Hi,
Customer have inquiries regarding AM654x, details below.
"
Board is connected to Windows via Code Composer. Pdk 7.3.0.54 is used. The board is initiated via the gel script and the ccsLoadDmsc\\am65xx\\launch.js script.
I am trying to send data between two PRUs for a switching environment(64-1522 byte). There are multiple ways to do that.
I would like to achieve a low latency and possible a high bandwidth.
My current approach is to use the global DDR memory or write to the internal DRAM memory of each PRU directly.
For this approach I must use one or multiple of the following methods: Interrupts, spinlock and/or (cylce memory addresses).
What I like to do is to connect two devices via PSI-L proxy interface directly. This is already done by the dual-ethernet firmware to arm host.
From the datasheet setup configuration is done via CFG_PROXY at address 31F78000h. I tried to test how to set up a connection, but I couldn't write to these registers.
I would like to cfg it directly from PRU. But I also tried to set them by a baremetal implementation from R5F and A53 using this simple C code:
int volatile * const ptr = (int *) 0x0031F78100; //0x45b02024 //0x31F78100
*ptr = 0x00004104;
I don't know if I must change any firewall settings. I tried to set them but was also not able to change them.
My question therefore are the following:
Latency and Bandwidth wise what can I expect by using the PSI-L approach?
How good are multiple transmission at the same time handled?
And lastly how can I connect a source with a destination?
A side question I have is access to local DRAM (64kbyte) on every ICSSG. Can it be accessed locally via a data movement accelerator?
As I could not determine a real speed improvement to just accessing DDR.
"
Thank you in advance.
Regards,
Maynard