Skip to content

Instantly share code, notes, and snippets.

@skaae

skaae/gist:aacb3ea0c1a5223c5dee

Created January 16, 2015 20:25
Show Gist options
  • Save skaae/aacb3ea0c1a5223c5dee to your computer and use it in GitHub Desktop.
Save skaae/aacb3ea0c1a5223c5dee to your computer and use it in GitHub Desktop.
Download ZIP
Raw
gistfile1.txt
Function profiling
==================
Message: experiment.py:196
Time in 1 calls to Function.__call__: 7.953641e+00s
Time in Function.fn.__call__: 7.953413e+00s (99.997%)
Time in thunks: 7.929524e+00s (99.697%)
Total compile time: 1.766550e+02s
Number of Apply nodes: 1214
Theano Optimizer time: 1.688680e+02s
Theano validate time: 2.550483e+00s
Theano Linker time (includes C, CUDA code generation/compiling): 7.392569e+00s
Class
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Class name>
88.0% 88.0% 6.978s 1.16e+00s Py 6 6 theano.scan_module.scan_op.Scan
4.9% 92.9% 0.390s 2.17e-02s C 18 18 theano.sandbox.cuda.blas.GpuDot22
1.5% 94.4% 0.118s 1.23e-03s C 96 96 theano.sandbox.cuda.basic_ops.GpuElemwise
1.4% 95.8% 0.109s 1.36e-02s C 8 8 theano.sandbox.cuda.basic_ops.GpuCAReduce
1.3% 97.0% 0.100s 2.62e-03s Py 38 38 theano.sandbox.cuda.basic_ops.GpuReshape
1.0% 98.0% 0.077s 2.20e-03s C 35 35 theano.sandbox.cuda.basic_ops.GpuIncSubtensor
0.7% 98.7% 0.055s 1.07e-03s C 51 51 theano.sandbox.cuda.basic_ops.GpuAlloc
0.5% 99.2% 0.042s 7.04e-03s C 6 6 theano.sandbox.cuda.basic_ops.HostFromGpu
0.3% 99.5% 0.022s 7.34e-03s Py 3 3 theano.tensor.basic.Split
0.2% 99.7% 0.017s 5.73e-03s C 3 3 theano.sandbox.cuda.basic_ops.GpuJoin
0.2% 99.9% 0.013s 1.20e-03s C 11 11 theano.sandbox.cuda.basic_ops.GpuFromHost
0.0% 99.9% 0.002s 2.30e-03s C 1 1 theano.tensor.nnet.nnet.SoftmaxGrad
0.0% 99.9% 0.002s 2.06e-03s C 1 1 theano.sandbox.cuda.nnet.GpuSoftmaxWithBias
0.0% 100.0% 0.001s 1.24e-03s C 1 1 theano.sandbox.cuda.blas.GpuGemm
0.0% 100.0% 0.001s 1.11e-03s C 1 1 theano.tensor.elemwise.Sum
0.0% 100.0% 0.001s 1.57e-06s C 507 507 theano.tensor.elemwise.Elemwise
0.0% 100.0% 0.000s 2.56e-06s C 113 113 theano.sandbox.cuda.basic_ops.GpuDimShuffle
0.0% 100.0% 0.000s 2.44e-06s C 82 82 theano.sandbox.cuda.basic_ops.GpuSubtensor
0.0% 100.0% 0.000s 1.82e-06s C 89 89 theano.compile.ops.Shape_i
0.0% 100.0% 0.000s 1.30e-05s Py 12 12 theano.compile.ops.Rebroadcast
... (remaining 3 Classes account for 0.00%(0.00s) of the runtime)
Ops
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Op name>
39.2% 39.2% 3.110s 3.11e+00s Py 1 1 forall_inplace,gpu,grad_of_scan_fn}
23.2% 62.4% 1.839s 1.84e+00s Py 1 1 forall_inplace,gpu,grad_of_scan_fn}
16.6% 79.0% 1.312s 1.31e+00s Py 1 1 forall_inplace,gpu,grad_of_scan_fn}
4.9% 83.9% 0.390s 2.17e-02s C 18 18 GpuDot22
3.9% 87.8% 0.307s 3.07e-01s Py 1 1 forall_inplace,gpu,scan_fn}
2.9% 90.7% 0.229s 2.29e-01s Py 1 1 forall_inplace,gpu,scan_fn}
2.3% 92.9% 0.180s 1.80e-01s Py 1 1 forall_inplace,gpu,scan_fn}
1.4% 94.3% 0.107s 1.79e-02s C 6 6 GpuCAReduce{add}{1,1,0}
1.2% 95.5% 0.099s 3.81e-03s Py 26 26 GpuReshape{2}
0.7% 96.2% 0.055s 1.07e-03s C 51 51 GpuAlloc{memset_0=True}
0.5% 96.7% 0.042s 7.04e-03s C 6 6 HostFromGpu
0.5% 97.2% 0.037s 6.16e-03s C 6 6 GpuIncSubtensor{Inc;:int64:}
0.4% 97.6% 0.033s 4.17e-03s C 8 8 GpuElemwise{add,no_inplace}
0.3% 98.0% 0.026s 2.19e-03s C 12 12 GpuIncSubtensor{InplaceInc;int64::}
0.3% 98.2% 0.022s 7.34e-03s Py 3 3 Split{2}
0.3% 98.5% 0.022s 9.13e-04s C 24 24 GpuElemwise{Add}[(0, 0)]
0.2% 98.7% 0.017s 5.73e-03s C 3 3 GpuJoin
0.2% 98.9% 0.015s 1.62e-03s C 9 9 GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}
0.2% 99.1% 0.014s 6.80e-04s C 20 20 GpuElemwise{Composite{[sub(mul(i0, i1), mul(i2, i3))]}}[(0, 1)]
0.2% 99.3% 0.013s 1.20e-03s C 11 11 GpuFromHost
... (remaining 126 Ops account for 0.75%(0.06s) of the runtime)
Apply
------
<% time> <sum %> <apply time> <time per call> <#call> <id> <Apply name>
39.2% 39.2% 3.110s 3.11e+00s 1 1119 forall_inplace,gpu,grad_of_scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{tanh,no_inplace}.0, GpuElemwise{mul,no_inplace}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{mul,no_inplace}.0, GpuElemwise{sub,no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},
23.2% 62.4% 1.839s 1.84e+00s 1 1175 forall_inplace,gpu,grad_of_scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{tanh,no_inplace}.0, GpuElemwise{mul,no_inplace}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{mul,no_inplace}.0, GpuElemwise{sub,no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},
16.6% 79.0% 1.312s 1.31e+00s 1 1063 forall_inplace,gpu,grad_of_scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{tanh,no_inplace}.0, GpuElemwise{mul,no_inplace}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{mul,no_inplace}.0, GpuElemwise{sub,no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},
3.9% 82.8% 0.307s 3.07e-01s 1 405 forall_inplace,gpu,scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuElemwise{sub,no_inplace}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, W_hid_to_gates_fwd, W_hid_to_gates_bck, GpuDimShuffle{x,0}.0, GpuDimShuffle{x,0}.
2.9% 85.7% 0.229s 2.29e-01s 1 680 forall_inplace,gpu,scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuElemwise{sub,no_inplace}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, W_hid_to_gates_fwd, W_hid_to_gates_bck, GpuDimShuffle{x,0}.0, GpuDimShuffle{x,0}.
