chainer_computational_cost (used from a blog post in daily.belltail.jp)

cost_vgg.py

import chainer
import chainer.links as L
import numpy as np
import chainer_computational_cost as c3

net = L.VGG16Layers()
x = np.random.random((1, 3, 224, 224)).astype(np.float32)  # dummy input
with chainer.no_backprop_mode(), chainer.using_config('train', False):
    with c3.ComputationalCostHook(fma_1flop=True) as cch:
        y = net(x)
        cch.show_summary_report(mode='table')    # ←

stdout.txt

+-----------------------+----------+--------+---------+----------+--------+
|      Layer type       | # Layers | GFLOPs | MemRead | MemWrite | MemR+W |
|                       |          |        |   GiB   |   GiB    |  GiB   |
+=======================+==========+========+=========+==========+========+
| Convolution2DFunction | 13       | 15.347 | 0.089   | 0.05     | 0.139  |
+-----------------------+----------+--------+---------+----------+--------+
| ReLU                  | 15       | 0.014  | 0.05    | 0.05     | 0.101  |
+-----------------------+----------+--------+---------+----------+--------+
| MaxPooling2D          | 5        | 0.005  | 0.023   | 0.006    | 0.029  |
+-----------------------+----------+--------+---------+----------+--------+
| Reshape               | 1        | 0.0    | 0.0     | 0.0      | 0.0    |
+-----------------------+----------+--------+---------+----------+--------+
| LinearFunction        | 3        | 0.124  | 0.461   | 0.0      | 0.461  |
+-----------------------+----------+--------+---------+----------+--------+
| Softmax               | 1        | 0.0    | 0.0     | 0.0      | 0.0    |
+-----------------------+----------+--------+---------+----------+--------+
| total                 | 38       | 15.488 | 0.623   | 0.107    | 0.729  |
+-----------------------+----------+--------+---------+----------+--------+