Andrei Alexandrescu: Writing Fast Code [video] | Hacker News Reader

Andrei Alexandrescu: Writing Fast Code [video]

youtube.com

172 pointsvmorgulis24 comments

Second part of the talk is at: https://www.youtube.com/watch?v=3_FXy3cT5C8

Every time I see things like this a part of my brain wonders, "How would these people describe what they're doing in sprint tasks and give estimates in points."

Thanks for the link, I always enjoy a good talk.

I use Pocket to save text to read later. Sadly, I don't know any similar program/service for video. I would love to have a service where I could send videos from youtube or a file or a torrent, and it would save the oldest one on my phone, so that I can watch it offline on my phone when going on the train.

If someone makes this, I'm def willing to pay you $5/month for it, and I'm sure many others are as well.

Such technology already exists and is embedded in your browser. It's called bookmarks.

He means offline viewing of online videos in a convenient way, like iOS Safari "Reading List" (Saves pages for offline viewing), or Pocket, which does much of the same but for text.

Effectively you want to turn arbitrary link submissions into a podcast feed of their multimedia (probably rehosted). Then you could reuse a normal podcast client to configure when items are automatically downloaded to your phone, do playback, maintain played-status, etc. Podcasts aren't necessarily audio-only - there are video podcasts and video-capable podcast clients out there.

See https://huffduffer.com/ for an (audio) example of the web service part.

Thats a very good idea! RSS is just nothing more than updating some xml. Super easy to make and can host on static website.

i think we only need offline availability because mobile data cost is still high

i really hope they mobile networks start providing unlimited data

its just better for everyone, we wont need to make redundant copies, and as the downvoted comment suggest we will only need a good bookmarking app

I think that we also need it in cases where the data is made unavailable at some future time or when a mobile connection isn't available.

Download it using youtube-dl and copy it to your phone using ssh. Should be trivially automatable.

Been thinking about doing something like that exactly...

You might look at youtube red, which is included with the google music streaming service. Will let you download videos to your mobile device for offline viewing. Combined with your youtube "to watch" list, it works pretty well. Unfortunately doesn't support an "add on desktop -> auto download mobile" workflow.

I think that Instapaper (https://www.instapaper.com/) can handle videos, although I haven't tried it out much myself.

Oh! Didn't know about that. Should sign up and check out the features.

It doesn't download. It just has UI for video; so does Pocket.

Couldn't you just sync a folder with your phone (via dropbox or a thousand other apps) and just save whatever webpage the video is on to that folder. If you do it your way, you'd have to worry about text that's associated with the video that the user may or may not want.

I think the main problem is, for YouTube, it violates ToS so the app usually gets blocked.

Open up the link and look for "Add to". You can then create your own playlist and put any videos in there for later viewing.

I hope this helps.

I was impressed by the trick of writing full cache lines being twice as fast as writing smaller quantities of data.

Still, measure and take the min when you measure are the two main take always from this video.

I was only vaguely aware of this thanks to graphics programming, where write-combined access to GPU memory means you'll be more than twice as fast by writing entire cachelines without holes (because reading write-combined cachelines is effectively uncached? Horribly slow.)

Documentation for 3D rendering APIs tend to have warnings about this as a result, and performance talks tend to cover it as well. "memcpy the (cachline aligned and write combined) site from orbit - it's the only way to be sure"

Good talk, thanks for sharing.

I found the parts about lowering data dependency to achieve additional parallelism within the cores particularly enlightening.

Two unit32_t vs one uint64_t was interesting and new to me. Also unsigned faster than signed. Lots of good stuff here!

"Simplistic not simple". This one's a keeper.

What's wrong with this audience? This dude is hilarious and they don't laugh.

Reply on news.ycombinator.com