一种多标量乘法计算方法、装置、计算机设备及程序产品
By implementing a distributed shared memory mechanism, efficient computation of multi-scalar multiplication is achieved in GPUs or GPU clusters, solving the problem of time-consuming multi-scalar multiplication in zero-knowledge proofs and improving computation speed.
CN120104317BActive Publication Date: 2026-07-17ZHEJIANG ANT SECRET TECH CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG ANT SECRET TECH CO LTD
- Filing Date
- 2025-02-12
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
In existing zero-knowledge proofs, multi-scalar multiplication operations are time-consuming, becoming the main bottleneck in the computational speed of zero-knowledge proofs.
Method used
By employing a distributed shared memory mechanism, global offsets are calculated through counter aggregation between thread blocks, avoiding global atomic operations and achieving efficient bucketed storage and computation of point sequences and scalar sequences.
Benefits of technology
It accelerates the execution speed of multiscalar multiplication, reduces the impact of global atomic operations, and improves the computational efficiency of GPUs or GPU clusters.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN120104317B_ABST
Abstract
本说明书提供了一种多标量乘法计算方法。多标量乘法涉及的点序列和标量序列一一对位相乘。该方法应用于目标GPU，目标GPU部署有多个线程块，每个线程块负责了点序列划分成的多个子序列中的一个子序列。方法执行过程中，任意的第一线程块确定分配的第一子序列中的各个点对应的分桶，并针对各个分桶，确定第一子序列中落入该分桶的点的数量和。各个线程块基于分布式共享内存机制，针对同一标量值对应的、来自不同线程块的分桶，聚合不同的计数器，得到各个线程块的各个分桶在目标全局内存中的全局偏移量。并根据全局偏移量，将点存储在全局内存中，并基于上述分桶结果执行计算操作，得到多标量乘法的计算结果。无需全局原子操作，实现了加速。
Need to check novelty before this filing date? Find Prior Art