Skip to content
This repository has been archived by the owner on Dec 16, 2022. It is now read-only.

Performance issue of running a single convolutional layer with different sparsity #21

Open
yushansu opened this issue Aug 27, 2019 · 7 comments

Comments

@yushansu
Copy link

Issue summary

I’m measuring the performance of the inference of a single sparse convolutional layer using SkimCaffe. However, the results show that:

  1. The effective flops achieved is only around 845 GFlops.
  2. The effective flops doesn’t change when the sparsity changes.

I get the FLOPs value from SkimCaffe’s output. Is this the right way to get the FLOPs? The dumped outputs are at the end of this post.
I was wondering why the FLOPs is low? And also why the FLOPs doesn’t change with the sparsity?

Here are some more details:
CPU: Intel clx (Xeon Platinum 8280)

KMP affinity:
export OMP_NUM_THREADS=28
export KMP_AFFINITY=granularity=fine,compact,1

Convolution size:
Input:1*3*224*224
Conv layer: 3*64*7*7
Batch size: 16

The command to run the convolution layer:
build/tools/caffe.bin test -model my-conv.prototxt -weights my-conv.caffemodel

System configuration

Operating system: CentOS Linux 7 (Core)
Compiler: icc version 19.0.4.243 (gcc version 4.8.5 compatibility)
BLAS: OpenBLAS 0.2.15
Python version (for pycaffe): Python 3.6.8

my-conv.prototxt

name: "ResNet-50"
layer {
name: "data"
type: "Data"
top: "data"
include {
phase: TEST
}
transform_param {
mirror: false
crop_size: 224
mean_value: 104
mean_value: 117
mean_value: 123
}
data_param {
source: "~/SkimCaffe/examples/imagenet/ilsvrc12_val_lmdb"
batch_size: 16
backend: LMDB
}
}
#input: "data"
#input_dim: 1
#input_dim: 3
#input_dim: 224
#input_dim: 224
layer {
bottom: "data"
name: "conv1"
#top: "conv1"
top: "conv1"
type: "Convolution"
convolution_param {
num_output: 64
kernel_size: 7
pad: 3
stride: 2
}
}

SkimCaffe's output when sparsity is 0.5 (50% of the weights are non-zeros)

