ncnn/docs/developer-guide/binaryop-broadcasting.md

• broadcasting rule

broadcasting rule

ncnn BinaryOp accepts blobs with different shape

C = BinaryOp(A, B)

shape notation convention is [w], [w,h], [w,h,c], [w,h,d,c]

• binaryop with scalar and scalar-like

type	A	B	C
0	[2]	scalar / [1]	[2]
1	[2,3]	scalar / [1] / [1,1]	[2,3]
2	[2,3,4]	scalar / [1] / [1,1] / [1,1,1]	[2,3,4]
3	[2,3,4,5]	scalar / [1] / [1,1] / [1,1,1] / [1,1,1,1]	[2,3,4,5]

• no broadcast

type	A	B	C
4	[2]	[2]	[2]
5	[2,3]	[2,3]	[2,3]
6	[2,3,4]	[2,3,4]	[2,3,4]
7	[2,3,4,5]	[2,3,4,5]	[2,3,4,5]

• broadcast B for inner axis

type	A	B	C
8	[2,3]	[3] / [1,3]	[2,3]
9	[2,3,4]	[4] / [1,1,4]	[2,3,4]
10	[2,3,4]	[3,4] / [1,3,4]	[2,3,4]
11	[2,3,4,5]	[5] / [1,1,1,5]	[2,3,4,5]
12	[2,3,4,5]	[4,5] / [1,1,4,5]	[2,3,4,5]
13	[2,3,4,5]	[3,4,5] / [1,3,4,5]	[2,3,4,5]

• broadcast B for outer axis

type	A	B	C
14	[2,3]	[2,1]	[2,3]
15	[2,3,4]	[2,1,1]	[2,3,4]
16	[2,3,4]	[2,3,1]	[2,3,4]
17	[2,3,4,5]	[2,1,1,1]	[2,3,4,5]
18	[2,3,4,5]	[2,3,1,1]	[2,3,4,5]
19	[2,3,4,5]	[2,3,4,1]	[2,3,4,5]

• some special broadcasting rule exists for model compatibility

special type	A	B	C
20	[2,3,4]	[2,1,4]	[2,3,4]