microbenchmarks | JMH projects http :

For code binary compatibility, the "non-tensor-core" members of the Turing family have hardware in the SM that will process tensor core instructions, albeit at a relatively low throughput, compared to a tensor core unit.

Run these commands:

Let me answer my own question:

This tweak of your vec_repl() gives a small speedup for larger examples:

Under the hood, streams are much more complicated than plain arrays. Compilers will get better, but currently, sequential for loops should be faster than stream operations.

This article has some background about stream pipelines, which are used to implement streams. It can help to understand the complexity behind it.

The advantage of streams is that the code can be clearer and it is easier to parallelize it.

Definitely just a merge conflict so dont be panic. Run git status and post the result first.

You are maybe accidentally edit flutter code.

To fix it, go to your flutter folder, it should be /Users/admin/Desktop/SDK/flutter

and check any changes by run

Go to your flutter sdk folder.

Then run the following:

You are measuring empty benchmarks, not empty methods. In other words, measuring the minimal infrastructure code that handles the benchmark itself. This is easy to dissect, because you'd expect only a few instructions on the hot path. JMH's -prof perfasm or -prof xperfasm would give you those hottest instructions in seconds.

I think the effect is due to Thread-Local Handshakes (JEP 312), see:

8u191: 0.389 ± 0.029 ns/op [so far so good]