Skip to content

P
pytensor
提交 4be478ff authored 作者: James Bergstra's avatar James Bergstra
浏览文件
操作

fixed test of canonize that was modifying global mode variables

上级 88b0c613
隐藏空白字符变更
内嵌 并排
正在显示 包含 214 行增加 和 191 行删除
theano/tensor/tests/test_opt.py
浏览文件 @ 4be478ff
...@@ -246,16 +246,20 @@ class test_canonize(unittest.TestCase): ...@@ -246,16 +246,20 @@ class test_canonize(unittest.TestCase):
#We must be sure that the Canonizer is working, but that we don't have other #We must be sure that the Canonizer is working, but that we don't have other
# optimisation that could hide bug in the Canonizer as local_elemwise_fusion # optimisation that could hide bug in the Canonizer as local_elemwise_fusion
mode=compile.mode.predefined_modes[compile.mode.default_mode] mode=compile.mode.predefined_modes[compile.mode.default_mode]
mode._optimizer=gof.Query(["canonicalize"]) old_optimizer = mode._optimizer
mode._optimizer=mode._optimizer.excluding('local_elemwise_fusion') try:
for id, [g, sym_inputs, val_inputs, nb_elemwise, out_dtype] in enumerate(cases): mode._optimizer=gof.Query(["canonicalize"])
f = compile.function(list(sym_inputs), g, mode._optimizer=mode._optimizer.excluding('local_elemwise_fusion')
#we need the optimisation enabled, debug do this. for id, [g, sym_inputs, val_inputs, nb_elemwise, out_dtype] in enumerate(cases):
mode=mode) f = compile.function(list(sym_inputs), g,
#we need the optimisation enabled, debug do this.
out = f(*val_inputs) mode=mode)
assert(len(f.maker.env.toposort())==nb_elemwise)
assert(out_dtype==out.dtype) out = f(*val_inputs)
assert(len(f.maker.env.toposort())==nb_elemwise)
assert(out_dtype==out.dtype)
finally:
mode._optimizer = old_optimizer
def test_elemwise_multiple_inputs_optimisation2(self): def test_elemwise_multiple_inputs_optimisation2(self):
""" """
...@@ -367,130 +371,134 @@ class test_canonize(unittest.TestCase): ...@@ -367,130 +371,134 @@ class test_canonize(unittest.TestCase):
#We must be sure that the Canonizer is working, but that we don't have other #We must be sure that the Canonizer is working, but that we don't have other
# optimisation that could hide bug in the Canonizer as local_elemwise_fusion # optimisation that could hide bug in the Canonizer as local_elemwise_fusion
mode=compile.mode.predefined_modes[compile.mode.default_mode] mode=compile.mode.predefined_modes[compile.mode.default_mode]
mode._optimizer=gof.Query(["canonicalize"]) old_optimizer = mode._optimizer
mode._optimizer=mode._optimizer.excluding('local_elemwise_fusion') try:
mode._optimizer=gof.Query(["canonicalize"])
#test x / x -> 1 mode._optimizer=mode._optimizer.excluding('local_elemwise_fusion')
for id, (g, sym_inputs, val_inputs, out_dtype) in enumerate([(fx/fx,[fx],[fxv],'float32'),
(dx/dx,[dx],[dxv],'float64'), #test x / x -> 1
(fv/fv,[fv],[fvv],'float32'), for id, (g, sym_inputs, val_inputs, out_dtype) in enumerate([(fx/fx,[fx],[fxv],'float32'),
(dv/dv,[dv],[dvv],'float64'), (dx/dx,[dx],[dxv],'float64'),
]): (fv/fv,[fv],[fvv],'float32'),
f = compile.function(list(sym_inputs), g, (dv/dv,[dv],[dvv],'float64'),
mode=mode) ]):
out = f(*val_inputs) f = compile.function(list(sym_inputs), g,
