ucs_status_t uct_dc_mlx5_ep_tag_eager_zcopy(uct_ep_h tl_ep, uct_tag_t tag,
uint64_t imm, const uct_iov_t *iov,
size_t iovcnt, unsigned flags,
uct_completion_t *comp)
{
uct_dc_mlx5_iface_t *iface = ucs_derived_of(tl_ep->iface, uct_dc_mlx5_iface_t);
uct_dc_mlx5_ep_t *ep = ucs_derived_of(tl_ep, uct_dc_mlx5_ep_t);
uint32_t app_ctx, ib_imm;
int opcode;
UCT_CHECK_IOV_SIZE(iovcnt, UCT_RC_MLX5_TM_EAGER_ZCOPY_MAX_IOV(UCT_IB_MLX5_AV_FULL_SIZE),
"uct_dc_mlx5_ep_tag_eager_zcopy");
UCT_RC_CHECK_ZCOPY_DATA(sizeof(struct ibv_exp_tmh),
uct_iov_total_length(iov, iovcnt),
iface->super.super.super.config.seg_size);
UCT_DC_MLX5_CHECK_RES(iface, ep);
UCT_RC_MLX5_FILL_TM_IMM(imm, app_ctx, ib_imm, opcode, MLX5_OPCODE_SEND, _IMM);
uct_dc_mlx5_iface_zcopy_post(iface, ep, opcode|UCT_RC_MLX5_OPCODE_FLAG_TM,
iov, iovcnt, 0, "", 0, 0, 0, tag, app_ctx,
ib_imm, comp, MLX5_WQE_CTRL_SOLICITED);
UCT_TL_EP_STAT_OP(&ep->super, TAG, ZCOPY,
uct_iov_total_length(iov, iovcnt));
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;
}
