Hi,
We have observed that both read and write access to shared memory is costly both in terms of time and CPU usage. We have seen this issue in two scenarios
1. We have some data on the shared buffer, we are trying to parse this buffer to perform some operations. We see that this operation takes around 21ms and 50% of the CPU. Where as if we were to memcopy the shared buffer to a linux buffer, and do the same parsing, it takes around 2ms and 2-3% of the CPU.
2. When we use shared memory buffer for the h264 decoder (our own decoder running on ARM) output, we see that the decode time is around 150ms. Where as if linux buffers are used for decoder output, we see that it operates at sub 16ms range.
This looks like some cashing issue. Can you please help us understand why is this difference seen?
Regards,
Apoorva