DNNL_aarch64 aborts, when TF+DNNL with a simple imagenet/resnet50 benchmark

[kawakami-k]

using TF+DNNL with a simple imagenet/resnet50 benchmark fails , see the following error output:

Train on 32 samples
bad err=17 in Xbyak::Error
terminate called after throwing an instance of 'Xbyak::Error'
  what():  illegal immediate parameter (condition error)
-----------------------------------------
index:   mnemonic location of define
   14:        eor (external/mkl_dnn/src/cpu/xbyak_aarch64/xbyak_aarch64/xbyak_aarch64_mnemonic.h:4726)
   15:      ptrue (external/mkl_dnn/src/cpu/xbyak_aarch64/xbyak_aarch64/xbyak_aarch64_mnemonic.h:15821)
   16:        add (external/mkl_dnn/src/cpu/xbyak_aarch64/xbyak_aarch64/xbyak_aarch64_mnemonic.h:40)
   17:       ld4w (external/mkl_dnn/src/cpu/xbyak_aarch64/xbyak_aarch64/xbyak_aarch64_mnemonic.h:18759)
   18:       ld3w (external/mkl_dnn/src/cpu/xbyak_aarch64/xbyak_aarch64/xbyak_aarch64_mnemonic.h:18751) <---- Error[1]    17441 abort (core dumped)  python3.6 -m benchmarker --mode=training --framework=tensorflow