cuda-fixnum
CSIRO RDS Repository
View Archive InfoField | Value | |
Title |
cuda-fixnum
|
|
Creator |
Hamish Ivey-Law
|
|
Subject |
Applied Discrete Mathematics
Data Encryption Pure Mathematics not elsewhere classified |
|
Description |
cuda-fixnum is a fixed-precision SIMD library that targets CUDA. It provides the apparatus necessary to easily create efficient functions that operate on vectors of n-bit integers, where n can be much larger than the size of a usual machine or device register. Currently supported values of n are 32, 64, 128, 256, 512, 1024, and 2048 (larger values will be possible in a forthcoming release). The primary use case for fast arithmetic of numbers in the range covered by cuda-fixnum is in cryptography and computational number theory; in particular it can form an integral part in accelerating homomorphic encryption primitives as used in privacy-preserving machine learning. As such, special attention is given to support modular arithmetic; this is used in an example implementation of the Paillier additively homomorphic encryption scheme and of elliptic curve scalar multiplication. Future releases will provide additional support for operations useful to implementing Ring-LWE-based somewhat homomorphic encryption schemes. |
|
Publisher |
CSIRO
|
|
Date |
2018-08-13
|
|
Type |
—
|
|
Format |
—
|
|
Identifier |
csiro:26753
|
|
Language |
—
|
|
Coverage |
—
|
|
Rights |
—
|
|