提交 290fc8bc authored 作者: Saizheng Zhang's avatar Saizheng Zhang

profile: more changes

上级 96a257ad
...@@ -31,7 +31,6 @@ from six import iteritems ...@@ -31,7 +31,6 @@ from six import iteritems
from theano.gof import graph from theano.gof import graph
from theano.configparser import AddConfigVar, BoolParam, IntParam, StrParam from theano.configparser import AddConfigVar, BoolParam, IntParam, StrParam
theano_imported_time = time.time() theano_imported_time = time.time()
config = theano.config config = theano.config
...@@ -1302,16 +1301,16 @@ class ProfileStats(object): ...@@ -1302,16 +1301,16 @@ class ProfileStats(object):
print("-----------------", file=file) print("-----------------", file=file)
self.optimizer_profile[0].print_profile(file, self.optimizer_profile[0].print_profile(file,
self.optimizer_profile[1]) self.optimizer_profile[1])
self.print_tips() self.print_tips(file)
def print_tips(self): def print_tips(self, file):
print("""Here are tips to potentially make your code run faster print("""Here are tips to potentially make your code run faster
(if you think of new ones, suggest them on the mailing list). (if you think of new ones, suggest them on the mailing list).
Test them first, as they are not guaranteed to always provide a speedup.""") Test them first, as they are not guaranteed to always provide a speedup.""", file = file)
from theano import tensor as T
from theano.tensor.raw_random import RandomFunction RandomFunction = theano.tensor.raw_random.RandomFunction
import theano scal = theano.scalar
import theano.scalar as scal T = theano.tensor
scalar_op_amdlibm_no_speed_up = [scal.LT, scal.GT, scal.LE, scal.GE, scalar_op_amdlibm_no_speed_up = [scal.LT, scal.GT, scal.LE, scal.GE,
scal.EQ, scal.NEQ, scal.InRange, scal.EQ, scal.NEQ, scal.InRange,
scal.Switch, scal.OR, scal.XOR, scal.Switch, scal.OR, scal.XOR,
...@@ -1356,7 +1355,7 @@ class ProfileStats(object): ...@@ -1356,7 +1355,7 @@ class ProfileStats(object):
return True return True
elif s_op.__class__ not in scalar_op_amdlibm_no_speed_up: elif s_op.__class__ not in scalar_op_amdlibm_no_speed_up:
print("We don't know if amdlibm will accelerate " print("We don't know if amdlibm will accelerate "
"this scalar op.", s_op) "this scalar op.", s_op , file = file)
return False return False
def exp_float32_op(op): def exp_float32_op(op):
...@@ -1369,28 +1368,28 @@ class ProfileStats(object): ...@@ -1369,28 +1368,28 @@ class ProfileStats(object):
printed_tip = False printed_tip = False
# tip 1 # tip 1
if config.floatX == 'float64': if config.floatX == 'float64':
print(" - Try the Theano flag floatX=float32") print(" - Try the Theano flag floatX=float32", file = file)
printed_tip = True printed_tip = True
# tip 2 # tip 2
if not config.lib.amdlibm and any([amdlibm_speed_up(a.op) for i, a if not config.lib.amdlibm and any([amdlibm_speed_up(a.op) for a
in self.apply_time]): in self.apply_time]):
print(" - Try installing amdlibm and set the Theano flag " print(" - Try installing amdlibm and set the Theano flag "
"lib.amdlibm=True. This speeds up only some Elemwise " "lib.amdlibm=True. This speeds up only some Elemwise "
"operation.") "operation.", file = file)
printed_tip = True printed_tip = True
# tip 3 # tip 3
if not config.lib.amdlibm and any([exp_float32_op(a.op) and if not config.lib.amdlibm and any([exp_float32_op(a.op) and
a.inputs[0].dtype == 'float32' a.inputs[0].dtype == 'float32'
for i, a in self.apply_time]): for a in self.apply_time]):
print(" - With the default gcc libm, exp in float32 is slower " print(" - With the default gcc libm, exp in float32 is slower "
"than in float64! Try Theano flag floatX=float64, or " "than in float64! Try Theano flag floatX=float64, or "
"install amdlibm and set the theano flags lib.amdlibm=True") "install amdlibm and set the theano flags lib.amdlibm=True", file = file)
printed_tip = True printed_tip = True
# tip 4 # tip 4
for a, t in iteritems(self.apply_time): for a in self.apply_time:
node = a node = a
if (isinstance(node.op, T.Dot) and if (isinstance(node.op, T.Dot) and
all([len(i.type.broadcastable) == 2 all([len(i.type.broadcastable) == 2
...@@ -1399,25 +1398,25 @@ class ProfileStats(object): ...@@ -1399,25 +1398,25 @@ class ProfileStats(object):
" dot22 (which is faster). Make sure the inputs are " " dot22 (which is faster). Make sure the inputs are "
"float32 or float64, and are the same for both inputs. " "float32 or float64, and are the same for both inputs. "
"Currently they are: %s" % "Currently they are: %s" %
[i.type for i in node.inputs]) [i.type for i in node.inputs], file = file)
printed_tip = True printed_tip = True
# tip 5 # tip 5
for a, t in iteritems(self.apply_time): for a in self.apply_time:
node = a node = a
if isinstance(node.op, RandomFunction): if isinstance(node.op, RandomFunction):
printed_tip = True printed_tip = True
print(" - Replace the default random number generator by " print(" - Replace the default random number generator by "
"'from theano.sandbox.rng_mrg import MRG_RandomStreams " "'from theano.sandbox.rng_mrg import MRG_RandomStreams "
"as RandomStreams', as this is is faster. It is still " "as RandomStreams', as this is is faster. It is still "
"experimental, but seems to work correctly.") "experimental, but seems to work correctly.", file = file)
if config.device.startswith("gpu"): if config.device.startswith("gpu"):
print(" - MRG_RandomStreams is the only random number" print(" - MRG_RandomStreams is the only random number"
" generator supported on the GPU.") " generator supported on the GPU.", file = file)
break break
if not printed_tip: if not printed_tip:
print(" Sorry, no tip for today.") print(" Sorry, no tip for today.", file = file)
if False: # old code still to be ported from ProfileMode if False: # old code still to be ported from ProfileMode
......
This source diff could not be displayed because it is too large. You can view the blob instead.
Markdown 格式
0%
您添加了 0 到此讨论。请谨慎行事。
请先完成此评论的编辑!
注册 或者 后发表评论