#optimized #multiplication #x86-64 #multiscalar #pasta #cuda #aarch64

pasta-msm

Optimized multiscalar multiplicaton for Pasta moduli for x86_64 and aarch64

5 releases

0.1.4 May 5, 2023
0.1.3 Aug 10, 2022
0.1.2 Jun 26, 2022
0.1.1 May 26, 2022
0.1.0 May 19, 2022

#1 in #pasta

Download history 126/week @ 2024-06-13 146/week @ 2024-06-20 87/week @ 2024-06-27 41/week @ 2024-07-04 128/week @ 2024-07-11 100/week @ 2024-07-18 97/week @ 2024-07-25 128/week @ 2024-08-01 115/week @ 2024-08-08 136/week @ 2024-08-15 111/week @ 2024-08-22 101/week @ 2024-08-29 102/week @ 2024-09-05 112/week @ 2024-09-12 190/week @ 2024-09-19 137/week @ 2024-09-26

552 downloads per month
Used in 7 crates (4 directly)

Apache-2.0

17KB
323 lines

Pasta Multi-Scalar Multiplication

This is an initial version with a list of planned improvements:

  • parallelize;
  • break down scalars to signed digits to half the buckets' integration complexity;
  • switch to alternative bucket point representation with faster addition formula;
  • migrate CUDA implementation;

To compile CUDA support ensure that you have nvcc, Nvidia CUDA compiler, on your program search path. Minimal installation suffices. For example on Ubuntu it would be sufficient to install cuda-minimal-build-11-7 instead of complete cuda package. If your laptop is equipped with a Turing+ controller, you're likely to have to compile with --features=cuda-mobile. Caveat lector. CUDA implementation does not adapt for the actual load yet, so that some results would be suboptimal.

Dependencies

~1.2–2.4MB
~51K SLoC