2.3% 88.0% 0.180s 1.80e-01s 1 972 forall_inplace,gpu,scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuElemwise{sub,no_inplace}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, W_hid_to_gates_fwd, W_hid_to_gates_bck, GpuDimShuffle{x,0}.0, GpuDimShuffle{x,0}.
0.4% 88.4% 0.035s 3.55e-02s 1 1146 GpuDot22(GpuReshape{2}.0, GpuDimShuffle{1,0}.0)
0.4% 88.9% 0.035s 3.55e-02s 1 490 GpuDot22(GpuReshape{2}.0, GpuReshape{2}.0)
0.4% 89.3% 0.035s 3.54e-02s 1 501 GpuDot22(GpuReshape{2}.0, GpuReshape{2}.0)
0.4% 89.8% 0.035s 3.47e-02s 1 1149 GpuDot22(GpuReshape{2}.0, GpuDimShuffle{1,0}.0)
0.4% 90.2% 0.032s 3.16e-02s 1 1150 GpuDot22(GpuDimShuffle{1,0}.0, GpuReshape{2}.0)
0.4% 90.6% 0.032s 3.16e-02s 1 1147 GpuDot22(GpuDimShuffle{1,0}.0, GpuReshape{2}.0)
0.4% 90.9% 0.029s 2.94e-02s 1 1090 GpuDot22(GpuReshape{2}.0, GpuDimShuffle{1,0}.0)
0.4% 91.3% 0.029s 2.92e-02s 1 1141 GpuCAReduce{add}{1,1,0}(GpuIncSubtensor{InplaceInc;int64::}.0)
0.4% 91.7% 0.028s 2.83e-02s 1 1139 GpuCAReduce{add}{1,1,0}(GpuIncSubtensor{InplaceInc;int64::}.0)
0.3% 92.0% 0.027s 2.71e-02s 1 1093 GpuDot22(GpuReshape{2}.0, GpuDimShuffle{1,0}.0)
0.3% 92.3% 0.023s 2.29e-02s 1 1110 HostFromGpu(GpuDimShuffle{1,0,2}.0)
0.3% 92.6% 0.022s 2.21e-02s 1 778 GpuDot22(GpuReshape{2}.0, GpuReshape{2}.0)
0.3% 92.9% 0.022s 2.20e-02s 1 768 GpuDot22(GpuReshape{2}.0, GpuReshape{2}.0)
0.3% 93.1% 0.020s 2.01e-02s 1 1094 GpuDot22(GpuDimShuffle{1,0}.0, GpuReshape{2}.0)
... (remaining 1194 Apply instances account for 6.89%(0.55s) of the runtime)
Scan Op profiling ( scan_fn )
==================
Message: None
Time in 1 calls of the op (for a total of 216 steps) 3.071730e-01s
Total time spent in calling the VM 2.666941e-01s (86.822%)
Total overhead (computing slices..) 4.047894e-02s (13.178%)
Class
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Class name>
83.9% 83.9% 0.220s 5.09e-04s C 432 2 theano.sandbox.cuda.blas.GpuGemm
15.8% 99.7% 0.042s 4.81e-05s C 864 4 theano.sandbox.cuda.basic_ops.GpuElemwise
0.3% 100.0% 0.001s 7.63e-07s C 864 4 theano.sandbox.cuda.basic_ops.GpuSubtensor
... (remaining 0 Classes account for 0.00%(0.00s) of the runtime)
Ops
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Op name>
83.9% 83.9% 0.220s 5.09e-04s C 432 2 GpuGemm{no_inplace}
5.1% 89.0% 0.013s 6.20e-05s C 216 1 GpuElemwise{Composite{[add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4)))]},no_inplace}
4.2% 93.2% 0.011s 5.06e-05s C 216 1 GpuElemwise{Composite{[add(mul(add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4))), i5), mul(i0, i6))]},no_inplace}
3.3% 96.5% 0.009s 4.01e-05s C 216 1 GpuElemwise{Composite{[add(mul(i0, i1), mul(i2, i3))]},no_inplace}
3.3% 99.7% 0.009s 3.95e-05s C 216 1 GpuElemwise{Composite{[mul(scalar_sigmoid(mul(i0, i1)), tanh(i2))]},no_inplace}
0.3% 100.0% 0.001s 7.63e-07s C 864 4 GpuSubtensor{::, int64:int64:}
... (remaining 0 Ops account for 0.00%(0.00s) of the runtime)
Apply
------
<% time> <sum %> <apply time> <time per call> <#call> <id> <Apply name>
43.2% 43.2% 0.113s 5.24e-04s 216 0 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_fwd[t-1][cuda], W_hid_to_gates_fwd_copy[cuda], TensorConstant{1.0})
40.7% 83.9% 0.107s 4.94e-04s 216 1 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_bck[t-1][cuda], W_hid_to_gates_bck_copy[cuda], TensorConstant{1.0})
5.1% 89.0% 0.013s 6.20e-05s 216 6 GpuElemwise{Composite{[add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4)))]},no_inplace}(cell_init_fwd[t-1][cuda], <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0)
4.2% 93.2% 0.011s 5.06e-05s 216 7 GpuElemwise{Composite{[add(mul(add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4))), i5), mul(i0, i6))]},no_inplace}(cell_init_bck[t-1][cuda], <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, col)>, <CudaNdarrayType(float32, col)>)
3.3% 96.5% 0.009s 4.01e-05s 216 9 GpuElemwise{Composite{[add(mul(i0, i1), mul(i2, i3))]},no_inplace}(GpuElemwise{Composite{[mul(scalar_sigmoid(mul(i0, i1)), tanh(i2))]},no_inplace}.0, <CudaNdarrayType(float32, col)>, hid_init_bck[t-1][cuda], <CudaNdarrayType(float32, col)>)
3.3% 99.7% 0.009s 3.95e-05s 216 8 GpuElemwise{Composite{[mul(scalar_sigmoid(mul(i0, i1)), tanh(i2))]},no_inplace}(cell_init_fwd[t-1][cuda], <CudaNdarrayType(float32, row)>, GpuElemwise{Composite{[add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4)))]},no_inplace}.0)
0.1% 99.8% 0.000s 1.22e-06s 216 3 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{0}, Constant{156})
0.1% 99.9% 0.000s 9.18e-07s 216 5 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{0}, Constant{156})
0.0% 100.0% 0.000s 4.90e-07s 216 2 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{312}, Constant{468})
0.0% 100.0% 0.000s 4.19e-07s 216 4 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{312}, Constant{468})
... (remaining 0 Apply instances account for 0.00%(0.00s) of the runtime)
Scan Op profiling ( scan_fn )
==================
Message: None
Time in 1 calls of the op (for a total of 216 steps) 2.288671e-01s
Total time spent in calling the VM 1.899178e-01s (82.982%)
Total overhead (computing slices..) 3.894925e-02s (17.018%)
Class
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Class name>
75.8% 75.8% 0.139s 3.23e-04s C 432 2 theano.sandbox.cuda.blas.GpuGemm
23.9% 99.7% 0.044s 5.09e-05s C 864 4 theano.sandbox.cuda.basic_ops.GpuElemwise
0.3% 100.0% 0.001s 6.65e-07s C 864 4 theano.sandbox.cuda.basic_ops.GpuSubtensor
