• Locked Locked
  • Replies 4 replies
  • Subscribers 119 subscribers
  • Views 3668 views
  • Users 0 members are here
2025 survey

We are running a survey to help us improve the experience for all of our members. If you see the survey appear, please take the time to tell us about your experience if you can.

Options
Related
How was your experience today?
This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Questions about code benchmark difference between A8 and A15

Adam Yao
Note: This was originally posted on 1st May 2013 at http://forums.arm.com

[size="3"]Hi all,
     I tested some protocol stack code's benchmark in Cortex A8 and A15. The code is same and I enabled L1/L2 cache and branch prediction in both CPUs (I placed all the codes and data in DDR and configurated DDR's MMU attribute to WB&WA for both L1 and L2 cache). Basicly A15's result is bettern than A8, but there is something I can not explain. The resut is as below:
                                                                    A8                                     A15
Overall cycles                                           71821                              30057
L1I cache miss                                           469                                   431
L1D cache miss                                      1015                                   139
L2 cache read miss                                    78                                    584
L2 cache write miss                                  247                                      26
mis-predicted/no predicted branch        469                                     217
predicted branch                                     1015                                 11385

What confused me are:
[/size](1) Why A15's L1D cache miss number is much lower than A8, both CPU's L1D cache is 32K? So what cause A15's L1D performance improvement?
(2) Why A15's L2 cache read miss number is much higher than A8?
(3) For branch prediction, why the sum of predicted and not predicted branch number is not equal in A8 and A15 and why A15's number is about 10 times A8's number?

Wait for a reasonable explanation.
Thank you.
Parents
  • Note: This was originally posted on 3rd May 2013 at http://forums.arm.com

    Hello,
         Thanks a lot for your reply.
          I run the test code only in one core of A15 cluster. After some investigation, I have some dobut about the PMU event nubmer of my test code. Do you know whether the PMU event number is the same for A8 and A15. I can find the event number in A15's ARM TRM, e.g. 0x52 for L2 cache read miss, 0x53 for L2 cache write miss, I can not find the similar event number in A8, so I use the same number as A15.
          Will it cause problem?

    Thank you.
Reply
  • Note: This was originally posted on 3rd May 2013 at http://forums.arm.com

    Hello,
         Thanks a lot for your reply.
          I run the test code only in one core of A15 cluster. After some investigation, I have some dobut about the PMU event nubmer of my test code. Do you know whether the PMU event number is the same for A8 and A15. I can find the event number in A15's ARM TRM, e.g. 0x52 for L2 cache read miss, 0x53 for L2 cache write miss, I can not find the similar event number in A8, so I use the same number as A15.
          Will it cause problem?

    Thank you.
Children
No data