\I0826 17:23:39.227200 364882 caffe.cpp:295] Use CPU.
I0826 17:23:39.237113 364882 cpu_info.cpp:452] Processor speed [MHz]: 2700
I0826 17:23:39.237139 364882 cpu_info.cpp:455] Total number of sockets: 2
I0826 17:23:39.237146 364882 cpu_info.cpp:458] Total number of CPU cores: 56
I0826 17:23:39.237151 364882 cpu_info.cpp:461] Total number of processors: 112
I0826 17:23:39.237155 364882 cpu_info.cpp:464] GPU is used: no
I0826 17:23:39.237160 364882 cpu_info.cpp:467] OpenMP environmental variables are specified: yes
I0826 17:23:39.237165 364882 cpu_info.cpp:470] OpenMP thread bind allowed: no
I0826 17:23:39.237169 364882 cpu_info.cpp:473] Number of OpenMP threads: 28
I0826 17:23:39.242393 364882 net.cpp:78] Initializing net from parameters:
name: "ResNet-50"
state {
phase: TEST
level: 0
stage: ""
}
layer {
name: "data"
type: "Data"
top: "data"
include {
phase: TEST
}
transform_param {
mirror: false
crop_size: 224
mean_value: 104
mean_value: 117
mean_value: 123
}
data_param {
source: "/nfs_home/yushansu/SkimCaffe/examples/imagenet/ilsvrc12_val_lmdb"
batch_size: 16
backend: LMDB
}
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
convolution_param {
num_output: 64
pad: 3
kernel_size: 7
stride: 2
}
}
I0826 17:23:39.242787 364882 layer_factory.hpp:77] Creating layer data
I0826 17:23:39.243247 364882 net.cpp:120] Creating Layer data
I0826 17:23:39.243268 364882 net.cpp:428] data -> data
I0826 17:23:39.247416 364885 db_lmdb.cpp:37] Opened lmdb /nfs_home/yushansu/SkimCaffe/examples/imagenet/ilsvrc12_val_lmdb
I0826 17:23:39.247437 364885 db_lmdb.cpp:40] Map size is 10038173696
I0826 17:23:39.248174 364882 data_layer.cpp:41] output data size: 16,3,224,224
I0826 17:23:39.267680 364882 net.cpp:170] Setting up data
I0826 17:23:39.267704 364882 net.cpp:177] Top shape: 16 3 224 224 (2408448)
I0826 17:23:39.267743 364882 net.cpp:185] Memory required for data: 9633792
I0826 17:23:39.267757 364882 layer_factory.hpp:77] Creating layer conv1
I0826 17:23:39.267802 364882 net.cpp:120] Creating Layer conv1
I0826 17:23:39.267812 364882 net.cpp:454] conv1 <- data
I0826 17:23:39.267827 364882 net.cpp:428] conv1 -> conv1
I0826 17:23:39.328923 364882 net.cpp:170] Setting up conv1
I0826 17:23:39.328961 364882 net.cpp:177] Top shape: 16 64 112 112 (12845056)
I0826 17:23:39.328980 364882 net.cpp:185] Memory required for data: 61014016
I0826 17:23:39.329057 364882 net.cpp:248] conv1 does not need backward computation.
I0826 17:23:39.329069 364882 net.cpp:248] data does not need backward computation.
I0826 17:23:39.329075 364882 net.cpp:290] This network produces output conv1
I0826 17:23:39.329092 364882 net.cpp:303] Network initialization done.
I0826 17:23:39.334676 364882 base_conv_layer.cpp:66] layer conv1 has sparsity of 0.5
I0826 17:23:39.335335 364882 base_conv_layer.cpp:368] ConvolutionParameter ConvMode: DEFAULT
I0826 17:23:39.335383 364882 caffe.cpp:301] Running for 50 iterations.
I0826 17:23:39.399005 364882 conv_layer.cpp:114] conv1 wall clock-time 0.054523 padding-time 0
I0826 17:23:39.399077 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 2117.29 avg 1113.66 mFlops-per-file 236.028 GF/s 300.257
I0826 17:23:39.557459 364882 net.cpp:655] Test time of data 9.041 ms ( 14.1998 % )
I0826 17:23:39.557515 364882 net.cpp:655] Test time of conv1 54.629 ms ( 85.8002 % )
I0826 17:23:39.557528 364882 caffe.cpp:330] Total forwarding time: 63.67 ms
I0826 17:23:39.571211 364882 conv_layer.cpp:114] conv1 wall clock-time 0.0100009 padding-time 0
I0826 17:23:39.571274 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 783.475 avg 436.182 mFlops-per-file 236.028 GF/s 811.425
I0826 17:23:39.597092 364882 net.cpp:655] Test time of data 3.63 ms ( 26.5002 % )
I0826 17:23:39.597131 364882 net.cpp:655] Test time of conv1 10.068 ms ( 73.4998 % )
I0826 17:23:39.597143 364882 caffe.cpp:330] Total forwarding time: 13.698 ms
I0826 17:23:39.610417 364882 conv_layer.cpp:114] conv1 wall clock-time 0.0095818 padding-time 0
I0826 17:23:39.610472 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 762.877 avg 417.005 mFlops-per-file 236.028 GF/s 833.333
I0826 17:23:39.636723 364882 net.cpp:655] Test time of data 3.66 ms ( 27.385 % )
I0826 17:23:39.636766 364882 net.cpp:655] Test time of conv1 9.705 ms ( 72.615 % )
I0826 17:23:39.636777 364882 caffe.cpp:330] Total forwarding time: 13.365 ms
I0826 17:23:39.650087 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00960612 padding-time 0
I0826 17:23:39.650146 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 767.088 avg 417.37 mFlops-per-file 236.028 GF/s 828.759
I0826 17:23:39.675879 364882 net.cpp:655] Test time of data 3.674 ms ( 27.533 % )
I0826 17:23:39.675917 364882 net.cpp:655] Test time of conv1 9.67 ms ( 72.467 % )
I0826 17:23:39.675928 364882 caffe.cpp:330] Total forwarding time: 13.344 ms
I0826 17:23:39.689330 364882 conv_layer.cpp:114] conv1 wall clock-time 0.009588 padding-time 0
I0826 17:23:39.689388 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 771.799 avg 417.632 mFlops-per-file 236.028 GF/s 823.7
I0826 17:23:39.715143 364882 net.cpp:655] Test time of data 3.785 ms ( 28.1706 % )
I0826 17:23:39.715183 364882 net.cpp:655] Test time of conv1 9.651 ms ( 71.8294 % )
I0826 17:23:39.715194 364882 caffe.cpp:330] Total forwarding time: 13.436 ms
I0826 17:23:39.728513 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00958014 padding-time 0
I0826 17:23:39.728569 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 772.278 avg 417.77 mFlops-per-file 236.028 GF/s 823.189
I0826 17:23:39.754333 364882 net.cpp:655] Test time of data 3.705 ms ( 27.759 % )
I0826 17:23:39.754371 364882 net.cpp:655] Test time of conv1 9.642 ms ( 72.241 % )
I0826 17:23:39.754382 364882 caffe.cpp:330] Total forwarding time: 13.347 ms
I0826 17:23:39.767673 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00955915 padding-time 0
I0826 17:23:39.767729 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 762.981 avg 416.588 mFlops-per-file 236.028 GF/s 833.22
I0826 17:23:39.793504 364882 net.cpp:655] Test time of data 3.7 ms ( 27.7778 % )
I0826 17:23:39.793545 364882 net.cpp:655] Test time of conv1 9.62 ms ( 72.2222 % )
I0826 17:23:39.793556 364882 caffe.cpp:330] Total forwarding time: 13.32 ms
I0826 17:23:39.806905 364882 conv_layer.cpp:114] conv1 wall clock-time 0.009583 padding-time 0
I0826 17:23:39.806962 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 770.659 avg 416.97 mFlops-per-file 236.028 GF/s 824.919
I0826 17:23:39.832778 364882 net.cpp:655] Test time of data 3.738 ms ( 27.933 % )
I0826 17:23:39.832820 364882 net.cpp:655] Test time of conv1 9.644 ms ( 72.067 % )
I0826 17:23:39.832831 364882 caffe.cpp:330] Total forwarding time: 13.382 ms
I0826 17:23:39.846145 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00957799 padding-time 0
I0826 17:23:39.846200 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 780.595 avg 416.957 mFlops-per-file 236.028 GF/s 814.418
I0826 17:23:39.871975 364882 net.cpp:655] Test time of data 3.706 ms ( 27.7728 % )
I0826 17:23:39.872025 364882 net.cpp:655] Test time of conv1 9.638 ms ( 72.2272 % )
I0826 17:23:39.872035 364882 caffe.cpp:330] Total forwarding time: 13.344 ms
I0826 17:23:39.885311 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00957417 padding-time 0
I0826 17:23:39.885367 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 766.77 avg 417.177 mFlops-per-file 236.028 GF/s 829.103
I0826 17:23:39.911111 364882 net.cpp:655] Test time of data 3.671 ms ( 27.587 % )
I0826 17:23:39.911149 364882 net.cpp:655] Test time of conv1 9.636 ms ( 72.413 % )
I0826 17:23:39.911160 364882 caffe.cpp:330] Total forwarding time: 13.307 ms
I0826 17:23:39.924415 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00956297 padding-time 0
I0826 17:23:39.924471 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 777.499 avg 416.566 mFlops-per-file 236.028 GF/s 817.662
I0826 17:23:39.950246 364882 net.cpp:655] Test time of data 3.66 ms ( 27.5478 % )
I0826 17:23:39.950285 364882 net.cpp:655] Test time of conv1 9.626 ms ( 72.4522 % )
I0826 17:23:39.950296 364882 caffe.cpp:330] Total forwarding time: 13.286 ms
I0826 17:23:39.963610 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00960588 padding-time 0
I0826 17:23:39.963711 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 784.482 avg 418.655 mFlops-per-file 236.028 GF/s 810.383
I0826 17:23:39.989496 364882 net.cpp:655] Test time of data 3.671 ms ( 27.4303 % )
I0826 17:23:39.989537 364882 net.cpp:655] Test time of conv1 9.712 ms ( 72.5697 % )
I0826 17:23:39.989547 364882 caffe.cpp:330] Total forwarding time: 13.383 ms
I0826 17:23:40.002790 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00954103 padding-time 0
I0826 17:23:40.002846 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 768.256 avg 417.488 mFlops-per-file 236.028 GF/s 827.499
I0826 17:23:40.028591 364882 net.cpp:655] Test time of data 3.669 ms ( 27.6467 % )
I0826 17:23:40.028630 364882 net.cpp:655] Test time of conv1 9.602 ms ( 72.3533 % )
I0826 17:23:40.028640 364882 caffe.cpp:330] Total forwarding time: 13.271 ms
I0826 17:23:40.041913 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00955105 padding-time 0
I0826 17:23:40.041970 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 768.066 avg 417.941 mFlops-per-file 236.028 GF/s 827.704
I0826 17:23:40.067750 364882 net.cpp:655] Test time of data 3.694 ms ( 27.7619 % )
I0826 17:23:40.067790 364882 net.cpp:655] Test time of conv1 9.612 ms ( 72.2381 % )
I0826 17:23:40.067800 364882 caffe.cpp:330] Total forwarding time: 13.306 ms
I0826 17:23:40.081027 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00956106 padding-time 0
I0826 17:23:40.081081 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 766.727 avg 416.938 mFlops-per-file 236.028 GF/s 829.149
I0826 17:23:40.106850 364882 net.cpp:655] Test time of data 3.636 ms ( 27.427 % )
I0826 17:23:40.106890 364882 net.cpp:655] Test time of conv1 9.621 ms ( 72.573 % )
I0826 17:23:40.106901 364882 caffe.cpp:330] Total forwarding time: 13.257 ms
I0826 17:23:40.120138 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00957394 padding-time 0
I0826 17:23:40.120193 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 775.212 avg 417.522 mFlops-per-file 236.028 GF/s 820.074
I0826 17:23:40.145938 364882 net.cpp:655] Test time of data 3.632 ms ( 27.3741 % )
I0826 17:23:40.145975 364882 net.cpp:655] Test time of conv1 9.636 ms ( 72.6259 % )
I0826 17:23:40.145985 364882 caffe.cpp:330] Total forwarding time: 13.268 ms
I0826 17:23:40.159260 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00957012 padding-time 0
I0826 17:23:40.159314 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 769.267 avg 417.958 mFlops-per-file 236.028 GF/s 826.411
I0826 17:23:40.185092 364882 net.cpp:655] Test time of data 3.674 ms ( 27.6158 % )
I0826 17:23:40.185130 364882 net.cpp:655] Test time of conv1 9.63 ms ( 72.3842 % )
I0826 17:23:40.185142 364882 caffe.cpp:330] Total forwarding time: 13.304 ms
I0826 17:23:40.198388 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00955677 padding-time 0
I0826 17:23:40.198442 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 768.003 avg 417.199 mFlops-per-file 236.028 GF/s 827.772
I0826 17:23:40.224225 364882 net.cpp:655] Test time of data 3.659 ms ( 27.5589 % )
I0826 17:23:40.224263 364882 net.cpp:655] Test time of conv1 9.618 ms ( 72.4411 % )
I0826 17:23:40.224274 364882 caffe.cpp:330] Total forwarding time: 13.277 ms
I0826 17:23:40.237462 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00952601 padding-time 0
I0826 17:23:40.237516 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 766.96 avg 414.539 mFlops-per-file 236.028 GF/s 828.897
I0826 17:23:40.266093 364882 net.cpp:655] Test time of data 3.63 ms ( 27.4605 % )
I0826 17:23:40.266481 364882 net.cpp:655] Test time of conv1 9.589 ms ( 72.5395 % )
I0826 17:23:40.266527 364882 caffe.cpp:330] Total forwarding time: 13.219 ms
I0826 17:23:40.281864 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00981688 padding-time 0
I0826 17:23:40.281920 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 788.28 avg 433.313 mFlops-per-file 236.028 GF/s 806.479
I0826 17:23:40.307670 364882 net.cpp:655] Test time of data 5.214 ms ( 34.5481 % )
I0826 17:23:40.307752 364882 net.cpp:655] Test time of conv1 9.878 ms ( 65.4519 % )
I0826 17:23:40.307765 364882 caffe.cpp:330] Total forwarding time: 15.092 ms
I0826 17:23:40.321373 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00954103 padding-time 0
I0826 17:23:40.321426 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 763.249 avg 418.376 mFlops-per-file 236.028 GF/s 832.927
I0826 17:23:40.347199 364882 net.cpp:655] Test time of data 4.04 ms ( 29.6166 % )
I0826 17:23:40.347239 364882 net.cpp:655] Test time of conv1 9.601 ms ( 70.3834 % )
I0826 17:23:40.347249 364882 caffe.cpp:330] Total forwarding time: 13.641 ms
I0826 17:23:40.361042 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00956511 padding-time 0
I0826 17:23:40.361097 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 774.089 avg 417.683 mFlops-per-file 236.028 GF/s 821.263
I0826 17:23:40.386855 364882 net.cpp:655] Test time of data 4.156 ms ( 30.1509 % )
I0826 17:23:40.386894 364882 net.cpp:655] Test time of conv1 9.628 ms ( 69.8491 % )
I0826 17:23:40.386905 364882 caffe.cpp:330] Total forwarding time: 13.784 ms
I0826 17:23:40.400305 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00955296 padding-time 0
