From 0f7c40713f2430744b03baf9f6a3ef89a466cc22 Mon Sep 17 00:00:00 2001
From: tensor-tang <jian.j.tang@intel.com>
Date: Thu, 21 Sep 2017 22:59:33 +0800
Subject: [PATCH] add vgg and script for mkldnn benchmark

---
 benchmark/paddle/image/provider.py   |   2 +-
 benchmark/paddle/image/run.mkldnn.sh |  46 ++++++++++++
 benchmark/paddle/image/vgg.py        | 103 +++++++++++++++++++++++++++
 3 files changed, 150 insertions(+), 1 deletion(-)
 create mode 100755 benchmark/paddle/image/run.mkldnn.sh
 create mode 100644 benchmark/paddle/image/vgg.py

diff --git a/benchmark/paddle/image/provider.py b/benchmark/paddle/image/provider.py
index 1ac47212b5a..4703944c872 100644
--- a/benchmark/paddle/image/provider.py
+++ b/benchmark/paddle/image/provider.py
@@ -22,5 +22,5 @@ def initHook(settings, height, width, color, num_class, **kwargs):
 def process(settings, file_list):
     for i in xrange(1024):
         img = np.random.rand(1, settings.data_size).reshape(-1, 1).flatten()
-        lab = random.randint(0, settings.num_class)
+        lab = random.randint(0, settings.num_class - 1)
         yield img.astype('float32'), int(lab)
diff --git a/benchmark/paddle/image/run.mkldnn.sh b/benchmark/paddle/image/run.mkldnn.sh
new file mode 100755
index 00000000000..03a87afbc30
--- /dev/null
+++ b/benchmark/paddle/image/run.mkldnn.sh
@@ -0,0 +1,46 @@
+set -e
+
+function train() {
+  topology=$1
+  bs=$2
+  thread=1
+  if [ $3 ]; then
+    thread=$3
+  fi
+  if [ $thread -eq 1 ]; then
+    use_mkldnn=1
+    log="logs/${topology}-mkldnn-${bs}.log"
+  else
+    use_mkldnn=0
+    log="logs/${topology}-${thread}mklml-${bs}.log"
+  fi
+  args="batch_size=${bs}"
+  config="${topology}.py"
+  paddle train --job=time \
+    --config=$config \
+    --use_mkldnn=$use_mkldnn \
+    --use_gpu=False \
+    --trainer_count=$thread \
+    --log_period=10 \
+    --test_period=100 \
+    --config_args=$args \
+    2>&1 | tee ${log} 
+}
+
+if [ ! -d "train.list" ]; then
+  echo " " > train.list
+fi
+if [ ! -d "logs" ]; then
+  mkdir logs
+fi
+
+#========= mkldnn =========#
+# vgg
+train vgg 64 
+train vgg 128
+train vgg 256
+
+#========== mklml ===========#
+train vgg 64 16
+train vgg 128 16
+train vgg 256 16
diff --git a/benchmark/paddle/image/vgg.py b/benchmark/paddle/image/vgg.py
new file mode 100644
index 00000000000..69e4a4cddb4
--- /dev/null
+++ b/benchmark/paddle/image/vgg.py
@@ -0,0 +1,103 @@
+#!/usr/bin/env python
+from paddle.trainer_config_helpers import *
+
+height = 224
+width = 224
+num_class = 1000
+batch_size = get_config_arg('batch_size', int, 64)
+layer_num = get_config_arg('layer_num', int, 16)
+
+args = {'height': height, 'width': width, 'color': True, 'num_class': num_class}
+define_py_data_sources2(
+    "train.list", None, module="provider", obj="process", args=args)
+
+settings(
+    batch_size=batch_size,
+    learning_rate=0.01 / batch_size,
+    learning_method=MomentumOptimizer(0.9),
+    regularization=L2Regularization(0.0005 * batch_size))
+
+img = data_layer(name='image', size=height * width * 3)
+
+
+def vgg_network(vgg_num=3):
+    tmp = img_conv_group(
+        input=img,
+        num_channels=3,
+        conv_padding=1,
+        conv_num_filter=[64, 64],
+        conv_filter_size=3,
+        conv_act=ReluActivation(),
+        pool_size=2,
+        pool_stride=2,
+        pool_type=MaxPooling())
+
+    tmp = img_conv_group(
+        input=tmp,
+        conv_num_filter=[128, 128],
+        conv_padding=1,
+        conv_filter_size=3,
+        conv_act=ReluActivation(),
+        pool_stride=2,
+        pool_type=MaxPooling(),
+        pool_size=2)
+
+    channels = []
+    for i in range(vgg_num):
+        channels.append(256)
+    tmp = img_conv_group(
+        input=tmp,
+        conv_num_filter=channels,
+        conv_padding=1,
+        conv_filter_size=3,
+        conv_act=ReluActivation(),
+        pool_stride=2,
+        pool_type=MaxPooling(),
+        pool_size=2)
+    channels = []
+    for i in range(vgg_num):
+        channels.append(512)
+    tmp = img_conv_group(
+        input=tmp,
+        conv_num_filter=channels,
+        conv_padding=1,
+        conv_filter_size=3,
+        conv_act=ReluActivation(),
+        pool_stride=2,
+        pool_type=MaxPooling(),
+        pool_size=2)
+    tmp = img_conv_group(
+        input=tmp,
+        conv_num_filter=channels,
+        conv_padding=1,
+        conv_filter_size=3,
+        conv_act=ReluActivation(),
+        pool_stride=2,
+        pool_type=MaxPooling(),
+        pool_size=2)
+
+    tmp = fc_layer(
+        input=tmp,
+        size=4096,
+        act=ReluActivation(),
+        layer_attr=ExtraAttr(drop_rate=0.5))
+
+    tmp = fc_layer(
+        input=tmp,
+        size=4096,
+        act=ReluActivation(),
+        layer_attr=ExtraAttr(drop_rate=0.5))
+
+    return fc_layer(input=tmp, size=num_class, act=SoftmaxActivation())
+
+
+if layer_num == 16:
+    vgg = vgg_network(3)
+elif layer_num == 19:
+    vgg = vgg_network(4)
+else:
+    print("Wrong layer number.")
+
+lab = data_layer('label', num_class)
+loss = cross_entropy(input=vgg, label=lab)
+outputs(loss)
-- 
GitLab