dynamic.in.cpp 9.94 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294
//*****************************************************************************
// Copyright 2017-2019 Intel Corporation
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//*****************************************************************************

#include "gtest/gtest.h"
#include "ngraph/ngraph.hpp"
#include "util/all_close_f.hpp"
#include "util/test_control.hpp"
#include "util/test_tools.hpp"

using namespace std;
using namespace ngraph;

static string s_manifest = "${MANIFEST}";

NGRAPH_TEST(${BACKEND_NAME}, create_dynamic_backend)
{
    auto backend = runtime::Backend::create("${BACKEND_NAME}", true);
    ASSERT_NE(backend, nullptr);
    ASSERT_TRUE(backend->supports_dynamic_tensors());
}

NGRAPH_TEST(${BACKEND_NAME}, create_dynamic_tensor)
{
    auto backend = runtime::Backend::create("${BACKEND_NAME}", true);
    auto t = backend->create_dynamic_tensor(element::f32, PartialShape{2, Dimension::dynamic(), 3});
    ASSERT_TRUE(t->get_partial_shape().same_scheme(PartialShape{2, Dimension::dynamic(), 3}));
}

NGRAPH_TEST(${BACKEND_NAME}, dynamic_abc)
{
    //
    // Create a graph for f(a,b,c) = (a+b)*c, where a, b, c all have shape {2,?,3}.
    //
    auto a = make_shared<op::Parameter>(element::f32, PartialShape{2, Dimension::dynamic(), 3});
    auto b = make_shared<op::Parameter>(element::f32, PartialShape{2, Dimension::dynamic(), 3});
    auto c = make_shared<op::Parameter>(element::f32, PartialShape{2, Dimension::dynamic(), 3});

    auto a_plus_b_times_c = (a + b) * c;

    auto f = make_shared<Function>(NodeVector{a_plus_b_times_c}, ParameterVector{a, b, c});

    //
    // Get a backend with dynamic support, and compile f.
    //
    auto backend = runtime::Backend::create("${BACKEND_NAME}", true);

    auto ex = backend->compile(f);

    //
    // Create a dynamic output tensor with shape {2,?,3}.
    //
    auto t_r =
        backend->create_dynamic_tensor(element::f32, PartialShape{2, Dimension::dynamic(), 3});

    //
    // For each of n=[0,...,5), run the compiled executable against a test vector of shape
    // {2,n,3}, and check the results.
    //
    for (size_t middle_dim = 0; middle_dim < 5; middle_dim++)
    {
        // Fill in some test input values, which we'll use for a, b, and c.
        vector<float> inputs(2 * middle_dim * 3);
        for (size_t i = 0; i < 2 * middle_dim * 3; i++)
        {
            inputs[i] = i;
        }

        // Create static tensors for the inputs and copy data.
        auto t_a = backend->create_tensor(element::f32, Shape{2, middle_dim, 3});
        auto t_b = backend->create_tensor(element::f32, Shape{2, middle_dim, 3});
        auto t_c = backend->create_tensor(element::f32, Shape{2, middle_dim, 3});

        copy_data(t_a, inputs);
        copy_data(t_b, inputs);
        copy_data(t_c, inputs);

        // Call ex, writing result into t_r (note we're using the same t_r from outside the loop.)
        ex->call_with_validate({t_r}, {t_a, t_b, t_c});

        // After call, t_r should have a shape of {2,n,3}.
        ASSERT_EQ(t_r->get_shape(), (Shape{2, middle_dim, 3}));

        // Read out the results, and compare them against expected values.
        auto results = read_vector<float>(t_r);

        vector<float> expected_values(2 * middle_dim * 3);
        for (size_t i = 0; i < 2 * middle_dim * 3; i++)
        {
            expected_values[i] = (i + i) * i;
        }

        EXPECT_TRUE(test::all_close_f(results, expected_values));
    }
}

static void axpy_test(const PartialShape& input_pshape, const std::vector<Shape>& input_shapes)
{
    auto a = make_shared<op::Parameter>(element::f32, input_pshape);
    auto x = make_shared<op::Parameter>(element::f32, input_pshape);
    auto y = make_shared<op::Parameter>(element::f32, input_pshape);

    auto axpy = a * x + y;

    auto f = make_shared<Function>(NodeVector{axpy}, ParameterVector{a, x, y});
    auto backend = runtime::Backend::create("${BACKEND_NAME}", true);
    auto ex = backend->compile(f);

    auto t_r = backend->create_dynamic_tensor(element::f32, input_pshape);

    for (auto& shape : input_shapes)
    {
        vector<float> inputs(shape_size(shape));
        for (size_t i = 0; i < shape_size(shape); i++)
        {
            inputs[i] = i;
        }

        auto t_a = backend->create_tensor(element::f32, shape);
        auto t_x = backend->create_tensor(element::f32, shape);
        auto t_y = backend->create_tensor(element::f32, shape);

        copy_data(t_a, inputs);
        copy_data(t_x, inputs);
        copy_data(t_y, inputs);

        ex->call_with_validate({t_r}, {t_a, t_x, t_y});

        ASSERT_EQ(t_r->get_shape(), shape);

        auto results = read_vector<float>(t_r);

        vector<float> expected_values(shape_size(shape));
        for (size_t i = 0; i < shape_size(shape); i++)
        {
            expected_values[i] = (i * i) + i;
        }

        EXPECT_TRUE(test::all_close_f(results, expected_values));
    }
}

NGRAPH_TEST(${BACKEND_NAME}, dynamic_axpy)
{
    // Test with shape {?, 3, 3}.
    axpy_test(PartialShape{Dimension::dynamic(), 3, 3}, {Shape{2, 3, 3}, Shape{5, 3, 3}});

