How to measure running time properly?

Support
1

I am building a Sobel filter by myself as an exercise.
Here is my block design

Screenshot%20from%202019-10-16%2022-52-53
Screenshot%20from%202019-10-16%2022-52-531139×198 22.1 KB

The input and output are 1080x1920 gray scale image. I use the High Performance Port to transfer image data, and use GPIO port to control the Sobel Filter IP.
Here is my port addresses
Screenshot%20from%202019-10-16%2023-06-37

On the python driver code, I firstly load the image to a continuous memory on DRAM, and pass read and write addresses to PL, then trigger the IP to run by hw_sobel.write(0x00, 0x1).

My problem is, I don’t know how to measure the running time of my design properly. I currently use %timeit hw_sobel.write(0x00, 0x1), but it gives me unreasonable result,

The slowest run took 4.26 times longer than the fastest. This could mean that an intermediate result is being cached. 10000 loops, best of 3: 29.8 µs per loop.

I think the reason is, hw_sobel.write(0x00, 0x1) is running in an asynchronized manner with IP operating. And I need some suggestions for measuring running time in such case.

Thanks.

2

hw_sobel.write(0x00, 0x1) is only doing an MMIO register write. This is what timeit will measure, not the time for your function to complete.

You would need to poll on the control register to check for ap_done.

Cathal

1 Like