uct.h

#ifndef UCT_H_
#define UCT_H_

#include <uct/api/uct_def.h>
#include <uct/api/tl.h>
#include <uct/api/version.h>
#include <ucs/async/async_fwd.h>
#include <ucs/config/types.h>
#include <ucs/datastruct/callbackq.h>
#include <ucs/type/status.h>
#include <ucs/type/thread_mode.h>
#include <ucs/type/cpu_set.h>
#include <ucs/stats/stats_fwd.h>
#include <ucs/sys/compiler_def.h>

#include <sys/socket.h>
#include <stdio.h>
#include <sched.h>

BEGIN_C_DECLS

typedef struct uct_md_resource_desc {
    char                     md_name[UCT_MD_NAME_MAX]; 
} uct_md_resource_desc_t;


typedef enum {
    UCT_DEVICE_TYPE_NET,     
    UCT_DEVICE_TYPE_SHM,     
    UCT_DEVICE_TYPE_ACC,     
    UCT_DEVICE_TYPE_SELF,    
    UCT_DEVICE_TYPE_LAST
} uct_device_type_t;


typedef struct uct_tl_resource_desc {
    char                     tl_name[UCT_TL_NAME_MAX];   
    char                     dev_name[UCT_DEVICE_NAME_MAX]; 
    uct_device_type_t        dev_type;     
} uct_tl_resource_desc_t;

#define UCT_TL_RESOURCE_DESC_FMT              "%s/%s"
#define UCT_TL_RESOURCE_DESC_ARG(_resource)   (_resource)->tl_name, (_resource)->dev_name


typedef enum uct_atomic_op {
    UCT_ATOMIC_OP_ADD,   
    UCT_ATOMIC_OP_AND,   
    UCT_ATOMIC_OP_OR,    
    UCT_ATOMIC_OP_XOR,   
    UCT_ATOMIC_OP_SWAP,  
    UCT_ATOMIC_OP_CSWAP, 
    UCT_ATOMIC_OP_LAST
} uct_atomic_op_t;


