On Wed, Apr 08, 2026 at 04:34:32PM +0100, Bruce Richardson wrote: > On Mon, Apr 06, 2026 at 10:11:25AM +0530, Shaiq Wani wrote: > > The IDPF split-queue completion queue was using ci_tx_queue, a structure > > designed for TX descriptor queues, wasting ~96 bytes per completion > > queue. Additionally, the CQ-only fields (compl_ring, txqs, tx_start_qid, > > expected_gen_id) bloated ci_tx_queue for every other Intel driver that > > shares it. > > > > Introduce struct idpf_complq with exactly the fields needed by the > > completion queue. This brings the CQ allocation down from ~150 bytes > > to ~48 bytes. > > > > Suggested-by: Bruce Richardson <[email protected]> > > Signed-off-by: Shaiq Wani <[email protected]> > > --- > > drivers/net/intel/common/tx.h | 11 +++-------- > > drivers/net/intel/cpfl/cpfl_ethdev.h | 2 +- > > drivers/net/intel/cpfl/cpfl_rxtx.c | 11 ++++++----- > > drivers/net/intel/idpf/idpf_common_rxtx.c | 4 ++-- > > drivers/net/intel/idpf/idpf_common_rxtx.h | 16 +++++++++++++++- > > drivers/net/intel/idpf/idpf_common_rxtx_avx2.c | 4 ++-- > > drivers/net/intel/idpf/idpf_common_rxtx_avx512.c | 4 ++-- > > drivers/net/intel/idpf/idpf_rxtx.c | 2 +- > > 8 files changed, 32 insertions(+), 22 deletions(-) > > > > diff --git a/drivers/net/intel/common/tx.h b/drivers/net/intel/common/tx.h > > index 283bd58d5d..9da9366046 100644 > > --- a/drivers/net/intel/common/tx.h > > +++ b/drivers/net/intel/common/tx.h > > @@ -114,6 +114,7 @@ struct ci_tx_desc { > > > > /* forward declaration of the common intel (ci) queue structure */ > > struct ci_tx_queue; > > +struct idpf_complq; > > > > Minor nit: You probably want to have a "struct idpf_flex_tx_sched_desc;" line > here > too, since you reference that struct below also. > If no other issues with this, I can make that change on apply.
Actually, since no other pointer types within the tx_queue struct have early references, we can instead just remove the new complq one too for consistency. > > Acked-by: Bruce Richardson <[email protected]> > Applied to dpdk-next-net-intel. Thanks, /Bruce