... (remaining 0 Classes account for 0.00%(0.00s) of the runtime)
Ops
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Op name>
75.8% 75.8% 0.139s 3.23e-04s C 432 2 GpuGemm{no_inplace}
6.9% 82.7% 0.013s 5.87e-05s C 216 1 GpuElemwise{Composite{[add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4)))]},no_inplace}
6.9% 89.5% 0.013s 5.84e-05s C 216 1 GpuElemwise{Composite{[add(mul(add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4))), i5), mul(i0, i6))]},no_inplace}
5.4% 94.9% 0.010s 4.63e-05s C 216 1 GpuElemwise{Composite{[mul(scalar_sigmoid(mul(i0, i1)), tanh(i2))]},no_inplace}
4.7% 99.7% 0.009s 4.03e-05s C 216 1 GpuElemwise{Composite{[add(mul(i0, i1), mul(i2, i3))]},no_inplace}
0.3% 100.0% 0.001s 6.65e-07s C 864 4 GpuSubtensor{::, int64:int64:}
... (remaining 0 Ops account for 0.00%(0.00s) of the runtime)
Apply
------
<% time> <sum %> <apply time> <time per call> <#call> <id> <Apply name>
41.7% 41.7% 0.077s 3.55e-04s 216 1 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_bck[t-1][cuda], W_hid_to_gates_bck_copy[cuda], TensorConstant{1.0})
34.1% 75.8% 0.063s 2.90e-04s 216 0 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_fwd[t-1][cuda], W_hid_to_gates_fwd_copy[cuda], TensorConstant{1.0})
6.9% 82.7% 0.013s 5.87e-05s 216 6 GpuElemwise{Composite{[add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4)))]},no_inplace}(cell_init_fwd[t-1][cuda], <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0)
6.9% 89.5% 0.013s 5.84e-05s 216 7 GpuElemwise{Composite{[add(mul(add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4))), i5), mul(i0, i6))]},no_inplace}(cell_init_bck[t-1][cuda], <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, col)>, <CudaNdarrayType(float32, col)>)
5.4% 94.9% 0.010s 4.63e-05s 216 8 GpuElemwise{Composite{[mul(scalar_sigmoid(mul(i0, i1)), tanh(i2))]},no_inplace}(cell_init_fwd[t-1][cuda], <CudaNdarrayType(float32, row)>, GpuElemwise{Composite{[add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4)))]},no_inplace}.0)
4.7% 99.7% 0.009s 4.03e-05s 216 9 GpuElemwise{Composite{[add(mul(i0, i1), mul(i2, i3))]},no_inplace}(GpuElemwise{Composite{[mul(scalar_sigmoid(mul(i0, i1)), tanh(i2))]},no_inplace}.0, <CudaNdarrayType(float32, col)>, hid_init_bck[t-1][cuda], <CudaNdarrayType(float32, col)>)
0.1% 99.8% 0.000s 9.83e-07s 216 3 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{0}, Constant{300})
0.1% 99.9% 0.000s 9.48e-07s 216 5 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{0}, Constant{300})
0.0% 100.0% 0.000s 3.75e-07s 216 4 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{600}, Constant{900})
0.0% 100.0% 0.000s 3.52e-07s 216 2 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{600}, Constant{900})
... (remaining 0 Apply instances account for 0.00%(0.00s) of the runtime)
Scan Op profiling ( scan_fn )
==================
Message: None
Time in 1 calls of the op (for a total of 216 steps) 1.795299e-01s
Total time spent in calling the VM 1.442864e-01s (80.369%)
Total overhead (computing slices..) 3.524351e-02s (19.631%)
Class
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Class name>
72.3% 72.3% 0.101s 2.34e-04s C 432 2 theano.sandbox.cuda.blas.GpuGemm
27.3% 99.6% 0.038s 4.43e-05s C 864 4 theano.sandbox.cuda.basic_ops.GpuElemwise
0.4% 100.0% 0.001s 6.75e-07s C 864 4 theano.sandbox.cuda.basic_ops.GpuSubtensor
... (remaining 0 Classes account for 0.00%(0.00s) of the runtime)
Ops
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Op name>
72.3% 72.3% 0.101s 2.34e-04s C 432 2 GpuGemm{no_inplace}
9.9% 82.1% 0.014s 6.40e-05s C 216 1 GpuElemwise{Composite{[add(mul(add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4))), i5), mul(i0, i6))]},no_inplace}
7.1% 89.3% 0.010s 4.63e-05s C 216 1 GpuElemwise{Composite{[add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4)))]},no_inplace}
5.7% 95.0% 0.008s 3.72e-05s C 216 1 GpuElemwise{Composite{[mul(scalar_sigmoid(mul(i0, i1)), tanh(i2))]},no_inplace}
4.6% 99.6% 0.006s 2.98e-05s C 216 1 GpuElemwise{Composite{[add(mul(i0, i1), mul(i2, i3))]},no_inplace}
0.4% 100.0% 0.001s 6.75e-07s C 864 4 GpuSubtensor{::, int64:int64:}
... (remaining 0 Ops account for 0.00%(0.00s) of the runtime)
Apply
------
<% time> <sum %> <apply time> <time per call> <#call> <id> <Apply name>
36.4% 36.4% 0.051s 2.36e-04s 216 1 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_bck[t-1][cuda], W_hid_to_gates_bck_copy[cuda], TensorConstant{1.0})
35.9% 72.3% 0.050s 2.33e-04s 216 0 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_fwd[t-1][cuda], W_hid_to_gates_fwd_copy[cuda], TensorConstant{1.0})
9.9% 82.1% 0.014s 6.40e-05s 216 7 GpuElemwise{Composite{[add(mul(add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4))), i5), mul(i0, i6))]},no_inplace}(cell_init_bck[t-1][cuda], <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, col)>, <CudaNdarrayType(float32, col)>)
7.1% 89.3% 0.010s 4.63e-05s 216 6 GpuElemwise{Composite{[add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4)))]},no_inplace}(cell_init_fwd[t-1][cuda], <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, row)>, GpuSubtensor{::, int64:int64:}.0)
5.7% 95.0% 0.008s 3.72e-05s 216 8 GpuElemwise{Composite{[mul(scalar_sigmoid(mul(i0, i1)), tanh(i2))]},no_inplace}(cell_init_fwd[t-1][cuda], <CudaNdarrayType(float32, row)>, GpuElemwise{Composite{[add(mul(scalar_sigmoid(mul(i0, i1)), i0), mul(scalar_sigmoid(add(i2, mul(i0, i3))), tanh(i4)))]},no_inplace}.0)