assert (out==numpy.ones(shp, dtype=out_dtype)).all() mode=mode)
topo=f.maker.env.toposort() out = f(*val_inputs)
assert len(topo)==1 assert (out==numpy.ones(shp, dtype=out_dtype)).all()
assert isinstance(topo[0].op,(T.Elemwise,)) topo=f.maker.env.toposort()
assert isinstance(topo[0].op.scalar_op,theano.scalar.basic.Second) assert len(topo)==1
assert len(topo[0].inputs)==2 assert isinstance(topo[0].op,(T.Elemwise,))
assert(out_dtype==out.dtype) assert isinstance(topo[0].op.scalar_op,theano.scalar.basic.Second)
assert len(topo[0].inputs)==2
#test (x * y) / x -> y assert(out_dtype==out.dtype)
for id,(g, sym_inputs, val_inputs, nb_elemwise, out_dtype) in enumerate([
((dx*dy)/dx,[dx,dy],[dxv,dyv],0,'float64'), #test (x * y) / x -> y
((fx*fy)/fx,[fx,fy],[fxv,fyv],0,'float32'), for id,(g, sym_inputs, val_inputs, nb_elemwise, out_dtype) in enumerate([
((dv*dy)/dv,[dv,dy],[dvv,dyv],0,'float64'), ((dx*dy)/dx,[dx,dy],[dxv,dyv],0,'float64'),
((fv*fy)/fv,[fv,fy],[fvv,fyv],0,'float32'), ((fx*fy)/fx,[fx,fy],[fxv,fyv],0,'float32'),
#must broadcast as their is a dimshuffle in the computation ((dv*dy)/dv,[dv,dy],[dvv,dyv],0,'float64'),
((dx*dv)/dx,[dx,dv],[dxv,dvv],1,'float64'), ((fv*fy)/fv,[fv,fy],[fvv,fyv],0,'float32'),
#topo: [Elemwise{second,no_inplace}(x, <TensorType(float64, row)>)] #must broadcast as their is a dimshuffle in the computation
((fx*fv)/fx,[fx,fv],[fxv,fvv],1,'float32') ((dx*dv)/dx,[dx,dv],[dxv,dvv],1,'float64'),
#topo: [Elemwise{second,no_inplace}(x, <TensorType(float32, row)>)] #topo: [Elemwise{second,no_inplace}(x, <TensorType(float64, row)>)]
]): ((fx*fv)/fx,[fx,fv],[fxv,fvv],1,'float32')
f = compile.function(list(sym_inputs), g, #topo: [Elemwise{second,no_inplace}(x, <TensorType(float32, row)>)]
mode=mode) ]):
out = f(*val_inputs) f = compile.function(list(sym_inputs), g,
assert numpy.allclose(out,val_inputs[1]) mode=mode)
topo=f.maker.env.toposort() out = f(*val_inputs)
assert len(topo)==nb_elemwise assert numpy.allclose(out,val_inputs[1])
assert(out_dtype==out.dtype) topo=f.maker.env.toposort()
assert len(topo)==nb_elemwise
#test x / y / x -> 1 / y assert(out_dtype==out.dtype)
for id,(g, sym_inputs, val_inputs, nb_elemwise, out_dtype) in enumerate([
((dx/dy)/dx,[dx,dy],[dxv,dyv],1,'float64'), #test x / y / x -> 1 / y
((fx/fy)/fx,[fx,fy],[fxv,fyv],1,'float32'), for id,(g, sym_inputs, val_inputs, nb_elemwise, out_dtype) in enumerate([
((dv/dy)/dv,[dv,dy],[dvv,dyv],1,'float64'), ((dx/dy)/dx,[dx,dy],[dxv,dyv],1,'float64'),
((fv/fy)/fv,[fv,fy],[fvv,fyv],1,'float32'), ((fx/fy)/fx,[fx,fy],[fxv,fyv],1,'float32'),
#must broadcast as their is a dimshuffle in the computation ((dv/dy)/dv,[dv,dy],[dvv,dyv],1,'float64'),
((fv/fy)/fv,[fv,fy],[fvv,fyv],1,'float32'),
((dx/dv)/dx,[dx,dv],[dxv,dvv],2,'float64'), #must broadcast as their is a dimshuffle in the computation
#topo: [Elemwise{inv,no_inplace}(<TensorType(float64, row)>), Elemwise{second,no_inplace}(x, Elemwise{inv,no_inplace}.0)]
((fx/fv)/fx,[fx,fv],[fxv,fvv],2,'float32'), ((dx/dv)/dx,[dx,dv],[dxv,dvv],2,'float64'),
