mirror of https://github.com/alibaba/MNN.git
56 lines
1.7 KiB
C++
56 lines
1.7 KiB
C++
//
|
|
// CPUQuantizedReshape.cpp
|
|
// MNN
|
|
//
|
|
// Created by MNN on 2018/08/09.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include "CPUQuantizedReshape.hpp"
|
|
#include "CommonOptFunction.h"
|
|
#include "Macro.h"
|
|
|
|
namespace MNN {
|
|
|
|
CPUQuantizedReshape::CPUQuantizedReshape(const MNN::Op *op, Backend *b) : MNN::Execution(b) {
|
|
auto param = op->main_as_QuantizedReshape();
|
|
mIstflite = param->modelFormat() == MNN::ModeFormat_TFLITE;
|
|
}
|
|
|
|
ErrorCode CPUQuantizedReshape::onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
|
|
return NO_ERROR;
|
|
}
|
|
|
|
ErrorCode CPUQuantizedReshape::onExecute(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
|
|
MNN_ASSERT(3 == inputs.size() || 4 == inputs.size() || 1 == inputs.size());
|
|
MNN_ASSERT(3 == outputs.size() || inputs.size() == 1);
|
|
|
|
auto &input = inputs[0]->buffer();
|
|
auto &output = outputs[0]->buffer();
|
|
|
|
MNN_ASSERT(input.dimensions <= 1 || input.dim[1].flags == 0);
|
|
|
|
if (input.dimensions <= 1 || input.dim[1].flags == 0) {
|
|
::memcpy(output.host, input.host, inputs[0]->size());
|
|
}
|
|
|
|
if (mIstflite == false) {
|
|
((float *)(outputs[1]->buffer().host))[0] = inputs[2]->host<float>()[0];
|
|
((float *)(outputs[2]->buffer().host))[0] = inputs[3]->host<float>()[0];
|
|
}
|
|
|
|
return NO_ERROR;
|
|
}
|
|
|
|
class CPUQuantizedReshapeCreator : 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 CPUQuantizedReshape(op, backend);
|
|
}
|
|
};
|
|
|
|
REGISTER_CPU_OP_CREATOR(CPUQuantizedReshapeCreator, OpType_QuantizedReshape);
|
|
|
|
} // namespace MNN
|