Designing Fusion Patterns

Designing the Fusion Solution

The following uses the many-to-one fusion pattern DecodeBboxV2ScopeFusionPass as an example to describe how to design and develop a scope fusion pattern.

This fusion pattern aims to fuse all small operators under Decode scope into a large operator DecodeBboxV2, as shown in Figure 1.

Figure 1 Fusion diagram

Specifying the Prototype of the Fused Operator

The prototype of the DecodeBboxV2 fused operator is as follows.

1
2
3
4
5
6
7
8

REG_OP(DecodeBboxV2)
    .INPUT(boxes, TensorType({DT_FLOAT16,DT_FLOAT}))
    .INPUT(anchors, TensorType({DT_FLOAT16,DT_FLOAT}))
    .OUTPUT(y, TensorType({DT_FLOAT16,DT_FLOAT}))
    .ATTR(scales, ListFloat, {1.0, 1.0, 1.0, 1.0})
    .ATTR(decode_clip, Float, 0.0)
    .ATTR(reversed_box, Bool, false)
    .OP_END_FACTORY_REG(DecodeBboxV2)

Specifying the Fusion Relationship

As shown in Figure 1:

• Input 0 of transpose (corresponding to Cast_1 in the above figure) is input 0 (boxes) of DecodeBboxV2.
• Input 0 of get_center_coordinates_and_sizes_transpose (corresponding to Cast0 in the above figure) is input 1 (anchors) of DecodeBboxV2.
• The output of transpose_1 (corresponding to Cast_2 in the above figure) is used as output y of DecodeBboxV2.