I am researching on Memory systems in heterogeneous systems at Arizona State University. As far as ARM processor is concerned, there is Performance Monitoring Unit to monitor the impact of application/process on memory. But to have an actual overview, we need to consider the request both from GPU, AP and CP. As far as Mali is concerned, I am unable to find any registers or tracing methodology to know their memory request. Is there any methodology that can be implemented or simulated somehow or can be accessed through some development board.
Incase of any query or discrepancy, please let me know. I will be more than happy to answer.
Have a great day
Thanks and Warm Regards
Davesh Shingari
The Mali integration into the ARM DS-5 Streamline profiler includes many performance counters which can be used to profile various memory related performance counters.
HTH, Pete
Thanks for the reply.
Yes, Streamline provide counters to get info about the L2 cache miss and hit rate. But I am more specifically inclined towards the memory access pattern of the GPU i.e. the addresses accessed by the GPU.
This isn't something available on a development board (not from the GPU itself at least); the amount of data you would need to store would be huge - a typical SoC can consume between 6 and 12 GB/s of bandwidth so you'd need a very large, and very fast trace buffer.