4.6% 99.6% 0.006s 2.98e-05s 216 9 GpuElemwise{Composite{[add(mul(i0, i1), mul(i2, i3))]},no_inplace}(GpuElemwise{Composite{[mul(scalar_sigmoid(mul(i0, i1)), tanh(i2))]},no_inplace}.0, <CudaNdarrayType(float32, col)>, hid_init_bck[t-1][cuda], <CudaNdarrayType(float32, col)>)
0.1% 99.7% 0.000s 9.63e-07s 216 3 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{0}, Constant{102})
0.1% 99.9% 0.000s 8.41e-07s 216 5 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{0}, Constant{102})
0.1% 99.9% 0.000s 4.72e-07s 216 2 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{204}, Constant{306})
0.1% 100.0% 0.000s 4.23e-07s 216 4 GpuSubtensor{::, int64:int64:}(GpuGemm{no_inplace}.0, Constant{204}, Constant{306})
... (remaining 0 Apply instances account for 0.00%(0.00s) of the runtime)
Scan Op profiling ( grad_of_scan_fn )
==================
Message: None
Time in 1 calls of the op (for a total of 216 steps) 1.312062e+00s
Total time spent in calling the VM 1.180757e+00s (89.992%)
Total overhead (computing slices..) 1.313052e-01s (10.008%)
Class
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Class name>
39.9% 39.9% 0.433s 2.51e-04s C 1728 8 theano.sandbox.cuda.blas.GpuGemm
33.9% 73.8% 0.369s 3.41e-05s C 10800 50 theano.sandbox.cuda.basic_ops.GpuElemwise
8.5% 82.3% 0.092s 2.13e-04s C 432 2 theano.sandbox.cuda.blas.GpuDot22
8.1% 90.3% 0.088s 4.05e-05s C 2160 10 theano.sandbox.cuda.basic_ops.GpuCAReduce
7.8% 98.1% 0.084s 2.17e-05s C 3888 18 theano.sandbox.cuda.basic_ops.GpuIncSubtensor
1.8% 99.8% 0.019s 4.42e-05s C 432 2 theano.sandbox.cuda.basic_ops.GpuAlloc
0.1% 99.9% 0.001s 1.20e-06s C 864 4 theano.sandbox.cuda.basic_ops.GpuSubtensor
0.1% 100.0% 0.001s 7.25e-07s C 864 4 theano.compile.ops.Shape_i
... (remaining 0 Classes account for 0.00%(0.00s) of the runtime)
Ops
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Op name>
27.5% 27.5% 0.299s 2.77e-04s C 1080 5 GpuGemm{no_inplace}
12.4% 39.9% 0.134s 2.07e-04s C 648 3 GpuGemm{inplace}
8.5% 48.4% 0.092s 2.13e-04s C 432 2 GpuDot22
8.1% 56.4% 0.088s 4.05e-05s C 2160 10 GpuCAReduce{add}{1,0}
7.6% 64.0% 0.082s 4.22e-05s C 1944 9 GpuElemwise{mul,no_inplace}
4.7% 68.7% 0.051s 3.95e-05s C 1296 6 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}
3.4% 72.1% 0.037s 1.70e-05s C 2160 10 GpuElemwise{Mul}[(0, 0)]
3.3% 75.4% 0.036s 1.67e-05s C 2160 10 GpuIncSubtensor{InplaceInc;int64:int64:}
2.7% 78.1% 0.030s 6.89e-05s C 432 2 GpuElemwise{Composite{[scalar_sigmoid(add(i0, i1))]},no_inplace}
2.3% 80.4% 0.025s 5.82e-05s C 432 2 GpuIncSubtensor{Inc;::, int64:int64:}
2.3% 82.7% 0.025s 3.79e-05s C 648 3 GpuElemwise{Composite{[mul(mul(i0, i1), i2)]},no_inplace}
2.1% 84.8% 0.023s 1.77e-05s C 1296 6 GpuIncSubtensor{InplaceInc;::, int64:int64:}
1.9% 86.7% 0.021s 4.83e-05s C 432 2 GpuElemwise{Add}[(0, 0)]
1.8% 88.5% 0.019s 4.42e-05s C 432 2 GpuAlloc{memset_0=True}
1.4% 89.9% 0.015s 3.48e-05s C 432 2 GpuElemwise{Tanh}[(0, 0)]
1.0% 90.9% 0.011s 5.04e-05s C 216 1 GpuElemwise{Composite{[add(mul(i0, i1), mul(i2, i3))]},no_inplace}
1.0% 91.8% 0.011s 4.87e-05s C 216 1 GpuElemwise{Composite{[add(add(add(add(mul(i0, i1), mul(i2, i3)), mul(i4, i5)), mul(i6, i7)), i8)]},no_inplace}
1.0% 92.8% 0.011s 4.87e-05s C 216 1 GpuElemwise{sub,no_inplace}
0.9% 93.7% 0.010s 2.34e-05s C 432 2 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]}}[(0, 0)]
0.9% 94.6% 0.010s 4.42e-05s C 216 1 GpuElemwise{Composite{[add(i0, add(i1, i1))]},no_inplace}
... (remaining 13 Ops account for 5.38%(0.06s) of the runtime)
Apply
------
<% time> <sum %> <apply time> <time per call> <#call> <id> <Apply name>
7.3% 7.3% 0.079s 3.66e-04s 216 73 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_fwd[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, TensorConstant{1.0})
5.7% 13.0% 0.062s 2.88e-04s 216 57 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_bck[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, TensorConstant{1.0})
4.9% 17.9% 0.053s 2.46e-04s 216 4 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_fwd[t-1][cuda], W_hid_to_gates_fwd_copy[cuda], TensorConstant{1.0})
4.9% 22.8% 0.053s 2.45e-04s 216 74 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_fwd_copy.T_replace[cuda], TensorConstant{1.0})
4.8% 27.5% 0.052s 2.40e-04s 216 0 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_bck[t-1][cuda], W_hid_to_gates_bck_copy[cuda], TensorConstant{1.0})
4.5% 32.1% 0.049s 2.28e-04s 216 67 GpuDot22(GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_fwd_copy.T_replace[cuda])
4.2% 36.3% 0.046s 2.12e-04s 216 58 GpuGemm{inplace}(GpuElemwise{mul,no_inplace}.0, TensorConstant{1.0}, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_bck_copy.T_replace[cuda], TensorConstant{1.0})
4.2% 40.4% 0.045s 2.10e-04s 216 76 GpuGemm{inplace}(GpuDot22.0, TensorConstant{1.0}, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_fwd_copy.T_replace[cuda], TensorConstant{1.0})
4.0% 44.4% 0.043s 2.00e-04s 216 77 GpuGemm{inplace}(GpuDot22.0, TensorConstant{1.0}, hid_init_fwd[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, TensorConstant{1.0})
3.9% 48.4% 0.043s 1.98e-04s 216 68 GpuDot22(hid_init_fwd[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0)
1.4% 49.8% 0.015s 7.06e-05s 216 27 GpuElemwise{Composite{[scalar_sigmoid(add(i0, i1))]},no_inplace}(GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, matrix)>)
1.3% 51.1% 0.015s 6.73e-05s 216 28 GpuElemwise{Composite{[scalar_sigmoid(add(i0, i1))]},no_inplace}(GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, matrix)>)