I0826 17:23:40.400359 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 765.94 avg 416.965 mFlops-per-file 236.028 GF/s 830.001
I0826 17:23:40.426134 364882 net.cpp:655] Test time of data 3.818 ms ( 28.4225 % )
I0826 17:23:40.426172 364882 net.cpp:655] Test time of conv1 9.615 ms ( 71.5775 % )
I0826 17:23:40.426182 364882 caffe.cpp:330] Total forwarding time: 13.433 ms
I0826 17:23:40.439442 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00953698 padding-time 0
I0826 17:23:40.439496 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 762.682 avg 418.315 mFlops-per-file 236.028 GF/s 833.547
I0826 17:23:40.465315 364882 net.cpp:655] Test time of data 3.696 ms ( 27.8041 % )
I0826 17:23:40.465353 364882 net.cpp:655] Test time of conv1 9.597 ms ( 72.1959 % )
I0826 17:23:40.465363 364882 caffe.cpp:330] Total forwarding time: 13.293 ms
I0826 17:23:40.478651 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00956917 padding-time 0
I0826 17:23:40.478705 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 768.386 avg 417.647 mFlops-per-file 236.028 GF/s 827.359
I0826 17:23:40.504456 364882 net.cpp:655] Test time of data 3.688 ms ( 27.6939 % )
I0826 17:23:40.504494 364882 net.cpp:655] Test time of conv1 9.629 ms ( 72.3061 % )
I0826 17:23:40.504504 364882 caffe.cpp:330] Total forwarding time: 13.317 ms
I0826 17:23:40.517815 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00957298 padding-time 0
I0826 17:23:40.517868 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 775.285 avg 417.851 mFlops-per-file 236.028 GF/s 819.997
I0826 17:23:40.543712 364882 net.cpp:655] Test time of data 3.709 ms ( 27.8036 % )
I0826 17:23:40.543753 364882 net.cpp:655] Test time of conv1 9.631 ms ( 72.1964 % )
I0826 17:23:40.543764 364882 caffe.cpp:330] Total forwarding time: 13.34 ms
I0826 17:23:40.557160 364882 conv_layer.cpp:114] conv1 wall clock-time 0.0095911 padding-time 0
I0826 17:23:40.557214 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 777.245 avg 417.398 mFlops-per-file 236.028 GF/s 817.928
I0826 17:23:40.583015 364882 net.cpp:655] Test time of data 3.778 ms ( 28.1373 % )
I0826 17:23:40.583055 364882 net.cpp:655] Test time of conv1 9.649 ms ( 71.8627 % )
I0826 17:23:40.583065 364882 caffe.cpp:330] Total forwarding time: 13.427 ms
I0826 17:23:40.596320 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00955701 padding-time 0
I0826 17:23:40.596374 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 769.677 avg 418.434 mFlops-per-file 236.028 GF/s 825.972
I0826 17:23:40.622519 364882 net.cpp:655] Test time of data 3.67 ms ( 27.621 % )
I0826 17:23:40.622560 364882 net.cpp:655] Test time of conv1 9.617 ms ( 72.379 % )
I0826 17:23:40.622570 364882 caffe.cpp:330] Total forwarding time: 13.287 ms
I0826 17:23:40.635856 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00956488 padding-time 0
I0826 17:23:40.635910 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 775.789 avg 416.091 mFlops-per-file 236.028 GF/s 819.464
I0826 17:23:40.661743 364882 net.cpp:655] Test time of data 3.693 ms ( 27.6257 % )
I0826 17:23:40.661785 364882 net.cpp:655] Test time of conv1 9.675 ms ( 72.3743 % )
I0826 17:23:40.661795 364882 caffe.cpp:330] Total forwarding time: 13.368 ms
I0826 17:23:40.675067 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00956297 padding-time 0
I0826 17:23:40.675122 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 762.887 avg 416.594 mFlops-per-file 236.028 GF/s 833.323
I0826 17:23:40.700861 364882 net.cpp:655] Test time of data 3.68 ms ( 27.6609 % )
I0826 17:23:40.700898 364882 net.cpp:655] Test time of conv1 9.624 ms ( 72.3391 % )
I0826 17:23:40.700908 364882 caffe.cpp:330] Total forwarding time: 13.304 ms
I0826 17:23:40.714176 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00957799 padding-time 0
I0826 17:23:40.714229 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 777.839 avg 416.767 mFlops-per-file 236.028 GF/s 817.304
I0826 17:23:40.739976 364882 net.cpp:655] Test time of data 3.662 ms ( 27.5359 % )
I0826 17:23:40.740020 364882 net.cpp:655] Test time of conv1 9.637 ms ( 72.4641 % )
I0826 17:23:40.740032 364882 caffe.cpp:330] Total forwarding time: 13.299 ms
I0826 17:23:40.753279 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00955009 padding-time 0
I0826 17:23:40.753332 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 765.156 avg 417.465 mFlops-per-file 236.028 GF/s 830.851
I0826 17:23:40.779093 364882 net.cpp:655] Test time of data 3.669 ms ( 27.6301 % )
I0826 17:23:40.779131 364882 net.cpp:655] Test time of conv1 9.61 ms ( 72.3699 % )
I0826 17:23:40.779141 364882 caffe.cpp:330] Total forwarding time: 13.279 ms
I0826 17:23:40.792373 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00956798 padding-time 0
I0826 17:23:40.792426 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 778.89 avg 416.655 mFlops-per-file 236.028 GF/s 816.201
I0826 17:23:40.818192 364882 net.cpp:655] Test time of data 3.632 ms ( 27.3886 % )
I0826 17:23:40.818233 364882 net.cpp:655] Test time of conv1 9.629 ms ( 72.6114 % )
I0826 17:23:40.818243 364882 caffe.cpp:330] Total forwarding time: 13.261 ms
I0826 17:23:40.831501 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00956392 padding-time 0
I0826 17:23:40.831557 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 768.512 avg 416.287 mFlops-per-file 236.028 GF/s 827.224
I0826 17:23:40.857972 364882 net.cpp:655] Test time of data 3.667 ms ( 27.5859 % )
I0826 17:23:40.858053 364882 net.cpp:655] Test time of conv1 9.626 ms ( 72.4141 % )
I0826 17:23:40.858064 364882 caffe.cpp:330] Total forwarding time: 13.293 ms
I0826 17:23:40.871491 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00957894 padding-time 0
I0826 17:23:40.871546 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 777.421 avg 418.265 mFlops-per-file 236.028 GF/s 817.743
I0826 17:23:40.897763 364882 net.cpp:655] Test time of data 3.818 ms ( 28.3719 % )
I0826 17:23:40.897801 364882 net.cpp:655] Test time of conv1 9.639 ms ( 71.6281 % )
I0826 17:23:40.897811 364882 caffe.cpp:330] Total forwarding time: 13.457 ms
I0826 17:23:40.911106 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00957513 padding-time 0
I0826 17:23:40.911160 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 779.785 avg 418.021 mFlops-per-file 236.028 GF/s 815.264
I0826 17:23:40.936878 364882 net.cpp:655] Test time of data 3.694 ms ( 27.7182 % )
I0826 17:23:40.936916 364882 net.cpp:655] Test time of conv1 9.633 ms ( 72.2818 % )
I0826 17:23:40.936928 364882 caffe.cpp:330] Total forwarding time: 13.327 ms
I0826 17:23:40.950183 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00953889 padding-time 0
I0826 17:23:40.950234 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 760.889 avg 416.466 mFlops-per-file 236.028 GF/s 835.511
I0826 17:23:40.975996 364882 net.cpp:655] Test time of data 3.689 ms ( 27.7682 % )
I0826 17:23:40.976038 364882 net.cpp:655] Test time of conv1 9.596 ms ( 72.2318 % )
I0826 17:23:40.976104 364882 caffe.cpp:330] Total forwarding time: 13.285 ms
I0826 17:23:40.989352 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00952291 padding-time 0
I0826 17:23:40.989408 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 764.975 avg 416.476 mFlops-per-file 236.028 GF/s 831.049
I0826 17:23:41.015156 364882 net.cpp:655] Test time of data 3.686 ms ( 27.7769 % )
I0826 17:23:41.015194 364882 net.cpp:655] Test time of conv1 9.584 ms ( 72.2231 % )
I0826 17:23:41.015204 364882 caffe.cpp:330] Total forwarding time: 13.27 ms
I0826 17:23:41.028442 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00955391 padding-time 0
I0826 17:23:41.028498 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 770.845 avg 415.482 mFlops-per-file 236.028 GF/s 824.72
I0826 17:23:41.054263 364882 net.cpp:655] Test time of data 3.655 ms ( 27.5413 % )
I0826 17:23:41.054301 364882 net.cpp:655] Test time of conv1 9.616 ms ( 72.4587 % )
I0826 17:23:41.054311 364882 caffe.cpp:330] Total forwarding time: 13.271 ms
I0826 17:23:41.067581 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00951982 padding-time 0
I0826 17:23:41.067636 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 762.466 avg 416.505 mFlops-per-file 236.028 GF/s 833.783
I0826 17:23:41.093394 364882 net.cpp:655] Test time of data 3.723 ms ( 27.9904 % )
I0826 17:23:41.093433 364882 net.cpp:655] Test time of conv1 9.578 ms ( 72.0096 % )
I0826 17:23:41.093443 364882 caffe.cpp:330] Total forwarding time: 13.301 ms
I0826 17:23:41.106710 364882 conv_layer.cpp:114] conv1 wall clock-time 0.009552 padding-time 0
I0826 17:23:41.106763 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 768.075 avg 415.164 mFlops-per-file 236.028 GF/s 827.694
I0826 17:23:41.132522 364882 net.cpp:655] Test time of data 3.687 ms ( 27.7281 % )
I0826 17:23:41.132561 364882 net.cpp:655] Test time of conv1 9.61 ms ( 72.2719 % )
I0826 17:23:41.132571 364882 caffe.cpp:330] Total forwarding time: 13.297 ms
I0826 17:23:41.145793 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00954604 padding-time 0
I0826 17:23:41.145844 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 766.646 avg 416.322 mFlops-per-file 236.028 GF/s 829.237
I0826 17:23:41.171620 364882 net.cpp:655] Test time of data 3.648 ms ( 27.5279 % )
I0826 17:23:41.171658 364882 net.cpp:655] Test time of conv1 9.604 ms ( 72.4721 % )
I0826 17:23:41.171669 364882 caffe.cpp:330] Total forwarding time: 13.252 ms
I0826 17:23:41.184911 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00955582 padding-time 0
I0826 17:23:41.184960 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 770.429 avg 417.252 mFlops-per-file 236.028 GF/s 825.165
I0826 17:23:41.210716 364882 net.cpp:655] Test time of data 3.656 ms ( 27.5488 % )
I0826 17:23:41.210757 364882 net.cpp:655] Test time of conv1 9.615 ms ( 72.4512 % )
I0826 17:23:41.210767 364882 caffe.cpp:330] Total forwarding time: 13.271 ms
I0826 17:23:41.223989 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00956202 padding-time 0
I0826 17:23:41.224050 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 770.548 avg 417.554 mFlops-per-file 236.028 GF/s 825.038
I0826 17:23:41.249781 364882 net.cpp:655] Test time of data 3.632 ms ( 27.3906 % )
I0826 17:23:41.249822 364882 net.cpp:655] Test time of conv1 9.628 ms ( 72.6094 % )
I0826 17:23:41.249832 364882 caffe.cpp:330] Total forwarding time: 13.26 ms
I0826 17:23:41.263079 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00957084 padding-time 0
I0826 17:23:41.263130 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 775.745 avg 417.931 mFlops-per-file 236.028 GF/s 819.511
I0826 17:23:41.288861 364882 net.cpp:655] Test time of data 3.648 ms ( 27.4782 % )
I0826 17:23:41.288899 364882 net.cpp:655] Test time of conv1 9.628 ms ( 72.5218 % )
I0826 17:23:41.288909 364882 caffe.cpp:330] Total forwarding time: 13.276 ms
I0826 17:23:41.302129 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00957108 padding-time 0
I0826 17:23:41.302182 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 776.288 avg 418.354 mFlops-per-file 236.028 GF/s 818.937
I0826 17:23:41.328037 364882 net.cpp:655] Test time of data 3.622 ms ( 27.2208 % )
I0826 17:23:41.328078 364882 net.cpp:655] Test time of conv1 9.684 ms ( 72.7792 % )
I0826 17:23:41.328088 364882 caffe.cpp:330] Total forwarding time: 13.306 ms
I0826 17:23:41.341290 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00953603 padding-time 0
I0826 17:23:41.341343 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 765.844 avg 415.29 mFlops-per-file 236.028 GF/s 830.105
I0826 17:23:41.367089 364882 net.cpp:655] Test time of data 3.637 ms ( 27.4864 % )
I0826 17:23:41.367128 364882 net.cpp:655] Test time of conv1 9.595 ms ( 72.5136 % )
I0826 17:23:41.367139 364882 caffe.cpp:330] Total forwarding time: 13.232 ms
I0826 17:23:41.380353 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00954413 padding-time 0
I0826 17:23:41.380407 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 771.473 avg 416.335 mFlops-per-file 236.028 GF/s 824.048
I0826 17:23:41.406165 364882 net.cpp:655] Test time of data 3.645 ms ( 27.5136 % )
I0826 17:23:41.406206 364882 net.cpp:655] Test time of conv1 9.603 ms ( 72.4864 % )
I0826 17:23:41.406217 364882 caffe.cpp:330] Total forwarding time: 13.248 ms
I0826 17:23:41.419467 364882 conv_layer.cpp:114] conv1 wall clock-time 0.0095489 padding-time 0
I0826 17:23:41.419518 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 767.572 avg 417.422 mFlops-per-file 236.028 GF/s 828.236
I0826 17:23:41.445269 364882 net.cpp:655] Test time of data 3.675 ms ( 27.6732 % )
I0826 17:23:41.445307 364882 net.cpp:655] Test time of conv1 9.605 ms ( 72.3268 % )
I0826 17:23:41.445317 364882 caffe.cpp:330] Total forwarding time: 13.28 ms
I0826 17:23:41.458581 364882 conv_layer.cpp:114] conv1 wall clock-time 0.00956202 padding-time 0
I0826 17:23:41.458636 364882 conv_layer.cpp:124] conv1 K-cycles-per-file max 769.138 avg 415.628 mFlops-per-file 236.028 GF/s 826.55
I0826 17:23:41.484419 364882 net.cpp:655] Test time of data 3.669 ms ( 27.6072 % )
I0826 17:23:41.484455 364882 net.cpp:655] Test time of conv1 9.621 ms ( 72.3928 % )
I0826 17:23:41.484465 364882 caffe.cpp:330] Total forwarding time: 13.29 ms
I0826 17:23:41.484474 364882 caffe.cpp:333] Loss: 0
I0826 17:23:46.853449 364882 caffe.cpp:350] Total-images-processed: 800
I0826 17:23:46.853484 364882 caffe.cpp:353] conv1 K-cycles-per-file 797.846 mFlops-per-file 236.028 GF/s 796.809
freq = 2.69346e+09

