Tom Forsyth

Graphics coder and HW architect, previously at Muckyfoot, RAD Game Tools, Valve, Oculus, Intel, Rec Room and Riot.

Public Key

npub1t8zuvyvpadmxuvul6ndgt9z3pnywuuawjel7g2v04lypcuvxr82qjagw60

NIP-05 Address

Profile Code

nprofile1qqs9n3wxzxq7kanwxw0afk59j3gsej8wwwhfvlly9x86ljquwxrpn4qpz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0dsx94flh

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub1t8zuvyvpadmxuvul6ndgt9z3pnywuuawjel7g2v04lypcuvxr82qjagw60

Show more details

Last Notes

2026-08-09 19:52:41 GMT

- reply

@nprofile…faww You have to multiply it by distance first.

2026-08-04 05:44:45 GMT

People forget that procrastination is a superpower. I can pack for a 10-day trip in under 5 minutes.
Of course there is also a cost. I am incapable of unpacking from a 10-day trip in anything less than 5 weeks.

2026-07-29 00:51:10 GMT

Me: "So you remember when Half Life 2 came out..."
Coworker: "Er... hang on... 2004... no, I wasn't sentient at the time."
Me: <crumbles to dust>

2026-07-10 08:48:23 GMT

- reply

So maybe MAYBE something good will come of all this AI bollocks. Just like something good came of all that BASIC bollocks. Which is the love of coding. Real coding. Not using BASIC - that was just the seed. Flattening the learning curve. Getting the love and the need, if not the details, into kids.

2026-06-22 20:21:18 GMT

Very pleasing that "height", "normal" and "albedo" all have the same number of letters. Come on "height" and "width", get your shit together!

2026-06-21 20:07:28 GMT

- reply

@nprofile…sm4q You do also want a reasonably "hugging" mesh, so the raytracer starts pretty close the actual surface. This is why rasterisers are still so much much faster than generic raytracers (and probably always will be - it's not compute, it's memory bandwidth). I'm hoping this combines the strengths of both.

2026-06-21 18:45:53 GMT

- reply

@nprofile…sm4q I'm not too worried about that part. I'm not trying to reduce triangle counts from where they are today (well, maybe 4x or something). I zoom way close in here just to show it's all working, but in a real game it would be just finer surface detail. The meshes would still be 10ktris for a person, etc.

2026-06-21 06:12:38 GMT

- reply

Here's some bad lighting. My lighting model is terrible - please ignore - but there's some indirect light that is never shadowed, and a direct light that is shadowed by shooting a second ray from where it hits the surface, off towards a sun that is fixed in world space.
https://cdn.masto.host/mastodongamedevplace/media_attachments/files/116/786/616/880/733/446/original/2716aa72c33be40d.mp4

2026-06-20 04:43:19 GMT

- reply

Big update! Changed the geometry to a tube with rounded ends, and some bones animating it. Same ugly heightfield, which also doesn't wrap properly (at the purple parts) - but it copes just fine. You can see what's going on though.
https://cdn.masto.host/mastodongamedevplace/media_attachments/files/116/780/603/237/066/986/original/023e8a1d023f8d35.mp4

2026-06-12 05:27:22 GMT

- reply

@nprofile…duaf Yeah I assume it's a legacy of needing Power Of Two sizes for so long.
I always wanted an extension where every even line of texels was shifted right half a texel coordinate. So you could have hexagonal/triangle based textures. The N64 almost had that, but not quite.

2026-06-10 15:24:50 GMT

- reply

The raytracer isn't the clever part - the one I use is dumb as rocks and not very efficient - it's just placeholder. There's tons of great Shadertoy/demoscene raytracers you could drop in that are way better.
The part I've been working on is the continuously-curved distortion. That's the key to taking all the brilliant volumetric/SDF-style rendering tech we have in demos, and getting it into the traditional boned-mesh art pipeline.

2026-06-10 15:21:19 GMT

- reply

I haven't done normal-map lighting yet, but that would all work in the standard way. But because this is fundamentally a raytrace through a volume, you could trace a ray back out towards the sun and get self-shadowing. Also, the surface could be fuzzy, or slightly translucent, or procedurally hairy.

2026-06-10 15:20:59 GMT

- reply

The mesh has the standard normal/tangent/bitangent at each vertex, those are skinned in the usual way, and those decide how the heightfield is distorted.
The nice thing is this is all just the same authoring pipeline as we have today. But as well as a normal map, you get a heightfield/SDF.

