Rainer Eschrich
Rainer Eschrich
Hi, I profiled the on graalvm 24ea16 with visual vm in JIT mode and I seems that most time (94+%) is used up by: FloatTensor.vectorDot (). Hope this helps.
There must be something wrong with the Graalvm 24 EA at least on the M1 Mac. I tested the benchmarks from this site:https://alexklibisz.com/2023/02/25/accelerating-vector-operations-jvm-jdk-incubator-vector-project-panama#complete-benchmark-results and got MUCH worse results using JEP338...
Any idea how I can build it for NI?