SkimCaffe's output when sparsity is 0.7 (70% of the weights are non-zeros)

I0826 17:22:56.127554 364794 caffe.cpp:295] Use CPU.
I0826 17:22:56.136904 364794 cpu_info.cpp:452] Processor speed [MHz]: 2700
I0826 17:22:56.136929 364794 cpu_info.cpp:455] Total number of sockets: 2
I0826 17:22:56.136934 364794 cpu_info.cpp:458] Total number of CPU cores: 56
I0826 17:22:56.136940 364794 cpu_info.cpp:461] Total number of processors: 112
I0826 17:22:56.136943 364794 cpu_info.cpp:464] GPU is used: no
I0826 17:22:56.136948 364794 cpu_info.cpp:467] OpenMP environmental variables are specified: yes
I0826 17:22:56.136952 364794 cpu_info.cpp:470] OpenMP thread bind allowed: no
I0826 17:22:56.136957 364794 cpu_info.cpp:473] Number of OpenMP threads: 28
I0826 17:22:56.143631 364794 net.cpp:78] Initializing net from parameters:
name: "ResNet-50"
state {
phase: TEST
level: 0
stage: ""
}
layer {
name: "data"
type: "Data"
top: "data"
include {
phase: TEST
}
transform_param {
mirror: false
crop_size: 224
mean_value: 104
mean_value: 117
mean_value: 123
}
data_param {
source: "/nfs_home/yushansu/SkimCaffe/examples/imagenet/ilsvrc12_val_lmdb"
batch_size: 16
backend: LMDB
}
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
convolution_param {
num_output: 64
pad: 3
kernel_size: 7
stride: 2
}
}
I0826 17:22:56.144038 364794 layer_factory.hpp:77] Creating layer data
I0826 17:22:56.144498 364794 net.cpp:120] Creating Layer data
I0826 17:22:56.144517 364794 net.cpp:428] data -> data
I0826 17:22:56.165719 364797 db_lmdb.cpp:37] Opened lmdb /nfs_home/yushansu/SkimCaffe/examples/imagenet/ilsvrc12_val_lmdb
I0826 17:22:56.165738 364797 db_lmdb.cpp:40] Map size is 10038173696
I0826 17:22:56.166476 364794 data_layer.cpp:41] output data size: 16,3,224,224
I0826 17:22:56.193086 364794 net.cpp:170] Setting up data
I0826 17:22:56.193110 364794 net.cpp:177] Top shape: 16 3 224 224 (2408448)
I0826 17:22:56.193150 364794 net.cpp:185] Memory required for data: 9633792
I0826 17:22:56.193166 364794 layer_factory.hpp:77] Creating layer conv1
I0826 17:22:56.193208 364794 net.cpp:120] Creating Layer conv1
I0826 17:22:56.193219 364794 net.cpp:454] conv1 <- data
I0826 17:22:56.193233 364794 net.cpp:428] conv1 -> conv1
I0826 17:22:56.247386 364794 net.cpp:170] Setting up conv1
I0826 17:22:56.247426 364794 net.cpp:177] Top shape: 16 64 112 112 (12845056)
I0826 17:22:56.247443 364794 net.cpp:185] Memory required for data: 61014016
I0826 17:22:56.247519 364794 net.cpp:248] conv1 does not need backward computation.
I0826 17:22:56.247531 364794 net.cpp:248] data does not need backward computation.
I0826 17:22:56.247536 364794 net.cpp:290] This network produces output conv1
I0826 17:22:56.247555 364794 net.cpp:303] Network initialization done.
I0826 17:22:56.253495 364794 base_conv_layer.cpp:66] layer conv1 has sparsity of 0.700043
I0826 17:22:56.254216 364794 base_conv_layer.cpp:368] ConvolutionParameter ConvMode: DEFAULT
I0826 17:22:56.254261 364794 caffe.cpp:301] Running for 50 iterations.
I0826 17:22:56.315704 364794 conv_layer.cpp:114] conv1 wall clock-time 0.0523908 padding-time 0
I0826 17:22:56.315775 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 1682.65 avg 947.043 mFlops-per-file 236.028 GF/s 377.812
I0826 17:22:56.473810 364794 net.cpp:655] Test time of data 8.992 ms ( 14.6233 % )
I0826 17:22:56.473865 364794 net.cpp:655] Test time of conv1 52.499 ms ( 85.3767 % )
I0826 17:22:56.473878 364794 caffe.cpp:330] Total forwarding time: 61.491 ms
I0826 17:22:56.487519 364794 conv_layer.cpp:114] conv1 wall clock-time 0.0100031 padding-time 0
I0826 17:22:56.487578 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 780.549 avg 436.695 mFlops-per-file 236.028 GF/s 814.46
I0826 17:22:56.515791 364794 net.cpp:655] Test time of data 3.58 ms ( 26.2233 % )
I0826 17:22:56.515830 364794 net.cpp:655] Test time of conv1 10.072 ms ( 73.7767 % )
I0826 17:22:56.515842 364794 caffe.cpp:330] Total forwarding time: 13.652 ms
I0826 17:22:56.529036 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00954103 padding-time 0
I0826 17:22:56.529091 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 743.617 avg 417.113 mFlops-per-file 236.028 GF/s 854.91
I0826 17:22:56.557291 364794 net.cpp:655] Test time of data 3.619 ms ( 27.2454 % )
I0826 17:22:56.557333 364794 net.cpp:655] Test time of conv1 9.664 ms ( 72.7546 % )
I0826 17:22:56.557345 364794 caffe.cpp:330] Total forwarding time: 13.283 ms
I0826 17:22:56.570549 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00953102 padding-time 0
I0826 17:22:56.570606 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 746.516 avg 417.004 mFlops-per-file 236.028 GF/s 851.59
I0826 17:22:56.598744 364794 net.cpp:655] Test time of data 3.635 ms ( 27.4775 % )
I0826 17:22:56.598784 364794 net.cpp:655] Test time of conv1 9.594 ms ( 72.5225 % )
I0826 17:22:56.598795 364794 caffe.cpp:330] Total forwarding time: 13.229 ms
I0826 17:22:56.611979 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00950789 padding-time 0
I0826 17:22:56.612035 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 741.533 avg 415.769 mFlops-per-file 236.028 GF/s 857.313
I0826 17:22:56.640594 364794 net.cpp:655] Test time of data 3.644 ms ( 27.5706 % )
I0826 17:22:56.640635 364794 net.cpp:655] Test time of conv1 9.573 ms ( 72.4294 % )
I0826 17:22:56.640646 364794 caffe.cpp:330] Total forwarding time: 13.217 ms
I0826 17:22:56.653890 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00951719 padding-time 0
I0826 17:22:56.653945 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 746.022 avg 416.095 mFlops-per-file 236.028 GF/s 852.154
I0826 17:22:56.682415 364794 net.cpp:655] Test time of data 3.695 ms ( 27.8448 % )
I0826 17:22:56.682454 364794 net.cpp:655] Test time of conv1 9.575 ms ( 72.1552 % )
I0826 17:22:56.682466 364794 caffe.cpp:330] Total forwarding time: 13.27 ms
I0826 17:22:56.695732 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00949001 padding-time 0
I0826 17:22:56.695791 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 746.19 avg 415.359 mFlops-per-file 236.028 GF/s 851.962
I0826 17:22:56.723908 364794 net.cpp:655] Test time of data 3.748 ms ( 28.1762 % )
I0826 17:22:56.723947 364794 net.cpp:655] Test time of conv1 9.554 ms ( 71.8238 % )
I0826 17:22:56.723956 364794 caffe.cpp:330] Total forwarding time: 13.302 ms
I0826 17:22:56.737150 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00948095 padding-time 0
I0826 17:22:56.737206 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 747.255 avg 415.4 mFlops-per-file 236.028 GF/s 850.749
I0826 17:22:56.765367 364794 net.cpp:655] Test time of data 3.682 ms ( 27.8412 % )
I0826 17:22:56.765405 364794 net.cpp:655] Test time of conv1 9.543 ms ( 72.1588 % )
I0826 17:22:56.765415 364794 caffe.cpp:330] Total forwarding time: 13.225 ms
I0826 17:22:56.778590 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00950193 padding-time 0
I0826 17:22:56.778646 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 749.24 avg 414.806 mFlops-per-file 236.028 GF/s 848.494
I0826 17:22:56.806838 364794 net.cpp:655] Test time of data 3.638 ms ( 27.5543 % )
I0826 17:22:56.806877 364794 net.cpp:655] Test time of conv1 9.565 ms ( 72.4457 % )
I0826 17:22:56.806888 364794 caffe.cpp:330] Total forwarding time: 13.203 ms
I0826 17:22:56.820128 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00952196 padding-time 0
I0826 17:22:56.820184 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 749.964 avg 416.067 mFlops-per-file 236.028 GF/s 847.675
I0826 17:22:56.848294 364794 net.cpp:655] Test time of data 3.686 ms ( 27.7748 % )
I0826 17:22:56.848335 364794 net.cpp:655] Test time of conv1 9.585 ms ( 72.2252 % )
I0826 17:22:56.848345 364794 caffe.cpp:330] Total forwarding time: 13.271 ms
I0826 17:22:56.861531 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00949812 padding-time 0
I0826 17:22:56.861587 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 749.973 avg 414.959 mFlops-per-file 236.028 GF/s 847.664
I0826 17:22:56.889752 364794 net.cpp:655] Test time of data 3.648 ms ( 27.6175 % )
I0826 17:22:56.889793 364794 net.cpp:655] Test time of conv1 9.561 ms ( 72.3825 % )
I0826 17:22:56.889804 364794 caffe.cpp:330] Total forwarding time: 13.209 ms
I0826 17:22:56.902994 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00951385 padding-time 0
I0826 17:22:56.903059 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 747.811 avg 416.108 mFlops-per-file 236.028 GF/s 850.116
I0826 17:22:56.931216 364794 net.cpp:655] Test time of data 3.604 ms ( 27.3465 % )
I0826 17:22:56.931257 364794 net.cpp:655] Test time of conv1 9.575 ms ( 72.6535 % )
I0826 17:22:56.931267 364794 caffe.cpp:330] Total forwarding time: 13.179 ms
I0826 17:22:56.944460 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00951791 padding-time 0
I0826 17:22:56.944516 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 753.797 avg 416.355 mFlops-per-file 236.028 GF/s 843.364
I0826 17:22:56.972712 364794 net.cpp:655] Test time of data 3.643 ms ( 27.5526 % )
I0826 17:22:56.972750 364794 net.cpp:655] Test time of conv1 9.579 ms ( 72.4474 % )
I0826 17:22:56.972761 364794 caffe.cpp:330] Total forwarding time: 13.222 ms
I0826 17:22:56.985934 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00949001 padding-time 0
I0826 17:22:56.985987 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 746.127 avg 416.211 mFlops-per-file 236.028 GF/s 852.034
I0826 17:22:57.014142 364794 net.cpp:655] Test time of data 3.65 ms ( 27.6473 % )
I0826 17:22:57.014183 364794 net.cpp:655] Test time of conv1 9.552 ms ( 72.3527 % )
I0826 17:22:57.014194 364794 caffe.cpp:330] Total forwarding time: 13.202 ms
I0826 17:22:57.027410 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00953913 padding-time 0
I0826 17:22:57.027465 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 750.23 avg 415.303 mFlops-per-file 236.028 GF/s 847.374
I0826 17:22:57.055609 364794 net.cpp:655] Test time of data 3.645 ms ( 27.5157 % )
I0826 17:22:57.055646 364794 net.cpp:655] Test time of conv1 9.602 ms ( 72.4843 % )
I0826 17:22:57.055657 364794 caffe.cpp:330] Total forwarding time: 13.247 ms
I0826 17:22:57.068832 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00953197 padding-time 0
I0826 17:22:57.068887 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 751.308 avg 416.88 mFlops-per-file 236.028 GF/s 846.158
I0826 17:22:57.097012 364794 net.cpp:655] Test time of data 3.614 ms ( 27.3643 % )
I0826 17:22:57.097051 364794 net.cpp:655] Test time of conv1 9.593 ms ( 72.6357 % )
I0826 17:22:57.097062 364794 caffe.cpp:330] Total forwarding time: 13.207 ms
I0826 17:22:57.110214 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00947404 padding-time 0
I0826 17:22:57.110270 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 745.881 avg 415.197 mFlops-per-file 236.028 GF/s 852.315
I0826 17:22:57.138409 364794 net.cpp:655] Test time of data 3.645 ms ( 27.6534 % )
I0826 17:22:57.138449 364794 net.cpp:655] Test time of conv1 9.536 ms ( 72.3466 % )
I0826 17:22:57.138459 364794 caffe.cpp:330] Total forwarding time: 13.181 ms
I0826 17:22:57.151625 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00949693 padding-time 0
I0826 17:22:57.151680 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 749.773 avg 415.311 mFlops-per-file 236.028 GF/s 847.891
I0826 17:22:57.182847 364794 net.cpp:655] Test time of data 3.637 ms ( 27.5635 % )
I0826 17:22:57.183236 364794 net.cpp:655] Test time of conv1 9.558 ms ( 72.4365 % )
I0826 17:22:57.183279 364794 caffe.cpp:330] Total forwarding time: 13.195 ms
I0826 17:22:57.198654 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00978494 padding-time 0
I0826 17:22:57.198709 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 787.515 avg 430.274 mFlops-per-file 236.028 GF/s 807.256
I0826 17:22:57.226840 364794 net.cpp:655] Test time of data 5.321 ms ( 35.0851 % )
I0826 17:22:57.226877 364794 net.cpp:655] Test time of conv1 9.845 ms ( 64.9149 % )
I0826 17:22:57.226888 364794 caffe.cpp:330] Total forwarding time: 15.166 ms
I0826 17:22:57.240298 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00951314 padding-time 0
I0826 17:22:57.240352 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 751.939 avg 414.779 mFlops-per-file 236.028 GF/s 845.449
I0826 17:22:57.268481 364794 net.cpp:655] Test time of data 3.866 ms ( 28.7627 % )
I0826 17:22:57.268564 364794 net.cpp:655] Test time of conv1 9.575 ms ( 71.2373 % )
I0826 17:22:57.268576 364794 caffe.cpp:330] Total forwarding time: 13.441 ms
I0826 17:22:57.282107 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00952196 padding-time 0
I0826 17:22:57.282158 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 754.022 avg 415.741 mFlops-per-file 236.028 GF/s 843.113
I0826 17:22:57.310261 364794 net.cpp:655] Test time of data 3.979 ms ( 29.3458 % )
I0826 17:22:57.310300 364794 net.cpp:655] Test time of conv1 9.58 ms ( 70.6542 % )
I0826 17:22:57.310312 364794 caffe.cpp:330] Total forwarding time: 13.559 ms
I0826 17:22:57.323701 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00951123 padding-time 0
I0826 17:22:57.323755 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 752.182 avg 417.116 mFlops-per-file 236.028 GF/s 845.175
I0826 17:22:57.351897 364794 net.cpp:655] Test time of data 3.815 ms ( 28.5042 % )
I0826 17:22:57.351934 364794 net.cpp:655] Test time of conv1 9.569 ms ( 71.4958 % )
I0826 17:22:57.351945 364794 caffe.cpp:330] Total forwarding time: 13.384 ms
I0826 17:22:57.365120 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00949788 padding-time 0