#topo:[Elemwise{inv,no_inplace}(<TensorType(float32, row)>), Elemwise{second,no_inplace}(x, Elemwise{inv,no_inplace}.0)] #topo: [Elemwise{inv,no_inplace}(<TensorType(float64, row)>), Elemwise{second,no_inplace}(x, Elemwise{inv,no_inplace}.0)]
]): ((fx/fv)/fx,[fx,fv],[fxv,fvv],2,'float32'),
f = compile.function(list(sym_inputs), g, #topo:[Elemwise{inv,no_inplace}(<TensorType(float32, row)>), Elemwise{second,no_inplace}(x, Elemwise{inv,no_inplace}.0)]
mode=mode) ]):
out = f(*val_inputs) f = compile.function(list(sym_inputs), g,
assert numpy.allclose(out,(1/val_inputs[1])) mode=mode)
topo=f.maker.env.toposort() out = f(*val_inputs)
assert len(topo)==nb_elemwise assert numpy.allclose(out,(1/val_inputs[1]))
assert isinstance(topo[0].op,(T.Elemwise,)) topo=f.maker.env.toposort()
assert isinstance(topo[0].op.scalar_op,(theano.scalar.basic.Inv, theano.scalar.basic.TrueDiv)) assert len(topo)==nb_elemwise
assert(out_dtype==out.dtype) assert isinstance(topo[0].op,(T.Elemwise,))
assert isinstance(topo[0].op.scalar_op,(theano.scalar.basic.Inv, theano.scalar.basic.TrueDiv))
#test (a / b) * (b / c) * (c / d) -> a / d assert(out_dtype==out.dtype)
for id,(g, sym_inputs, val_inputs, out_dtype) in enumerate([
((dx / dy) * (dy / dz) * (dz / dw),[dx,dy,dz,dw],[dxv,dyv,dzv,dwv],'float64'), #test (a / b) * (b / c) * (c / d) -> a / d
((fx / fy) * (fy / fz) * (fz / fw),[fx,fy,fz,fw],[fxv,fyv,fzv,fwv],'float32'), for id,(g, sym_inputs, val_inputs, out_dtype) in enumerate([
((dv / dy) * (dy / dz) * (dz / dw),[dv,dy,dz,dw],[dvv,dyv,dzv,dwv],'float64'), ((dx / dy) * (dy / dz) * (dz / dw),[dx,dy,dz,dw],[dxv,dyv,dzv,dwv],'float64'),
((fv / fy) * (fy / fz) * (fz / fw),[fv,fy,fz,fw],[fvv,fyv,fzv,fwv],'float32'), ((fx / fy) * (fy / fz) * (fz / fw),[fx,fy,fz,fw],[fxv,fyv,fzv,fwv],'float32'),
((dx / dv) * (dv / dz) * (dz / dw),[dx,dv,dz,dw],[dxv,dvv,dzv,dwv],'float64'), ((dv / dy) * (dy / dz) * (dz / dw),[dv,dy,dz,dw],[dvv,dyv,dzv,dwv],'float64'),
((fx / fv) * (fv / fz) * (fz / fw),[fx,fv,fz,fw],[fxv,fvv,fzv,fwv],'float32'), ((fv / fy) * (fy / fz) * (fz / fw),[fv,fy,fz,fw],[fvv,fyv,fzv,fwv],'float32'),
((dx / dy) * (dy / dv) * (dv / dw),[dx,dy,dv,dw],[dxv,dyv,dvv,dwv],'float64'), ((dx / dv) * (dv / dz) * (dz / dw),[dx,dv,dz,dw],[dxv,dvv,dzv,dwv],'float64'),
((fx / fy) * (fy / fv) * (fv / fw),[fx,fy,fv,fw],[fxv,fyv,fvv,fwv],'float32'), ((fx / fv) * (fv / fz) * (fz / fw),[fx,fv,fz,fw],[fxv,fvv,fzv,fwv],'float32'),
((dx / dy) * (dy / dz) * (dz / dv),[dx,dy,dz,dv],[dxv,dyv,dzv,dvv],'float64'), ((dx / dy) * (dy / dv) * (dv / dw),[dx,dy,dv,dw],[dxv,dyv,dvv,dwv],'float64'),
((fx / fy) * (fy / fz) * (fz / fv),[fx,fy,fz,fv],[fxv,fyv,fzv,fvv],'float32'), ((fx / fy) * (fy / fv) * (fv / fw),[fx,fy,fv,fw],[fxv,fyv,fvv,fwv],'float32'),
]): ((dx / dy) * (dy / dz) * (dz / dv),[dx,dy,dz,dv],[dxv,dyv,dzv,dvv],'float64'),
f = compile.function(list(sym_inputs), g, ((fx / fy) * (fy / fz) * (fz / fv),[fx,fy,fz,fv],[fxv,fyv,fzv,fvv],'float32'),
mode=mode) ]):
out = f(*val_inputs) f = compile.function(list(sym_inputs), g,
assert numpy.allclose(out,(val_inputs[0]/val_inputs[3])) mode=mode)
topo=f.maker.env.toposort() out = f(*val_inputs)
assert len(topo)==1 assert numpy.allclose(out,(val_inputs[0]/val_inputs[3]))
assert isinstance(topo[0].op,(T.Elemwise,)) topo=f.maker.env.toposort()
