TbeAippConv2dAddRelu6MulMulFusionPass
Description
Fuses the following pattern structure into one operator.
Constraints
- For the Conv2D operator, fusion is disabled when strides is [1, 1], pad is [0, 0, 0, 0], and kernel is 1 x 1.
- Fusion is disabled when the AIPP resizing function is enabled.
- Fusion is disabled when the AIPP mode is set to dynamic.
- Fusion is disabled when the AIPP padding function is enabled.
- If DMA is enabled for the Conv2D operator, fusion is not supported.
- Fusion is canceled if the minimum tiling is provided and L1 still cannot accommodate the AIPP processing result.
- Fusion is canceled if the convolutional kernel height is less than or equal to the sum of the values of Conv2d pad and AIPP pad in either the upper or lower direction.
Applicable Products
Parent topic: UB Fusion Patterns