ReshapeTransposeFusionPass

Fuses the Reshape and TransposeD operators that fit the graph fusion pattern into the ConfusionTransposeD operator.

Restrictions on the Reshape node:

Fusion is not performed if the input shape is dynamic.
Fusion is not performed if the input dimension is 1 and the shape is dynamic.
Fusion is not performed if the input dimension is 1 and the shape is not divisible by 16.
Fusion is not performed if the input dimension is greater than or equal to 2 and either of the last two dimensions has a dynamic shape.
Fusion is not performed if the input dimension is greater than or equal to 2 and either of the last two dimensions is not divisible by 16.
Fusion is not performed if the input is an empty tensor.

Restrictions on TransposeD:

Supported input data types: float16, float32, int8, int16, int32, int64, uint8, uint16, uint32, and uint64.
The input dimension must not exceed 8, and the input shape must match the output shape.

Fusion is not performed if the output shape is dynamic.
Fusion is not performed if the output dimension is 1 and the shape is dynamic.
Fusion is not performed if the output dimension is 1 and the shape is not divisible by 16.
Fusion is not performed if the output dimension is greater than or equal to 2 and either of the last two dimensions has a dynamic shape.
Fusion is not performed if the output dimension is greater than or equal to 2 and either of the last two dimensions is not divisible by 16.

Atlas A2 training product/Atlas A2 inference product

Parent topic: Graph Fusion Patterns