I'm running into cases where I have to allgather extra bytes(say 4 bytes), which make the data not perfectly 32byte or 64byte aligned. While doing this, substantial ...

Due to this byte alignment rule, case 2 does not support zero-copy, but case 1 does.

I can do this statically using an '__align(8)' compiler specific directive, but that doesn't really work well for the dynamic objects I'm creating. The alignment is forced by the co-processor I'm ...