I was just browsing yesterday for some sci-fi desktop interfaces [0, 1].
I was just browsing yesterday for some sci-fi desktop interfaces [0, 1].
And I would describe Arcan et. al. as a whole as curiously interesting. The only thing I know of like it is PicoGUI, a 2000-era display manager that could go as low as 1bpp on a ~20MHz 68k running uClinux (specifically this: https://www.flickr.com/photos/micahdowty/albums/721576270325...), while also being able to handle 24bpp and even some bits of OpenGL. It was abandoned in 2003, and the author did a writeup of the worst of the design fails (some were... notable) at http://picogui.org/papers/ghost-of-picogui-past.html.
While I would firmly categorize PGUI as a subset of your work, I thought I'd mention it because I think it tried to reach similar goals to what you've done, and you may potentially find the source code vaguely interesting to scavenge through on a rainy day (although it might take eg a Debian 4 VM to build cleanly without fuss). http://svn.navi.cx/picogui/trunk/ https://sourceforge.net/projects/pgui/files/picogui/ https://sourceforge.net/projects/pgui/files/OldFiles/ (hidden directory I found?)
I have no idea if this will be interesting, I figured I'd mention it just in case.
On-topic - definitely a cool project and one that I haven't heard about actually, added to my queue of weekly source-code readings.
I'd also consider DirectFB similar, and another project that I just have in my vaguest of memories (think they used a project-name like Cairo that got google-CVed out of existence from the more recent one) from the 2001-02ish era. They tried to attack X back then, but fell victims to the driver situation (but they had arbitrarily rotated windows!).
One of my stronger personal influences is BeOS though - Compare https://www.youtube.com/watch?v=BsVydyC8ZGQ to https://www.youtube.com/watch?v=3O40cPUqLbU&feature=youtu.be... :)
I've been meaning to poke around PicoGUI myself - I personally love stuff that's tiny and efficient, always looking out for things like that. (I just found a bunch of old versions of Contiki, the ones with the GUI stack; the non-broken ones were fun to play with: http://hitmen.c02.at/html/tools_contiki.html) Very cool to hear I recommended something relevant! ^^
I've heard of DirectFB, but my understanding is that it just tames framebuffers, as opposed to dealing with everything below the toolkit level.
I don't recall anything called Cairo myself, but as for attacking X I do vaguely recall a company that made a closed-source alternate Linux display stack+desktop environment; it was very rudimentary and went nowhere because of that.
I like BeOS too. I keep meaning to install it (and OS/2... and QNX... and...), just preferably on real hardware. I have a bunch of old stuff hanging around here that I hope to use once I have a little file server and I can free up my dozens of old HDDs :P
And your video (I watched the other one in 2008 :D still remember it) was really awesome, and led to git cloning and compila--"wait it's done already?! Nice."
Now my main request is, please update the documentation on GitHub (particularly the quickstart instructions) so we all aren't stuck with just welcome.lua (which I initially thought was a builtin options screen then double-checked to see if I'd specified the args wrong, lol). I (and probably everyone else) want(s) to play with the stuff you're demoing in your videos!
Another tiny thing I'd mention about the video (it was the first thing I noticed actually) was that it would be a) awesome to watch and b) a great demo of your engine's graphics pipeline performance, if you have the media player update its graph at like 60fps. Or at least 24fps. Just a thought.
For my favorite reference of what fast VU looks like, I recommend rezound (http://rezound.sf.net/) - run it once, then edit ~/.rezound/registry.dat and set meterUpdateTime (in Meters {}) to something like 4 (it's a delay in microseconds I think, 4 is nice and doesn't flood X with updates too fast on my machine). The weird knob thing to the right of the VU (very bottom-right) adjusts the frequency response.
Another possible source of fast VU updates is the Linux port of Open Cubic Player (http://stian.cubic.org/project-ocp.php) - alt+c, set framerate to 60 or 120fps, set font to 4x4 (after loading music :D), and the result there looks nice, too - although it responds best (for me) with a non-fullscreen window. (This one's a bit of a project to learn all the shortcuts for, IMO.)
I had even forgot there were instructions like that, outside the technical bits I'm probably the worst person to write helpful guides as the workflow etc. is just so strongly internalized that most of it 'feels' obvious.
Now don't look at the code for generating the FFT (or anything else in the _decode frameserver for that matter), the reason it doesn't update smoother is just how much I don't get along with "lib"vlc, but I added it more as a novelty (spoiler, FFT precision is murdered and packed into a texture and the rest is a shader, it doesn't even synch well).
I know what you mean by the obviousness thing. I however am sitting here with no idea how even to desktop with Arcan. I wouldn't mind finding out though.
And I was wondering if it was updating the FFT so slowly because of... yeah, something like that. I see. I have to acknowledge and agree that it does indeed not sync well. Fixing this sounds like a large pile of boringness; I can see that the other display components update pretty quickly, at least. (But now I'm wondering, is the video of a VNC server? I thought it was SDL. How are you updating the screen so quickly?!)
I think the hardest I've pushed the shared memory interface is basic computer vision (filtering, 9-segment display OCR, tracking an x-y plotter and some glowing devices) on 8-bit mono 2x1000fps 320x180-or-so cameras. Even then most time was spent waiting for synchronization bottlenecks because of OpenGL2.1 limitations.
you thinking about https://youtu.be/bQlHnW2qCh0?t=1m28s ? so the round-trip time there is gpu-composite -> readback -> vnc-server -> vnc->client -> back to gpu.
I had a major derp, however: I forgot durdan and arcan aren't, err, the same thing... let's just say I just installed durdan, and successfully played around for a bit. It's a bit slow on my frankensystem (GPU is older than motherboard+CPU... don't ask >.>) but still very cool.
