enable VGG with MKLDNN layers #4310
Conversation
benchmark/paddle/image/run.mkldnn.sh Outdated
| #========== mklml ===========# | ||
| train vgg 64 16 | ||
| train vgg 128 16 | ||
| train vgg 256 16 |
There was a problem hiding this comment.
41行和47行的注释不明白：一个是用mkldnn，一个是没用？
另外，最后一个layer_num的参数，既然传了16进去，前面的也传19进去吧。
There was a problem hiding this comment.
这里比较的是，在WITH_MKLDNN和WITH_MKLML都打开的前提下，使用MKLDNN与不使用MKLDNN的区别。
最后一个参数不是layer的层数，是trainer_count的个数。mkldnn默认用一个，所以没写。
| @@ -0,0 +1,50 @@ | |||
| set -e | |||
There was a problem hiding this comment.
这个文件名从run.mkldnn.sh改成run_mkldnn.sh
| else | ||
| use_mkldnn=0 | ||
| log="logs/${topology}-${thread}mklml-${bs}.log" | ||
| fi |
There was a problem hiding this comment.
14-20行：thread=1的时候，用mkldnn？否则不用？
那看41-50行，都是用了mkldnn。
trainer_count和use_mkldnn都做为一个参数，在41-50行的时候传进去会比较清楚。
There was a problem hiding this comment.
因为当使用MKLDNN的时候只需要一个trainer，MKLDNN会自己使用多线程进行计算，所以才是现在看到的这样。不过我会改一个把接口都留好的，也方便以后大家测试。
| ${CMAKE_CURRENT_BINARY_DIR}/test_CompareMKLDNNandCPU | ||
| --config_file_a=trainer/tests/sample_trainer_config_simple_net.conf --use_mkldnn_a=True | ||
| --config_file_b=trainer/tests/sample_trainer_config_simple_net.conf --use_mkldnn_b=False | ||
| --use_gpu=False |
There was a problem hiding this comment.
这个是因为目前使用对比的时候是不希望使用gpu的。
| | ||
| DEFINE_string(config_file_a, "", "config of one network to compare"); | ||
| DEFINE_string(config_file_b, "", "config of another network to compare"); | ||
| DEFINE_bool(use_mkldnn_a, false, "whether to use mkldnn to run network"); |
There was a problem hiding this comment.
whether to use mkldnn to run config_file_a
whether to use mkldnn to run config_file_b
| @@ -0,0 +1,50 @@ | |||
| set -e | |||
| else | ||
| use_mkldnn=0 | ||
| log="logs/${topology}-${thread}mklml-${bs}.log" | ||
| fi |
There was a problem hiding this comment.
因为当使用MKLDNN的时候只需要一个trainer，MKLDNN会自己使用多线程进行计算，所以才是现在看到的这样。不过我会改一个把接口都留好的，也方便以后大家测试。
benchmark/paddle/image/run.mkldnn.sh Outdated
| #========== mklml ===========# | ||
| train vgg 64 16 | ||
| train vgg 128 16 | ||
| train vgg 256 16 |
There was a problem hiding this comment.
这里比较的是，在WITH_MKLDNN和WITH_MKLML都打开的前提下，使用MKLDNN与不使用MKLDNN的区别。
最后一个参数不是layer的层数，是trainer_count的个数。mkldnn默认用一个，所以没写。
| ${CMAKE_CURRENT_BINARY_DIR}/test_CompareMKLDNNandCPU | ||
| --config_file_a=trainer/tests/sample_trainer_config_simple_net.conf --use_mkldnn_a=True | ||
| --config_file_b=trainer/tests/sample_trainer_config_simple_net.conf --use_mkldnn_b=False | ||
| --use_gpu=False |
There was a problem hiding this comment.
这个是因为目前使用对比的时候是不希望使用gpu的。
| | ||
| DEFINE_string(config_file_a, "", "config of one network to compare"); | ||
| DEFINE_string(config_file_b, "", "config of another network to compare"); | ||
| DEFINE_bool(use_mkldnn_a, false, "whether to use mkldnn to run network"); |
| thread=`nproc` | ||
| log="logs/${topology}-${thread}mklml-${bs}.log" | ||
| else | ||
| echo "Wrong input $3, use True or False." |
| fi | ||
| | ||
| #========= mkldnn =========# | ||
| # vgg |
2 participants
Related #4309