MNN/source/backend/cpu/CPUGather.cpp

60 lines
1.8 KiB
C++

//
// CPUGather.cpp
// MNN
//
// Created by MNN on 2018/08/02.
// Copyright © 2018, Alibaba Group Holding Limited
//
#include "backend/cpu/CPUGather.hpp"
#include "backend/cpu/CPUBackend.hpp"
#include "backend/cpu/compute/CommonOptFunction.h"
#include "core/Macro.h"
namespace MNN {
CPUGather::CPUGather(Backend *b, const MNN::Op *op) : MNN::Execution(b), mOp(op) {
// nothing to do
}
ErrorCode CPUGather::onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
return NO_ERROR;
}
ErrorCode CPUGather::onExecute(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
auto embedding = inputs[0];
auto indices = inputs[1];
auto output = outputs[0];
auto bytes = embedding->buffer().type.bytes();
const size_t indicesCount = indices->elementSize();
const auto limit = embedding->length(0);
auto outputData = output->host<uint8_t>();
const auto *inputData = embedding->host<uint8_t>();
const int firstDimStride = embedding->buffer().dim[0].stride * bytes;
const int *indicesData = indices->host<int32_t>();
for (int i = 0; i < indicesCount; i++) {
if (indicesData[i] < 0 || indicesData[i] > limit) {
return INPUT_DATA_ERROR;
}
memcpy(outputData + i * firstDimStride, inputData + firstDimStride * indicesData[i],
firstDimStride);
}
return NO_ERROR;
}
class CPUGatherCreator : public CPUBackend::Creator {
public:
virtual Execution *onCreate(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs,
const MNN::Op *op, Backend *backend) const override {
return new CPUGather(backend, op);
}
};
REGISTER_CPU_OP_CREATOR(CPUGatherCreator, OpType_Gather);
} // namespace MNN