Functions | |
INT | bm_match_event (short int event_id, short int trigger_mask, EVENT_HEADER *pevent) |
INT | bm_open_buffer (char *buffer_name, INT buffer_size, INT *buffer_handle) |
INT | bm_close_buffer (INT buffer_handle) |
INT | bm_close_all_buffers (void) |
INT | bm_set_cache_size (INT buffer_handle, INT read_size, INT write_size) |
INT | bm_compose_event (EVENT_HEADER *event_header, short int event_id, short int trigger_mask, DWORD size, DWORD serial) |
INT | bm_request_event (HNDLE buffer_handle, short int event_id, short int trigger_mask, INT sampling_type, HNDLE *request_id, void(*func)(HNDLE, HNDLE, EVENT_HEADER *, void *)) |
INT | bm_remove_event_request (INT buffer_handle, INT request_id) |
INT | bm_delete_request (INT request_id) |
INT | bm_send_event (INT buffer_handle, void *source, INT buf_size, INT async_flag) |
INT | bm_flush_cache (INT buffer_handle, INT async_flag) |
INT | bm_receive_event (INT buffer_handle, void *destination, INT *buf_size, INT async_flag) |
INT | bm_skip_event (INT buffer_handle) |
INT | bm_push_event (char *buffer_name) |
INT | bm_check_buffers () |
INT | bm_empty_buffers () |
|
Check if any requested event is waiting in a buffer
Definition at line 6817 of file midas.c. Referenced by cm_yield(). |
|
Close all open buffers
Definition at line 4105 of file midas.c. Referenced by cm_disconnect_experiment(), and cm_set_client_info(). |
|
Closes an event buffer previously opened with bm_open_buffer().
Definition at line 3994 of file midas.c. Referenced by bm_close_all_buffers(), and source_unbooking(). |
|
Compose a Midas event header. An event header can usually be set-up manually or through this routine. If the data size of the event is not known when the header is composed, it can be set later with event_header->data-size = <...> Following structure is created at the beginning of an event typedef struct { short int event_id; short int trigger_mask; DWORD serial_number; DWORD time_stamp; DWORD data_size; } EVENT_HEADER; char event[1000]; bm_compose_event((EVENT_HEADER *)event, 1, 0, 100, 1); *(event+sizeof(EVENT_HEADER)) = <...>
Definition at line 5080 of file midas.c. Referenced by cm_msg(), cm_msg1(), and source_scan(). |
|
Deletes an event request previously done with bm_request_event(). When an event request gets deleted, events of that requested type are not received any more. When a buffer is closed via bm_close_buffer(), all event requests from that buffer are deleted automatically
Definition at line 5387 of file midas.c. Referenced by bm_close_buffer(), and source_unbooking(). |
|
Clears event buffer and cache. If an event buffer is large and a consumer is slow in analyzing events, events are usually received some time after they are produced. This effect is even more experienced if a read cache is used (via bm_set_cache_size()). When changes to the hardware are made in the experience, the consumer will then still analyze old events before any new event which reflects the hardware change. Users can be fooled by looking at histograms which reflect the hardware change many seconds after they have been made. To overcome this potential problem, the analyzer can call bm_empty_buffers() just after the hardware change has been made which skips all old events contained in event buffers and read caches. Technically this is done by forwarding the read pointer of the client. No events are really deleted, they are still visible to other clients like the logger. Note that the front-end also contains write buffers which can delay the delivery of events. The standard front-end framework mfe.c reduces this effect by flushing all buffers once every second.
Definition at line 7139 of file midas.c. Referenced by handFlush(), source_booking(), and source_unbooking(). |
|
Empty write cache. This function should be used if events in the write cache should be visible to the consumers immediately. It should be called at the end of each run, otherwise events could be kept in the write buffer and will flow to the data of the next run.
Definition at line 5803 of file midas.c. Referenced by bm_send_event(), close_buffers(), scan_fragment(), scheduler(), send_event(), and tr_stop(). |
|
Check if an event matches a given event request by the event id and trigger mask
Definition at line 3710 of file midas.c. Referenced by bm_flush_cache(), bm_push_event(), bm_receive_event(), and bm_send_event(). |
|
Open an event buffer. Two default buffers are created by the system. The "SYSTEM" buffer is used to exchange events and the "SYSMSG" buffer is used to exchange system messages. The name and size of the event buffers is defined in midas.h as EVENT_BUFFER_NAME and EVENT_BUFFER_SIZE. Following example opens the "SYSTEM" buffer, requests events with ID 1 and enters a main loop. Events are then received in process_event() #include <stdio.h> #include "midas.h" void process_event(HNDLE hbuf, HNDLE request_id, EVENT_HEADER *pheader, void *pevent) { printf("Received event #%d\r", pheader->serial_number); } main() { INT status, request_id; HNDLE hbuf; status = cm_connect_experiment("pc810", "Sample", "Simple Analyzer", NULL); if (status != CM_SUCCESS) return 1; bm_open_buffer(EVENT_BUFFER_NAME, EVENT_BUFFER_SIZE, &hbuf); bm_request_event(hbuf, 1, TRIGGER_ALL, GET_ALL, request_id, process_event); do { status = cm_yield(1000); } while (status != RPC_SHUTDOWN && status != SS_ABORT); cm_disconnect_experiment(); return 0; }
Definition at line 3772 of file midas.c. Referenced by cm_msg(), cm_msg1(), cm_msg_register(), register_equipment(), and source_booking(). |
|
Check a buffer if an event is available and call the dispatch function if found.