1.2% 52.3% 0.013s 6.14e-05s 216 13 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}(<CudaNdarrayType(float32, matrix)>, cell_init_fwd[t-1][cuda], <CudaNdarrayType(float32, matrix)>, <CudaNdarrayType(float32, matrix)>)
1.2% 53.5% 0.013s 5.89e-05s 216 24 GpuElemwise{mul,no_inplace}(GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}.0, <CudaNdarrayType(float32, row)>)
1.2% 54.6% 0.013s 5.84e-05s 216 48 GpuIncSubtensor{Inc;::, int64:int64:}(GpuAlloc{memset_0=True}.0, GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}.0, Constant{0}, Constant{102})
1.2% 55.8% 0.013s 5.80e-05s 216 46 GpuIncSubtensor{Inc;::, int64:int64:}(GpuAlloc{memset_0=True}.0, GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}.0, Constant{0}, Constant{102})
1.1% 56.9% 0.012s 5.73e-05s 216 85 GpuElemwise{Add}[(0, 0)](GpuGemm{inplace}.0, GpuGemm{no_inplace}.0)
1.1% 58.0% 0.012s 5.36e-05s 216 80 GpuCAReduce{add}{1,0}(GpuElemwise{Mul}[(0, 0)].0)
1.0% 59.0% 0.011s 5.04e-05s 216 64 GpuElemwise{Composite{[add(mul(i0, i1), mul(i2, i3))]},no_inplace}(GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}.0, <CudaNdarrayType(float32, row)>, GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]}}[(0, 0)].0, <CudaNdarrayType(float32, row)>)
1.0% 60.0% 0.011s 4.93e-05s 216 21 GpuAlloc{memset_0=True}(CudaNdarrayConstant{[[ 0.]]}, Shape_i{0}.0, Shape_i{1}.0)
... (remaining 78 Apply instances account for 40.01%(0.43s) of the runtime)
Scan Op profiling ( grad_of_scan_fn )
==================
Message: None
Time in 1 calls of the op (for a total of 216 steps) 3.109816e+00s
Total time spent in calling the VM 2.880412e+00s (92.623%)
Total overhead (computing slices..) 2.294040e-01s (7.377%)
Class
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Class name>
55.4% 55.4% 1.540s 8.91e-04s C 1728 8 theano.sandbox.cuda.blas.GpuGemm
18.9% 74.3% 0.524s 4.85e-05s C 10800 50 theano.sandbox.cuda.basic_ops.GpuElemwise
15.0% 89.2% 0.416s 9.62e-04s C 432 2 theano.sandbox.cuda.blas.GpuDot22
6.2% 95.4% 0.171s 4.41e-05s C 3888 18 theano.sandbox.cuda.basic_ops.GpuIncSubtensor
3.5% 98.9% 0.097s 4.51e-05s C 2160 10 theano.sandbox.cuda.basic_ops.GpuCAReduce
1.0% 99.9% 0.028s 6.57e-05s C 432 2 theano.sandbox.cuda.basic_ops.GpuAlloc
0.0% 100.0% 0.001s 1.41e-06s C 864 4 theano.sandbox.cuda.basic_ops.GpuSubtensor
0.0% 100.0% 0.001s 1.15e-06s C 864 4 theano.compile.ops.Shape_i
... (remaining 0 Classes account for 0.00%(0.00s) of the runtime)
Ops
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Op name>
35.6% 35.6% 0.989s 9.16e-04s C 1080 5 GpuGemm{no_inplace}
19.8% 55.4% 0.551s 8.50e-04s C 648 3 GpuGemm{inplace}
15.0% 70.4% 0.416s 9.62e-04s C 432 2 GpuDot22
3.5% 73.9% 0.097s 4.51e-05s C 2160 10 GpuCAReduce{add}{1,0}
3.0% 76.9% 0.084s 4.33e-05s C 1944 9 GpuElemwise{mul,no_inplace}
3.0% 79.9% 0.084s 6.49e-05s C 1296 6 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}
2.9% 82.8% 0.080s 1.85e-04s C 432 2 GpuIncSubtensor{Inc;::, int64:int64:}
2.3% 85.1% 0.064s 1.48e-04s C 432 2 GpuElemwise{Add}[(0, 0)]
1.8% 86.9% 0.050s 2.31e-05s C 2160 10 GpuIncSubtensor{InplaceInc;int64:int64:}
1.7% 88.6% 0.047s 2.15e-05s C 2160 10 GpuElemwise{Mul}[(0, 0)]
1.5% 90.1% 0.042s 3.22e-05s C 1296 6 GpuIncSubtensor{InplaceInc;::, int64:int64:}
1.2% 91.3% 0.035s 8.00e-05s C 432 2 GpuElemwise{Composite{[scalar_sigmoid(add(i0, i1))]},no_inplace}
1.1% 92.4% 0.030s 1.39e-04s C 216 1 GpuElemwise{Composite{[add(add(i0, i1), i2)]}}[(0, 0)]
1.0% 93.4% 0.028s 6.57e-05s C 432 2 GpuAlloc{memset_0=True}
0.9% 94.3% 0.024s 3.76e-05s C 648 3 GpuElemwise{Composite{[mul(mul(i0, i1), i2)]},no_inplace}
0.8% 95.1% 0.023s 1.09e-04s C 216 1 GpuElemwise{Composite{[add(i0, add(i1, i1))]},no_inplace}
0.6% 95.8% 0.018s 4.17e-05s C 432 2 GpuElemwise{Tanh}[(0, 0)]
0.5% 96.3% 0.015s 6.97e-05s C 216 1 GpuElemwise{Composite{[add(add(add(add(mul(i0, i1), mul(i2, i3)), mul(i4, i5)), mul(i6, i7)), i8)]},no_inplace}
0.5% 96.8% 0.013s 2.94e-05s C 432 2 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]}}[(0, 0)]
0.4% 97.2% 0.010s 4.82e-05s C 216 1 GpuElemwise{Composite{[add(add(add(add(add(i0, i1), mul(i2, i3)), add(add(i4, i5), mul(i6, i3))), add(add(add(i7, i8), i9), mul(i10, i3))), i11)]}}[(0, 0)]
... (remaining 13 Ops account for 2.84%(0.08s) of the runtime)
Apply
------
<% time> <sum %> <apply time> <time per call> <#call> <id> <Apply name>
13.0% 13.0% 0.360s 1.67e-03s 216 73 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_fwd[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, TensorConstant{1.0})
11.8% 24.8% 0.328s 1.52e-03s 216 57 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_bck[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, TensorConstant{1.0})
10.9% 35.6% 0.303s 1.40e-03s 216 77 GpuGemm{inplace}(GpuDot22.0, TensorConstant{1.0}, hid_init_fwd[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, TensorConstant{1.0})
10.5% 46.1% 0.291s 1.35e-03s 216 68 GpuDot22(hid_init_fwd[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0)
4.7% 50.8% 0.131s 6.08e-04s 216 74 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_fwd_copy.T_replace[cuda], TensorConstant{1.0})
4.5% 55.3% 0.125s 5.78e-04s 216 67 GpuDot22(GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_fwd_copy.T_replace[cuda])
4.5% 59.8% 0.124s 5.75e-04s 216 58 GpuGemm{inplace}(GpuElemwise{mul,no_inplace}.0, TensorConstant{1.0}, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_bck_copy.T_replace[cuda], TensorConstant{1.0})