        /* Active message capabilities */
#define UCT_IFACE_FLAG_AM_SHORT       UCS_BIT(0)  
#define UCT_IFACE_FLAG_AM_BCOPY       UCS_BIT(1)  
#define UCT_IFACE_FLAG_AM_ZCOPY       UCS_BIT(2)  
#define UCT_IFACE_FLAG_PENDING        UCS_BIT(3)  
        /* PUT capabilities */
#define UCT_IFACE_FLAG_PUT_SHORT      UCS_BIT(4)  
#define UCT_IFACE_FLAG_PUT_BCOPY      UCS_BIT(5)  
#define UCT_IFACE_FLAG_PUT_ZCOPY      UCS_BIT(6)  
        /* GET capabilities */
#define UCT_IFACE_FLAG_GET_SHORT      UCS_BIT(8)  
#define UCT_IFACE_FLAG_GET_BCOPY      UCS_BIT(9)  
#define UCT_IFACE_FLAG_GET_ZCOPY      UCS_BIT(10) 
        /* Atomic operations domain */
#define UCT_IFACE_FLAG_ATOMIC_CPU     UCS_BIT(30) 
#define UCT_IFACE_FLAG_ATOMIC_DEVICE  UCS_BIT(31) 
        /* Error handling capabilities */
#define UCT_IFACE_FLAG_ERRHANDLE_SHORT_BUF    UCS_BIT(32) 
#define UCT_IFACE_FLAG_ERRHANDLE_BCOPY_BUF    UCS_BIT(33) 
#define UCT_IFACE_FLAG_ERRHANDLE_ZCOPY_BUF    UCS_BIT(34) 
#define UCT_IFACE_FLAG_ERRHANDLE_AM_ID        UCS_BIT(35) 
#define UCT_IFACE_FLAG_ERRHANDLE_REMOTE_MEM   UCS_BIT(36) 
#define UCT_IFACE_FLAG_ERRHANDLE_BCOPY_LEN    UCS_BIT(37) 
#define UCT_IFACE_FLAG_ERRHANDLE_PEER_FAILURE UCS_BIT(38) 
#define UCT_IFACE_FLAG_EP_CHECK               UCS_BIT(39) 
        /* Connection establishment */
#define UCT_IFACE_FLAG_CONNECT_TO_IFACE       UCS_BIT(40) 
#define UCT_IFACE_FLAG_CONNECT_TO_EP          UCS_BIT(41) 
#define UCT_IFACE_FLAG_CONNECT_TO_SOCKADDR    UCS_BIT(42) 
        /* Special transport flags */
#define UCT_IFACE_FLAG_AM_DUP         UCS_BIT(43) 
        /* Callback invocation */
#define UCT_IFACE_FLAG_CB_SYNC        UCS_BIT(44) 
#define UCT_IFACE_FLAG_CB_ASYNC       UCS_BIT(45) 
        /* Event notification */
#define UCT_IFACE_FLAG_EVENT_SEND_COMP UCS_BIT(46) 
#define UCT_IFACE_FLAG_EVENT_RECV      UCS_BIT(47) 
#define UCT_IFACE_FLAG_EVENT_RECV_SIG  UCS_BIT(48) 
        /* Tag matching operations */
#define UCT_IFACE_FLAG_TAG_EAGER_SHORT UCS_BIT(50) 
#define UCT_IFACE_FLAG_TAG_EAGER_BCOPY UCS_BIT(51) 
#define UCT_IFACE_FLAG_TAG_EAGER_ZCOPY UCS_BIT(52) 
#define UCT_IFACE_FLAG_TAG_RNDV_ZCOPY  UCS_BIT(53) 
typedef enum {
    UCT_ALLOC_METHOD_THP,  
    UCT_ALLOC_METHOD_MD,   
    UCT_ALLOC_METHOD_HEAP, 
    UCT_ALLOC_METHOD_MMAP, 
    UCT_ALLOC_METHOD_HUGE, 
    UCT_ALLOC_METHOD_LAST,
    UCT_ALLOC_METHOD_DEFAULT = UCT_ALLOC_METHOD_LAST 
} uct_alloc_method_t;


enum uct_iface_event_types {
    UCT_EVENT_SEND_COMP     = UCS_BIT(0), 
    UCT_EVENT_RECV          = UCS_BIT(1), 
    UCT_EVENT_RECV_SIG      = UCS_BIT(2)  
};


enum uct_flush_flags {
    UCT_FLUSH_FLAG_LOCAL    = 0,            
    UCT_FLUSH_FLAG_CANCEL   = UCS_BIT(0)    
};


enum uct_progress_types {
    UCT_PROGRESS_SEND        = UCS_BIT(0),  
    UCT_PROGRESS_RECV        = UCS_BIT(1),  
    UCT_PROGRESS_THREAD_SAFE = UCS_BIT(7)   
};


enum uct_msg_flags {
    UCT_SEND_FLAG_SIGNALED = UCS_BIT(0) 
};


enum uct_cb_flags {
    UCT_CB_FLAG_RESERVED = UCS_BIT(1), 
    UCT_CB_FLAG_ASYNC    = UCS_BIT(2)  
};


enum uct_iface_open_mode {
   UCT_IFACE_OPEN_MODE_DEVICE          = UCS_BIT(0),  
   UCT_IFACE_OPEN_MODE_SOCKADDR_SERVER = UCS_BIT(1),  
   UCT_IFACE_OPEN_MODE_SOCKADDR_CLIENT = UCS_BIT(2)   
};


typedef enum {
   UCT_SOCKADDR_ACC_LOCAL,  
   UCT_SOCKADDR_ACC_REMOTE  
} uct_sockaddr_accessibility_t;


enum {
    UCT_MD_FLAG_ALLOC      = UCS_BIT(0),  
    UCT_MD_FLAG_REG        = UCS_BIT(1),  
    UCT_MD_FLAG_NEED_MEMH  = UCS_BIT(2),  
    UCT_MD_FLAG_NEED_RKEY  = UCS_BIT(3),  
    UCT_MD_FLAG_ADVISE     = UCS_BIT(4),  
    UCT_MD_FLAG_FIXED      = UCS_BIT(5),  
    UCT_MD_FLAG_RKEY_PTR   = UCS_BIT(6),  
    UCT_MD_FLAG_SOCKADDR   = UCS_BIT(7)   
};

