Building for KNL, get not declared in scope errors about caffe HOT 6 CLOSED

I found this post:
http://mingw-w64-public.narkive.com/0yRqe2R8/problem-with-cpuid-when-compiling-xerces-c-3-1-0
I wonder if you have a similar issue? Are you including <intrin.h> prior to the <cpuid.h> included in the file "xbyak/xbyak_util.h" ?

from caffe.

Thanks very much for your report, we are working on reproducing your problem and will reply to you as soon as we find a solution

from caffe.

Intel Distribution of Caffe switched to Intel(R) Machine Learning Scaling Library (MLSL) in order to provide multi-node training functionality. Pure MPI multi-node solution is experimental code only and we have no plans to continue development on it. This means future releases of Intel Distribution of Caffe may brake or even remove pure Intel MPI multi-node. Please update to latest master branch of Intel Caffe and follow this instruction which shall explain you how to prepare to and how to build the latest Intel Caffe with MLSL. Plase note this is MLSL Beta version, full Gold quality is expected end of Q1'2017. Let us know if you need more information or have issues building/funning latest Intel Caffe MN.

from caffe.

@apwojcik Hello, I try to run distributed resnet50 with mlsl, but failied due to MAX_COMPUTE_OP defined in mlsl is 400 and too small to run resnet50.

(106827): /localdisk/jenkins/mlsl-build/src/mlsl_impl.h:ComputeOpImpl:568: ASSERT 'op_count < MAX_COMPUTE_OP' FAILED: too many operations (current count 400, limit 400)

This should be mlsl library issue. We hope to run distributed resnet 50 and even resnet101 with intel caffe.
Thanks for paying attention to this issue.

from caffe.

Any updates on the MAX_COMPUTE_OP issue? I also encounter it with distributed training of googlenet_v2.

from caffe.

The MAX_COMPUTE_OP will be fixed in MLSL 2017 Gold release planed end of Q1'17.

from caffe.