by support »
Hello,
The 32-bit size of the DWORD is related to the TLP formation. That has to do with how the data runs on the actual bus.
As for the 128-bit endpoint, it's no different from any other endpoint. It's just that its interface with the user logic has a 128-bit wide data vector in order to facilitate data transfers at very high rates.
So it all boils down to how those DWORDs are packed in the 128-bit vector, which differs from one PCIe front end to another (and is often configurable). Please refer to your PCIe block's user manual for that piece of info.
Regards,
Eli