/*
 * @ingroup UCT_MD
 * @brief  Memory types
 */
typedef enum {
    UCT_MD_MEM_TYPE_HOST = 0,      
    UCT_MD_MEM_TYPE_CUDA,          
    UCT_MD_MEM_TYPE_CUDA_MANAGED,  
    UCT_MD_MEM_TYPE_LAST
} uct_memory_type_t;


enum uct_md_mem_flags {
    UCT_MD_MEM_FLAG_NONBLOCK    = UCS_BIT(0), 
    UCT_MD_MEM_FLAG_FIXED       = UCS_BIT(1), 
    UCT_MD_MEM_FLAG_LOCK        = UCS_BIT(2), 
    UCT_MD_MEM_FLAG_HIDE_ERRORS = UCS_BIT(3), 
    /* memory access flags */
    UCT_MD_MEM_ACCESS_REMOTE_PUT    = UCS_BIT(5), 
    UCT_MD_MEM_ACCESS_REMOTE_GET    = UCS_BIT(6), 
    UCT_MD_MEM_ACCESS_REMOTE_ATOMIC = UCS_BIT(7), 
    UCT_MD_MEM_ACCESS_ALL =  (UCT_MD_MEM_ACCESS_REMOTE_PUT|
                              UCT_MD_MEM_ACCESS_REMOTE_GET|
                              UCT_MD_MEM_ACCESS_REMOTE_ATOMIC),

    UCT_MD_MEM_ACCESS_RMA = (UCT_MD_MEM_ACCESS_REMOTE_PUT|
                             UCT_MD_MEM_ACCESS_REMOTE_GET)
};


typedef enum {
    UCT_MADV_NORMAL  = 0,  
    UCT_MADV_WILLNEED      
} uct_mem_advice_t;


/*
 * @ingroup UCT_RESOURCE
 * @brief Linear growth specification: f(x) = overhead + growth * x
 *
 *  This structure specifies a linear function which is used as basis for time
 * estimation of various UCT operations. This information can be used to select
 * the best performing combination of UCT operations.
 */
typedef struct uct_linear_growth {
    double                   overhead;  
    double                   growth;    
} uct_linear_growth_t;


struct uct_iface_attr {
    struct {
        struct {
            size_t           max_short;  
            size_t           max_bcopy;  
            size_t           min_zcopy;  
            size_t           max_zcopy;  
            size_t           opt_zcopy_align; 
            size_t           align_mtu;       
            size_t           max_iov;    
        } put;                           
        struct {
            size_t           max_short;  
            size_t           max_bcopy;  
            size_t           min_zcopy;  
            size_t           max_zcopy;  
            size_t           opt_zcopy_align; 
            size_t           align_mtu;       
            size_t           max_iov;    
        } get;                           
        struct {
            size_t           max_short;  
            size_t           max_bcopy;  
            size_t           min_zcopy;  
            size_t           max_zcopy;  
            size_t           opt_zcopy_align; 
            size_t           align_mtu;       
            size_t           max_hdr;    
            size_t           max_iov;    
        } am;                            
        struct {
            struct {
                size_t       min_recv;   
                size_t       max_zcopy;  
                size_t       max_iov;    
                size_t       max_outstanding; 
            } recv;

