# Copyright (c) 2016-present, Facebook, Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
##############################################################################

## @package convnet_benchmarks
# Module caffe2.experiments.python.convnet_benchmarks




"""
Benchmark for common convnets.

(NOTE: Numbers below prior with missing parameter=update step, TODO to update)

Speed on Titan X, with 10 warmup steps and 10 main steps and with different
versions of cudnn, are as follows (time reported below is per-batch time,
forward / forward+backward):

                    CuDNN V3        CuDNN v4
                    AlexNet         32.5 / 108.0    27.4 /  90.1
                    OverFeat       113.0 / 342.3    91.7 / 276.5
                    Inception      134.5 / 485.8   125.7 / 450.6
                    VGG (batch 64) 200.8 / 650.0   164.1 / 551.7

Speed on Inception with varied batch sizes and CuDNN v4 is as follows:

Batch Size   Speed per batch     Speed per image
16             22.8 /  72.7         1.43 / 4.54
32             38.0 / 127.5         1.19 / 3.98
64             67.2 / 233.6         1.05 / 3.65
128            125.7 / 450.6         0.98 / 3.52

Speed on Tesla M40, which 10 warmup steps and 10 main steps and with cudnn
v4, is as follows:

AlexNet         68.4 / 218.1
OverFeat       210.5 / 630.3
Inception      300.2 / 1122.2
VGG (batch 64) 405.8 / 1327.7

(Note that these numbers involve a "full" backprop, i.e. the gradient
with respect to the input image is also computed.)

To get the numbers, simply run:

for MODEL in AlexNet OverFeat Inception; do
PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \
    --batch_size 128 --model $MODEL --forward_only True
done
for MODEL in AlexNet OverFeat Inception; do
PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \
    --batch_size 128 --model $MODEL
done
PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \
    --batch_size 64 --model VGGA --forward_only True
PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \
    --batch_size 64 --model VGGA

for BS in 16 32 64 128; do
PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \
    --batch_size $BS --model Inception --forward_only True
PYTHONPATH=../gen:$PYTHONPATH python convnet_benchmarks.py \
    --batch_size $BS --model Inception
done

Note that VGG needs to be run at batch 64 due to memory limit on the backward
pass.
"""

import argparse
import time

from caffe2.python import cnn, workspace, core

import caffe2.python.SparseTransformer as SparseTransformer  # type: ignore[import]


def MLP(order):
    model = cnn.CNNModelHelper()
    d = 256
    depth = 20
    width = 3
    for i in range(depth):
        for j in range(width):
            current = "fc_{}_{}".format(i, j) if i > 0 else "data"
            next_ = "fc_{}_{}".format(i + 1, j)
            model.FC(
                current, next_,
                dim_in=d, dim_out=d,
                weight_init=model.XavierInit,
                bias_init=model.XavierInit)
            model.Sum(["fc_{}_{}".format(depth, j)
                       for j in range(width)], ["sum"])
            model.FC("sum", "last",
                     dim_in=d, dim_out=1000,
                     weight_init=model.XavierInit,
                     bias_init=model.XavierInit)
            xent = model.LabelCrossEntropy(["last", "label"], "xent")
            model.AveragedLoss(xent, "loss")
            return model, d


def AlexNet(order):
    model = cnn.CNNModelHelper(order, name="alexnet",
                               use_cudnn=True, cudnn_exhaustive_search=True)
    conv1 = model.Conv(
        "data",
        "conv1",
        3,
        64,
        11,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        stride=4,
        pad=2
    )

