优化蛮力矩阵向量乘法

Question

矩阵是已知的编译时间。相邻行之间的变化很小。计算它的目的是找到最大索引。

我想在 gpu 上并行化它。分解会有帮助吗？

这里是

N=16

的矩阵。我的实际大小是

N=1024

（零真的是

-1

s）：

1100110011001100
1000110011001100
1001110011001100
1001100011001100
1001100111001100
1001100110001100
0001100110001100
0011100110011100
0011000110011100
0011001110011100
0111001110011100
0110001100011000
1110001100011000
1110011100011000
1100011100011000
1100011100111000
1100011000111000
1100111000111000
1000111000111000
0000111000111000
0001111000111000
0001110000111000
0001110001111000
0011110001111000
0011100001110000
0111100001110000
0111100011110000
0111000011110000
1111000011110000
1110000011110000
1110000111110000
1100000111100000
1100001111100000
1000001111100000
0000001111100000
0000011111100000
0000011111000000
0000111111000000
0001111111000000
0001111110000000
0011111110000000
0111111110000000
0111111100000000
1111111100000000