SameInputConv2dFixpipePass

Description

Fuses multiple Conv2D+AscendDequant+Relu+AscendQuant operators into one Conv2D+AscendDequant+Relu+AscendQuant+Split operator.

Before: After:

Restrictions

• The specifications and attributes of the first Conv2D (conv2d_0 and conv2d_1 in the figure) of each channel must be the same, and the value of output_channel must be a multiple of 32.
• Only the static input shape (fmap, filter, bias) is supported.
• Only the first Conv2D groups of each channel can be 1.

Availability

Atlas 200I/500 A2 inference product

Atlas A2 training product/Atlas A2 inference product

Atlas A3 training product/Atlas A3 inference product