In order to calculate the cycle count of our C7x code on the SK-AM62A-LP board, we used an inbuilt "cycles()" function. However, we obtained a high MCPS value. Can anyone suggest an alternative method/step for calculating the MCPS of a C7x code using the CCS IDE?
Thanks,
Betsy Varughese
Yes, sure. In order to profile our code we followed the below methods,
1. Using the __time32() and __time64() functions in the time.h header file.
Observations:
From the header file itself, we understood that the resolution was only expressed in seconds (Screenshot attached below). So, In our audio frame processing, most of the frames are reported as 0's or 1's. Is there any way to express the time in microseconds or nanoseconds, so that we get more precise values?
2. Using the clock() : _CODE_ACCESS clock_t clock(void);
Observations:
1. Our team has noticed that there are multiple clock configuration setup options within CCS (screenshot attached). We are seeking guidance on which specific option would be most suitable for our profiling purposes. Additionally, we would greatly appreciate any documentation or resources that could provide further information on the topic.
2. The clocks consumed by cycles() itself are approximately 3000, which is a very large value.
3. Memory alignment has also been checked, and it is found to be normal
We have observed higher MCPS values in both cases, which is unexpected. Therefore, we are seeking guidance on the optimal method to conduct profiling of our c7x code using CCS on the AM62A-LP board. We would greatly appreciate any suggestions or recommendations that can help us address this issue effectively.
Note: I have attached the sample profiling code for your reference.
Clock configuration setup options available in CCS :
Thanks & Regards,
Betsy Varughese
