#pragma once
|
|
#include "caffe2/core/context.h"
|
#include "caffe2/core/logging.h"
|
#include "caffe2/core/operator.h"
|
#include "caffe2/operators/filler_op.h"
|
#include "caffe2/utils/cast.h"
|
#include "caffe2/utils/math.h"
|
|
namespace caffe2 {
|
|
template <class Context>
|
class GivenTensorByteStringToUInt8FillOp final : public FillerOp<Context> {
|
public:
|
USE_OPERATOR_CONTEXT_FUNCTIONS;
|
explicit GivenTensorByteStringToUInt8FillOp(const OperatorDef& operator_def, Workspace* ws)
|
: FillerOp<Context>(operator_def, ws) {
|
const ArgumentHelper helper(operator_def);
|
if (!helper.HasArgument("dtype")) {
|
Extract();
|
} else {
|
auto dtype = cast::GetCastDataType(helper, "dtype");
|
switch (dtype) {
|
case TensorProto_DataType_STRING:
|
Extract();
|
break;
|
case TensorProto_DataType_UNDEFINED:
|
CAFFE_THROW("Cannot have undefined 'dtype' argument");
|
default:
|
CAFFE_THROW("Unexpected 'dtype' argument value: ", dtype);
|
}
|
}
|
}
|
|
bool Fill(Tensor* output) override {
|
DCHECK_EQ(output->numel(), values_.numel())
|
<< "output size: " << output->numel()
|
<< " given size: " << values_.numel();
|
auto* data = output->template mutable_data<uint8_t>();
|
const uint8_t* values_data = values_.template data<uint8_t>();
|
if (output->numel()) {
|
context_.template CopySameDevice<uint8_t>(
|
output->numel(), values_data, data);
|
}
|
return true;
|
}
|
|
private:
|
void Extract() {
|
auto source_values = this->template GetRepeatedArgument<string>("values");
|
DCHECK_EQ(source_values.size(), 1)
|
<< "expected size: 1 "
|
<< " given size: " << source_values.size();
|
|
auto str = source_values[0];
|
ReinitializeTensor(&values_, {static_cast<int64_t>(str.size())}, at::dtype<uint8_t>().device(CPU));
|
uint8_t* values_data = values_.template mutable_data<uint8_t>();
|
for (int i = 0; i < str.size(); i++) {
|
values_data[i] = static_cast<uint8_t>(str[i]);
|
}
|
}
|
|
Tensor values_;
|
};
|
} // namespace caffe2
|