            struct {
                  size_t     max_short;  
                  size_t     max_bcopy;  
                  size_t     max_zcopy;  
                  size_t     max_iov;    
            } eager;                     
            struct {
                  size_t     max_zcopy;  
                  size_t     max_hdr;    
                  size_t     max_iov;    
            } rndv;                      
        } tag;                           
        struct {
            uint64_t         op_flags;   
            uint64_t         fop_flags;  
        } atomic32, atomic64;            
        uint64_t             flags;      
    } cap;                               
    size_t                   device_addr_len;
    size_t                   iface_addr_len; 
    size_t                   ep_addr_len;    
    size_t                   max_conn_priv;  
    /*
     * The following fields define expected performance of the communication
     * interface, this would usually be a combination of device and system
     * characteristics and determined at run time.
     */
    double                   overhead;     
    double                   bandwidth;    
    uct_linear_growth_t      latency;      
    uint8_t                  priority;     
};


struct uct_iface_params {
    ucs_cpu_set_t                                cpu_mask;
    uint64_t                                     open_mode;
    union {
        struct {
            const char                           *tl_name;  
            const char                           *dev_name; 
        } device;
        struct {
            ucs_sock_addr_t                      listen_sockaddr;
            void                                 *conn_request_arg;
            uct_sockaddr_conn_request_callback_t conn_request_cb;
            uint32_t                             cb_flags;
        } sockaddr;
    } mode;

    ucs_stats_node_t                             *stats_root;
    size_t                                       rx_headroom;

    void                                         *err_handler_arg;
    uct_error_handler_t                          err_handler;
    uint32_t                                     err_handler_flags;

    void                                         *eager_arg;
    uct_tag_unexp_eager_cb_t                     eager_cb;
    void                                         *rndv_arg;
    uct_tag_unexp_rndv_cb_t                      rndv_cb;
};


struct uct_md_attr {
    struct {
        size_t               max_alloc; 
        size_t               max_reg;   
        uint64_t             flags;     
        uint64_t             reg_mem_types; 
        uct_memory_type_t    mem_type;  
    } cap;

    uct_linear_growth_t      reg_cost;  
    char                     component_name[UCT_MD_COMPONENT_NAME_MAX]; 
    size_t                   rkey_packed_size; 
    cpu_set_t                local_cpus;    
};


typedef struct uct_allocated_memory {
    void                     *address; 
    size_t                   length;   
    uct_alloc_method_t       method;   
    uct_memory_type_t        mem_type; 
    uct_md_h                 md;       
    uct_mem_h                memh;     
} uct_allocated_memory_t;


typedef struct uct_rkey_bundle {
    uct_rkey_t               rkey;    
    void                     *handle; 
    void                     *type;   
} uct_rkey_bundle_t;


struct uct_completion {
    uct_completion_callback_t func;    
    int                       count;   
};


struct uct_pending_req {
    uct_pending_callback_t    func;   
    char                      priv[UCT_PENDING_REQ_PRIV_LEN]; 
};


struct uct_tag_context {
    void (*tag_consumed_cb)(uct_tag_context_t *self);

     void (*completed_cb)(uct_tag_context_t *self, uct_tag_t stag, uint64_t imm,
                          size_t length, ucs_status_t status);

     void (*rndv_cb)(uct_tag_context_t *self, uct_tag_t stag, const void *header,
                     unsigned header_length, ucs_status_t status);

     char priv[UCT_TAG_PRIV_LEN];
};


extern const char *uct_alloc_method_names[];


ucs_status_t uct_query_md_resources(uct_md_resource_desc_t **resources_p,
                                    unsigned *num_resources_p);

void uct_release_md_resource_list(uct_md_resource_desc_t *resources);


ucs_status_t uct_md_open(const char *md_name, const uct_md_config_t *config,
                         uct_md_h *md_p);

void uct_md_close(uct_md_h md);


ucs_status_t uct_md_query_tl_resources(uct_md_h md,
                                       uct_tl_resource_desc_t **resources_p,
                                       unsigned *num_resources_p);


void uct_release_tl_resource_list(uct_tl_resource_desc_t *resources);


ucs_status_t uct_worker_create(ucs_async_context_t *async,
                               ucs_thread_mode_t thread_mode,
                               uct_worker_h *worker_p);


void uct_worker_destroy(uct_worker_h worker);


void uct_worker_progress_register_safe(uct_worker_h worker, ucs_callback_t func,
                                       void *arg, unsigned flags,
                                       uct_worker_cb_id_t *id_p);