assert isinstance(topo[0].op.scalar_op,theano.scalar.basic.TrueDiv) assert len(topo)==1
assert len(topo[0].inputs)==2 assert isinstance(topo[0].op,(T.Elemwise,))
assert(out_dtype==out.dtype) assert isinstance(topo[0].op.scalar_op,theano.scalar.basic.TrueDiv)
assert len(topo[0].inputs)==2
#test (2.0 * x) / (4.0 * y) -> (0.5 * x) / y assert(out_dtype==out.dtype)
for id,(g, sym_inputs, val_inputs, out_dtype) in enumerate([
(((2.0*dx)/(4.0*dy)),[dx,dy],[dxv,dyv],'float64'), #test (2.0 * x) / (4.0 * y) -> (0.5 * x) / y
(((2.0*fx)/(4.0*fy)),[fx,fy],[fxv,fyv],'float32'), for id,(g, sym_inputs, val_inputs, out_dtype) in enumerate([
(((2.0*dv)/(4.0*dy)),[dv,dy],[dvv,dyv],'float64'), (((2.0*dx)/(4.0*dy)),[dx,dy],[dxv,dyv],'float64'),
(((2.0*fv)/(4.0*fy)),[fv,fy],[fvv,fyv],'float32'), (((2.0*fx)/(4.0*fy)),[fx,fy],[fxv,fyv],'float32'),
(((2.0*dx)/(4.0*dv)),[dx,dv],[dxv,dvv],'float64'), (((2.0*dv)/(4.0*dy)),[dv,dy],[dvv,dyv],'float64'),
(((2.0*fx)/(4.0*fv)),[fx,fv],[fxv,fvv],'float32'), (((2.0*fv)/(4.0*fy)),[fv,fy],[fvv,fyv],'float32'),
]): (((2.0*dx)/(4.0*dv)),[dx,dv],[dxv,dvv],'float64'),
f = compile.function(list(sym_inputs), g, (((2.0*fx)/(4.0*fv)),[fx,fv],[fxv,fvv],'float32'),
mode=mode) ]):
out = f(*val_inputs) f = compile.function(list(sym_inputs), g,
assert numpy.allclose(out,(0.5*val_inputs[0]/val_inputs[1])) mode=mode)
topo=f.maker.env.toposort() out = f(*val_inputs)
assert len(topo)==2 assert numpy.allclose(out,(0.5*val_inputs[0]/val_inputs[1]))
assert isinstance(topo[0].op,(T.Elemwise,)) topo=f.maker.env.toposort()
assert isinstance(topo[0].op.scalar_op,theano.scalar.basic.Mul) assert len(topo)==2
assert len(topo[0].inputs)==2 assert isinstance(topo[0].op,(T.Elemwise,))
assert isinstance(topo[1].op,(T.Elemwise,)) assert isinstance(topo[0].op.scalar_op,theano.scalar.basic.Mul)
assert isinstance(topo[1].op.scalar_op,theano.scalar.basic.TrueDiv) assert len(topo[0].inputs)==2
assert len(topo[1].inputs)==2 assert isinstance(topo[1].op,(T.Elemwise,))
assert(out_dtype==out.dtype) assert isinstance(topo[1].op.scalar_op,theano.scalar.basic.TrueDiv)
assert len(topo[1].inputs)==2
#test 2 * x / 2 -> x assert(out_dtype==out.dtype)
for id,(g, sym_inputs, val_inputs, out_dtype) in enumerate([
((2*dx)/2,[dx],[dxv],'float64'), #test 2 * x / 2 -> x
((2*fx)/2,[fx],[fxv],'float32'), for id,(g, sym_inputs, val_inputs, out_dtype) in enumerate([
((2*dv)/2,[dv],[dvv],'float64'), ((2*dx)/2,[dx],[dxv],'float64'),
((2*fv)/2,[fv],[fvv],'float32'), ((2*fx)/2,[fx],[fxv],'float32'),
]): ((2*dv)/2,[dv],[dvv],'float64'),
f = compile.function(list(sym_inputs), g, ((2*fv)/2,[fv],[fvv],'float32'),
mode=mode) ]):
out = f(*val_inputs) f = compile.function(list(sym_inputs), g,
assert numpy.allclose(out,val_inputs[0]) mode=mode)
topo=f.maker.env.toposort() out = f(*val_inputs)
assert len(topo)==0 assert numpy.allclose(out,val_inputs[0])
assert(out_dtype==out.dtype) topo=f.maker.env.toposort()
assert len(topo)==0
assert(out_dtype==out.dtype)
finally:
mode._optimizer = old_optimizer
def test_multiple_case_that_fail(self): def test_multiple_case_that_fail(self):
