[dpdk-dev] [PATCH v3 0/7] Performance optimizations for mlx5 and mlx4

Bruce Richardson bruce.richardson at intel.com
Wed Mar 9 17:28:09 CET 2016


On Thu, Mar 03, 2016 at 03:27:10PM +0100, Adrien Mazarguil wrote:
> This patchset improves the mlx5 PMD performance by doing better prefetching,
> by reordering internal structure fields and by removing a few unnecessary
> operations.
> 
> Note: should be applied after "Add flow director and RX VLAN stripping
> support" to avoid conflicts.
> 
> Changes in v3:
> - None, submitted again due to dependency with previous patchset.
> 
> Changes in v2:
> - Rebased patchset on top of dpdk-next-net/rel_16_04.
> - Fixed missing update for receive function in rxq_rehash().
> - Added a commit to register memory on page boundaries instead of mempool
>   object boundaries for better performance (mlx4 and mlx5).
> 
> Adrien Mazarguil (1):
>   mlx: use aligned memory to register regions
> 
> Nelio Laranjeiro (6):
>   mlx5: prefetch next TX mbuf header and data
>   mlx5: reorder TX/RX queue structure
>   mlx5: remove one indirection level from RX/TX functions
>   mlx5: process offload flags only when requested
>   mlx5: avoid lkey retrieval for inlined packets
>   mlx5: free buffers immediately after completion
>
Applied to dpdk-next-net/rel_16_04

/Bruce



More information about the dev mailing list