void uct_worker_progress_unregister_safe(uct_worker_h worker,
                                         uct_worker_cb_id_t *id_p);


ucs_status_t uct_md_iface_config_read(uct_md_h md, const char *tl_name,
                                      const char *env_prefix, const char *filename,
                                      uct_iface_config_t **config_p);


void uct_config_release(void *config);


ucs_status_t uct_config_get(void *config, const char *name, char *value,
                            size_t max);


ucs_status_t uct_config_modify(void *config, const char *name, const char *value);


ucs_status_t uct_iface_open(uct_md_h md, uct_worker_h worker,
                            const uct_iface_params_t *params,
                            const uct_iface_config_t *config,
                            uct_iface_h *iface_p);


void uct_iface_close(uct_iface_h iface);


ucs_status_t uct_iface_query(uct_iface_h iface, uct_iface_attr_t *iface_attr);


ucs_status_t uct_iface_get_device_address(uct_iface_h iface, uct_device_addr_t *addr);


ucs_status_t uct_iface_get_address(uct_iface_h iface, uct_iface_addr_t *addr);


int uct_iface_is_reachable(const uct_iface_h iface, const uct_device_addr_t *dev_addr,
                           const uct_iface_addr_t *iface_addr);


ucs_status_t uct_ep_check(const uct_ep_h ep, unsigned flags,
                          uct_completion_t *comp);


ucs_status_t uct_iface_event_fd_get(uct_iface_h iface, int *fd_p);


ucs_status_t uct_iface_event_arm(uct_iface_h iface, unsigned events);


ucs_status_t uct_iface_mem_alloc(uct_iface_h iface, size_t length, unsigned flags,
                                 const char *name, uct_allocated_memory_t *mem);


void uct_iface_mem_free(const uct_allocated_memory_t *mem);


ucs_status_t uct_iface_set_am_handler(uct_iface_h iface, uint8_t id,
                                      uct_am_callback_t cb, void *arg, uint32_t flags);


ucs_status_t uct_iface_set_am_tracer(uct_iface_h iface, uct_am_tracer_t tracer,
                                     void *arg);


ucs_status_t uct_iface_accept(uct_iface_h iface,
                              uct_conn_request_h conn_request);


ucs_status_t uct_iface_reject(uct_iface_h iface,
                              uct_conn_request_h conn_request);


ucs_status_t uct_ep_create(uct_iface_h iface, uct_ep_h *ep_p);


ucs_status_t uct_ep_create_connected(uct_iface_h iface,
                                     const uct_device_addr_t *dev_addr,
                                     const uct_iface_addr_t *iface_addr,
                                     uct_ep_h *ep_p);


void uct_ep_destroy(uct_ep_h ep);


ucs_status_t uct_ep_get_address(uct_ep_h ep, uct_ep_addr_t *addr);


ucs_status_t uct_ep_connect_to_ep(uct_ep_h ep, const uct_device_addr_t *dev_addr,
                                  const uct_ep_addr_t *ep_addr);


ucs_status_t uct_ep_create_sockaddr(uct_iface_h iface,
                                    const ucs_sock_addr_t *sockaddr,
                                    uct_sockaddr_priv_pack_callback_t pack_cb,
                                    void *arg, uint32_t cb_flags, uct_ep_h *ep_p);


ucs_status_t uct_md_query(uct_md_h md, uct_md_attr_t *md_attr);


ucs_status_t uct_md_mem_alloc(uct_md_h md, size_t *length_p, void **address_p,
                              unsigned flags, const char *name, uct_mem_h *memh_p);


ucs_status_t uct_md_mem_free(uct_md_h md, uct_mem_h memh);


ucs_status_t uct_md_mem_advise(uct_md_h md, uct_mem_h memh, void *addr,
                               size_t length, uct_mem_advice_t advice);


ucs_status_t uct_md_mem_reg(uct_md_h md, void *address, size_t length,
                            unsigned flags, uct_mem_h *memh_p);


