1 //
2 //  ShapeBinaryOp.cpp
3 //  MNN
4 //
5 //  Created by MNN on 2019/01/10.
6 //  Copyright © 2018, Alibaba Group Holding Limited
7 //
8 
9 #include "shape/SizeComputer.hpp"
10 #include "core/Macro.h"
11 #include <vector>
12 namespace MNN {
13 class BinaryOpComputer : public SizeComputer {
14 public:
outputBool(int operation)15     static bool outputBool(int operation) {
16         if (operation == BinaryOpOperation_GREATER_EQUAL) {
17             return true;
18         }
19         if (operation == BinaryOpOperation_GREATER) {
20             return true;
21         }
22         if (operation == BinaryOpOperation_LESS) {
23             return true;
24         }
25         if (operation == BinaryOpOperation_LESS_EQUAL) {
26             return true;
27         }
28         if (operation == BinaryOpOperation_EQUAL) {
29             return true;
30         }
31         if (operation == BinaryOpOperation_NOTEQUAL) {
32             return true;
33         }
34         return false;
35     }
onComputeSize(const Op * op,const std::vector<Tensor * > & inputs,const std::vector<Tensor * > & outputs) const36     virtual bool onComputeSize(const Op* op, const std::vector<Tensor*>& inputs,
37                                const std::vector<Tensor*>& outputs) const override {
38         MNN_ASSERT(2 == inputs.size());
39         MNN_ASSERT(1 == outputs.size());
40         // set output type & format
41         auto input0 = inputs[0], input1 = inputs[1], output = outputs[0];
42         auto &buffer = output->buffer();
43         const auto opType = op->main_as_BinaryOp()->opType();
44         if (outputBool(opType)) {
45             buffer.type = halide_type_of<int32_t>();
46         } else {
47             buffer.type = input0->getType();
48         }
49         if (input0->getType().code != input1->getType().code) {
50             MNN_PRINT("Error for binary op: input0's type != input1's type\n");
51             return false;
52         }
53 
54         if (input0->dimensions() < input1->dimensions()) {
55             auto temp = input0;
56             input0 = input1;
57             input1 = temp;
58         }
59         TensorUtils::getDescribe(output)->dimensionFormat = TensorUtils::getDescribe(input0)->dimensionFormat;
60         return SizeComputer::computeBroadCastDims(op, inputs, outputs);
61     }
62 };
63 
64 REGISTER_SHAPE(BinaryOpComputer, OpType_BinaryOp);
65 } // namespace MNN
66