Skip to content

MLBEDSW-10832: Align ElementwiseMul quantization scaling with reference

Max Bergfelt requested to merge MLBEDSW-10832 into main

Adjusted quantization scaling of ElementwiseMul to bit-exactly match the scaling calculation of QuantizedConvolutionMultipler in the reference. Scaling of BatchMatMul is consequently aligned.

Change-Id: I00cc55d04ed84acdd8785e9f3610f6198423c65e Signed-off-by: Max Bergfelt max.bergfelt@arm.com

Merge request reports