ucs_status_t uct_md_mem_dereg(uct_md_h md, uct_mem_h memh);


int uct_md_is_mem_type_owned(uct_md_h md, void *addr, size_t length);

ucs_status_t uct_mem_alloc(void *addr, size_t min_length, unsigned flags,
                           uct_alloc_method_t *methods, unsigned num_methods,
                           uct_md_h *mds, unsigned num_mds, const char *name,
                           uct_allocated_memory_t *mem);


ucs_status_t uct_mem_free(const uct_allocated_memory_t *mem);

ucs_status_t uct_md_config_read(const char *name, const char *env_prefix,
                                const char *filename,
                                uct_md_config_t **config_p);



int uct_md_is_sockaddr_accessible(uct_md_h md, const ucs_sock_addr_t *sockaddr,
                                  uct_sockaddr_accessibility_t mode);


ucs_status_t uct_md_mkey_pack(uct_md_h md, uct_mem_h memh, void *rkey_buffer);


ucs_status_t uct_rkey_unpack(const void *rkey_buffer, uct_rkey_bundle_t *rkey_ob);


ucs_status_t uct_rkey_ptr(uct_rkey_bundle_t *rkey_ob, uint64_t remote_addr,
                          void **addr_p);


ucs_status_t uct_rkey_release(const uct_rkey_bundle_t *rkey_ob);


UCT_INLINE_API unsigned uct_worker_progress(uct_worker_h worker)
{
    return ucs_callbackq_dispatch(&worker->progress_q);
}


UCT_INLINE_API ucs_status_t uct_iface_flush(uct_iface_h iface, unsigned flags,
                                            uct_completion_t *comp)
{
    return iface->ops.iface_flush(iface, flags, comp);
}

UCT_INLINE_API ucs_status_t uct_iface_fence(uct_iface_h iface, unsigned flags)
{
    return iface->ops.iface_fence(iface, flags);
}

UCT_INLINE_API void uct_iface_release_desc(void *desc)
{
    uct_recv_desc_t *release_desc = uct_recv_desc(desc);
    release_desc->cb(release_desc, desc);
}


UCT_INLINE_API ucs_status_t uct_ep_put_short(uct_ep_h ep, const void *buffer, unsigned length,
                                             uint64_t remote_addr, uct_rkey_t rkey)
{
    return ep->iface->ops.ep_put_short(ep, buffer, length, remote_addr, rkey);
}


UCT_INLINE_API ssize_t uct_ep_put_bcopy(uct_ep_h ep, uct_pack_callback_t pack_cb,
                                        void *arg, uint64_t remote_addr,
                                        uct_rkey_t rkey)
{
    return ep->iface->ops.ep_put_bcopy(ep, pack_cb, arg, remote_addr, rkey);
}


UCT_INLINE_API ucs_status_t uct_ep_put_zcopy(uct_ep_h ep,
                                             const uct_iov_t *iov, size_t iovcnt,
                                             uint64_t remote_addr, uct_rkey_t rkey,
                                             uct_completion_t *comp)
{
    return ep->iface->ops.ep_put_zcopy(ep, iov, iovcnt, remote_addr, rkey, comp);
}


UCT_INLINE_API ucs_status_t uct_ep_get_short(uct_ep_h ep, void *buffer, unsigned length,
                                             uint64_t remote_addr, uct_rkey_t rkey)
{
    return ep->iface->ops.ep_get_short(ep, buffer, length, remote_addr, rkey);
}


UCT_INLINE_API ucs_status_t uct_ep_get_bcopy(uct_ep_h ep, uct_unpack_callback_t unpack_cb,
                                             void *arg, size_t length,
                                             uint64_t remote_addr, uct_rkey_t rkey,
                                             uct_completion_t *comp)
{
    return ep->iface->ops.ep_get_bcopy(ep, unpack_cb, arg, length, remote_addr,
                                       rkey, comp);
}


