In the layers.pyConv2DBatchnorm makes a transpose if it is the resource strategy, but only for Vivado and VivadoAccelerator backends. Note that Conv2DBatchnorm inherits from Conv2D, and Conv2D makes the same transpose in:
I think therefore the transpose is doubly-applied. I am therefore removing the backend-specific transpose in the layers.py, which regardless does not belong there.
Type of change
[x] Bug fix (non-breaking change that fixes an issue)
Tests
I assume there's a Conv2DBatchnorm test that will catch this change, but maybe not.
Description
In the
layers.py
Conv2DBatchnorm
makes a transpose if it is the resource strategy, but only forVivado
andVivadoAccelerator
backends. Note thatConv2DBatchnorm
inherits fromConv2D
, andConv2D
makes the same transpose in:https://github.com/fastmachinelearning/hls4ml/blob/main/hls4ml/backends/vivado/passes/resource_strategy.py#L29-L32
I think therefore the transpose is doubly-applied. I am therefore removing the backend-specific transpose in the
layers.py
, which regardless does not belong there.Type of change
Tests
I assume there's a
Conv2DBatchnorm
test that will catch this change, but maybe not.Test Configuration:
Checklist
pre-commit
on the files I edited or added.