gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Trying a new mutation operator

Browse Source
This commit is contained in:
Philippe Tillet
2014-09-10 11:10:19 -04:00
parent 46567ff812
commit 539efa029f
4 changed files with 57 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
autotune/config.ini
View File

@@ -1,10 +1,10 @@
#will save the archive into /tmp/name-of-operation.dat
tmp-folder = /tmp/
[vector-axpy]
devices = 0
precision = all
size = 10000000
#~ [vector-axpy]
#~ devices = 0
#~ precision = all
#~ size = 10000000
#~ [matrix-axpy]
#~ devices = 0
@@ -19,6 +19,6 @@ size = 10000000
[matrix-product]
devices = 0
precision = all
precision = single
layout = NT
size = 1536, 1536, 1536

2
autotune/python/autotune.py
View File

@@ -71,7 +71,7 @@ def do_tuning(config_fname, spec_fname, viennacl_root):
p = config[operation]
confdevices = p['devices']
devices = utils.DEVICES_PRESETS[confdevices] if confdevices in utils.DEVICES_PRESETS else [utils.all_devices[int(i)] for i in confdevices]
precisions = ['single', 'double'] if 'all' in p['precision'] else p['precision']
precisions = map_to_list((str, p['precision']))
datatypes = [DATATYPES[k] for k in precisions]
s = map_to_list((int, p['size']))

57
autotune/python/genetic_operators.py
View File

@@ -53,10 +53,13 @@ class GeneticOperators(object):
return (new_x, new_y)
def repair(self,func):
def repair_impl(child):
D = odict(zip(self.parameter_names, child))
dummy_template = self.build_template(self.ParameterType(*D.values()))
FetchingPolicy = vcl.atidlas.FetchingPolicy;
D['local-size-0'] = max(1, D['local-size-0'])
D['local-size-1'] = max(1, D['local-size-1'])
if 'local-size-1' not in D:
D['local-size-0'] = min(D['local-size-0'], self.device.max_work_group_size)
elif D['local-size-0']*D['local-size-1'] > self.device.max_work_group_size:
@@ -67,11 +70,13 @@ class GeneticOperators(object):
if self.ParameterType is vcl.atidlas.MatrixProductTemplate.Parameters:
if dummy_template.A_trans != 'N' and dummy_template.B_trans != 'T':
D['simd-width'] = 1
D['mS'] = max(D['mS'], D['simd-width'])
D['mS'] = D['mS'] - D['mS']%D['simd-width']
D['kL'] = max(1, D['kL'])
D['kS'] = max(1, D['kS'])
D['mS'] = max(D['mS'], D['simd-width'])
D['nS'] = max(D['nS'], D['simd-width'])
D['mS'] = D['mS'] - D['mS']%D['simd-width']
D['nS'] = D['nS'] - D['nS']%D['simd-width']
@@ -100,16 +105,48 @@ class GeneticOperators(object):
for i in range(len(child)):
if child[i] != new_child[i]:
child[i] = new_child[i]
return offspring
return wrappper
def mutate(self, individual, indpb):
for i in range(len(individual)):
if random.random() < indpb:
j = self.parameters[i].index(individual[i])
j = max(0,min(random.randint(j-2, j+2),len(self.parameters[i])-1))
individual[i] = self.parameters[i][j]
def mutate(self, individual):
p = random.random()
coef = random.choice([2, 4])
multiply_or_divide = random.choice([lambda x:x/coef, lambda x:x*coef])
if p < 0.1 :
individual[0] = random.choice(self.parameters[0])
elif p < 0.2:
idx = random.choice([1, 4])
nidx = 4 if idx==1 else 1
individual[idx] = individual[idx]*coef
individual[nidx] = individual[nidx]/coef
elif p < 0.3:
idx = random.choice([1, 4])
individual[idx] = multiply_or_divide(individual[idx])
if idx==1:
individual[9] = multiply_or_divide(individual[9])
elif p < 0.4:
idx = random.choice([3, 6])
nidx = 6 if idx==3 else 3
individual[idx] = individual[idx]*coef
individual[nidx] = individual[nidx]/coef
elif p < 0.5:
idx = random.choice([3, 6])
individual[idx] = multiply_or_divide(individual[idx])
if idx==3:
individual[10] = multiply_or_divide(individual[10])
elif p < 0.6:
individual[2] = multiply_or_divide(individual[2])
elif p < 0.7:
individual[4] = multiply_or_divide(individual[4])
elif p < 0.8:
individual[7] = random.choice([x for x in self.parameters[7] if x!=individual[7]])
elif p < 0.9:
individual[8] = random.choice([x for x in self.parameters[8] if x!=individual[8]])
elif p < 1:
idx = random.choice([9, 10])
nidx = 10 if idx==9 else 9
individual[idx] = individual[idx]*coef
individual[nidx] = individual[nidx]/coef
return individual,
def evaluate(self, individual):

9
autotune/python/optimize.py
View File

@@ -60,15 +60,14 @@ def genetic(statement, context, TemplateType, build_template, parameter_names, a
toolbox = base.Toolbox()
toolbox.register("individual", tools.initIterate, creator.Individual, gen.init)
toolbox.register("population", tools.initRepeat, list, toolbox.individual)
toolbox.decorate("population", gen.repair)
toolbox.register("evaluate", gen.evaluate)
toolbox.register("mate", tools.cxUniform, indpb=0.1)
toolbox.register("mate", tools.cxTwoPoint)
toolbox.decorate("mate", gen.repair)
toolbox.register("mutate", gen.mutate, indpb=0.1)
toolbox.register("mutate", gen.mutate)
toolbox.decorate("mutate", gen.repair)
toolbox.register("select", tools.selNSGA2)
pop = toolbox.population(n=70)
pop = toolbox.population(n=30)
hof = tools.HallOfFame(1)
best_performer = lambda x: max([compute_perf(hof[0].fitness.values[0]) for t in x])
@@ -81,4 +80,4 @@ def genetic(statement, context, TemplateType, build_template, parameter_names, a
stats.register("max (" + perf_metric + ")", lambda x: max([compute_perf(hof[0].fitness.values[0]) for t in x]))
stats.register("profile ", lambda x: '(%s)'%','.join(map(str,hof[0])))
pop = eaMuPlusLambda(pop, toolbox, 70, 100, cxpb=0.1, mutpb=0.1, maxtime='5m0s', maxgen=1000, halloffame=hof, compute_perf=compute_perf, perf_metric=perf_metric)
pop = eaMuPlusLambda(pop, toolbox, 30, 50, cxpb=0.2, mutpb=0.3, maxtime='5m0s', maxgen=200, halloffame=hof, compute_perf=compute_perf, perf_metric=perf_metric)