finished first pass on i3-8100B #3
Conversation
|
M1 numbers on MBP M1 runs twice as fast as old i3 chip. Mainly because of larger cache? |
| for acc := range accChan { | ||
| accSum += acc | ||
| count += 1 | ||
| if count == 4 { |
There was a problem hiding this comment.
https://gobyexample.com/range-over-channels fwiw I was doing this break too but then I saw you can close a channel and ranging over a closed channel stops once it's empty
There was a problem hiding this comment.
Oh I see. I knew you could could stop when closed, but I didn't think about closing the channel being currently looped over. Thanks!
Before I had a separate go routine wait and close the channel, but I think I set it up wrong and it ended up with some race conditions.
| accChan <- (acc0 + acc1) + (acc2 + acc3) | ||
| } | ||
|
|
||
| func procAccPayStdDev (i, stop int, payments PaymentsSlice, accChan chan<- float64) { |
|
update std dev to calc in one pass: Andrews-Mini:metrics andrewzheng$ go test -bench=. |
|
new M1 benchmark: goos: darwin |
Andrews-Mini:metrics andrewzheng$ go test -bench=.
goos: darwin
goarch: amd64
pkg: bradfieldcs/csi/systems/mem-hier-2
cpu: Intel(R) Core(TM) i3-8100B CPU @ 3.60GHz
BenchmarkMetrics/Average_age-4 43518 28287 ns/op
BenchmarkMetrics/Average_payment-4 5706 200055 ns/op
BenchmarkMetrics/Payment_stddev-4 2706 423003 ns/op
PASS
ok bradfieldcs/csi/systems/mem-hier-2 5.895s
Considering further unrolling of standard deviation to take advantage of long latency in float division.