Question

使用名为MARS 4.5的模拟器我试图提高此代码的缓存性能。这是汇编程序的子部分，使用Sieve of Eratosthenes算法计算素数。

由于某种原因，sw（存储字）的缓存命中率为25％，其中程序的其余部分在其当前状态下平均为约50％。我已经尝试重新排列一些东西，但我无法弄清楚是什么导致了这个瓶颈。需要做些什么来提高缓存命中率？

inner:  add $t2, $s2, 0 # save the bottom of stack address to $t2
        mul $t3, $t1, 4 # calculate the number of bytes to jump over
        sub $t2, $t2, $t3   # subtract them from bottom of stack address
        add $t2, $t2, 8 # add 2 words - we started counting at 2!

        sw  $s0, ($t2)  # store 1's -> it's not a prime number!

        add $t1, $t1, $t0   # do this for every multiple of $t0
        bgt $t1, $t9, outer # every multiple done? go back to outer loop

        j   inner       # some multiples left? go back to inner loop

Answer 1

我能够通过修改程序来存储字节而不是单词来解决这个问题。这增加了缓存中的存储块数量，从而提高了命中率。

inner:  add $t2, $s2, 0 # save the bottom of stack address to $t2
    addi $t3, $t1, 1 # add one byte
    sub $t2, $t2, $t3   # subtract them from bottom of stack address
    add $t2, $t2, 2 # add 2 bytes - we started counting at 2!

    sb  $s0, ($t2)  # store 1's -> it's not a prime number!

    add $t1, $t1, $t0   # do this for every multiple of $t0
    bgt $t1, $t9, outer # every multiple done? go back to outer loop

    j   inner       # some multiples left? go back to inner loop

如何提高此MIPS代码的缓存性能

1 个答案: