What's New in Compiler Explorer? - Matt Godbolt - C++ on Sea 2023

[@Gabriel38196]

What I absolutely love about compiler explorer, is that I tend to use it exactly like I use the python repl. Just fire it and experiment away. The fact that it is a very low commitment (zero setup required, no ide needed, can link code to other people) and that I instantly see the feedback, is exactly what the cpp world needed. Thanks Matt!

[@JeremyCoppin]

this is just the most phenomenal tool. i cannot believe that any professional programmer can work effectively without it.

[@driedurchin]

Well, most programmers have the luxury of targeting incredibly fast computers operated by users who don't care all that much about performance, and simply having -O3 works for them.

[@HaykKarapetyanGP]

I couldn't stop saying 'WOW' while I was watching this. This is mindblowing. 🚀👍💪

[@cpponsea]

Pleased to receive your comment describing how much you valued this presentation. Great to hear!

[@coreC..]

Great talk. I was always wondering about that parsing step. It's great to see it demonstrated in Compiler Explorer. (edit) 6502/6510 old dude too here..

[@cpponsea]

Very pleased to hear that you enjoyed Matt Godbolt's presentation!

[@xdman2956]

The "Add new..." was breakthrough from me. I was lost in the new version of the explorer but now I understand how to use it

[@jmparis83]

Fantastic tool! Thanks Matt (and the team). I love the 6502 "touch". BTW I prefer an Apple II over the BBC 😋

[@MattGodbolt]

Apple II is ace too, just not part of my childhood! I've never touched an Apple II, sadly, so it doesn't have the same feeling for me.

[@GRHmedia]

Seriously good work.

[@yorailevi6747]

amazing amazing amazing

[@DedmenMiller]

15:30 was what i needed last week. But i was trying to run a benchmark and quickbench didnt have it. I ended up spending 20 minutes to set up google benchmark locally just to settle an argument.

[@tolkienfan1972]

Auto vectorization won't work for summing an array of floats, because addtion of floats isn't associative. Understanding your machine and hand optimizing is still important to many fields

[@RepChris]

You can get the compiler to ignore that issue and still auto vectorize float loops. In gcc for instance with " -fassociative-math" or "-ffast-math" for more general optimizations. There probably is some attribute or something you can use to tell it to ignore those things for a specific function too. (Looked it up, there is; just use __attribute__((optimize ("associative-math"))) Complemented with whatever other options you want/need for that function only. Arguably it isnt fully automatic, but the only thing stopping the compiler from violating IEEE 754 is you not telling it that it may. Certainly a lot less involved than actually hand optimizing. You may also want to provide the compiler some information about what SIMD extensions the architecture you are targeting has access to, to allow the compiler to go nuts.

[@TheHackysack]

A lot of your videos all seem to link to the same repository? Edit: Ok, it just clicked why that is. But still, the only slides that seem to be available are from 1 video (which just so happened to be the first one I watched, which is why I was confused)

[@cpponsea]

Thank you for your comment. We shall attempt to rectify this as soon as possible, to ensure that the Speakers' slides are available.

[@nXqd]

god like tool for people who cares about performance :)

[@xarcaz]

Mason... is that Mason Murner of M++Meekly fame?

[@MattGodbolt]

Mmmmmmaybe

[@pw1169]

If std::accumulate and a for loop generate identical assembly code, why would you ever bother using accumulate? lmao

[@spaghettiking653]

Because it's more idiomatic. The for loop is full of implementation details, whereas the accumulate is abstract and has less potential for fuck-ups.

[@driedurchin]

@@spaghettiking653 One downside is it really sucks to debug std::algorithms as opposed to "regular" loops.

[@binary132]

For loops have a lot of potential footguns. Sometimes they’re still the right thing.