Failure on Intel Xeon CPUs about sycl-ml HOT 3 CLOSED

Hello,
So it seems the reduction is failing. There are two things you can try:

• Compile with -DSYCL_PSTL_USE_OLD_ALGO to use a different kind of algorithm in SyclParallelSTL. Please make sure to clean the build directory (in particular the .sycl files) and you can re-run cmake with -DCMAKE_CXX_FLAGS="-DSYCL_PSTL_USE_OLD_ALGO" -DCOMPUTECPP_USER_FLAGS="-DSYCL_PSTL_USE_OLD_ALGO" to add the option for both host and device compiler.
• Check if the reduction sample in ComputeCpp-sdk passes. If not there may be an issue with the local barrier.

As for why only 2 CPUs are used, I am not sure. I am wondering whether each CPU is seen as a different device. If that's the case it would be an issue as Eigen and SYCL-ML are built to use only one device for now. It would help if you can paste here the output of computecpp_info.

from sycl-ml.

reduction in the samples worked fine. I then recompiled SYCL-ML with the suggested flags and it worked fine.
Looks like the code tops out to 2 cores only initially. Once the PCA step/training starts, it utilizes all available cores.

from sycl-ml.

Great, that's good to hear!
I'm closing this issue then, feel free to comment in here or create a new one if you have more issues.

from sycl-ml.