Why dense is not optimized for arm cpu?

Questions
#1

https://github.com/apache/tvm/blob/main/python/tvm/relay/op/strategy/arm_cpu.py#L507

I want to port BERT to Pixel6 CPU with AutoTVM. But I found that the dense operator is not even optimized for arm CPUs. It’s so weird that the most commonly seen dense operator is not optimized. I would like to ask that is there any way to enable dense operator on arm CPU with AutoTVM?

1 Like
#2

It’s already enabled, it’s just that nobody has looked at dense performance on ARM in detail.

1 Like