ucs_status_t uct_rc_verbs_ep_am_zcopy(uct_ep_h tl_ep, uint8_t id, const void *header,
unsigned header_length, const uct_iov_t *iov,
size_t iovcnt, uct_completion_t *comp)
{
uct_rc_verbs_iface_t *iface = ucs_derived_of(tl_ep->iface, uct_rc_verbs_iface_t);
uct_rc_verbs_ep_t *ep = ucs_derived_of(tl_ep, uct_rc_verbs_ep_t);
uct_rc_iface_send_desc_t *desc = NULL;
struct ibv_sge sge[UCT_IB_MAX_IOV]; /* First sge is reserved for the header */
struct ibv_send_wr wr;
int send_flags;
size_t sge_cnt;
UCT_CHECK_IOV_SIZE(iovcnt, uct_ib_iface_get_max_iov(&iface->super.super) - 1,
"uct_rc_verbs_ep_am_zcopy");
UCT_RC_CHECK_AM_ZCOPY(id, header_length, uct_iov_total_length(iov, iovcnt),
iface->verbs_common.config.short_desc_size,
iface->super.super.config.seg_size);
UCT_RC_CHECK_RES(&iface->super, &ep->super);
UCT_RC_CHECK_FC_WND(&iface->super, &ep->super, id);
UCT_RC_IFACE_GET_TX_AM_ZCOPY_DESC(&iface->super, &iface->verbs_common.short_desc_mp,
desc, id, header, header_length, comp, &send_flags);
sge[0].length = sizeof(uct_rc_hdr_t) + header_length;
sge_cnt = uct_ib_verbs_sge_fill_iov(sge + 1, iov, iovcnt);
UCT_RC_VERBS_FILL_AM_ZCOPY_WR_IOV(wr, sge, (sge_cnt + 1), wr.opcode);
UCT_TL_EP_STAT_OP(&ep->super.super, AM, ZCOPY,
(header_length + uct_iov_total_length(iov, iovcnt)));
uct_rc_verbs_ep_post_send_desc(ep, &wr, desc, send_flags);
UCT_RC_UPDATE_FC_WND(&iface->super, &ep->super, id);
return UCS_INPROGRESS;
}
ucs_status_t uct_rc_verbs_ep_tag_eager_zcopy(uct_ep_h tl_ep, uct_tag_t tag,
uint64_t imm, const uct_iov_t *iov,
size_t iovcnt, uct_completion_t *comp)
{
uct_rc_verbs_iface_t *iface = ucs_derived_of(tl_ep->iface, uct_rc_verbs_iface_t);
uct_rc_verbs_ep_t *ep = ucs_derived_of(tl_ep, uct_rc_verbs_ep_t);
uct_rc_iface_send_desc_t *desc;
struct ibv_sge sge[UCT_IB_MAX_IOV];
struct ibv_send_wr wr;
int send_flags;
size_t sge_cnt;
uint32_t app_ctx;
UCT_CHECK_IOV_SIZE(iovcnt, 1ul, "uct_rc_verbs_ep_tag_eager_zcopy");
UCT_RC_CHECK_ZCOPY_DATA(iface->tm.eager_hdr_size,
uct_iov_total_length(iov, iovcnt),
iface->super.super.config.seg_size);
UCT_RC_CHECK_RES(&iface->super, &ep->super);
sge_cnt = uct_ib_verbs_sge_fill_iov(sge + 1, iov, iovcnt);
UCT_RC_VERBS_FILL_TM_IMM(wr, imm, app_ctx);
UCT_RC_VERBS_GET_TM_ZCOPY_DESC(iface, &iface->verbs_common.short_desc_mp,
desc, tag, app_ctx, comp, &send_flags, sge[0]);
wr.num_sge = sge_cnt + 1;
wr.sg_list = sge;
uct_rc_verbs_ep_post_send_desc(ep, &wr, desc, send_flags);
return UCS_INPROGRESS;
}
static inline ucs_status_t
uct_rc_verbs_ep_rdma_zcopy(uct_rc_verbs_ep_t *ep, const uct_iov_t *iov,
size_t iovcnt, uint64_t remote_addr, uct_rkey_t rkey,
uct_completion_t *comp, int opcode)
{
uct_rc_verbs_iface_t *iface = ucs_derived_of(ep->super.super.super.iface,
uct_rc_verbs_iface_t);
struct ibv_sge sge[UCT_IB_MAX_IOV];
struct ibv_send_wr wr;
size_t sge_cnt;
UCT_RC_CHECK_RES(&iface->super, &ep->super);
sge_cnt = uct_ib_verbs_sge_fill_iov(sge, iov, iovcnt);
UCT_SKIP_ZERO_LENGTH(sge_cnt);
UCT_RC_VERBS_FILL_RDMA_WR_IOV(wr, wr.opcode, opcode, sge, sge_cnt, remote_addr, rkey);
wr.next = NULL;
uct_rc_verbs_ep_post_send(iface, ep, &wr, IBV_SEND_SIGNALED);
uct_rc_txqp_add_send_comp(&iface->super, &ep->super.txqp, comp, ep->txcnt.pi);
return UCS_INPROGRESS;
}
