1fi_trigger(3) Libfabric v1.6.1 fi_trigger(3)
2
6 fi_trigger - Triggered operations
7
9 #include <rdma/fi_trigger.h>
10
12 Triggered operations allow an application to queue a data transfer
13 request that is deferred until a specified condition is met. A typical
14 use is to send a message only after receiving all input data.
15 A triggered operation may be requested by specifying the FI_TRIGGER
17 flag as part of the operation. Alternatively, an endpoint alias may be
18 created and configured with the FI_TRIGGER flag. Such an endpoint is
19 referred to as a trigger-able endpoint. All data transfer operations
20 on a trigger-able endpoint are deferred.
21 Any data transfer operation is potentially trigger-able, subject to
23 provider constraints. Trigger-able endpoints are initialized such that
24 only those interfaces supported by the provider which are trigger-able
25 are available.
26 Triggered operations require that applications use struct fi_trig‐
28 gered_context as their per operation context parameter, or if the
29 provider requires the FI_CONTEXT2 mode, struct fi_trigger_context2.
30 The use of struct fi_triggered_context[2] replaces struct fi_con‐
31 text[2], if required by the provider. Although struct fi_trig‐
32 gered_context[2] is not opaque to the application, the contents of the
33 structure may be modified by the provider once it has been submitted as
34 an operation. This structure has similar requirements as struct
35 fi_context[2]. It must be allocated by the application and remain
36 valid until the corresponding operation completes or is successfully
37 canceled.
38 Struct fi_triggered_context[2] is used to specify the condition that
40 must be met before the triggered data transfer is initiated. If the
41 condition is met when the request is made, then the data transfer may
42 be initiated immediately. The format of struct fi_triggered_context[2]
43 is described below.
44 struct fi_triggered_context {
46 enum fi_trigger_event event_type; /* trigger type */
47 union {
48 struct fi_trigger_threshold threshold;
49 void *internal[3]; /* reserved */
50 } trigger;
51 };
52 struct fi_triggered_context2 {
54 enum fi_trigger_event event_type; /* trigger type */
55 union {
56 struct fi_trigger_threshold threshold;
57 void *internal[7]; /* reserved */
58 } trigger;
59 };
60 The triggered context indicates the type of event assigned to the trig‐
62 ger, along with a union of trigger details that is based on the event
63 type.
64 TRIGGER EVENTS
66 The following trigger events are defined.
67 FI_TRIGGER_THRESHOLD : This indicates that the data transfer operation
69 will be deferred until an event counter crosses an application speci‐
70 fied threshold value. The threshold is specified using struct fi_trig‐
71 ger_threshold:
72 struct fi_trigger_threshold {
74 struct fid_cntr *cntr; /* event counter to check */
75 size_t threshold; /* threshold value */
76 };
77 Threshold operations are triggered in the order of the threshold val‐
79 ues. This is true even if the counter increments by a value greater
80 than 1. If two triggered operations have the same threshold, they will
81 be triggered in the order in which they were submitted to the endpoint.
82
84 The following feature and description are enhancements to triggered
85 operation support, but should be considered experimental. Until the
86 experimental tag is removed, the interfaces, semantics, and data struc‐
87 tures defined below may change between library versions.
88 The deferred work queue interface is designed as primitive constructs
90 that can be used to implement application-level collective operations.
91 They are a more advanced form of triggered operation. They allow an
92 application to queue operations to a deferred work queue that is asso‐
93 ciated with the domain. Note that the deferred work queue is a concep‐
94 tual construct, rather than an implementation requirement. Deferred
95 work requests consist of three main components: an event or condition
96 that must first be met, an operation to perform, and a completion noti‐
97 fication.
98 Because deferred work requests are posted directly to the domain, they
100 can support a broader set of conditions and operations. Deferred work
101 requests are submitted using struct fi_deferred_work. That structure,
102 along with the corresponding operation structures (referenced through
103 the op union) used to describe the work must remain valid until the
104 operation completes or is canceled. The format of the deferred work
105 request is as follows:
106 struct fi_deferred_work {
108 struct fi_context2 context;
109 uint64_t threshold;
111 struct fid_cntr *triggering_cntr;
112 struct fid_cntr *completion_cntr;
113 enum fi_trigger_op op_type;
115 union {
117 struct fi_op_msg *msg;
118 struct fi_op_tagged *tagged;
119 struct fi_op_rma *rma;
120 struct fi_op_atomic *atomic;
121 struct fi_op_fetch_atomic *fetch_atomic;
122 struct fi_op_compare_atomic *compare_atomic;
123 struct fi_op_cntr *cntr;
124 } op;
125 };
126 Once a work request has been posted to the deferred work queue, it will
128 remain on the queue until the triggering counter (success plus error
129 counter values) has reached the indicated threshold. If the triggering
130 condition has already been met at the time the work request is queued,
131 the operation will be initiated immediately.
132 On the completion of a deferred data transfer, the specified completion
134 counter will be incremented by one. Note that deferred counter opera‐
135 tions do not update the completion counter; only the counter specified
136 through the fi_op_cntr is modified. The completion_cntr field must be
137 NULL for counter operations.
138 Because deferred work targets support of collective communication oper‐
140 ations, posted work requests do not generate any completions at the
141 endpoint by default. For example, completed operations are not written
142 to the EP's completion queue or update the EP counter (unless the EP
143 counter is explicitly referenced as the completion_cntr). An applica‐
144 tion may request EP completions by specifying the FI_COMPLETION flag as
145 part of the operation.
146 It is the responsibility of the application to detect and handle situa‐
148 tions that occur which could result in a deferred work request's condi‐
149 tion not being met. For example, if a work request is dependent upon
150 the successful completion of a data transfer operation, which fails,
151 then the application must cancel the work request.
152 To submit a deferred work request, applications should use the domain's
154 fi_control function with command FI_QUEUE_WORK and struct
155 fi_deferred_work as the fi_control arg parameter. To cancel a deferred
156 work request, use fi_control with command FI_CANCEL_WORK and the corre‐
157 sponding struct fi_deferred_work to cancel. The fi_control command
158 FI_FLUSH_WORK will cancel all queued work requests. FI_FLUSH_WORK may
159 be used to flush all work queued to the domain, or may be used to can‐
160 cel all requests waiting on a specific triggering_cntr.
161 Deferred work requests are not acted upon by the provider until the
163 associated event has occurred; although, certain validation checks may
164 still occur when a request is submitted. Referenced data buffers are
165 not read or otherwise accessed. But the provider may validate fabric
166 objects, such as endpoints and counters, and that input parameters fall
167 within supported ranges. If a specific request is not supported by the
168 provider, it will fail the operation with -FI_ENOSYS.
169
171 fi_getinfo(3), fi_endpoint(3), fi_alias(3), fi_cntr(3)
172
174 OpenFabrics.
175Libfabric Programmer's Manual 2017-12-01 fi_trigger(3)