Definition at line 6545 of file midas.c. Referenced by bm_check_buffers(). |
|
Receives events directly. This function is an alternative way to receive events without a main loop. It can be used in analysis systems which actively receive events, rather than using callbacks. A analysis package could for example contain its own command line interface. A command like "receive 1000 events" could make it necessary to call bm_receive_event() 1000 times in a row to receive these events and then return back to the command line prompt. The according bm_request_event() call contains NULL as the callback routine to indicate that bm_receive_event() is called to receive events. #include <stdio.h> #include "midas.h" void process_event(EVENT_HEADER *pheader) { printf("Received event #%d\r", pheader->serial_number); } main() { INT status, request_id; HNDLE hbuf; char event_buffer[1000]; status = cm_connect_experiment("", "Sample", "Simple Analyzer", NULL); if (status != CM_SUCCESS) return 1; bm_open_buffer(EVENT_BUFFER_NAME, EVENT_BUFFER_SIZE, &hbuf); bm_request_event(hbuf, 1, TRIGGER_ALL, GET_ALL, request_id, NULL); do { size = sizeof(event_buffer); status = bm_receive_event(hbuf, event_buffer, &size, ASYNC); if (status == CM_SUCCESS) process_event((EVENT_HEADER *) event_buffer); <...do something else...> status = cm_yield(0); } while (status != RPC_SHUTDOWN && status != SS_ABORT); cm_disconnect_experiment(); return 0; }
Definition at line 6165 of file midas.c. Referenced by handFlush(), and source_scan(). |
|
Delete a previously placed request for a specific event type in the client structure of the buffer refereced by buffer_handle.
Definition at line 5314 of file midas.c. Referenced by bm_delete_request(). |
|
Place an event request based on certain characteristics. Multiple event requests can be placed for each buffer, which are later identified by their request ID. They can contain different callback routines. Example see bm_open_buffer() and bm_receive_event()
Definition at line 5246 of file midas.c. Referenced by cm_msg_register(), and source_booking(). |
|
Sends an event to a buffer. This function check if the buffer has enough space for the event, then copies the event to the buffer in shared memory. If clients have requests for the event, they are notified via an UDP packet. char event[1000]; // create event with ID 1, trigger mask 0, size 100 bytes and serial number 1 bm_compose_event((EVENT_HEADER *) event, 1, 0, 100, 1); // set first byte of event *(event+sizeof(EVENT_HEADER)) = <...> #include <stdio.h> #include "midas.h" main() { INT status, i; HNDLE hbuf; char event[1000]; status = cm_connect_experiment("", "Sample", "Producer", NULL); if (status != CM_SUCCESS) return 1; bm_open_buffer(EVENT_BUFFER_NAME, EVENT_BUFFER_SIZE, &hbuf); // create event with ID 1, trigger mask 0, size 100 bytes and serial number 1 bm_compose_event((EVENT_HEADER *) event, 1, 0, 100, 1); // set event data for (i=0 ; i<100 ; i++) *(event+sizeof(EVENT_HEADER)+i) = i; // send event bm_send_event(hbuf, event, 100+sizeof(EVENT_HEADER), SYNC); cm_disconnect_experiment(); return 0; }
Definition at line 5451 of file midas.c. Referenced by cm_msg(), cm_msg1(), rpc_send_event(), and send_event(). |
|
Modifies buffer cache size. Without a buffer cache, events are copied to/from the shared memory event by event. To protect processed from accessing the shared memory simultaneously, semaphores are used. Since semaphore operations are CPU consuming (typically 50-100us) this can slow down the data transfer especially for small events. By using a cache the number of semaphore operations is reduced dramatically. Instead writing directly to the shared memory, the events are copied to a local cache buffer. When this buffer is full, it is copied to the shared memory in one operation. The same technique can be used when receiving events. The drawback of this method is that the events have to be copied twice, once to the cache and once from the cache to the shared memory. Therefore it can happen that the usage of a cache even slows down data throughput on a given environment (computer type, OS type, event size). The cache size has therefore be optimized manually to maximize data throughput.
Definition at line 4983 of file midas.c. Referenced by register_equipment(). |
|
Skip all events in current buffer. Useful for single event displays to see the newest events
|