I0826 17:22:57.365175 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 749.323 avg 415.686 mFlops-per-file 236.028 GF/s 848.4
I0826 17:22:57.393313 364794 net.cpp:655] Test time of data 3.647 ms ( 27.6204 % )
I0826 17:22:57.393350 364794 net.cpp:655] Test time of conv1 9.557 ms ( 72.3796 % )
I0826 17:22:57.393362 364794 caffe.cpp:330] Total forwarding time: 13.204 ms
I0826 17:22:57.406569 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00950909 padding-time 0
I0826 17:22:57.406620 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 748.487 avg 416.18 mFlops-per-file 236.028 GF/s 849.347
I0826 17:22:57.434756 364794 net.cpp:655] Test time of data 3.668 ms ( 27.7165 % )
I0826 17:22:57.434794 364794 net.cpp:655] Test time of conv1 9.566 ms ( 72.2835 % )
I0826 17:22:57.434805 364794 caffe.cpp:330] Total forwarding time: 13.234 ms
I0826 17:22:57.447983 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00950408 padding-time 0
I0826 17:22:57.448043 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 762.542 avg 417.31 mFlops-per-file 236.028 GF/s 833.692
I0826 17:22:57.476189 364794 net.cpp:655] Test time of data 3.644 ms ( 27.5789 % )
I0826 17:22:57.476225 364794 net.cpp:655] Test time of conv1 9.569 ms ( 72.4211 % )
I0826 17:22:57.476236 364794 caffe.cpp:330] Total forwarding time: 13.213 ms
I0826 17:22:57.489393 364794 conv_layer.cpp:114] conv1 wall clock-time 0.0094831 padding-time 0
I0826 17:22:57.489449 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 748.82 avg 416.688 mFlops-per-file 236.028 GF/s 848.97
I0826 17:22:57.517606 364794 net.cpp:655] Test time of data 3.645 ms ( 27.6346 % )
I0826 17:22:57.517643 364794 net.cpp:655] Test time of conv1 9.545 ms ( 72.3654 % )
I0826 17:22:57.517654 364794 caffe.cpp:330] Total forwarding time: 13.19 ms
I0826 17:22:57.530827 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00949812 padding-time 0
I0826 17:22:57.530881 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 751.337 avg 415.418 mFlops-per-file 236.028 GF/s 846.126
I0826 17:22:57.559113 364794 net.cpp:655] Test time of data 3.647 ms ( 27.6162 % )
I0826 17:22:57.559155 364794 net.cpp:655] Test time of conv1 9.559 ms ( 72.3838 % )
I0826 17:22:57.559165 364794 caffe.cpp:330] Total forwarding time: 13.206 ms
I0826 17:22:57.572379 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00955486 padding-time 0
I0826 17:22:57.572432 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 757.086 avg 418.006 mFlops-per-file 236.028 GF/s 839.701
I0826 17:22:57.600639 364794 net.cpp:655] Test time of data 3.628 ms ( 27.3977 % )
I0826 17:22:57.600679 364794 net.cpp:655] Test time of conv1 9.614 ms ( 72.6023 % )
I0826 17:22:57.600690 364794 caffe.cpp:330] Total forwarding time: 13.242 ms
I0826 17:22:57.613874 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00949693 padding-time 0
I0826 17:22:57.613983 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 746.421 avg 414.199 mFlops-per-file 236.028 GF/s 851.699
I0826 17:22:57.642489 364794 net.cpp:655] Test time of data 3.656 ms ( 27.5488 % )
I0826 17:22:57.642532 364794 net.cpp:655] Test time of conv1 9.615 ms ( 72.4512 % )
I0826 17:22:57.642544 364794 caffe.cpp:330] Total forwarding time: 13.271 ms
I0826 17:22:57.655705 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00948501 padding-time 0
I0826 17:22:57.655759 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 749.226 avg 416.444 mFlops-per-file 236.028 GF/s 848.509
I0826 17:22:57.683912 364794 net.cpp:655] Test time of data 3.649 ms ( 27.6565 % )
I0826 17:22:57.683953 364794 net.cpp:655] Test time of conv1 9.545 ms ( 72.3435 % )
I0826 17:22:57.683962 364794 caffe.cpp:330] Total forwarding time: 13.194 ms
I0826 17:22:57.697171 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00948405 padding-time 0
I0826 17:22:57.697223 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 746.028 avg 415.069 mFlops-per-file 236.028 GF/s 852.147
I0826 17:22:57.725365 364794 net.cpp:655] Test time of data 3.695 ms ( 27.9142 % )
I0826 17:22:57.725404 364794 net.cpp:655] Test time of conv1 9.542 ms ( 72.0858 % )
I0826 17:22:57.725415 364794 caffe.cpp:330] Total forwarding time: 13.237 ms
I0826 17:22:57.738567 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00949192 padding-time 0
I0826 17:22:57.738620 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 751.016 avg 416.358 mFlops-per-file 236.028 GF/s 846.487
I0826 17:22:57.767297 364794 net.cpp:655] Test time of data 3.632 ms ( 27.5506 % )
I0826 17:22:57.767371 364794 net.cpp:655] Test time of conv1 9.551 ms ( 72.4494 % )
I0826 17:22:57.767381 364794 caffe.cpp:330] Total forwarding time: 13.183 ms
I0826 17:22:57.780930 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00968695 padding-time 0
I0826 17:22:57.780982 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 794.998 avg 419.104 mFlops-per-file 236.028 GF/s 799.657
I0826 17:22:57.809826 364794 net.cpp:655] Test time of data 3.831 ms ( 28.2148 % )
I0826 17:22:57.809866 364794 net.cpp:655] Test time of conv1 9.747 ms ( 71.7852 % )
I0826 17:22:57.809876 364794 caffe.cpp:330] Total forwarding time: 13.578 ms
I0826 17:22:57.823069 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00952005 padding-time 0
I0826 17:22:57.823122 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 748.847 avg 415.124 mFlops-per-file 236.028 GF/s 848.939
I0826 17:22:57.851263 364794 net.cpp:655] Test time of data 3.645 ms ( 27.5656 % )
I0826 17:22:57.851299 364794 net.cpp:655] Test time of conv1 9.578 ms ( 72.4344 % )
I0826 17:22:57.851308 364794 caffe.cpp:330] Total forwarding time: 13.223 ms
I0826 17:22:57.864498 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00947189 padding-time 0
I0826 17:22:57.864550 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 744.37 avg 415.602 mFlops-per-file 236.028 GF/s 854.046
I0826 17:22:57.892693 364794 net.cpp:655] Test time of data 3.683 ms ( 27.8741 % )
I0826 17:22:57.892733 364794 net.cpp:655] Test time of conv1 9.53 ms ( 72.1259 % )
I0826 17:22:57.892743 364794 caffe.cpp:330] Total forwarding time: 13.213 ms
I0826 17:22:57.905951 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00951886 padding-time 0
I0826 17:22:57.906002 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 752.625 avg 416.113 mFlops-per-file 236.028 GF/s 844.678
I0826 17:22:57.934128 364794 net.cpp:655] Test time of data 3.657 ms ( 27.6208 % )
I0826 17:22:57.934168 364794 net.cpp:655] Test time of conv1 9.583 ms ( 72.3792 % )
I0826 17:22:57.934178 364794 caffe.cpp:330] Total forwarding time: 13.24 ms
I0826 17:22:57.947403 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00952411 padding-time 0
I0826 17:22:57.947456 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 751.291 avg 416.451 mFlops-per-file 236.028 GF/s 846.178
I0826 17:22:57.975605 364794 net.cpp:655] Test time of data 3.669 ms ( 27.6843 % )
I0826 17:22:57.975641 364794 net.cpp:655] Test time of conv1 9.584 ms ( 72.3157 % )
I0826 17:22:57.975708 364794 caffe.cpp:330] Total forwarding time: 13.253 ms
I0826 17:22:57.988955 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00949717 padding-time 0
I0826 17:22:57.989020 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 746.678 avg 415.702 mFlops-per-file 236.028 GF/s 851.405
I0826 17:22:58.017169 364794 net.cpp:655] Test time of data 3.7 ms ( 27.8824 % )
I0826 17:22:58.017207 364794 net.cpp:655] Test time of conv1 9.57 ms ( 72.1176 % )
I0826 17:22:58.017217 364794 caffe.cpp:330] Total forwarding time: 13.27 ms
I0826 17:22:58.030375 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00948191 padding-time 0
I0826 17:22:58.030428 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 743.001 avg 414.876 mFlops-per-file 236.028 GF/s 855.619
I0826 17:22:58.058575 364794 net.cpp:655] Test time of data 3.647 ms ( 27.6539 % )
I0826 17:22:58.058614 364794 net.cpp:655] Test time of conv1 9.541 ms ( 72.3461 % )
I0826 17:22:58.058624 364794 caffe.cpp:330] Total forwarding time: 13.188 ms
I0826 17:22:58.071851 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00951099 padding-time 0
I0826 17:22:58.071903 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 751.514 avg 415.387 mFlops-per-file 236.028 GF/s 845.927
I0826 17:22:58.100001 364794 net.cpp:655] Test time of data 3.691 ms ( 27.8356 % )
I0826 17:22:58.100046 364794 net.cpp:655] Test time of conv1 9.569 ms ( 72.1644 % )
I0826 17:22:58.100056 364794 caffe.cpp:330] Total forwarding time: 13.26 ms
I0826 17:22:58.113258 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00947905 padding-time 0
I0826 17:22:58.113312 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 745.325 avg 415.183 mFlops-per-file 236.028 GF/s 852.951
I0826 17:22:58.141428 364794 net.cpp:655] Test time of data 3.695 ms ( 27.9247 % )
I0826 17:22:58.141469 364794 net.cpp:655] Test time of conv1 9.537 ms ( 72.0753 % )
I0826 17:22:58.141479 364794 caffe.cpp:330] Total forwarding time: 13.232 ms
I0826 17:22:58.154659 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00950599 padding-time 0
I0826 17:22:58.154709 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 753.428 avg 415.788 mFlops-per-file 236.028 GF/s 843.778
I0826 17:22:58.182811 364794 net.cpp:655] Test time of data 3.641 ms ( 27.5708 % )
I0826 17:22:58.182850 364794 net.cpp:655] Test time of conv1 9.565 ms ( 72.4292 % )
I0826 17:22:58.182860 364794 caffe.cpp:330] Total forwarding time: 13.206 ms
I0826 17:22:58.196020 364794 conv_layer.cpp:114] conv1 wall clock-time 0.009516 padding-time 0
I0826 17:22:58.196072 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 749.018 avg 415.867 mFlops-per-file 236.028 GF/s 848.746
I0826 17:22:58.224195 364794 net.cpp:655] Test time of data 3.61 ms ( 27.3796 % )
I0826 17:22:58.224231 364794 net.cpp:655] Test time of conv1 9.575 ms ( 72.6204 % )
I0826 17:22:58.224241 364794 caffe.cpp:330] Total forwarding time: 13.185 ms
I0826 17:22:58.237392 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00946498 padding-time 0
I0826 17:22:58.237444 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 745.481 avg 414.632 mFlops-per-file 236.028 GF/s 852.773
I0826 17:22:58.265622 364794 net.cpp:655] Test time of data 3.656 ms ( 27.7411 % )
I0826 17:22:58.265663 364794 net.cpp:655] Test time of conv1 9.523 ms ( 72.2589 % )
I0826 17:22:58.265673 364794 caffe.cpp:330] Total forwarding time: 13.179 ms
I0826 17:22:58.280547 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00970387 padding-time 0
I0826 17:22:58.280601 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 766.803 avg 417.548 mFlops-per-file 236.028 GF/s 829.06
I0826 17:22:58.308818 364794 net.cpp:655] Test time of data 5.108 ms ( 34.3487 % )
I0826 17:22:58.308858 364794 net.cpp:655] Test time of conv1 9.763 ms ( 65.6513 % )
I0826 17:22:58.308868 364794 caffe.cpp:330] Total forwarding time: 14.871 ms
I0826 17:22:58.322549 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00952101 padding-time 0
I0826 17:22:58.322604 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 748.321 avg 415.996 mFlops-per-file 236.028 GF/s 849.536
I0826 17:22:58.350795 364794 net.cpp:655] Test time of data 4.121 ms ( 29.96 % )
I0826 17:22:58.350837 364794 net.cpp:655] Test time of conv1 9.634 ms ( 70.04 % )
I0826 17:22:58.350847 364794 caffe.cpp:330] Total forwarding time: 13.755 ms
I0826 17:22:58.364291 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00951195 padding-time 0
I0826 17:22:58.364346 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 756.438 avg 416.656 mFlops-per-file 236.028 GF/s 840.42
I0826 17:22:58.392495 364794 net.cpp:655] Test time of data 3.901 ms ( 28.9521 % )
I0826 17:22:58.392534 364794 net.cpp:655] Test time of conv1 9.573 ms ( 71.0479 % )
I0826 17:22:58.392545 364794 caffe.cpp:330] Total forwarding time: 13.474 ms
I0826 17:22:58.405853 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00947094 padding-time 0
I0826 17:22:58.405903 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 747.888 avg 416.602 mFlops-per-file 236.028 GF/s 850.028
I0826 17:22:58.434072 364794 net.cpp:655] Test time of data 3.807 ms ( 28.5489 % )
I0826 17:22:58.434113 364794 net.cpp:655] Test time of conv1 9.528 ms ( 71.4511 % )
I0826 17:22:58.434124 364794 caffe.cpp:330] Total forwarding time: 13.335 ms
I0826 17:22:58.447350 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00948405 padding-time 0
I0826 17:22:58.447402 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 756.927 avg 416.053 mFlops-per-file 236.028 GF/s 839.877
I0826 17:22:58.475601 364794 net.cpp:655] Test time of data 3.712 ms ( 28.0066 % )
I0826 17:22:58.475639 364794 net.cpp:655] Test time of conv1 9.542 ms ( 71.9934 % )
I0826 17:22:58.475651 364794 caffe.cpp:330] Total forwarding time: 13.254 ms
I0826 17:22:58.488797 364794 conv_layer.cpp:114] conv1 wall clock-time 0.00947404 padding-time 0
I0826 17:22:58.488852 364794 conv_layer.cpp:124] conv1 K-cycles-per-file max 749.681 avg 415.028 mFlops-per-file 236.028 GF/s 847.995
I0826 17:22:58.516995 364794 net.cpp:655] Test time of data 3.645 ms ( 27.6597 % )
I0826 17:22:58.517038 364794 net.cpp:655] Test time of conv1 9.533 ms ( 72.3403 % )
I0826 17:22:58.517050 364794 caffe.cpp:330] Total forwarding time: 13.178 ms
I0826 17:22:58.517058 364794 caffe.cpp:333] Loss: 0
I0826 17:23:03.935611 364794 caffe.cpp:350] Total-images-processed: 800
I0826 17:23:03.935645 364794 caffe.cpp:353] conv1 K-cycles-per-file 770.74 mFlops-per-file 236.028 GF/s 824.825
freq = 2.69343e+09