2026-06-10 15:12:21 GMT

- reply

The heightfield is a fixed texture I drew in MSPaint - I'm not animating that. You could use any sort of SDF or other volumetric structure you want. The trick is to set up a curved (and animated) mapping between linear object/world space, and the space the heightfield/SDF was defined in.
More details at some point - lots more work to do (seams!) and got to write it all up and so on. But very exciting, had to post :-)

2026-06-09 06:21:26 GMT

Now and then, I find my wife's socks in my washing. Never the other way round. Been happening for years. Very odd.
My wife puts her (clean) socks on top of a dresser. Underneath the dresser is my pile of laundry. And today rather a lot of the socks had... jumped. Or maybe pushed by a furry someone?

2026-06-09 05:36:13 GMT

Recalling my joy at discovering that Germans say the word "yacht" EXACTLY as you would expect a German to say the word.
The respective Germans did not understand my tears of laughing joy, nor my absolute refusal to tell them how it is pronounced in English. I was too embarrassed for my language.

2026-06-05 07:06:13 GMT

- reply

@nprofile…aqfp It's what - 4 MIPS or so? That's twice as fast as an Amiga, with a quarter of the pixels. Luxury!

2026-06-04 21:53:45 GMT

Deciding on Pico-8 vs Tic-80 for teaching tweenagers to code. They're very similar systems - either is fine.
I think Pico-8's 128-pixel-wide screen is too small for coding on. Tic-80s is 240 which is nearly twice as wide. Yes, you can use an external text editor for both, but I think that's a distraction for people starting out.
Tic-80 has slightly better graphics - two layers, more colours. But I don't think that matters for the kids learning.
Otherwise very similar - anything I'm missing?

2026-05-28 05:32:20 GMT

It is January 20th, 2028, and Mitch McConnell is in his office getting ready for the second inauguration of President Harris. Celebrating her second landslide victory over a demoralized Mike Pence, she is poised to pass the 28th Amendment, adding six new states to the Union.

2026-05-27 22:03:05 GMT

- reply

@nprofile…c3sc Right but "d11 + b11" doesn't need a tangent-frame sign - it's already in there, depending on the relative angles of d11 and b11.

2026-05-25 08:29:46 GMT

- reply

@nprofile…3zxe
Int64 positions is for consistency across the world. It also happens to be quite a lot of range, but that's almost incidental:
https://tomforsyth1000.github.io/blog.wiki.html#%5B%5BA%20matter%20of%20precision%5D%5D
NTP is just a time protocol, fixed point 32.32 with the binary point at seconds. Again, consistent precision is the goal:
https://en.wikipedia.org/wiki/Network_Time_Protocol#Timestamps

2026-05-25 07:59:36 GMT

After much yak-shaving, I finally have a little template app with:
* D3D11 setup, shaders, rendering
* Dear ImGui integration
* An int64 time class based on NTP
* An int64 world position class
* A vectors-are-columns, row-major vector library.
* Inf far clip & reverse-Z projection matrices
* Flyable WASD camera
* Maya-style orbit camera
* Framerate-independent game sim loop with interpolation.
This is awesome for me. The question is whether I go the extra mile to put it up on GitHub etc.

2026-05-25 07:40:27 GMT

- reply

@nprofile…tkfs Funny thing is that optimising for a 2005-era vertex cache is still a very useful thing to do, even though (a) they never quite existed the way anybody modeled them and (b) they absolutely don't exist in that form any more.
But because the current vertex processing pipelines are optimised for existing content, which optimised a "thing that looks like a vertex cache", you still want to do that optimisation, even though that thing doesn't exist any more. Which is very odd.

2026-05-24 03:32:18 GMT

- reply

I think it is FASCINATING to have one of these services where they aggressively curate for "interesting" rather than eyeball time.
Oh yeah they have a kid's section too. Careful! Your kid might get edumacated.

2026-05-24 03:31:58 GMT

- reply

Also, it's fancy. It's smart. It's... erudite. I scrolled through their films, hadn't heard of half of them, but the ones I knew are fantastic.
If you pick a RANDOM film on this, I am 90% sure it will be amazing. It might not have what you were looking for, but it has what you need.

2026-05-24 03:31:32 GMT

Huh. My wife just showed me Kanopy. It's yet another video streaming system, BUT WAIT! It is free if you have a library card. Free as in coffee. Like actually free. There's a ticket system so you can't watch too many a month, but... free.

2026-05-20 20:07:26 GMT

- reply

@nprofile…d6ud DSPs: "of course! We have a bit-reverse built into every addressing mode"
CPUs: "why on earth would you want to do that?"

2026-05-20 20:05:12 GMT

- reply

@nprofile…d6ud It's frustrating how few instruction sets have an instruction to reverse the order of bits in a register. It's really simple to implement in HW, but if nobody thinks to ask for it, it's a huge pain to emulate.

2026-05-19 19:52:58 GMT

- reply

@nprofile…ahvu Good to know, thanks. I am also enjoying it - it has all the core Subnautica loops absolutely present and correct. There's a few niggles, but I'm sure they'll clean those up.