    relu1 = model.Relu(conv1, "conv1")
    pool1 = model.MaxPool(relu1, "pool1", kernel=3, stride=2)
    conv2 = model.Conv(
        pool1,
        "conv2",
        64,
        192,
        5,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=2
    )
    relu2 = model.Relu(conv2, "conv2")
    pool2 = model.MaxPool(relu2, "pool2", kernel=3, stride=2)
    conv3 = model.Conv(
        pool2,
        "conv3",
        192,
        384,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu3 = model.Relu(conv3, "conv3")
    conv4 = model.Conv(
        relu3,
        "conv4",
        384,
        256,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu4 = model.Relu(conv4, "conv4")
    conv5 = model.Conv(
        relu4,
        "conv5",
        256,
        256,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu5 = model.Relu(conv5, "conv5")
    pool5 = model.MaxPool(relu5, "pool5", kernel=3, stride=2)
    fc6 = model.FC(
        pool5, "fc6", 256 * 6 * 6, 4096, ('XavierFill', {}),
        ('ConstantFill', {})
    )
    relu6 = model.Relu(fc6, "fc6")
    fc7 = model.FC(
        relu6, "fc7", 4096, 4096, ('XavierFill', {}), ('ConstantFill', {})
    )
    relu7 = model.Relu(fc7, "fc7")
    fc8 = model.FC(
        relu7, "fc8", 4096, 1000, ('XavierFill', {}), ('ConstantFill', {})
    )
    pred = model.Softmax(fc8, "pred")
    xent = model.LabelCrossEntropy([pred, "label"], "xent")
    model.AveragedLoss(xent, "loss")
    return model, 224


def OverFeat(order):
    model = cnn.CNNModelHelper(order, name="overfeat",
                               use_cudnn=True, cudnn_exhaustive_search=True)
    conv1 = model.Conv(
        "data",
        "conv1",
        3,
        96,
        11,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        stride=4
    )
    relu1 = model.Relu(conv1, "conv1")
    pool1 = model.MaxPool(relu1, "pool1", kernel=2, stride=2)
    conv2 = model.Conv(
        pool1, "conv2", 96, 256, 5, ('XavierFill', {}), ('ConstantFill', {})
    )
    relu2 = model.Relu(conv2, "conv2")
    pool2 = model.MaxPool(relu2, "pool2", kernel=2, stride=2)
    conv3 = model.Conv(
        pool2,
        "conv3",
        256,
        512,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu3 = model.Relu(conv3, "conv3")
    conv4 = model.Conv(
        relu3,
        "conv4",
        512,
        1024,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu4 = model.Relu(conv4, "conv4")
    conv5 = model.Conv(
        relu4,
        "conv5",
        1024,
        1024,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu5 = model.Relu(conv5, "conv5")
    pool5 = model.MaxPool(relu5, "pool5", kernel=2, stride=2)
    fc6 = model.FC(
        pool5, "fc6", 1024 * 6 * 6, 3072, ('XavierFill', {}),
        ('ConstantFill', {})
    )
    relu6 = model.Relu(fc6, "fc6")
    fc7 = model.FC(
        relu6, "fc7", 3072, 4096, ('XavierFill', {}), ('ConstantFill', {})
    )
    relu7 = model.Relu(fc7, "fc7")
    fc8 = model.FC(
        relu7, "fc8", 4096, 1000, ('XavierFill', {}), ('ConstantFill', {})
    )
    pred = model.Softmax(fc8, "pred")
    xent = model.LabelCrossEntropy([pred, "label"], "xent")
    model.AveragedLoss(xent, "loss")
    return model, 231


def VGGA(order):
    model = cnn.CNNModelHelper(order, name='vgg-a',
                               use_cudnn=True, cudnn_exhaustive_search=True)
    conv1 = model.Conv(
        "data",
        "conv1",
        3,
        64,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu1 = model.Relu(conv1, "conv1")
    pool1 = model.MaxPool(relu1, "pool1", kernel=2, stride=2)
    conv2 = model.Conv(
        pool1,
        "conv2",
        64,
        128,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu2 = model.Relu(conv2, "conv2")
    pool2 = model.MaxPool(relu2, "pool2", kernel=2, stride=2)
    conv3 = model.Conv(
        pool2,
        "conv3",
        128,
        256,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu3 = model.Relu(conv3, "conv3")
    conv4 = model.Conv(
        relu3,
        "conv4",
        256,
        256,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu4 = model.Relu(conv4, "conv4")
    pool4 = model.MaxPool(relu4, "pool4", kernel=2, stride=2)
    conv5 = model.Conv(
        pool4,
        "conv5",
        256,
        512,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu5 = model.Relu(conv5, "conv5")
    conv6 = model.Conv(
        relu5,
        "conv6",
        512,
        512,
        3,
        ('XavierFill', {}),
        ('ConstantFill', {}),
        pad=1
    )
    relu6 = model.Relu(conv6, "conv6")
    pool6 = model.MaxPool(relu6, "pool6", kernel=2, stride=2)
    conv7 = model.Conv(
        pool6,
        "conv7",
        512,
        512,