A Foray into Go Assembly Programming | Hacker News Reader

A Foray into Go Assembly Programming

blog.sgmansfield.com

143 pointssgmansfield20 comments

Here's a good talk about the Go assembler backend: https://www.youtube.com/watch?v=KINIAgRpkDA

Spoiler: they're nearly to the point (maybe now they've achieved it fully) of autogenerating a new language backend from ISA documentation PDFs! Bring that up next time someone complains about Go not using LLVM.

Slides: https://talks.golang.org/2016/asm.slide

Does this make targeting new ISA's trivial?

Not really, writing a good enough formal ISA spec is quite hard. More so if the architecture is quirky.

PeachPy, an x86-64 assembler in Python, is often used to target Golang assembler without dealing with its weird syntax. PeachPy uses Intel assembly syntax, but can generate assembly listing for Golang toolchain as its output. Damian Gryski wrote a tutorial about using PeachPy for Golang assembly: https://blog.gopheracademy.com/advent-2016/peachpy/

Disclosure: I'm the author of PeachPy

"I literally copy-paste the middot every time. Who has a middot key?"

I did too, until I found.

http://middot.net/

As that site indicates, in X.org with the defaults, '·' is easily typed with 'Compose . -', 'Compose . ^' or 'Compose ^ .' (this should also work from the console in Debian, at least, which uses xkb bindings throughout).

I love the Compose key, and don't think I'll ever use a desktop without it.

Excellent read about Go Assembly and how to use it. I'd love to see more on this, especially since assembly isn't something that often used, but after reading your post I can appreciate the overhead of having to maintain it in the future.

Thank you for the time put in the post about this.

If writing assembly code to run in the Go environment, and you're changing pointers, you have to write garbage-collector aware code.[1][2] Go has concurrent garbage collection, which means there are some unusual constraints on code that manipulates pointers.

This is an area where proofs of correctness are considered necessary, because intuition fails. See [3]

[1] https://blog.golang.org/go15gc [2] https://golang.org/src/runtime/mbarrier.go [3] https://github.com/golang/proposal/blob/master/design/17503-...

Tried writing a version with no divisions. At least in my tests it was ~5x faster. Correctness not guaranteed. https://gist.github.com/karliss/6cc343f7359c2de9166f1551f884...

FYI, blog layout breaks down completely when trying to zoom text on an iPad, making it impossible to read.

Thanks for letting me know, I'll see if I can fix it. If I could trouble you to do so, can you provide a screenshot?

I'm not the OP, but I have an iPad too, and I can provide screenshots.

http://instantshare.virtivia.com:27080/1eikl92n5jtyc.PNG - default view

http://instantshare.virtivia.com:27080/o1wn6nbthvuu.PNG - after zooming in unsuccessfully, OR just scrolling to the right

http://instantshare.virtivia.com:27080/eiw3z2u1n52h.PNG - fully zoomed out

http://instantshare.virtivia.com:27080/1jsmbx8km815h.PNG - the culprit causing this, a long URL which isn't being wrapped by your CSS rules

To be honest, the same issue happens on desktop browsers, it's just not as noticeable because the default zoom kinda matches what you'd expect. But you can still scroll to the side, in a desktop browser. See:

http://instantshare.virtivia.com:27080/b69zdjyzbvfo.png

I think the CSS fix can be something like this:

    word-wrap: break-word;

http://instantshare.virtivia.com:27080/1l8rbeisu9a2a.png

However, different browsers behave differently™, so you may need to also add:

    word-wrap: break-word;
    white-space: pre-wrap;

To make it work in Firefox (but you might need to adjust other things, also play with word-break CSS property, and test in all browsers). I've dealt with this previously at https://github.com/shurcooL/play/issues/4 and other places that deal with displaying user generated content.

> In this case, a divide by 3 is equivalent to multiplying by 0x55555556 and then taking the top half of the output.

What the fuck. That's amazing!

In that case, fixed point arithmetic might blow your mind. :)

https://en.wikipedia.org/wiki/Fixed-point_arithmetic

In more general terms, take a look at Montgomery multiplication[1] which is where this trick originates. It's a common technique used by compilers to optimise away divisions.

[1] https://en.wikipedia.org/wiki/Montgomery_modular_multiplicat...

Couldn't he have compiled his go code with assembly output and edited the resulting file? Wouldn't that have saved a lot of struggle with the format?

go tool compile -S -S file.go

Sure, but that ruins the fun :)

Wait, so the Go assembler allows generating invalid opcodes?

Any way of using a more normal assembler with a non-GNU Go compiler?

This was probably not explained well. The compilation failed with an error saying that I was basically trying to do something that was not allowed.

For your second question: I don't believe there is a way to use a separate assembler.

Reply on news.ycombinator.com