32 bit arch optimization suggestion

[Morten Brørup]

I was wondering...

How widely is DPDK used with 32 bit architectures?

Both phys_addr_t and rte_iova_t are cast as uint64_t in rte_common.h. Should they be uint32_t on 32 bit arch?

If so, the rte_mbuf structure has 12 unused byte in the first cache line on 32 bit arch.

Now, moving the "next" (4 byte pointer) and "tx_offload" (8 byte flags) fields to the first cache line could improve performance significantly, because this modification would leave the second cache line unused in most cases.


Med venlig hilsen / Kind regards,
-Morten Brørup