Some questions about GPU acceleration

[summer-xrx]

Hello, I am very curious about GPU acceletarion. Now, I have the following questions:

1. Why does the concrete library support GPU acceleration, while other mainstream libraries such as google/fully-homomorphic-encryption, SEAL, OpenFHE, TenSEAL (based on SEAL), and secretflow do not support GPU acceleration?
2. Compared to the aforementioned libraries, what are the advantages of the concrete library?
3. Why can't GPU acceleration be implemented in those aforementioned libraries now? What are the technical challenges of using GPU to accelerate homomorphic encryption?
4. Why does the concrete library implement GPU acceleration while concrete-ml does not? Are there any technical challenges in implementing GPU acceleration in homomorphic machine learning?
   Looking forward to your answer, thank you!

[bcm-at-zama]

Dup of zama-ai/concrete-ml#839, we'll answer there