# script for controlling perfexplorer 2.0...
# import the necessary Java classes from the glue and rules packages
from edu.uoregon.tau.perfexplorer.glue import *
from edu.uoregon.tau.perfexplorer.rules import *

True = 1
False = 0

def loadRules():
	global ruleHarness
	print "Loading Rules..."
	ruleHarness = RuleHarness.useGlobalRules("rules/GeneralRules.drl")
	ruleHarness.addRules("rules/ApplicationRules.drl")
	ruleHarness.addRules("rules/MachineRules.drl")
	return 

def loaddata():
	print "loading the data..."
	#from home
	# Utilities.setSession("localhost:5432/perfdmf")
	# trial = TrialMeanResult(Utilities.getTrial("gtc_bench", "jaguar", "64"))
	#from office
	Utilities.setSession("apart")
	trial = TrialMeanResult(Utilities.getTrial("gtc", "jaguar", "512"))
	return trial

def extractNonCallpath(input):
	# extract the non-callpath events from the trial
	extractor = ExtractNonCallpathEventOperation(input)
	#extractor = ExtractCallpathEventOperation(input)
	return extractor.processData().get(0)

def getTop10andMain(input):
	print "Getting top 10 events (sorted by exclusive time)..."
	getTop10 = TopXEvents(input, input.getTimeMetric(), AbstractResult.EXCLUSIVE, 10)
	top10 = getTop10.processData().get(0)
	sorted = top10.getEvents()

	# also put main in here, if it wasn't selected
	extractor = ExtractEventOperation(input, input.getMainEvent())
	justMain = extractor.processData().get(0)
	merger = MergeTrialsOperation(top10)
	merger.addInput(justMain)
	outputs = merger.processData()
	return sorted, outputs.get(0)

def deriveMetrics(input):
	derivor = DeriveAllMetricsOperation(input)
	return derivor.processData().get(0);

def processEvent(input, event, type):
	value = input.getDataPoint(0, event, input.getTimeMetric(), type)

	# get the computation density for this routine
	metric1Value = input.getDataPoint(0, event, DerivedMetrics.MFLOP_RATE, type)

	# get the L1 access rate for this routine
	metric2Value = input.getDataPoint(0, event, DerivedMetrics.MEM_ACCESSES, type)

	# get the L1 hit ratio
	metric4Value = input.getDataPoint(0, event, DerivedMetrics.L1_HIT_RATE, type)

	# get the L2 access rate for this routine
	metric5Value = input.getDataPoint(0, event, DerivedMetrics.L2_ACCESSES, type)

	# get the L2 hit ratio
	metric7Value = input.getDataPoint(0, event, DerivedMetrics.L2_HIT_RATE, type)

	# get the communication density for this routine
	# group == MPI ? value : 0.0;
	if event.startswith("MPI_"):
		metric8Value = value
	else:
		metric8Value = 0.0;
	
	print event, value, metric1Value, metric2Value, metric4Value, metric5Value, metric7Value, metric8Value

print "--------------- JPython test script start ------------"

print "doing single trial analysis for gtc on jaguar"

# create a rulebase for processing
loadRules()

# load the trial
trial = loaddata()

# extract the non-callpath events
extracted = extractNonCallpath(trial)

# extract the top 10 events, along with main, and get the event names sorted by exclusive
sorted,top10 = getTop10andMain(extracted)
top10=extracted

# calculate all derived metrics
derived = deriveMetrics(top10)

#print "EventName, value, MFlop/s, L1 Accesses/s, L1 Hit Rate, L2 Accesses/s, L2 Hit Rate, Communication"

# process main
#processEvent(derived, derived.getMainEvent(), AbstractResult.INCLUSIVE)

# process the top 10 events
for event in sorted:
	#processEvent(derived, event, AbstractResult.EXCLUSIVE)
	MeanEventFact.compareEventToMain(derived, derived.getMainEvent(), derived, event)

RuleHarness.getInstance().processRules()

print "---------------- JPython test script end -------------"