Hello @nmeisburger, @uyongw & @iitkgpanshu,

The MLSys '21 paper doesn't seem to mention how many cores (and hence, threads) were used on each machine to gather data, but based on the README file in this repo, it seems that the experiments were performed with different number of cores (and hence, threads) for both the machines.

Besides the data reported in the paper, had you also compared performance (without BF16) on Cascade Lake & Cooper Lake by using equal number of cores for both?

I'm curious if you observed any improvement in AVX512 performance (besides BF16 support) in Cooper Lake over Cascade Lake, as Ice Lake SP (like Cooper Lake, it's also Xeon SP 3rd gen, but with 1 or 2 sockets, and 48 KB L1D cache on each core) reportedly has improvements pertaining to frequency (downclocking) when AVX512 instructions are used. Since GCP/AWS/Microsoft Azure don't have Cooper Lake, so it's not possible for me to gauge its performance.

Thank you!

https://github.com/IntelLabs/SLIDE_opt_ia/blob/29e40b45d89d62d50bc4a86df5b804b0594ce514/SLIDE/LSH.cpp#L115

Could you please explain the reason for changing log2() to log() ? Given that each of the K hash functions has log2(binsize) bits, is it not correct to shift the hash functions by those number of bits in order to combine them? Wouldn't log2() give a larger value for the shift length, hence increasing the range of the hash function?

Hi,

My name is André, and I'm trying to run some experiments with AVX512 and Deep Learning. I read your paper and found it's exciting. So I decided to test your optimized version of SLIDE, but I'm facing an error (below) during the compilation of the code.

[ 58%] Building CXX object CMakeFiles/SLIDE_LIB.dir/SLIDE/DataLayerOpt.cpp.o
/opt/intel/parallel_studio_xe_2020/compilers_and_libraries_2020.2.254/linux/bin/intel64/icpc  -DOPT_AVX512=1 -DOPT_AVX512_BF16=1 -DOPT_IA=1 -I${HOMEDIR}/SLIDE/SLIDE_opt_ia/bin/ep/include  -qopenmp   -std=c++14 -O2 -DNDEBUG -std=gnu++14 -o CMakeFiles/SLIDE_LIB.dir/SLIDE/DataLayerOpt.cpp.o -c ${HOMEDIR}/SLIDE/SLIDE_opt_ia/SLIDE/DataLayerOpt.cpp
In file included from ${HOMEDIR}/SLIDE/SLIDE_opt_ia/SLIDE/DataLayerOpt.cpp(5):
${HOMEDIR}/SLIDE/SLIDE_opt_ia/SLIDE/DataLayerOpt.h(27): error: expected a ";"
    DataLayerOpt() numRecords_{0}, numFeatures_ {0}, numLabels_ {0} {};
                   ^

compilation aborted for ${HOMEDIR}/SLIDE/SLIDE_opt_ia/SLIDE/DataLayerOpt.cpp (code 2)
make[2]: ** [CMakeFiles/SLIDE_LIB.dir/SLIDE/DataLayerOpt.cpp.o] Erro 2

The file DataLayerOpt.h is as following:

...
22   // labels
     23   std::vector<int> labelOffsets_;
     24   std::vector<int> labelLengths_;
     25   std::vector<int> labels_;
     26 
     27   DataLayerOpt() numRecords_{0}, numFeatures_ {0}, numLabels_ {0} {};
     28   void loadData(const std::string &srcFile);
     29 
     30   inline int lengthByRecordIndex(size_t n) {
     31     return lengths_[n];
     32   }
...

Any suggestion on how to solve this?

I used Intel PSXE 2020 and 2019, with both resulting in the same error. Below are the commands used to download and compile the code:

git clone https://github.com/RUSH-LAB/SLIDE.git
cd SLIDE
git submodule init
git submodule update

cd SLIDE_opt_ia

module load cmake/3.17.3 
module load intel_psxe/2020 

mkdir -p bin && cd bin 
# SKX/CLX (AVX512)
cmake .. -DCMAKE_CXX_COMPILER=icpc -DCMAKE_C_COMPILER=icc -DOPT_AVX512=1
# CPX (AVX512 + BF16)
cmake .. -DCMAKE_CXX_COMPILER=icpc -DCMAKE_C_COMPILER=icc -DOPT_AVX512=1 -DOPT_AVX512_BF16=1

VERBOSE=1 make

Thank you for your attention and best regards.