SkimCaffe's performance when sparsity is 0.98 (98% of the weights are non-zeros)

I0826 17:21:24.016013 364703 caffe.cpp:295] Use CPU.
I0826 17:21:24.025918 364703 cpu_info.cpp:452] Processor speed [MHz]: 2700
I0826 17:21:24.025944 364703 cpu_info.cpp:455] Total number of sockets: 2
I0826 17:21:24.025951 364703 cpu_info.cpp:458] Total number of CPU cores: 56
I0826 17:21:24.025955 364703 cpu_info.cpp:461] Total number of processors: 112
I0826 17:21:24.025961 364703 cpu_info.cpp:464] GPU is used: no
I0826 17:21:24.025965 364703 cpu_info.cpp:467] OpenMP environmental variables are specified: yes
I0826 17:21:24.025970 364703 cpu_info.cpp:470] OpenMP thread bind allowed: no
I0826 17:21:24.025974 364703 cpu_info.cpp:473] Number of OpenMP threads: 28
I0826 17:21:24.031673 364703 net.cpp:78] Initializing net from parameters:
name: "ResNet-50"
state {
phase: TEST
level: 0
stage: ""
}
layer {
name: "data"
type: "Data"
top: "data"
include {
phase: TEST
}
transform_param {
mirror: false
crop_size: 224
mean_value: 104
mean_value: 117
mean_value: 123
}
data_param {
source: "/nfs_home/yushansu/SkimCaffe/examples/imagenet/ilsvrc12_val_lmdb"
batch_size: 16
backend: LMDB
}
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
convolution_param {
num_output: 64
pad: 3
kernel_size: 7
stride: 2
}
}
I0826 17:21:24.032078 364703 layer_factory.hpp:77] Creating layer data
I0826 17:21:24.032537 364703 net.cpp:120] Creating Layer data
I0826 17:21:24.032557 364703 net.cpp:428] data -> data
I0826 17:21:24.036964 364706 db_lmdb.cpp:37] Opened lmdb /nfs_home/yushansu/SkimCaffe/examples/imagenet/ilsvrc12_val_lmdb
I0826 17:21:24.036985 364706 db_lmdb.cpp:40] Map size is 10038173696
I0826 17:21:24.037729 364703 data_layer.cpp:41] output data size: 16,3,224,224
I0826 17:21:24.056303 364703 net.cpp:170] Setting up data
I0826 17:21:24.056329 364703 net.cpp:177] Top shape: 16 3 224 224 (2408448)
I0826 17:21:24.056367 364703 net.cpp:185] Memory required for data: 9633792
I0826 17:21:24.056383 364703 layer_factory.hpp:77] Creating layer conv1
I0826 17:21:24.056428 364703 net.cpp:120] Creating Layer conv1
I0826 17:21:24.056439 364703 net.cpp:454] conv1 <- data
I0826 17:21:24.056453 364703 net.cpp:428] conv1 -> conv1
I0826 17:21:24.118024 364703 net.cpp:170] Setting up conv1
I0826 17:21:24.118062 364703 net.cpp:177] Top shape: 16 64 112 112 (12845056)
I0826 17:21:24.118078 364703 net.cpp:185] Memory required for data: 61014016
I0826 17:21:24.118151 364703 net.cpp:248] conv1 does not need backward computation.
I0826 17:21:24.118163 364703 net.cpp:248] data does not need backward computation.
I0826 17:21:24.118168 364703 net.cpp:290] This network produces output conv1
I0826 17:21:24.118187 364703 net.cpp:303] Network initialization done.
I0826 17:21:24.123944 364703 base_conv_layer.cpp:66] layer conv1 has sparsity of 0.980017
I0826 17:21:24.124778 364703 base_conv_layer.cpp:368] ConvolutionParameter ConvMode: DEFAULT
I0826 17:21:24.124826 364703 caffe.cpp:301] Running for 50 iterations.
I0826 17:21:24.187851 364703 conv_layer.cpp:114] conv1 wall clock-time 0.0539792 padding-time 0
I0826 17:21:24.187922 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 2215.64 avg 1168.17 mFlops-per-file 236.028 GF/s 286.929
I0826 17:21:24.346465 364703 net.cpp:655] Test time of data 8.981 ms ( 14.2402 % )
I0826 17:21:24.346521 364703 net.cpp:655] Test time of conv1 54.087 ms ( 85.7598 % )
I0826 17:21:24.346534 364703 caffe.cpp:330] Total forwarding time: 63.068 ms
I0826 17:21:24.360201 364703 conv_layer.cpp:114] conv1 wall clock-time 0.0100288 padding-time 0
I0826 17:21:24.360257 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 787.023 avg 437.865 mFlops-per-file 236.028 GF/s 807.767
I0826 17:21:24.386021 364703 net.cpp:655] Test time of data 3.581 ms ( 26.1884 % )
I0826 17:21:24.386059 364703 net.cpp:655] Test time of conv1 10.093 ms ( 73.8116 % )
I0826 17:21:24.386071 364703 caffe.cpp:330] Total forwarding time: 13.674 ms
I0826 17:21:24.399328 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00955606 padding-time 0
I0826 17:21:24.399381 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 753.909 avg 421.358 mFlops-per-file 236.028 GF/s 843.246
I0826 17:21:24.425209 364703 net.cpp:655] Test time of data 3.667 ms ( 27.4846 % )
I0826 17:21:24.425248 364703 net.cpp:655] Test time of conv1 9.675 ms ( 72.5154 % )
I0826 17:21:24.425259 364703 caffe.cpp:330] Total forwarding time: 13.342 ms
I0826 17:21:24.438416 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00950718 padding-time 0
I0826 17:21:24.438468 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 755.653 avg 417.648 mFlops-per-file 236.028 GF/s 841.3
I0826 17:21:24.464257 364703 net.cpp:655] Test time of data 3.618 ms ( 27.4403 % )
I0826 17:21:24.464298 364703 net.cpp:655] Test time of conv1 9.567 ms ( 72.5597 % )
I0826 17:21:24.464308 364703 caffe.cpp:330] Total forwarding time: 13.185 ms
I0826 17:21:24.477512 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00950289 padding-time 0
I0826 17:21:24.477566 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 750.251 avg 418.335 mFlops-per-file 236.028 GF/s 847.358
I0826 17:21:24.503337 364703 net.cpp:655] Test time of data 3.661 ms ( 27.6803 % )
I0826 17:21:24.503377 364703 net.cpp:655] Test time of conv1 9.565 ms ( 72.3197 % )
I0826 17:21:24.503388 364703 caffe.cpp:330] Total forwarding time: 13.226 ms
I0826 17:21:24.516638 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00952411 padding-time 0
I0826 17:21:24.516690 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 763.149 avg 418.463 mFlops-per-file 236.028 GF/s 833.037
I0826 17:21:24.542450 364703 net.cpp:655] Test time of data 3.695 ms ( 27.8217 % )
I0826 17:21:24.542490 364703 net.cpp:655] Test time of conv1 9.586 ms ( 72.1783 % )
I0826 17:21:24.542500 364703 caffe.cpp:330] Total forwarding time: 13.281 ms
I0826 17:21:24.555783 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00948811 padding-time 0
I0826 17:21:24.555836 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 755.145 avg 417.626 mFlops-per-file 236.028 GF/s 841.866
I0826 17:21:24.581579 364703 net.cpp:655] Test time of data 3.764 ms ( 28.2731 % )
I0826 17:21:24.581619 364703 net.cpp:655] Test time of conv1 9.549 ms ( 71.7269 % )
I0826 17:21:24.581629 364703 caffe.cpp:330] Total forwarding time: 13.313 ms
I0826 17:21:24.594795 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00949097 padding-time 0
I0826 17:21:24.594848 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 751.11 avg 416.833 mFlops-per-file 236.028 GF/s 846.389
I0826 17:21:24.621042 364703 net.cpp:655] Test time of data 3.645 ms ( 27.6241 % )
I0826 17:21:24.621084 364703 net.cpp:655] Test time of conv1 9.55 ms ( 72.3759 % )
I0826 17:21:24.621094 364703 caffe.cpp:330] Total forwarding time: 13.195 ms
I0826 17:21:24.634346 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00953102 padding-time 0
I0826 17:21:24.634399 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 767.299 avg 418.09 mFlops-per-file 236.028 GF/s 828.531
I0826 17:21:24.660146 364703 net.cpp:655] Test time of data 3.688 ms ( 27.7711 % )
I0826 17:21:24.660187 364703 net.cpp:655] Test time of conv1 9.592 ms ( 72.2289 % )
I0826 17:21:24.660197 364703 caffe.cpp:330] Total forwarding time: 13.28 ms
I0826 17:21:24.673354 364703 conv_layer.cpp:114] conv1 wall clock-time 0.0095129 padding-time 0
I0826 17:21:24.673408 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 762.574 avg 417.988 mFlops-per-file 236.028 GF/s 833.665
I0826 17:21:24.699149 364703 net.cpp:655] Test time of data 3.615 ms ( 27.4092 % )
I0826 17:21:24.699188 364703 net.cpp:655] Test time of conv1 9.574 ms ( 72.5908 % )
I0826 17:21:24.699200 364703 caffe.cpp:330] Total forwarding time: 13.189 ms
I0826 17:21:24.712431 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00955796 padding-time 0
I0826 17:21:24.712483 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 769.183 avg 419.551 mFlops-per-file 236.028 GF/s 826.502
I0826 17:21:24.738225 364703 net.cpp:655] Test time of data 3.641 ms ( 27.4585 % )
I0826 17:21:24.738262 364703 net.cpp:655] Test time of conv1 9.619 ms ( 72.5415 % )
I0826 17:21:24.738273 364703 caffe.cpp:330] Total forwarding time: 13.26 ms
I0826 17:21:24.751473 364703 conv_layer.cpp:114] conv1 wall clock-time 0.0095129 padding-time 0
I0826 17:21:24.751528 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 757.762 avg 417.623 mFlops-per-file 236.028 GF/s 838.959
I0826 17:21:24.777281 364703 net.cpp:655] Test time of data 3.605 ms ( 27.3562 % )
I0826 17:21:24.777320 364703 net.cpp:655] Test time of conv1 9.573 ms ( 72.6438 % )
I0826 17:21:24.777331 364703 caffe.cpp:330] Total forwarding time: 13.178 ms
I0826 17:21:24.790571 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00952601 padding-time 0
I0826 17:21:24.790621 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 754.903 avg 417.191 mFlops-per-file 236.028 GF/s 842.136
I0826 17:21:24.816355 364703 net.cpp:655] Test time of data 3.681 ms ( 27.7476 % )
I0826 17:21:24.816395 364703 net.cpp:655] Test time of conv1 9.585 ms ( 72.2524 % )
I0826 17:21:24.816406 364703 caffe.cpp:330] Total forwarding time: 13.266 ms
I0826 17:21:24.829635 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00953698 padding-time 0
I0826 17:21:24.829684 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 758.837 avg 419.158 mFlops-per-file 236.028 GF/s 837.77
I0826 17:21:24.855408 364703 net.cpp:655] Test time of data 3.661 ms ( 27.6219 % )
I0826 17:21:24.855444 364703 net.cpp:655] Test time of conv1 9.593 ms ( 72.3781 % )
I0826 17:21:24.855456 364703 caffe.cpp:330] Total forwarding time: 13.254 ms
I0826 17:21:24.868613 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00952196 padding-time 0
I0826 17:21:24.868665 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 759.614 avg 417.941 mFlops-per-file 236.028 GF/s 836.913
I0826 17:21:24.894402 364703 net.cpp:655] Test time of data 3.605 ms ( 27.3396 % )
I0826 17:21:24.894441 364703 net.cpp:655] Test time of conv1 9.581 ms ( 72.6604 % )
I0826 17:21:24.894452 364703 caffe.cpp:330] Total forwarding time: 13.186 ms
I0826 17:21:24.907644 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00948906 padding-time 0
I0826 17:21:24.907694 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 754.783 avg 416.915 mFlops-per-file 236.028 GF/s 842.27
I0826 17:21:24.933410 364703 net.cpp:655] Test time of data 3.673 ms ( 27.7879 % )
I0826 17:21:24.933451 364703 net.cpp:655] Test time of conv1 9.545 ms ( 72.2121 % )
I0826 17:21:24.933461 364703 caffe.cpp:330] Total forwarding time: 13.218 ms
I0826 17:21:24.946650 364703 conv_layer.cpp:114] conv1 wall clock-time 0.0095098 padding-time 0
I0826 17:21:24.946700 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 757.153 avg 417.732 mFlops-per-file 236.028 GF/s 839.633
I0826 17:21:24.972506 364703 net.cpp:655] Test time of data 3.644 ms ( 27.5789 % )
I0826 17:21:24.972549 364703 net.cpp:655] Test time of conv1 9.569 ms ( 72.4211 % )
I0826 17:21:24.972560 364703 caffe.cpp:330] Total forwarding time: 13.213 ms
I0826 17:21:24.985885 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00955892 padding-time 0
I0826 17:21:24.985939 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 765.506 avg 417.976 mFlops-per-file 236.028 GF/s 830.472
I0826 17:21:25.011703 364703 net.cpp:655] Test time of data 3.732 ms ( 27.9467 % )
I0826 17:21:25.011742 364703 net.cpp:655] Test time of conv1 9.622 ms ( 72.0533 % )
I0826 17:21:25.011754 364703 caffe.cpp:330] Total forwarding time: 13.354 ms
I0826 17:21:25.024919 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00949717 padding-time 0
I0826 17:21:25.024972 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 753.098 avg 416.911 mFlops-per-file 236.028 GF/s 844.154
I0826 17:21:25.053516 364703 net.cpp:655] Test time of data 3.639 ms ( 27.5786 % )
I0826 17:21:25.053897 364703 net.cpp:655] Test time of conv1 9.556 ms ( 72.4214 % )
I0826 17:21:25.053939 364703 caffe.cpp:330] Total forwarding time: 13.195 ms
I0826 17:21:25.069224 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00973082 padding-time 0
I0826 17:21:25.069278 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 786.113 avg 428.405 mFlops-per-file 236.028 GF/s 808.703
I0826 17:21:25.094991 364703 net.cpp:655] Test time of data 5.227 ms ( 34.8072 % )
I0826 17:21:25.095077 364703 net.cpp:655] Test time of conv1 9.79 ms ( 65.1928 % )
I0826 17:21:25.095088 364703 caffe.cpp:330] Total forwarding time: 15.017 ms
I0826 17:21:25.108749 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00953794 padding-time 0
I0826 17:21:25.108801 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 769.342 avg 418.624 mFlops-per-file 236.028 GF/s 826.331
I0826 17:21:25.134546 364703 net.cpp:655] Test time of data 4.09 ms ( 29.8867 % )
I0826 17:21:25.134583 364703 net.cpp:655] Test time of conv1 9.595 ms ( 70.1133 % )
I0826 17:21:25.134593 364703 caffe.cpp:330] Total forwarding time: 13.685 ms
I0826 17:21:25.148247 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00951314 padding-time 0
I0826 17:21:25.148299 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 758.723 avg 416.233 mFlops-per-file 236.028 GF/s 837.896
I0826 17:21:25.174031 364703 net.cpp:655] Test time of data 4.109 ms ( 30.0344 % )
I0826 17:21:25.174072 364703 net.cpp:655] Test time of conv1 9.572 ms ( 69.9656 % )
I0826 17:21:25.174082 364703 caffe.cpp:330] Total forwarding time: 13.681 ms
I0826 17:21:25.187445 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00952506 padding-time 0
