矩阵乘法是科学计算中最基本的操作,高效实现矩阵乘法可以加速许多应用。本文使用NVIDIA的CUDA在GPU上实现了一个高效的矩阵乘法。测试结果表明,在Geforce GTX260上,本文提出的矩阵乘法的速度是理论峰值的97%,跟CUBLAS库中的矩阵乘法相当。
Matrix multiplication is a basic operation in scientific computing. Efficient implementation of matrix multiplication can speed up many applications. In this paper, we implement an efficient matrix multiplication on GPU using NVIDINs CUDA. The experiment shows that our implementation is as fast as the implementation in CUBLAS, and the speed of our implementation can reach the peak speed's 97%, on Geforce GTX260.