4.5% 64.3% 0.124s 5.74e-04s 216 76 GpuGemm{inplace}(GpuDot22.0, TensorConstant{1.0}, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_fwd_copy.T_replace[cuda], TensorConstant{1.0})
3.8% 68.0% 0.104s 4.83e-04s 216 0 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_bck[t-1][cuda], W_hid_to_gates_bck_copy[cuda], TensorConstant{1.0})
2.4% 70.4% 0.065s 3.03e-04s 216 4 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_fwd[t-1][cuda], W_hid_to_gates_fwd_copy[cuda], TensorConstant{1.0})
2.0% 72.4% 0.057s 2.62e-04s 216 48 GpuIncSubtensor{Inc;::, int64:int64:}(GpuAlloc{memset_0=True}.0, GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}.0, Constant{0}, Constant{300})
1.8% 74.2% 0.051s 2.36e-04s 216 86 GpuElemwise{Add}[(0, 0)](GpuGemm{inplace}.0, GpuGemm{no_inplace}.0)
1.1% 75.3% 0.030s 1.39e-04s 216 72 GpuElemwise{Composite{[add(add(i0, i1), i2)]}}[(0, 0)](GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0)
0.8% 76.2% 0.023s 1.09e-04s 216 11 GpuElemwise{Composite{[add(i0, add(i1, i1))]},no_inplace}(<CudaNdarrayType(float32, vector)>, <CudaNdarrayType(float32, vector)>)
0.8% 77.0% 0.023s 1.07e-04s 216 46 GpuIncSubtensor{Inc;::, int64:int64:}(GpuAlloc{memset_0=True}.0, GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}.0, Constant{0}, Constant{300})
0.8% 77.8% 0.022s 1.03e-04s 216 8 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}(<CudaNdarrayType(float32, matrix)>, <CudaNdarrayType(float32, matrix)>, <CudaNdarrayType(float32, matrix)>, <CudaNdarrayType(float32, matrix)>)
0.8% 78.6% 0.022s 1.01e-04s 216 13 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}(<CudaNdarrayType(float32, matrix)>, cell_init_fwd[t-1][cuda], <CudaNdarrayType(float32, matrix)>, <CudaNdarrayType(float32, matrix)>)
0.7% 79.3% 0.018s 8.54e-05s 216 28 GpuElemwise{Composite{[scalar_sigmoid(add(i0, i1))]},no_inplace}(GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, matrix)>)
0.6% 79.8% 0.016s 7.46e-05s 216 27 GpuElemwise{Composite{[scalar_sigmoid(add(i0, i1))]},no_inplace}(GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, matrix)>)
0.5% 80.4% 0.015s 7.06e-05s 216 24 GpuElemwise{mul,no_inplace}(GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}.0, <CudaNdarrayType(float32, row)>)
... (remaining 78 Apply instances account for 19.62%(0.55s) of the runtime)
Scan Op profiling ( grad_of_scan_fn )
==================
Message: None
Time in 1 calls of the op (for a total of 216 steps) 1.839203e+00s
Total time spent in calling the VM 1.672098e+00s (90.914%)
Total overhead (computing slices..) 1.671045e-01s (9.086%)
Class
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Class name>
46.4% 46.4% 0.731s 4.23e-04s C 1728 8 theano.sandbox.cuda.blas.GpuGemm
26.7% 73.1% 0.421s 3.90e-05s C 10800 50 theano.sandbox.cuda.basic_ops.GpuElemwise
11.7% 84.8% 0.184s 4.26e-04s C 432 2 theano.sandbox.cuda.blas.GpuDot22
7.2% 92.1% 0.114s 2.93e-05s C 3888 18 theano.sandbox.cuda.basic_ops.GpuIncSubtensor
5.9% 98.0% 0.093s 4.31e-05s C 2160 10 theano.sandbox.cuda.basic_ops.GpuCAReduce
1.9% 99.9% 0.030s 6.95e-05s C 432 2 theano.sandbox.cuda.basic_ops.GpuAlloc
0.1% 99.9% 0.001s 1.26e-06s C 864 4 theano.sandbox.cuda.basic_ops.GpuSubtensor
0.1% 100.0% 0.001s 9.80e-07s C 864 4 theano.compile.ops.Shape_i
... (remaining 0 Classes account for 0.00%(0.00s) of the runtime)
Ops
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Op name>
30.0% 30.0% 0.472s 4.37e-04s C 1080 5 GpuGemm{no_inplace}
16.4% 46.4% 0.258s 3.99e-04s C 648 3 GpuGemm{inplace}
11.7% 58.1% 0.184s 4.26e-04s C 432 2 GpuDot22
5.9% 64.0% 0.093s 4.31e-05s C 2160 10 GpuCAReduce{add}{1,0}
4.8% 68.8% 0.075s 3.85e-05s C 1944 9 GpuElemwise{mul,no_inplace}
3.5% 72.3% 0.055s 4.25e-05s C 1296 6 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}
2.9% 75.2% 0.046s 2.15e-05s C 2160 10 GpuElemwise{Mul}[(0, 0)]
2.5% 77.7% 0.040s 3.07e-05s C 1296 6 GpuIncSubtensor{InplaceInc;::, int64:int64:}
2.5% 80.3% 0.040s 1.84e-05s C 2160 10 GpuIncSubtensor{InplaceInc;int64:int64:}
2.2% 82.4% 0.034s 7.91e-05s C 432 2 GpuIncSubtensor{Inc;::, int64:int64:}
2.0% 84.4% 0.031s 7.18e-05s C 432 2 GpuElemwise{Composite{[scalar_sigmoid(add(i0, i1))]},no_inplace}
1.9% 86.3% 0.030s 6.95e-05s C 432 2 GpuAlloc{memset_0=True}
1.7% 88.0% 0.026s 4.08e-05s C 648 3 GpuElemwise{Composite{[mul(mul(i0, i1), i2)]},no_inplace}
1.7% 89.6% 0.026s 1.21e-04s C 216 1 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), sub(i3, i2))]},no_inplace}
1.5% 91.1% 0.023s 1.06e-04s C 216 1 GpuElemwise{Composite{[add(mul(i0, i1), mul(i2, i3))]},no_inplace}
1.3% 92.4% 0.020s 4.57e-05s C 432 2 GpuElemwise{Add}[(0, 0)]
1.1% 93.4% 0.017s 7.95e-05s C 216 1 GpuElemwise{Composite{[add(add(i0, i1), i2)]}}[(0, 0)]
0.8% 94.2% 0.012s 2.80e-05s C 432 2 GpuElemwise{Tanh}[(0, 0)]
0.7% 94.9% 0.011s 2.49e-05s C 432 2 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]}}[(0, 0)]
0.7% 95.6% 0.010s 4.79e-05s C 216 1 GpuElemwise{Composite{[add(i0, add(i1, i1))]},no_inplace}
... (remaining 13 Ops account for 4.45%(0.07s) of the runtime)
Apply
------
<% time> <sum %> <apply time> <time per call> <#call> <id> <Apply name>
6.3% 6.3% 0.099s 4.57e-04s 216 74 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_fwd_copy.T_replace[cuda], TensorConstant{1.0})
6.1% 12.3% 0.096s 4.43e-04s 216 73 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_fwd[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, TensorConstant{1.0})