I0826 17:21:25.187495 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 761.121 avg 417.872 mFlops-per-file 236.028 GF/s 835.257
I0826 17:21:25.213222 364703 net.cpp:655] Test time of data 3.808 ms ( 28.4391 % )
I0826 17:21:25.213259 364703 net.cpp:655] Test time of conv1 9.582 ms ( 71.5609 % )
I0826 17:21:25.213270 364703 caffe.cpp:330] Total forwarding time: 13.39 ms
I0826 17:21:25.226568 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00955701 padding-time 0
I0826 17:21:25.226619 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 767.155 avg 417.951 mFlops-per-file 236.028 GF/s 828.687
I0826 17:21:25.252369 364703 net.cpp:655] Test time of data 3.704 ms ( 27.8057 % )
I0826 17:21:25.252408 364703 net.cpp:655] Test time of conv1 9.617 ms ( 72.1943 % )
I0826 17:21:25.252419 364703 caffe.cpp:330] Total forwarding time: 13.321 ms
I0826 17:21:25.265674 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00956392 padding-time 0
I0826 17:21:25.265722 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 763.967 avg 418.047 mFlops-per-file 236.028 GF/s 832.145
I0826 17:21:25.291438 364703 net.cpp:655] Test time of data 3.662 ms ( 27.5691 % )
I0826 17:21:25.291476 364703 net.cpp:655] Test time of conv1 9.621 ms ( 72.4309 % )
I0826 17:21:25.291487 364703 caffe.cpp:330] Total forwarding time: 13.283 ms
I0826 17:21:25.304739 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00952601 padding-time 0
I0826 17:21:25.304790 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 759.732 avg 418.309 mFlops-per-file 236.028 GF/s 836.783
I0826 17:21:25.330540 364703 net.cpp:655] Test time of data 3.696 ms ( 27.8313 % )
I0826 17:21:25.330577 364703 net.cpp:655] Test time of conv1 9.584 ms ( 72.1687 % )
I0826 17:21:25.330588 364703 caffe.cpp:330] Total forwarding time: 13.28 ms
I0826 17:21:25.343804 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00954795 padding-time 0
I0826 17:21:25.343856 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 761.824 avg 417.534 mFlops-per-file 236.028 GF/s 834.486
I0826 17:21:25.369642 364703 net.cpp:655] Test time of data 3.64 ms ( 27.4779 % )
I0826 17:21:25.369681 364703 net.cpp:655] Test time of conv1 9.607 ms ( 72.5221 % )
I0826 17:21:25.369691 364703 caffe.cpp:330] Total forwarding time: 13.247 ms
I0826 17:21:25.382923 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00952792 padding-time 0
I0826 17:21:25.382973 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 766.422 avg 417.104 mFlops-per-file 236.028 GF/s 829.48
I0826 17:21:25.408733 364703 net.cpp:655] Test time of data 3.672 ms ( 27.6965 % )
I0826 17:21:25.408772 364703 net.cpp:655] Test time of conv1 9.586 ms ( 72.3035 % )
I0826 17:21:25.408783 364703 caffe.cpp:330] Total forwarding time: 13.258 ms
I0826 17:21:25.422034 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00951815 padding-time 0
I0826 17:21:25.422130 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 754.746 avg 417.629 mFlops-per-file 236.028 GF/s 842.312
I0826 17:21:25.447921 364703 net.cpp:655] Test time of data 3.703 ms ( 27.792 % )
I0826 17:21:25.447960 364703 net.cpp:655] Test time of conv1 9.621 ms ( 72.208 % )
I0826 17:21:25.447970 364703 caffe.cpp:330] Total forwarding time: 13.324 ms
I0826 17:21:25.461280 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00956702 padding-time 0
I0826 17:21:25.461331 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 767.794 avg 418.775 mFlops-per-file 236.028 GF/s 827.998
I0826 17:21:25.487082 364703 net.cpp:655] Test time of data 3.713 ms ( 27.8357 % )
I0826 17:21:25.487114 364703 net.cpp:655] Test time of conv1 9.626 ms ( 72.1643 % )
I0826 17:21:25.487125 364703 caffe.cpp:330] Total forwarding time: 13.339 ms
I0826 17:21:25.500321 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00950098 padding-time 0
I0826 17:21:25.500372 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 758.904 avg 417.709 mFlops-per-file 236.028 GF/s 837.696
I0826 17:21:25.526160 364703 net.cpp:655] Test time of data 3.665 ms ( 27.7127 % )
I0826 17:21:25.526199 364703 net.cpp:655] Test time of conv1 9.56 ms ( 72.2873 % )
I0826 17:21:25.526209 364703 caffe.cpp:330] Total forwarding time: 13.225 ms
I0826 17:21:25.539425 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00955105 padding-time 0
I0826 17:21:25.539475 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 774.611 avg 419.359 mFlops-per-file 236.028 GF/s 820.71
I0826 17:21:25.565276 364703 net.cpp:655] Test time of data 3.635 ms ( 27.4505 % )
I0826 17:21:25.565315 364703 net.cpp:655] Test time of conv1 9.607 ms ( 72.5495 % )
I0826 17:21:25.565325 364703 caffe.cpp:330] Total forwarding time: 13.242 ms
I0826 17:21:25.578552 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00952601 padding-time 0
I0826 17:21:25.578601 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 757.121 avg 416.834 mFlops-per-file 236.028 GF/s 839.67
I0826 17:21:25.604383 364703 net.cpp:655] Test time of data 3.658 ms ( 27.6284 % )
I0826 17:21:25.604423 364703 net.cpp:655] Test time of conv1 9.582 ms ( 72.3716 % )
I0826 17:21:25.604432 364703 caffe.cpp:330] Total forwarding time: 13.24 ms
I0826 17:21:25.617971 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00987601 padding-time 0
I0826 17:21:25.618041 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 871.395 avg 425.229 mFlops-per-file 236.028 GF/s 729.556
I0826 17:21:25.644495 364703 net.cpp:655] Test time of data 3.635 ms ( 26.7555 % )
I0826 17:21:25.644534 364703 net.cpp:655] Test time of conv1 9.951 ms ( 73.2445 % )
I0826 17:21:25.644546 364703 caffe.cpp:330] Total forwarding time: 13.586 ms
I0826 17:21:25.658023 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00961494 padding-time 0
I0826 17:21:25.658082 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 777.987 avg 419.123 mFlops-per-file 236.028 GF/s 817.149
I0826 17:21:25.683818 364703 net.cpp:655] Test time of data 3.832 ms ( 28.36 % )
I0826 17:21:25.683858 364703 net.cpp:655] Test time of conv1 9.68 ms ( 71.64 % )
I0826 17:21:25.683868 364703 caffe.cpp:330] Total forwarding time: 13.512 ms
I0826 17:21:25.697114 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00951004 padding-time 0
I0826 17:21:25.697170 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 764.22 avg 416.777 mFlops-per-file 236.028 GF/s 831.87
I0826 17:21:25.722898 364703 net.cpp:655] Test time of data 3.708 ms ( 27.9259 % )
I0826 17:21:25.722936 364703 net.cpp:655] Test time of conv1 9.57 ms ( 72.0741 % )
I0826 17:21:25.722946 364703 caffe.cpp:330] Total forwarding time: 13.278 ms
I0826 17:21:25.736188 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00954604 padding-time 0
I0826 17:21:25.736239 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 759.753 avg 418.272 mFlops-per-file 236.028 GF/s 836.761
I0826 17:21:25.761998 364703 net.cpp:655] Test time of data 3.666 ms ( 27.6241 % )
I0826 17:21:25.762043 364703 net.cpp:655] Test time of conv1 9.605 ms ( 72.3759 % )
I0826 17:21:25.762109 364703 caffe.cpp:330] Total forwarding time: 13.271 ms
I0826 17:21:25.775310 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00953889 padding-time 0
I0826 17:21:25.775362 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 765.586 avg 418.959 mFlops-per-file 236.028 GF/s 830.385
I0826 17:21:25.801112 364703 net.cpp:655] Test time of data 3.619 ms ( 27.3814 % )
I0826 17:21:25.801151 364703 net.cpp:655] Test time of conv1 9.598 ms ( 72.6186 % )
I0826 17:21:25.801162 364703 caffe.cpp:330] Total forwarding time: 13.217 ms
I0826 17:21:25.814337 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00951695 padding-time 0
I0826 17:21:25.814389 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 765.408 avg 417.243 mFlops-per-file 236.028 GF/s 830.578
I0826 17:21:25.840119 364703 net.cpp:655] Test time of data 3.63 ms ( 27.4896 % )
I0826 17:21:25.840158 364703 net.cpp:655] Test time of conv1 9.575 ms ( 72.5104 % )
I0826 17:21:25.840168 364703 caffe.cpp:330] Total forwarding time: 13.205 ms
I0826 17:21:25.853355 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00954795 padding-time 0
I0826 17:21:25.853406 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 764.605 avg 418.635 mFlops-per-file 236.028 GF/s 831.451
I0826 17:21:25.879135 364703 net.cpp:655] Test time of data 3.611 ms ( 27.3209 % )
I0826 17:21:25.879174 364703 net.cpp:655] Test time of conv1 9.606 ms ( 72.6791 % )
I0826 17:21:25.879184 364703 caffe.cpp:330] Total forwarding time: 13.217 ms
I0826 17:21:25.892402 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00953102 padding-time 0
I0826 17:21:25.892454 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 766.902 avg 418.409 mFlops-per-file 236.028 GF/s 828.96
I0826 17:21:25.918195 364703 net.cpp:655] Test time of data 3.658 ms ( 27.6159 % )
I0826 17:21:25.918234 364703 net.cpp:655] Test time of conv1 9.588 ms ( 72.3841 % )
I0826 17:21:25.918246 364703 caffe.cpp:330] Total forwarding time: 13.246 ms
I0826 17:21:25.931411 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00952697 padding-time 0
I0826 17:21:25.931463 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 763.955 avg 417.207 mFlops-per-file 236.028 GF/s 832.158
I0826 17:21:25.957213 364703 net.cpp:655] Test time of data 3.608 ms ( 27.3458 % )
I0826 17:21:25.957252 364703 net.cpp:655] Test time of conv1 9.586 ms ( 72.6542 % )
I0826 17:21:25.957263 364703 caffe.cpp:330] Total forwarding time: 13.194 ms
I0826 17:21:25.970428 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00952005 padding-time 0
I0826 17:21:25.970477 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 763.133 avg 416.389 mFlops-per-file 236.028 GF/s 833.054
I0826 17:21:25.996209 364703 net.cpp:655] Test time of data 3.615 ms ( 27.405 % )
I0826 17:21:25.996248 364703 net.cpp:655] Test time of conv1 9.576 ms ( 72.595 % )
I0826 17:21:25.996259 364703 caffe.cpp:330] Total forwarding time: 13.191 ms
I0826 17:21:26.009418 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00950885 padding-time 0
I0826 17:21:26.009469 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 763.697 avg 418.914 mFlops-per-file 236.028 GF/s 832.439
I0826 17:21:26.035225 364703 net.cpp:655] Test time of data 3.618 ms ( 27.4424 % )
I0826 17:21:26.035264 364703 net.cpp:655] Test time of conv1 9.566 ms ( 72.5576 % )
I0826 17:21:26.035275 364703 caffe.cpp:330] Total forwarding time: 13.184 ms
I0826 17:21:26.048460 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00954199 padding-time 0
I0826 17:21:26.048512 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 773.112 avg 419.84 mFlops-per-file 236.028 GF/s 822.301
I0826 17:21:26.074259 364703 net.cpp:655] Test time of data 3.612 ms ( 27.3367 % )
I0826 17:21:26.074298 364703 net.cpp:655] Test time of conv1 9.601 ms ( 72.6633 % )
I0826 17:21:26.074311 364703 caffe.cpp:330] Total forwarding time: 13.213 ms
I0826 17:21:26.087534 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00954318 padding-time 0
I0826 17:21:26.087584 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 764.52 avg 418.444 mFlops-per-file 236.028 GF/s 831.543
I0826 17:21:26.113420 364703 net.cpp:655] Test time of data 3.648 ms ( 27.4286 % )
I0826 17:21:26.113461 364703 net.cpp:655] Test time of conv1 9.652 ms ( 72.5714 % )
I0826 17:21:26.113471 364703 caffe.cpp:330] Total forwarding time: 13.3 ms
I0826 17:21:26.126662 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00953794 padding-time 0
I0826 17:21:26.126711 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 765.613 avg 419.525 mFlops-per-file 236.028 GF/s 830.356
I0826 17:21:26.152494 364703 net.cpp:655] Test time of data 3.619 ms ( 27.3855 % )
I0826 17:21:26.152532 364703 net.cpp:655] Test time of conv1 9.596 ms ( 72.6145 % )
I0826 17:21:26.152542 364703 caffe.cpp:330] Total forwarding time: 13.215 ms
I0826 17:21:26.165742 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00953102 padding-time 0
I0826 17:21:26.165794 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 759.865 avg 417.329 mFlops-per-file 236.028 GF/s 836.637
I0826 17:21:26.191534 364703 net.cpp:655] Test time of data 3.64 ms ( 27.5153 % )
I0826 17:21:26.191576 364703 net.cpp:655] Test time of conv1 9.589 ms ( 72.4847 % )
I0826 17:21:26.191587 364703 caffe.cpp:330] Total forwarding time: 13.229 ms
I0826 17:21:26.204808 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00954795 padding-time 0
I0826 17:21:26.204859 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 768.451 avg 417.844 mFlops-per-file 236.028 GF/s 827.289
I0826 17:21:26.230626 364703 net.cpp:655] Test time of data 3.644 ms ( 27.5019 % )
I0826 17:21:26.230666 364703 net.cpp:655] Test time of conv1 9.606 ms ( 72.4981 % )
I0826 17:21:26.230676 364703 caffe.cpp:330] Total forwarding time: 13.25 ms
I0826 17:21:26.243921 364703 conv_layer.cpp:114] conv1 wall clock-time 0.00960302 padding-time 0
I0826 17:21:26.243973 364703 conv_layer.cpp:124] conv1 K-cycles-per-file max 778.147 avg 419.387 mFlops-per-file 236.028 GF/s 816.981
I0826 17:21:26.269690 364703 net.cpp:655] Test time of data 3.613 ms ( 27.2186 % )
I0826 17:21:26.269729 364703 net.cpp:655] Test time of conv1 9.661 ms ( 72.7814 % )
I0826 17:21:26.269739 364703 caffe.cpp:330] Total forwarding time: 13.274 ms
I0826 17:21:26.269748 364703 caffe.cpp:333] Loss: 0
I0826 17:21:31.687211 364703 caffe.cpp:350] Total-images-processed: 800
I0826 17:21:31.687244 364703 caffe.cpp:353] conv1 K-cycles-per-file 794.73 mFlops-per-file 236.028 GF/s 799.933
freq = 2.69346e+09

