@@ -161,13 +161,15 @@ def __init__(self, plugin_args=None):
161161 self .n_gpus = self .n_gpus_visible
162162 self .total_gpu_processors = self .n_gpus * self .n_gpu_proc
163163 else :
164+ #total gpu_processors = no.of GPUs * no.of threads per single GPU
164165 self .total_gpu_processors = self .n_gpus * self .n_gpu_proc
165166
166167
168+ #form a GPU queue first
167169 gpus = GPUtil .getGPUs ()
168170 self .gpu_q = {}
169-
170-
171+
172+ #initialize the queue, set all slots free
171173 slotno = 0
172174 for gpu in gpus :
173175 temp = {}
@@ -207,6 +209,7 @@ def gpu_count(self):
207209 return len (GPUtil .getGPUs ())
208210
209211 def gpu_has_free_slot (self ):
212+ #if a GPU has free slot, return True,its device-ID and the slot no.
210213 free = False
211214 devno = None
212215 slotno = None
@@ -223,6 +226,7 @@ def gpu_has_free_slot(self):
223226 return free ,devno ,slotno
224227
225228 def set_gpu_slot_busy (self ,slotno ,jobid ):
229+ #if a GPU has free slot, book it for a jobid,modify the queue and set its slotno busy
226230 devno = None
227231 for dk in self .gpu_q .keys ():
228232 for sk in self .gpu_q [dk ].keys ():
@@ -233,6 +237,7 @@ def set_gpu_slot_busy(self,slotno,jobid):
233237
234238
235239 def set_gpu_slot_free (self ,jobid ):
240+ #if a GPU task is finished, then set the slotno free in the queue
236241 devno = None
237242 for dk in self .gpu_q .keys ():
238243 for sdk in self .gpu_q [dk ].keys ():
@@ -243,7 +248,7 @@ def set_gpu_slot_free(self,jobid):
243248 return devno
244249
245250
246- #override, to set gpu slot free
251+ #override, to set gpu slot free, if the job was a gpu job
247252 def _task_finished_cb (self , jobid ):
248253 """ Extract outputs and assign to inputs of dependent tasks
249254
@@ -281,7 +286,7 @@ def _submit_job(self, node, devno=None, updatehash=False):
281286 run_node , (node , updatehash , self ._taskid , devno ),
282287 callback = self ._async_callback )
283288
284- logger .info ('MultiProc submitted task %s (taskid=%d).' ,
289+ logger .debug ('MultiProc submitted task %s (taskid=%d).' ,
285290 node .fullname , self ._taskid )
286291 return self ._taskid
287292
0 commit comments