...@@ -510,43 +518,48 @@ class test_canonize(unittest.TestCase): ...@@ -510,43 +518,48 @@ class test_canonize(unittest.TestCase):
#We must be sure that the Canonizer is working, but that we don't have other #We must be sure that the Canonizer is working, but that we don't have other
# optimisation that could hide bug in the Canonizer as local_elemwise_fusion # optimisation that could hide bug in the Canonizer as local_elemwise_fusion
mode=compile.mode.predefined_modes[compile.mode.default_mode] mode=compile.mode.predefined_modes[compile.mode.default_mode]
mode._optimizer=gof.Query(["canonicalize"]) old_optimizer = mode._optimizer
mode._optimizer=mode._optimizer.excluding('local_elemwise_fusion') try:
mode._optimizer=gof.Query(["canonicalize"])
#test fail! mode._optimizer=mode._optimizer.excluding('local_elemwise_fusion')
#test x / y / z -> x / (y * z)
for (g, sym_inputs, val_inputs, out_dtype) in [ #test fail!
((dx/dy)/dz,[dx,dy,dz],[dxv,dyv,dzv],'float64'), #test x / y / z -> x / (y * z)
((fx/fy)/fz,[fx,fy,fz],[fxv,fyv,fzv],'float32') for (g, sym_inputs, val_inputs, out_dtype) in [
]: ((dx/dy)/dz,[dx,dy,dz],[dxv,dyv,dzv],'float64'),
f = compile.function(list(sym_inputs), g, ((fx/fy)/fz,[fx,fy,fz],[fxv,fyv,fzv],'float32')
mode=mode) ]:
out = f(*val_inputs) f = compile.function(list(sym_inputs), g,
assert numpy.allclose(out,val_inputs[0]/val_inputs[1]/val_inputs[2]) mode=mode)
topo=f.maker.env.toposort() out = f(*val_inputs)
print topo assert numpy.allclose(out,val_inputs[0]/val_inputs[1]/val_inputs[2])
assert len(topo)==2 topo=f.maker.env.toposort()
assert isinstance(topo[0].op,(T.Elemwise,)) print topo
assert isinstance(topo[0].op.scalar_op,theano.scalar.basic.Inv) assert len(topo)==2
assert len(topo[0].inputs)==1 assert isinstance(topo[0].op,(T.Elemwise,))
assert(out_dtype==out.dtype) assert isinstance(topo[0].op.scalar_op,theano.scalar.basic.Inv)
assert len(topo[0].inputs)==1
#test x / (y / z) -> (x * z) / y assert(out_dtype==out.dtype)
for (g, sym_inputs, val_inputs, out_dtype) in [
(dx/(dy/dz),[dx,dy,dz],[dxv,dyv,dzv],'float64'), #test x / (y / z) -> (x * z) / y
(fx/(fy/fz),[fx,fy,fz],[fxv,fyv,fzv],'float32') for (g, sym_inputs, val_inputs, out_dtype) in [
]: (dx/(dy/dz),[dx,dy,dz],[dxv,dyv,dzv],'float64'),
f = compile.function(list(sym_inputs), g, (fx/(fy/fz),[fx,fy,fz],[fxv,fyv,fzv],'float32')
mode=mode) ]:
out = f(*val_inputs) f = compile.function(list(sym_inputs), g,
assert numpy.allclose(out,val_inputs[0]/(val_inputs[1]/val_inputs[2])) mode=mode)
topo=f.maker.env.toposort() out = f(*val_inputs)
print topo assert numpy.allclose(out,val_inputs[0]/(val_inputs[1]/val_inputs[2]))
assert len(topo)==2 topo=f.maker.env.toposort()
assert isinstance(topo[0].op,(T.Elemwise,)) print topo
assert isinstance(topo[0].op.scalar_op,theano.scalar.basic.Inv) assert len(topo)==2
assert len(topo[0].inputs)==1 assert isinstance(topo[0].op,(T.Elemwise,))
assert(out_dtype==out.dtype) assert isinstance(topo[0].op.scalar_op,theano.scalar.basic.Inv)
assert len(topo[0].inputs)==1
assert(out_dtype==out.dtype)
finally:
mode._optimizer = old_optimizer
def test_dont_merge_if_multiple_client(self): def test_dont_merge_if_multiple_client(self):