@jspark1105
Copy link
Contributor

Please set conv_mode to DIRECT_SCONV (example:

)

@jspark1105
Copy link
Contributor

BTW, can you share how you fixed the link error for undefined symbols in protobuf? protobuf 3.9.0 didn't work. Which version worked for you?

@jspark1105
Copy link
Contributor

Also, 50% is not high enough sparsity to get noticeable speedup. I'd first try with a higher sparsity like 90%.

@yushansu
Copy link
Author

yushansu commented Aug 28, 2019

BTW, can you share how you fixed the link error for undefined symbols in protobuf? protobuf 3.9.0 didn't work. Which version worked for you?

Sure. I had the issue of undefined symbols in protobuf when I built the Caffe:
.build_release/lib/libcaffe.so: undefined reference to google::protobuf::internal::WireFormatLite::WriteStringMaybeAliased(int, std::string const&, google::protobuf::io::CodedOutputStream*)'
I was using the protobuf 3.9.0 in miniconda. The issue was fixed by using the protobuf 3.9.0 I built from source (https://github.com/protocolbuffers/protobuf) instead of using the one in miniconda.

@yushansu
Copy link
Author

yushansu commented Aug 28, 2019

Please set conv_mode to DIRECT_SCONV (example:

)

Thank you for the suggestion. I set the conv_mode to DIRECT_SCONV. However the performance drops:

Sparsity 98%: 686 GFLOPs
Sparsity 70%: 110 GFLOPs
Sparsity 50%: 73 GFLOPs

I also get the warning:
W0827 21:55:07.189496 5115 math_functions_intel.cpp:926] Inefficient code path: kernel 7x7 image 224x224 pad 3x3 stride 2x2 dilation 1x1

Do the convolution parameters need to fall into a certain range in order for the direct convolution method to get a speedup?

Thanks,
Yushan

@jspark1105
Copy link
Contributor

Yes. Please see https://github.com/IntelLabs/SkimCaffe/blob/intel_scnn/src/caffe/util/math_functions_intel.cpp for what shapes are optimized. BTW, please first test performance with just a single thread because thread scalability may not be optimized for all settings.

@jspark1105
Copy link
Contributor

BTW lower sparsity means more non-zeros so it's expected to see perf drops with lower sparsity.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants