aboutsummaryrefslogtreecommitdiff
path: root/src/backends/reference/workloads/Gather.cpp
blob: b195003e04f84f131f765020a5cf47ce66b051bd (plain)
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
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//

#include "Gather.hpp"

#include "RefWorkloadUtils.hpp"

#include <backendsCommon/WorkloadData.hpp>

namespace armnn
{

template <typename T>
void Gather(const TensorInfo& paramsInfo,
            const TensorInfo& indicesInfo,
            const TensorInfo& outputInfo,
            const T* params,
            const int32_t* indices,
            T* output)
{
    const TensorShape& paramsShape = paramsInfo.GetShape();

    unsigned int paramsProduct = 1;
    for (unsigned int i = 1; i < paramsInfo.GetNumDimensions(); ++i)
    {
        paramsProduct = paramsProduct * paramsShape[i];
    }

    unsigned int outIndex = 0;
    for (unsigned int i = 0; i < indicesInfo.GetNumElements(); ++i)
    {
        unsigned int indx = boost::numeric_cast<unsigned int>(indices[i]);

        BOOST_ASSERT(indices[i] >= 0 && indx < paramsShape[0]);

        unsigned int startOffset = indx * paramsProduct;
        unsigned int endOffset = startOffset + paramsProduct;
        for (unsigned int j = startOffset; j < endOffset; ++j)
        {
            output[outIndex] = params[j];
            ++outIndex;
        }
    }

    BOOST_ASSERT(outIndex == outputInfo.GetNumElements());
}

template void Gather<float>(const TensorInfo& paramsInfo,
                            const TensorInfo& indicesInfo,
                            const TensorInfo& outputInfo,
                            const float* params,
                            const int32_t* indices,
                            float* output);

template void Gather<uint8_t>(const TensorInfo& paramsInfo,
                              const TensorInfo& indicesInfo,
                              const TensorInfo& outputInfo,
                              const uint8_t* params,
                              const int32_t* indices,
                              uint8_t* output);

} //namespace armnn