UCT_INLINE_API ucs_status_t uct_ep_get_zcopy(uct_ep_h ep,
                                             const uct_iov_t *iov, size_t iovcnt,
                                             uint64_t remote_addr, uct_rkey_t rkey,
                                             uct_completion_t *comp)
{
    return ep->iface->ops.ep_get_zcopy(ep, iov, iovcnt, remote_addr, rkey, comp);
}


UCT_INLINE_API ucs_status_t uct_ep_am_short(uct_ep_h ep, uint8_t id, uint64_t header,
                                            const void *payload, unsigned length)
{
    return ep->iface->ops.ep_am_short(ep, id, header, payload, length);
}


UCT_INLINE_API ssize_t uct_ep_am_bcopy(uct_ep_h ep, uint8_t id,
                                       uct_pack_callback_t pack_cb, void *arg,
                                       unsigned flags)
{
    return ep->iface->ops.ep_am_bcopy(ep, id, pack_cb, arg, flags);
}


UCT_INLINE_API ucs_status_t uct_ep_am_zcopy(uct_ep_h ep, uint8_t id,
                                            const void *header,
                                            unsigned header_length,
                                            const uct_iov_t *iov, size_t iovcnt,
                                            unsigned flags,
                                            uct_completion_t *comp)
{
    return ep->iface->ops.ep_am_zcopy(ep, id, header, header_length, iov, iovcnt,
                                      flags, comp);
}

UCT_INLINE_API ucs_status_t uct_ep_atomic_cswap64(uct_ep_h ep, uint64_t compare, uint64_t swap,
                                                  uint64_t remote_addr, uct_rkey_t rkey,
                                                  uint64_t *result, uct_completion_t *comp)
{
    return ep->iface->ops.ep_atomic_cswap64(ep, compare, swap, remote_addr, rkey, result, comp);
}


UCT_INLINE_API ucs_status_t uct_ep_atomic_cswap32(uct_ep_h ep, uint32_t compare, uint32_t swap,
                                                  uint64_t remote_addr, uct_rkey_t rkey,
                                                  uint32_t *result, uct_completion_t *comp)
{
    return ep->iface->ops.ep_atomic_cswap32(ep, compare, swap, remote_addr, rkey, result, comp);
}


UCT_INLINE_API ucs_status_t uct_ep_atomic32_post(uct_ep_h ep, uct_atomic_op_t opcode,
                                                 uint32_t value, uint64_t remote_addr,
                                                 uct_rkey_t rkey)
{
    return ep->iface->ops.ep_atomic32_post(ep, opcode, value, remote_addr, rkey);
}


UCT_INLINE_API ucs_status_t uct_ep_atomic64_post(uct_ep_h ep, uct_atomic_op_t opcode,
                                                 uint64_t value, uint64_t remote_addr,
                                                 uct_rkey_t rkey)
{
    return ep->iface->ops.ep_atomic64_post(ep, opcode, value, remote_addr, rkey);
}


UCT_INLINE_API ucs_status_t uct_ep_atomic32_fetch(uct_ep_h ep, uct_atomic_op_t opcode,
                                                  uint32_t value, uint32_t *result,
                                                  uint64_t remote_addr, uct_rkey_t rkey,
                                                  uct_completion_t *comp)
{
    return ep->iface->ops.ep_atomic32_fetch(ep, opcode, value, result,
                                            remote_addr, rkey, comp);
}


UCT_INLINE_API ucs_status_t uct_ep_atomic64_fetch(uct_ep_h ep, uct_atomic_op_t opcode,
                                                  uint64_t value, uint64_t *result,
                                                  uint64_t remote_addr, uct_rkey_t rkey,
                                                  uct_completion_t *comp)
{
    return ep->iface->ops.ep_atomic64_fetch(ep, opcode, value, result,
                                            remote_addr, rkey, comp);
}


UCT_INLINE_API ucs_status_t uct_ep_pending_add(uct_ep_h ep,
                                               uct_pending_req_t *req,
                                               unsigned flags)
{
    return ep->iface->ops.ep_pending_add(ep, req, flags);
}