    // Test with shape {?, ?, ?}.
    axpy_test(PartialShape::dynamic(3),
              {Shape{2, 3, 3}, Shape{5, 3, 3}, Shape{2, 5, 2}, Shape{8, 1, 8}});

    // Test with shape ?. (Rank unknown.)
    axpy_test(PartialShape::dynamic(),
              {Shape{2, 3, 3},
               Shape{5, 3, 3},
               Shape{2, 5, 2},
               Shape{8, 1, 8},
               Shape{5},
               Shape{8, 2},
               Shape{8, 2, 8, 2},
               Shape{2, 3, 4, 5, 2}});
}

static void to_vector_test(const PartialShape& input_pshape, const std::vector<Shape>& input_shapes)
{
    auto x = make_shared<op::Parameter>(element::f32, input_pshape);

    shared_ptr<Node> x_new_shape = make_shared<op::ShapeOf>(x);
    x_new_shape = make_shared<op::Product>(x_new_shape, AxisSet{0});
    x_new_shape = make_shared<op::Reshape>(x_new_shape, AxisVector{}, Shape{1});

    auto x_reshaped = make_shared<op::DynReshape>(x, x_new_shape);

    auto f = make_shared<Function>(NodeVector{x_reshaped}, ParameterVector{x});
    auto backend = runtime::Backend::create("${BACKEND_NAME}", true);
    auto ex = backend->compile(f);

    auto t_r = backend->create_dynamic_tensor(element::f32, PartialShape::dynamic(1));

    for (auto& shape : input_shapes)
    {
        vector<float> inputs(shape_size(shape));
        for (size_t i = 0; i < shape_size(shape); i++)
        {
            inputs[i] = i;
        }

        auto t_x = backend->create_tensor(element::f32, shape);

        copy_data(t_x, inputs);

        ex->call_with_validate({t_r}, {t_x});

        ASSERT_EQ(t_r->get_shape(), (Shape{shape_size(shape)}));

        auto results = read_vector<float>(t_r);

        EXPECT_TRUE(test::all_close_f(results, inputs));
    }
}

NGRAPH_TEST(${BACKEND_NAME}, dynamic_to_vector)
{
    // Test with shape {?, 3, 3}.
    to_vector_test(PartialShape{Dimension::dynamic(), 3, 3}, {Shape{2, 3, 3}, Shape{5, 3, 3}});

    // Test with shape {?, ?, ?}.
    to_vector_test(PartialShape::dynamic(3),
                   {Shape{2, 3, 3}, Shape{5, 3, 3}, Shape{2, 5, 2}, Shape{8, 1, 8}});

    // Test with shape ?. (Rank unknown.)
    to_vector_test(PartialShape::dynamic(),
                   {Shape{2, 3, 3},
                    Shape{5, 3, 3},
                    Shape{2, 5, 2},
                    Shape{8, 1, 8},
                    Shape{5},
                    Shape{8, 2},
                    Shape{8, 2, 8, 2},
                    Shape{2, 3, 4, 5, 2}});
}

static void reverse_shape_test(const PartialShape& input_pshape,
                               const std::vector<Shape>& input_shapes)
{
    auto x = make_shared<op::Parameter>(element::f32, input_pshape);

    shared_ptr<Node> x_new_shape = make_shared<op::ShapeOf>(x);
    x_new_shape = make_shared<op::Reverse>(x_new_shape, AxisSet{0});

    auto x_reshaped = make_shared<op::DynReshape>(x, x_new_shape);

    auto f = make_shared<Function>(NodeVector{x_reshaped}, ParameterVector{x});
    auto backend = runtime::Backend::create("${BACKEND_NAME}", true);
    auto ex = backend->compile(f);

    auto t_r = backend->create_dynamic_tensor(element::f32, PartialShape::dynamic());

    for (auto& shape : input_shapes)
    {
        vector<float> inputs(shape_size(shape));
        for (size_t i = 0; i < shape_size(shape); i++)
        {
            inputs[i] = i;
        }

        auto t_x = backend->create_tensor(element::f32, shape);

        copy_data(t_x, inputs);

        ex->call_with_validate({t_r}, {t_x});

        Shape expected_shape = shape;
        std::reverse(expected_shape.begin(), expected_shape.end());
        ASSERT_EQ(t_r->get_shape(), expected_shape);

        auto results = read_vector<float>(t_r);

        EXPECT_TRUE(test::all_close_f(results, inputs));
    }
}

NGRAPH_TEST(${BACKEND_NAME}, dynamic_reverse_shape)
{
    // Test with shape {?, 3, 3}.
    reverse_shape_test(PartialShape{Dimension::dynamic(), 3, 3}, {Shape{2, 3, 3}, Shape{5, 3, 3}});

    // Test with shape {?, ?, ?}.
    reverse_shape_test(PartialShape::dynamic(3),
                       {Shape{2, 3, 3}, Shape{5, 3, 3}, Shape{2, 5, 2}, Shape{8, 1, 8}});

    // Test with shape ?. (Rank unknown.)
    reverse_shape_test(PartialShape::dynamic(),
                       {Shape{2, 3, 3},
                        Shape{5, 3, 3},
                        Shape{2, 5, 2},
                        Shape{8, 1, 8},
                        Shape{5},
                        Shape{8, 2},
                        Shape{8, 2, 8, 2},
                        Shape{2, 3, 4, 5, 2}});
}