Fork me on GitHub
#cljs-dev
<
2018-09-01
>
dnolen00:09:34

actually have two more micro-opts, I actually got a basic compile of 8.8 seconds

dnolen00:09:25

maybe didn’t make as much difference I thought, but pushed

dnolen00:09:32

curious to see where cljs-perf ends up

dnolen00:09:47

there’s definitely more stuff to look at, but done for the evening

🍺 8
mfikes00:09:16

I’m running the latest and will post here when done. 🙂

mfikes00:09:02

Speedup of 1.17 for the last two commits, putting things at 1.95 relative to 1.10.339

👍 24
mfikes12:09:31

Interestingly, between 1.10.329 and 1.10.339 we had a 10% perf regression. I’m tempted to add intermediate builds between those two releases until the source of the regression is found. https://github.com/mfikes/cljs-perf

mfikes14:09:21

^ The difference is evidently writing out caches as edn vs. Transit JSON.

mfikes14:09:50

(With edn being faster.)

dnolen15:09:24

Huh but slower for reading I would bet - incremental builds

mfikes15:09:51

I’m digging deeper into it. Setting :cache-analysis-format :edn doesn’t speed master up, so I’m testing whether simply having Transit on the classpath is causing our dynaload code to somehow cause it.

dnolen17:09:33

I think have two more promising tweaks - not calling load-core in analyze-form and unrolling emits & emitln

john17:09:57

This work for you? clojure -Sdeps '{:deps {johnmn3/coal-mine2vec {:git/url "" :sha "46af3670c16b9a3b827c1e144b76c116dc45a4ce"}}}' "-m" "coal-mine2vec.is-to" map reduce for

mfikes17:09:51

@john Yes. Wondering what it does 🙂

john17:09:19

It does a word2vec on the coal mine 🙂

👍 4
dnolen17:09:12

pushed those two tweaks