""" test those case take from the comment in Canonizer """ test those case take from the comment in Canonizer
...@@ -571,10 +584,16 @@ def test_local_shape_lift_dot(): ...@@ -571,10 +584,16 @@ def test_local_shape_lift_dot():
for y in [fvector, fmatrix]: for y in [fvector, fmatrix]:
i = x() i = x()
j = y() j = y()
print 'I SHAPE', i.type.shape
print 'J SHAPE', j.type.shape
d = shape(dot(i,j)) d = shape(dot(i,j))
g = Env([i,j], [d]) if x is fvector and y is fvector:
gof.TopoOptimizer(gof.LocalOptGroup(local_shape_lift_dot), order='out_to_in').optimize(g) assert d == ()
assert pprint(g.outputs[0]) == args_to_result[(x,y)] else:
g = Env([i,j], [d])
gof.TopoOptimizer(gof.LocalOptGroup(local_shape_lift_dot), order='out_to_in').optimize(g)
print pprint(g.outputs[0]), args_to_result[(x,y)]
assert pprint(g.outputs[0]) == args_to_result[(x,y)]
# def test_plusmin(self): # def test_plusmin(self):
# x, y, z = inputs() # x, y, z = inputs()
...@@ -982,23 +1001,27 @@ class test_fusion(unittest.TestCase): ...@@ -982,23 +1001,27 @@ class test_fusion(unittest.TestCase):
#Follow up. Clinker do the same... second cause? #Follow up. Clinker do the same... second cause?
mode2=compile.Mode(linker(), copy.copy(compile.mode.OPT_FAST_RUN)) mode2=compile.Mode(linker(), copy.copy(compile.mode.OPT_FAST_RUN))
# mode2=copy.copy(compile.mode.predefined_modes['FAST_RUN']) # mode2=copy.copy(compile.mode.predefined_modes['FAST_RUN'])
mode2._optimizer=mode2._optimizer.excluding('local_elemwise_fusion') old_optimizer = mode2._optimizer
# mode2=compile.Mode(gof.OpWiseCLinker(allow_gc=True), compile.mode.OPT_FAST_COMPILE) try:
mode2._optimizer=mode2._optimizer.excluding('local_elemwise_fusion')
if s is None: # mode2=compile.Mode(gof.OpWiseCLinker(allow_gc=True), compile.mode.OPT_FAST_COMPILE)
s=slice(0,49)
#s=slice(49,59) if s is None:
nb_repeat=10 s=slice(0,49)
print "test with linker", str(linker) #s=slice(49,59)
times1=self.do(mode1, shared_fn, shp, gpu=gpu, nb_repeat=nb_repeat, assert_len_topo=False,slice=s) nb_repeat=10
times2=self.do(mode2, shared_fn, shp, gpu=gpu, nb_repeat=nb_repeat, assert_len_topo=False,slice=s) print "test with linker", str(linker)
print "times1 FAST_RUN optimisation" times1=self.do(mode1, shared_fn, shp, gpu=gpu, nb_repeat=nb_repeat, assert_len_topo=False,slice=s)
print times1, times1.min(), times1.max(), times1.sum() times2=self.do(mode2, shared_fn, shp, gpu=gpu, nb_repeat=nb_repeat, assert_len_topo=False,slice=s)
print "times2 FAST_RUN optimisation without local_elemwise_fusion" print "times1 FAST_RUN optimisation"
print times2, times2.min(), times2.max(), times2.sum() print times1, times1.min(), times1.max(), times1.sum()
d=times2/times1 print "times2 FAST_RUN optimisation without local_elemwise_fusion"
# d.sort() print times2, times2.min(), times2.max(), times2.sum()
print "times2/times1",d,d.min(), d.max(), d.mean(), d.std() d=times2/times1
# d.sort()
print "times2/times1",d,d.min(), d.max(), d.mean(), d.std()
finally:
mode2._optimizer = old_optimizer
def speed_fusion_gpu(self): def speed_fusion_gpu(self):
import theano_cuda_ndarray as tcn import theano_cuda_ndarray as tcn
......
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论