KonduitAI / deeplearning4j

Eclipse Deeplearning4j, ND4J, DataVec and more - deep learning & linear algebra for Java/Scala with GPUs + Spark
http://deeplearning4j.konduit.ai
Apache License 2.0
11 stars 7 forks source link

NEC compiler performance #314

Open raver119 opened 4 years ago

raver119 commented 4 years ago

NCC compiler performance is subpar if compared to regular gcc/clang/msvc compilers.

I.e.: NCC compiler

time ./buildnativeoperations.sh -o aurora -t -j 24 1>out.ok 2>out.err

real    86m6.058s
user    1351m31.968s
sys     11m27.982s

GCC 7.4 compiler:

time ./buildnativeoperations.sh -a avx2 -t -j 24 1>out.ok 2>out.err

real    2m40.058s
user    52m52.287s
sys 1m53.943s
raver119 commented 4 years ago

cc @mikishin ^^^