6.1% 18.4% 0.095s 4.41e-04s 216 4 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_fwd[t-1][cuda], W_hid_to_gates_fwd_copy[cuda], TensorConstant{1.0})
6.0% 24.4% 0.094s 4.34e-04s 216 0 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_bck[t-1][cuda], W_hid_to_gates_bck_copy[cuda], TensorConstant{1.0})
5.9% 30.3% 0.094s 4.33e-04s 216 67 GpuDot22(GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_fwd_copy.T_replace[cuda])
5.8% 36.0% 0.091s 4.20e-04s 216 68 GpuDot22(hid_init_fwd[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0)
5.6% 41.7% 0.089s 4.11e-04s 216 57 GpuGemm{no_inplace}(<CudaNdarrayType(float32, matrix)>, TensorConstant{1.0}, hid_init_bck[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, TensorConstant{1.0})
5.6% 47.3% 0.089s 4.10e-04s 216 58 GpuGemm{inplace}(GpuElemwise{mul,no_inplace}.0, TensorConstant{1.0}, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_bck_copy.T_replace[cuda], TensorConstant{1.0})
5.4% 52.7% 0.085s 3.95e-04s 216 76 GpuGemm{inplace}(GpuDot22.0, TensorConstant{1.0}, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, W_hid_to_gates_fwd_copy.T_replace[cuda], TensorConstant{1.0})
5.4% 58.1% 0.084s 3.91e-04s 216 77 GpuGemm{inplace}(GpuDot22.0, TensorConstant{1.0}, hid_init_fwd[t-1].T_replace[cuda], GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, TensorConstant{1.0})
1.7% 59.8% 0.026s 1.21e-04s 216 36 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), sub(i3, i2))]},no_inplace}(GpuElemwise{mul,no_inplace}.0, GpuElemwise{Tanh}[(0, 0)].0, GpuElemwise{Composite{[scalar_sigmoid(add(i0, i1))]},no_inplace}.0, CudaNdarrayConstant{[[ 1.]]})
1.5% 61.2% 0.023s 1.06e-04s 216 64 GpuElemwise{Composite{[add(mul(i0, i1), mul(i2, i3))]},no_inplace}(GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}.0, <CudaNdarrayType(float32, row)>, GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]}}[(0, 0)].0, <CudaNdarrayType(float32, row)>)
1.4% 62.6% 0.022s 1.01e-04s 216 48 GpuIncSubtensor{Inc;::, int64:int64:}(GpuAlloc{memset_0=True}.0, GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}.0, Constant{0}, Constant{156})
1.3% 63.9% 0.021s 9.61e-05s 216 17 GpuAlloc{memset_0=True}(CudaNdarrayConstant{[[ 0.]]}, Shape_i{0}.0, Shape_i{1}.0)
1.2% 65.1% 0.018s 8.56e-05s 216 27 GpuElemwise{Composite{[scalar_sigmoid(add(i0, i1))]},no_inplace}(GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, matrix)>)
1.1% 66.2% 0.017s 7.95e-05s 216 72 GpuElemwise{Composite{[add(add(i0, i1), i2)]}}[(0, 0)](GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0, GpuIncSubtensor{InplaceInc;::, int64:int64:}.0)
0.9% 67.1% 0.015s 6.79e-05s 216 13 GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]},no_inplace}(<CudaNdarrayType(float32, matrix)>, cell_init_fwd[t-1][cuda], <CudaNdarrayType(float32, matrix)>, <CudaNdarrayType(float32, matrix)>)
0.8% 68.0% 0.013s 6.12e-05s 216 49 GpuIncSubtensor{InplaceInc;::, int64:int64:}(GpuAlloc{memset_0=True}.0, GpuElemwise{Composite{[mul(mul(mul(i0, i1), i2), i3)]}}[(0, 3)].0, Constant{0}, Constant{156})
0.8% 68.8% 0.013s 5.85e-05s 216 86 GpuElemwise{Add}[(0, 0)](GpuGemm{inplace}.0, GpuGemm{no_inplace}.0)
0.8% 69.6% 0.013s 5.79e-05s 216 28 GpuElemwise{Composite{[scalar_sigmoid(add(i0, i1))]},no_inplace}(GpuSubtensor{::, int64:int64:}.0, <CudaNdarrayType(float32, matrix)>)
... (remaining 78 Apply instances account for 30.45%(0.48s) of the runtime)
Function profiling
==================
Message: experiment.py:197
Time in 0 calls to Function.__call__: 0.000000e+00s
Total compile time: 1.279212e+01s
Number of Apply nodes: 0
Theano Optimizer time: 1.090553e+01s
Theano validate time: 4.676583e-01s
Theano Linker time (includes C, CUDA code generation/compiling): 1.666232e+00s
Function profiling
==================
Message: experiment.py:198
Time in 0 calls to Function.__call__: 0.000000e+00s
Total compile time: 1.224835e+01s
Number of Apply nodes: 0
Theano Optimizer time: 1.069219e+01s
Theano validate time: 2.016317e+00s
Theano Linker time (includes C, CUDA code generation/compiling): 1.361080e+00s
Function profiling
==================
Message: Sum of all(3) printed profiles at exit excluding Scan op profile.
Time in 1 calls to Function.__call__: 7.953641e+00s
Time in Function.fn.__call__: 7.953413e+00s (99.997%)
Time in thunks: 7.929524e+00s (99.697%)
Total compile time: 2.016955e+02s
Number of Apply nodes: 1214
Theano Optimizer time: 1.904657e+02s
Theano validate time: 5.034458e+00s
Theano Linker time (includes C, CUDA code generation/compiling): 1.041988e+01s
Class
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Class name>
88.0% 88.0% 6.978s 1.16e+00s Py 6 6 theano.scan_module.scan_op.Scan
4.9% 92.9% 0.390s 2.17e-02s C 18 18 theano.sandbox.cuda.blas.GpuDot22
1.5% 94.4% 0.118s 1.23e-03s C 96 96 theano.sandbox.cuda.basic_ops.GpuElemwise
1.4% 95.8% 0.109s 1.36e-02s C 8 8 theano.sandbox.cuda.basic_ops.GpuCAReduce
1.3% 97.0% 0.100s 2.62e-03s Py 38 38 theano.sandbox.cuda.basic_ops.GpuReshape
1.0% 98.0% 0.077s 2.20e-03s C 35 35 theano.sandbox.cuda.basic_ops.GpuIncSubtensor
0.7% 98.7% 0.055s 1.07e-03s C 51 51 theano.sandbox.cuda.basic_ops.GpuAlloc
0.5% 99.2% 0.042s 7.04e-03s C 6 6 theano.sandbox.cuda.basic_ops.HostFromGpu
0.3% 99.5% 0.022s 7.34e-03s Py 3 3 theano.tensor.basic.Split
0.2% 99.7% 0.017s 5.73e-03s C 3 3 theano.sandbox.cuda.basic_ops.GpuJoin
0.2% 99.9% 0.013s 1.20e-03s C 11 11 theano.sandbox.cuda.basic_ops.GpuFromHost