UCT_INLINE_API void uct_ep_pending_purge(uct_ep_h ep,
                                         uct_pending_purge_callback_t cb,
                                         void *arg)
{
    ep->iface->ops.ep_pending_purge(ep, cb, arg);
}


UCT_INLINE_API ucs_status_t uct_ep_flush(uct_ep_h ep, unsigned flags,
                                         uct_completion_t *comp)
{
    return ep->iface->ops.ep_flush(ep, flags, comp);
}


UCT_INLINE_API ucs_status_t uct_ep_fence(uct_ep_h ep, unsigned flags)
{
    return ep->iface->ops.ep_fence(ep, flags);
}


UCT_INLINE_API ucs_status_t uct_ep_tag_eager_short(uct_ep_h ep, uct_tag_t tag,
                                                   const void *data, size_t length)
{
    return ep->iface->ops.ep_tag_eager_short(ep, tag, data, length);
}


UCT_INLINE_API ssize_t uct_ep_tag_eager_bcopy(uct_ep_h ep, uct_tag_t tag,
                                              uint64_t imm,
                                              uct_pack_callback_t pack_cb,
                                              void *arg, unsigned flags)
{
    return ep->iface->ops.ep_tag_eager_bcopy(ep, tag, imm, pack_cb, arg, flags);
}


UCT_INLINE_API ucs_status_t uct_ep_tag_eager_zcopy(uct_ep_h ep, uct_tag_t tag,
                                                   uint64_t imm,
                                                   const uct_iov_t *iov,
                                                   size_t iovcnt,
                                                   unsigned flags,
                                                   uct_completion_t *comp)
{
    return ep->iface->ops.ep_tag_eager_zcopy(ep, tag, imm, iov, iovcnt, flags,
                                             comp);
}


UCT_INLINE_API ucs_status_ptr_t uct_ep_tag_rndv_zcopy(uct_ep_h ep, uct_tag_t tag,
                                                      const void *header,
                                                      unsigned header_length,
                                                      const uct_iov_t *iov,
                                                      size_t iovcnt,
                                                      unsigned flags,
                                                      uct_completion_t *comp)
{
    return ep->iface->ops.ep_tag_rndv_zcopy(ep, tag, header, header_length,
                                            iov, iovcnt, flags, comp);
}


UCT_INLINE_API ucs_status_t uct_ep_tag_rndv_cancel(uct_ep_h ep, void *op)
{
    return ep->iface->ops.ep_tag_rndv_cancel(ep, op);
}


UCT_INLINE_API ucs_status_t uct_ep_tag_rndv_request(uct_ep_h ep, uct_tag_t tag,
                                                    const void* header,
                                                    unsigned header_length,
                                                    unsigned flags)
{
    return ep->iface->ops.ep_tag_rndv_request(ep, tag, header, header_length,
                                              flags);
}


UCT_INLINE_API ucs_status_t uct_iface_tag_recv_zcopy(uct_iface_h iface,
                                                     uct_tag_t tag,
                                                     uct_tag_t tag_mask,
                                                     const uct_iov_t *iov,
                                                     size_t iovcnt,
                                                     uct_tag_context_t *ctx)
{
    return iface->ops.iface_tag_recv_zcopy(iface, tag, tag_mask, iov, iovcnt, ctx);
}


UCT_INLINE_API ucs_status_t uct_iface_tag_recv_cancel(uct_iface_h iface,
                                                      uct_tag_context_t *ctx,
                                                      int force)
{
    return iface->ops.iface_tag_recv_cancel(iface, ctx, force);
}


UCT_INLINE_API void uct_iface_progress_enable(uct_iface_h iface, unsigned flags)
{
    iface->ops.iface_progress_enable(iface, flags);
}


UCT_INLINE_API void uct_iface_progress_disable(uct_iface_h iface, unsigned flags)
{
    iface->ops.iface_progress_disable(iface, flags);
}


UCT_INLINE_API unsigned uct_iface_progress(uct_iface_h iface)
{
    return iface->ops.iface_progress(iface);
}


END_C_DECLS

#endif