Incubator-mxnet: Intel MKL FATAL ERROR: Cannot load libmkl_avx512_mic.so or libmkl_def.so.

Created on 30 Nov 2017  路  4Comments  路  Source: apache/incubator-mxnet

I run into the problem when I run cifar-10 on Intel KNL.
The description is as following:
image 5
I have excuted it "source xx/xx/mklvars.sh intel64".

Bug Data-loading MKL

Most helpful comment

We also observed similar issue and found a bug in prepare_mkl.sh that would wrongly set flag USE_MKLML to 0 when building mxnet with mklml so libmxnet.so will link with incorrect library, would you try to patch below diff and retry?

diff --git a/prepare_mkl.sh b/prepare_mkl.sh
index 97a1e49..e048f59 100755
--- a/prepare_mkl.sh
+++ b/prepare_mkl.sh
@@ -115,7 +115,7 @@ if [ -z $MKLROOT ]; then
fi

# Check what MKL lib we have in MKLROOT
-if [ -z find $MKLROOT -name libmklml_gnu.so -o -name libmklml.dylib -print -quit ]; then
+if [ -z find $MKLROOT \( -name libmklml_gnu.so -o -name libmklml.dylib \) -print -quit ]; then
USE_MKLML=0
elif [ -z find $MKLROOT -name libmkl_core.so -print -quit ]; then
USE_MKLML=1

We plan to fix this issue and may submit pull request later.

All 4 comments

We also observed similar issue and found a bug in prepare_mkl.sh that would wrongly set flag USE_MKLML to 0 when building mxnet with mklml so libmxnet.so will link with incorrect library, would you try to patch below diff and retry?

diff --git a/prepare_mkl.sh b/prepare_mkl.sh
index 97a1e49..e048f59 100755
--- a/prepare_mkl.sh
+++ b/prepare_mkl.sh
@@ -115,7 +115,7 @@ if [ -z $MKLROOT ]; then
fi

# Check what MKL lib we have in MKLROOT
-if [ -z find $MKLROOT -name libmklml_gnu.so -o -name libmklml.dylib -print -quit ]; then
+if [ -z find $MKLROOT \( -name libmklml_gnu.so -o -name libmklml.dylib \) -print -quit ]; then
USE_MKLML=0
elif [ -z find $MKLROOT -name libmkl_core.so -print -quit ]; then
USE_MKLML=1

We plan to fix this issue and may submit pull request later.

Proposed Labels : "Bug", "MKL", "Data Loading"

Resolved by #9112 and the mklml has been replaced by MKL-DNN.
This should be closed. @sandeep-krishnamurthy

@wuzhijiexia could you try the new code?

This issue has been fixed in #9112. @marcoabreu could you help close this one?

Was this page helpful?
0 / 5 - 0 ratings