And wow, that's a pretty awesome usage of the SHMIF, cool. I wonder if porting Arcan to Vulkan would produce interesting results...
And I meant https://youtu.be/3O40cPUqLbU?t=279 - in particular the part I seeked to, where you play and record video and blit it onto a 3D surface so on and so forth - the VU is noticeably slower than everything else. I just thought it would be cool if the video was like "look: EVERYTHING is updating at 60fps!" - but it's cool. :P
About the video you linked, that's pretty incredible too... wow.
You might be able to reduce the fillrate cost by trying config/system/simple displaymode.. it removes a lot of features but you save at least a full extra renderpass.
I'm not sure if you need to shout at NVIDIA, Linux or Vulkan to make this possible, but there are so many awesome things people could do if this were possible...
I would totally recommend you shout at all the relevant mailing lists - even NVIDIA's, if it comes to that :P to get this supported.
And how are you managing qemu integration? What do you mean by that? o.o
I tried the simple displaymode, which seems to be a tad faster, but it's still glitchy - the issue specifically is that opening a fullscreen terminal (at 1600x1200) basically makes my mouse a slideshow, and there's noticeable typing lag too. Resizing the terminal down makes it go away: this is proportional to terminal size.
And as an aside, -w 1600 -h 1200 makes the mouse cursor go halfsize, -w <= 1599 and -h <= 1199 makes the cursor normal size. Curious.
Your GPU is most likely fill-rate limited. Are you running this entirely natively (EGL/GBM/KMS stack) or using SDL through X? In the latter case there'll be so many fullscreen sized buffer copies that your GPU cries. There's more special tricks I can do to get the fullscreen case to go faster, and it's on the near todo for Durden anyhow.
Also, the terminal emulator only really supports truetype fonts (there's a built-in fallback that is quick, but it's awful and only 7bit ascii..), which is damn expensive to render.
My GPU is everything-limited :P I just tried arcan on my integrated video, no issues there. Maybe an almost imperceptible bit of slowdown, but just that, almost imperceptible.
I'm using an ancient ATI X1300-series [1002:7183] fanless GPU I yanked from an oldish workstation so I can have 3 screens in a pinch. It runs two displays; my i3's HD 2000 runs the 3rd. (And I can't move Chrome between :0.0 and :0.1, which X won the fight over me having. Yey :3)
As for drivers, I'm using the radeon driver w/ KMS (switching between X and tty is instantaneous); the only issue is that I occasionally see framedrops due to driver bugs, but that's the only problem I have; the driver is very fast. (But it should be, the card's a decade old. :P)
I'm not sure if I'm running truly fullscreen, incidentally - I can see vestiges of i3 (my windowmanager) in the form of a 1px border underneath the bottom of the arcan window.
...So I just tried -f... and you don't support screens with different resolutions. My center display is 1600x1200, my left is 1280x1024. I get fullscreen on the left, and a cute arcan window (that I can't move my mouse into) on my center display. (The left and middle are on the ATI card.) With fullscreen it's still slowish (which I understand is to be expected at this point).
AFAIK I'm using SDL; I tried building with X11 but the build balked, so I conceded to the instructions on github :P
About truetype and terminal emulators, I had an idea a while ago: glyph caching. (((8x15=size)x3=24bpp)x256=ASCII)=92160 bytes. That's remarkably manageable. Unicode blows the 256 out of the water, but how much of Unicode will the average terminal session see? Certainly not all of it, so the cache eviction algorithm won't need to be particularly aggressive or smart.
The only really major catch is that #222222 will not antialias the same as #FFFFFF, and trying to make one look like the other will look either really dim or really pukey, so the caching system would also need to cache each color of each character that it sees. However, this is not actually totally the end of the world, since glyph tables don't actually take up all that much space, as I've just noted. Very mathematically inelegant, but quite possibly worth it in practice.
That's the point I would be stopping at; you may be interested in going full crazy with something like https://news.ycombinator.com/item?id=11440599 (this looks like a lot of fun to play with).
What's the builtin fallback font, out of curiosity?
Also, I want to clarify and emphasize that, running the terminal in arcan on my old ATI card, the size of the terminal window is directly proportionate to the input+video lag. As I resize the terminal smaller (in floating mode) it speeds up, as I resize it bigger it sl-ow-s ddoowwnn and gets stuttery and glitchy.
Lastly, I just discovered an extremely curious phenomenon I thought I'd mention. I wanted to screencap the terminal cursor to get the font size for the calculation above. After learning about "mouse lock > no" (good riddance :D) so I could take the screenshot, I soon found that the input lag I'd experienced was actually affecting my entire X session. Moving the arcan window offscreen seems to alleviate it; moving it back onscreen and eg running htop (inside a fullscreen terminal >:D) bogs everything down so badly that typing into Chrome (running on the i3 GPU, on :0.1) is very very very noticeably slow, and resizing xterm is... well I can sit there for 10-20 seconds watching it repaint. The weird thing: my CPU was near 0% utilization. This is either a bottleneck in my GPU, X or both (or X blocking on GPU bottlenecks).
I'm not highlighting any of this to make arcan look bad; I personally find it really interesting to throw code at suboptimal hardware and seeing what, if anything, can be done to speed it up - because if code runs well on worst-case scenario hardware, it'll fly on average kit. So I thought I'd mention the above in case it's interesting. If you have any ideas for benchmarks I can run or timing information I can collect I'm fine with supplying that. ^^
Clicking the "n minutes ago" thing opens a working reply form for me if I can't see the reply link, I'm not sure if it works in this case.
Either reply method works. :D