0.0% 99.9% 0.002s 2.30e-03s C 1 1 theano.tensor.nnet.nnet.SoftmaxGrad
0.0% 99.9% 0.002s 2.06e-03s C 1 1 theano.sandbox.cuda.nnet.GpuSoftmaxWithBias
0.0% 100.0% 0.001s 1.24e-03s C 1 1 theano.sandbox.cuda.blas.GpuGemm
0.0% 100.0% 0.001s 1.11e-03s C 1 1 theano.tensor.elemwise.Sum
0.0% 100.0% 0.001s 1.57e-06s C 507 507 theano.tensor.elemwise.Elemwise
0.0% 100.0% 0.000s 2.56e-06s C 113 113 theano.sandbox.cuda.basic_ops.GpuDimShuffle
0.0% 100.0% 0.000s 2.44e-06s C 82 82 theano.sandbox.cuda.basic_ops.GpuSubtensor
0.0% 100.0% 0.000s 1.82e-06s C 89 89 theano.compile.ops.Shape_i
0.0% 100.0% 0.000s 1.30e-05s Py 12 12 theano.compile.ops.Rebroadcast
... (remaining 3 Classes account for 0.00%(0.00s) of the runtime)
Ops
---
<% time> <sum %> <apply time> <time per call> <type> <#call> <#apply> <Op name>
39.2% 39.2% 3.110s 3.11e+00s Py 1 1 forall_inplace,gpu,grad_of_scan_fn}
23.2% 62.4% 1.839s 1.84e+00s Py 1 1 forall_inplace,gpu,grad_of_scan_fn}
16.6% 79.0% 1.312s 1.31e+00s Py 1 1 forall_inplace,gpu,grad_of_scan_fn}
4.9% 83.9% 0.390s 2.17e-02s C 18 18 GpuDot22
3.9% 87.8% 0.307s 3.07e-01s Py 1 1 forall_inplace,gpu,scan_fn}
2.9% 90.7% 0.229s 2.29e-01s Py 1 1 forall_inplace,gpu,scan_fn}
2.3% 92.9% 0.180s 1.80e-01s Py 1 1 forall_inplace,gpu,scan_fn}
1.4% 94.3% 0.107s 1.79e-02s C 6 6 GpuCAReduce{add}{1,1,0}
1.2% 95.5% 0.099s 3.81e-03s Py 26 26 GpuReshape{2}
0.7% 96.2% 0.055s 1.07e-03s C 51 51 GpuAlloc{memset_0=True}
0.5% 96.7% 0.042s 7.04e-03s C 6 6 HostFromGpu
0.5% 97.2% 0.037s 6.16e-03s C 6 6 GpuIncSubtensor{Inc;:int64:}
0.4% 97.6% 0.033s 4.17e-03s C 8 8 GpuElemwise{add,no_inplace}
0.3% 98.0% 0.026s 2.19e-03s C 12 12 GpuIncSubtensor{InplaceInc;int64::}
0.3% 98.2% 0.022s 7.34e-03s Py 3 3 Split{2}
0.3% 98.5% 0.022s 9.13e-04s C 24 24 GpuElemwise{Add}[(0, 0)]
0.2% 98.7% 0.017s 5.73e-03s C 3 3 GpuJoin
0.2% 98.9% 0.015s 1.62e-03s C 9 9 GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}
0.2% 99.1% 0.014s 6.80e-04s C 20 20 GpuElemwise{Composite{[sub(mul(i0, i1), mul(i2, i3))]}}[(0, 1)]
0.2% 99.3% 0.013s 1.20e-03s C 11 11 GpuFromHost
... (remaining 126 Ops account for 0.75%(0.06s) of the runtime)
Apply
------
<% time> <sum %> <apply time> <time per call> <#call> <id> <Apply name>
39.2% 39.2% 3.110s 3.11e+00s 1 1119 forall_inplace,gpu,grad_of_scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{tanh,no_inplace}.0, GpuElemwise{mul,no_inplace}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{mul,no_inplace}.0, GpuElemwise{sub,no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},
23.2% 62.4% 1.839s 1.84e+00s 1 1175 forall_inplace,gpu,grad_of_scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{tanh,no_inplace}.0, GpuElemwise{mul,no_inplace}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{mul,no_inplace}.0, GpuElemwise{sub,no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},
16.6% 79.0% 1.312s 1.31e+00s 1 1063 forall_inplace,gpu,grad_of_scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{tanh,no_inplace}.0, GpuElemwise{mul,no_inplace}.0, GpuDimShuffle{0,2,1}.0, GpuElemwise{mul,no_inplace}.0, GpuElemwise{sub,no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},no_inplace}.0, GpuElemwise{Composite{[scalar_sigmoid(mul(i0, i1))]},
3.9% 82.8% 0.307s 3.07e-01s 1 405 forall_inplace,gpu,scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuElemwise{sub,no_inplace}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, W_hid_to_gates_fwd, W_hid_to_gates_bck, GpuDimShuffle{x,0}.0, GpuDimShuffle{x,0}.
2.9% 85.7% 0.229s 2.29e-01s 1 680 forall_inplace,gpu,scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuElemwise{sub,no_inplace}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, W_hid_to_gates_fwd, W_hid_to_gates_bck, GpuDimShuffle{x,0}.0, GpuDimShuffle{x,0}.
2.3% 88.0% 0.180s 1.80e-01s 1 972 forall_inplace,gpu,scan_fn}(Elemwise{Composite{[minimum(minimum(i0, i1), i2)]}}.0, GpuElemwise{sub,no_inplace}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuSubtensor{int64:int64:int8}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, GpuIncSubtensor{InplaceSet;:int64:}.0, W_hid_to_gates_fwd, W_hid_to_gates_bck, GpuDimShuffle{x,0}.0, GpuDimShuffle{x,0}.
0.4% 88.4% 0.035s 3.55e-02s 1 1146 GpuDot22(GpuReshape{2}.0, GpuDimShuffle{1,0}.0)
0.4% 88.9% 0.035s 3.55e-02s 1 490 GpuDot22(GpuReshape{2}.0, GpuReshape{2}.0)
0.4% 89.3% 0.035s 3.54e-02s 1 501 GpuDot22(GpuReshape{2}.0, GpuReshape{2}.0)
0.4% 89.8% 0.035s 3.47e-02s 1 1149 GpuDot22(GpuReshape{2}.0, GpuDimShuffle{1,0}.0)
0.4% 90.2% 0.032s 3.16e-02s 1 1150 GpuDot22(GpuDimShuffle{1,0}.0, GpuReshape{2}.0)
0.4% 90.6% 0.032s 3.16e-02s 1 1147 GpuDot22(GpuDimShuffle{1,0}.0, GpuReshape{2}.0)
0.4% 90.9% 0.029s 2.94e-02s 1 1090 GpuDot22(GpuReshape{2}.0, GpuDimShuffle{1,0}.0)
0.4% 91.3% 0.029s 2.92e-02s 1 1141 GpuCAReduce{add}{1,1,0}(GpuIncSubtensor{InplaceInc;int64::}.0)
0.4% 91.7% 0.028s 2.83e-02s 1 1139 GpuCAReduce{add}{1,1,0}(GpuIncSubtensor{InplaceInc;int64::}.0)
0.3% 92.0% 0.027s 2.71e-02s 1 1093 GpuDot22(GpuReshape{2}.0, GpuDimShuffle{1,0}.0)
0.3% 92.3% 0.023s 2.29e-02s 1 1110 HostFromGpu(GpuDimShuffle{1,0,2}.0)
0.3% 92.6% 0.022s 2.21e-02s 1 778 GpuDot22(GpuReshape{2}.0, GpuReshape{2}.0)
0.3% 92.9% 0.022s 2.20e-02s 1 768 GpuDot22(GpuReshape{2}.0, GpuReshape{2}.0)
0.3% 93.1% 0.020s 2.01e-02s 1 1094 GpuDot22(GpuDimShuffle{1,0}.0, GpuReshape{2}.0)
... (remaining 1194 Apply instances account for 6.89%(0.55s) of the runtime)
bn09574:lasagne sorensonderby$
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment