On uniprocessor darwin machine, major performance hit for GOMAXPROCS > 1

[gopherbot]

by consalus:

(on darwin 386)

I created a simple test program (attached) to compare the cost of
enumerating integers with a channel and a generating goroutine to the old
fashioned loop way.

However, I noticed that the version that uses the channel

(invoke with 8.out -max 1000000 -method 0)

can be an order of magnitude slower with GOMAXPROCS > 1.

It consistently takes 0.200s or so with GOMAXPROCS=1,
but with GOMAXPROCS=2, it is often as much as 5s, but sometimes as low as
0.3s.

This isn't a terrible thing, since I shouldn't be setting GOMAXPROCS > 1 on
a uniprocessor machine, but I wouldn't expect the performance impact to be
so variable or significant.



What is the expected output? What do you see instead?

I'd expect a lesser and more consistent performance hit.

What is your $GOOS?  $GOARCH?

darwin 386

Which revision are you sync'ed to?  (hg log -l 1)

changeset:   3977:a522a4541e0e

Attachments:

1. try.go (742 bytes)

[agl]

Comment 1:

There's a lot that we need to do to the runtime! However, as a tool, the bug tracker 
isn't really suited to issues like this so I'm going to close it.
But please don't take this as a sign that we don't care!

Status changed to WontFix.