f,
4.f, 5.f, 6.f,
7.f, 8.f, 9.f,
10.f, 11.f, 12.f
});
set_values(input2, {
0, 1, 1, 0
});
topology topology;
topology.add(input_layout("InputDictionary", input1.get_layout()));
topology.add(input_layout("InputText", input2.get_layout()));
topology.add(
gather("gather", "InputDictionary", "InputText", axis, tensor(4, 1, 3, 2))
);
network network(engine, topology);
network.set_input_data("InputDictionary", input1);
network.set_input_data("InputText", input2);
auto outputs = network.execute();
auto output = outputs.at("gather").get_memory();
auto output_ptr = output.pointer<float>();
std::vector<float> expected_results = {
1.f, 2.f, 3.f, 4.f, 5.f, 6.f,
7.f, 8.f, 9.f, 10.f, 11.f, 12.f,
7.f, 8.f, 9.f, 10.f, 11.f, 12.f,
1.f, 2.f, 3.f, 4.f, 5.f, 6.f
};
ASSERT_EQ(expected_results.size(), output_ptr.size());
for (size_t i = 0; i < expected_results.size(); ++i) {
EXPECT_EQ(expected_results[i], output_ptr[i]) << " at i=" << i;
}
}
TEST(gather_gpu_fp32, d41_axisF) {
// Dictionary : 2x3x2x1
// Indexes : 4x1x1x1
// Axis : 0
// Output : 2x4x1x2
// Input values in fp32, indices in i32
// Indexes:
// 1, 0, 1, 2
//
// Dictionary:
// 1, 2, 3, 4, 5, 6,
// 7, 8, 9, 10, 11, 12
//
// Output:
// 3, 4, 1, 2, 3, 4, 5, 6,
// 9, 10, 7, 8, 9, 10, 11, 12
engine engine;
auto input1 = memory::allocate(engine, { data_types::f32, format::bfyx, { 2, 3, 1, 2 } }); // Dictionary
auto input2 = memory::allocate(engine, { data_types::i32, format::bfyx, { 4, 1, 1, 1 } }); // Indexes
auto axis = cldnn::gather::gather_axis::along_f;
set_values(input1, {
1.f, 2.f, 3.f, 4.f, 5.f, 6.f,
7.f, 8.f, 9.f, 10.f, 11.f, 12.f
});
set_values(input2, {
1, 0, 1, 2
});
topology topology;
topology.add(input_layout("InputDictionary", input1.get_layout()));
topology.add(input_layout("InputText", input2.get_layout()));
topology.add(
gather("gather", "InputDictionary", "InputText", axis, tensor(2, 4, 2, 1))
);
network network(engine, topology);
network.set_input_data("InputDictionary", input1);
network.set_input_data("InputText", input2);
auto outputs = network.execute();
auto output = outputs.at("gather").get_memory();
auto output_ptr = output.pointer<float>();
std::vector<float> expected_results = {
3.f, 4.f, 1.f, 2.f, 3.f, 4.f, 5.f, 6.f,
9.f, 10.f, 7.f, 8.f, 9.f, 10.f, 11.f, 12.f
};
ASSERT_EQ(expected_results.size(), output_ptr.size());
for (size_t i = 0; i < expected_results.size(); ++i) {
EXPECT_EQ(expected_results[i], output_ptr[i]) << " at i=" << i;
}
}
TEST(gather_gpu_fp32, d2_axisX) {
// Dictionary : 2x2x1x1
// Indexes : 2x1x1x1
// Axis : 0
// Output : 2x2x1x2
// Input values in fp32, indices in i32
// Indexes:
// 0, 0
//
// Dictionary:
// 1, 2, 3, 4
//
// Output:
// 1, 1, 2, 2, 3, 3, 4, 4
engine engine;
auto input1 = memory::allocate(engine, { data_types::f32, format::bfyx, { 2, 2, 1, 1 } }); // Dictionary
auto input2 = memory::allocate(engine, { data_types::i32, format::bfyx, { 2, 1, 1, 1 } }); // Indexes
auto axis = cldnn::gather::gather_axis::along_x;
set_values(input1, {
1.f, 2.f,
3.f, 4.f,
});
set_values(input2, {
0, 0
});
topology topology;
topology.add(input_layout("InputDictionary", input1.get_layout()));
topology.add(input_layout("InputText", input2.get_layout()));
topology.add(
gather("gather", "InputDictionary", "InputText", axis, tensor(2, 2, 2, 1))
);
network network(engine, topology);
network.set_input_data("InputDictionary", input1);
network.set_input_data("InputText", input2);
auto outputs = network.execute();
auto output = outputs.at("gather").get_memory();
auto output_ptr = output.pointer<float>();
std::vector<float> expected_results = {
1.f, 1.f, 2.f, 2.f,
3.f, 3.f, 4.f, 4.f
};
ASSERT_EQ(expected_results.size(), output_ptr.size());
for (size_t i = 0; i < expected_results.size(); ++i) {
EXPECT_EQ(expected_results[i], output_ptr[i]) << " at i=" << i;
}
}