The 04/22/2020 12:19, Will Deacon wrote:
>
> I wrote the silly harness below for the snippets given in [1] but I can't
> see any difference between the forwards and backwards versions on any arm64
> systems I have access to.
>
> Will
>
> --->8
>
> #include <errno.h>
> #include <stdio.h>
> #include <stdlib.h>
> #include <string.h>
> #include <time.h>
> #include <unistd.h>
>
> [...]
> int main(void)
> {
> void *buf;
> unsigned long long delta;
> struct timespec ts_start, ts_end;
>
> if (posix_memalign(&buf, PAGE_SIZE, BUF_SZ)) {
> perror("posix_memalign()");
> return -1;
> }
>
> memset(buf, 0xd, BUF_SZ);
>
> [...]
With this exact test code I also didn't observe any significant difference between
forward and backwards versions on SM8150.
------------------------------------------------------------------------------
Output on 8150 under controlled conditions(CPU0 & 6 turned on, CPUs set to max
frequency and DDR set to performance governor):
------------------------------------------------------------------------------
Forwards: took 0.319658 seconds
Backwards: took 0.320983 seconds
------------------------------------------------------------------------------
But when I used malloc instead of posix_memalign because that was the big difference
between this and our test code, I observed significant difference between
forward and backwards version on SM8150.
------------------------------------------------------------------------------
Output on 8150 under controlled conditions(CPU0 & 6 turned on, CPUs set to max
frequency and DDR set to performance governor):
------------------------------------------------------------------------------
Forwards: took 0.323157 seconds
Backwards: took 0.581638 seconds
------------------------------------------------------------------------------
I don't know the implementation differences between posix_memalign and
malloc which might lead to these results.
--
Prathu Baronia
OnePlus RnD
