xll_bit的博客

内存合并的基础是DRAM Burst。mei当我们从全局内存搬运数据到共享内存或者寄存器时，

前面已经写过了利用Fortran实现任意维度矩阵乘的CUDA实现，详见：https://blog.youkuaiyun.com/xll_bit/article/details/117551476?spm=1001.2014.3001.5501，今天更新一个

module reverse_m implicit none integer, device :: n_dcontains attributes(global) subroutine staticReverse(a, d) real :: d(:), a(:), result integer :: t, tr, t1, j, sa integer,parameter :: ra = 3 real, shared :: s(-ra+1:64 + ra) .

之前写过一篇矩阵乘，但只能是方阵，详见https://blog.youkuaiyun.com/xll_bit/article/details/103271788?spm=1001.2014.3001.5501这次

在学习CUDA的过程中，免不了要实现矩阵乘，从而对GPU的运作机制以及如何使用share memory有更深的理解。下面是用FORTRAN对简单矩阵乘和利用分块并将数据放入share memory实现矩阵乘的实现。! this program demonstates various memory optimzation techniques! applied to a matrix mul...