slower benchmarks about pytorch-fcn HOT 1 CLOSED

PyTorch implementation is faster for static inputs and slower for dynamic ones than Chainer one at test time.
(In the previous performance, Chainer one was much slower, but it was fixed via wkentaro/fcn#90.)
Below are tested using a single GPU, Titan X (Pascal).

% ./speedtest.py
==> Benchmark: gpu=1, times=1000, dynamic_input=False
==> Testing FCN32s with Chainer
Elapsed time: 48.98 [s / 1000 evals]
Hz: 20.42 [hz]
==> Testing FCN32s with PyTorch
Elapsed time: 45.15 [s / 1000 evals]
Hz: 22.15 [hz]

% ./speedtest.py --dynamic-input
==> Benchmark: gpu=0, times=1000, dynamic_input=True
==> Testing FCN32s with Chainer
Elapsed time: 48.83 [s / 1000 evals]
Hz: 20.48 [hz]
==> Testing FCN32s with PyTorch
Elapsed time: 57.00 [s / 1000 evals]
Hz: 17.55 [hz]