2026-05-18 05:52:37 GMT

- reply

@nprofile…3q9y You can literally put the skydome at infinity if you want. Just needs a little vertex shader magic.

2026-05-14 04:35:35 GMT

- reply

@nprofile…c9pu Compresses better. Saves drive space. You can take more pictures! Why so grumpy?

2026-05-14 04:23:20 GMT

- reply

@nprofile…c9pu "Future's so bright, I gotta wear shades."

2026-05-13 08:09:58 GMT

Does this seem factually accurate to folks? I mean aside from the totally arbitrary scores, obviously. Don't let this affect your decision in my poll or anything...
https://cdn.masto.host/mastodongamedevplace/media_attachments/files/116/566/247/405/500/630/original/573e0c154841d4ed.png

2026-05-09 19:17:13 GMT

- reply

@nprofile…26t9 I accept the argument that you always have a lot of range, and arguably "enough". But there seems to be no reason not to always use reverse-Z, and use it better.

2026-05-09 19:15:35 GMT

- reply

Related - thinking about float32 Z buffers, and most implementations are only using the [0,1] range of float32, so we're wasting all that lovely [1,inf] space that float32 can also express!
But then I realised - those two ranges have (approx) the same number of representable values in the float32 encoding.
We're also not using negative numbers, but those seem difficult to map.
So OK, we're only using a quarter of the float32 space - that's still ~30 bits. It's fine. It's fiiiiiiiine.

2026-05-09 18:13:32 GMT

- reply

@nprofile…26t9 With regular integer Z buffers which store (1/Z) as an integer, the problem is that the precision is all bunched up by the near clip plane.
Switching to floating point solves nothing - I get even more precision at the near plane, where I already had too much precision
But if you store "inverse Z" i.e. 1-(1/Z), so that 0.0 is at the far clip plane and 1.0 is at the near, then the two cancel out. My extra precision from using float32 is out where I want it - in the distance.

2026-05-09 18:01:18 GMT

- reply

@nprofile…3zxe I was being figurative - talking about the big round thing buried under Lake Geneva. Just smash that code together, pick up the pieces later.

2026-05-09 18:00:12 GMT

- reply

@nprofile…26t9 @nprofile…epna Wait wait - "inverse Z" in this case refers to using "one minus Z", not taking the reciprocal. Yes, of course you always interpolate 1/Z - that's how perspective correct interpolation works!

2026-05-09 07:55:25 GMT

- reply

@nprofile…epna Z interpolation, test, and ideally write should usually be before the pixel shader ever gets invoked (except in specific circumstances like shader-Z etc).

2026-05-09 07:37:23 GMT

- reply

@nprofile…epna (for the gods sakes, if someone KNOWS this is wrong, tell me!)

2026-05-09 07:35:58 GMT

- reply

@nprofile…epna Yeah, it's frustrating that people overcomplicate things. It's actually simple:
* Reversed Z doesn't do anything with int24 depth buffers (why would it? They're integers...)
* Reversed Z is mandatory for float32 buffers.
* You can use infinite far planes with either.
* OpenGL is wrong, fix it.

2026-05-09 07:26:55 GMT

- reply

@nprofile…epna Precision problems? In worldspace positions? If only someone had written a blog post about that... :-)

2026-05-09 06:05:41 GMT

- reply

@nprofile…0edt Right. I'm wondering if there's like "on OpenGL 6.66 it doesn't work because nasal demons" or something.

2026-05-09 05:57:54 GMT

- reply

@nprofile…c5ug Yeah that would be cool. I've sometimes thought about trying to structure a series of commits like this - not so they can actually be done separately (code probably won't compile mid-way through), just to tell the story.

2026-05-09 05:55:21 GMT

Is there any reason NOT to use a (nearly) infinite clip plane in D3D? You can just pass something big like 2^80 as the far clip plane into the standard projection generator and it spits out basically the right thing.
It's odd that more people don't know about "just set the far clip to infinity". Just don't know, or is there some real reason to avoid it?

2026-05-09 05:17:34 GMT

- reply

Update! I grabbed:
-Dear ImGui Win32+DX11: https://github.com/ocornut/imgui/wiki/Getting-Started
-https://gargaj.github.io/demos-for-dummies
-https://github.com/jobtalle/ccVector
-My own tiny library of misc stuff.
I put them in the supercollider, and it works great. Well... except the vector library - I don't really like that because it's a "To" lib not a "From" lib. I'd be happy to replace that.
This is a great starting point for my stuff. Vaguely thinking of publishing the combined mess - not sure.

2026-05-07 16:52:49 GMT

- reply

@nprofile…xmk3 I've met all sorts. And I gave up classifying them, so... maybe I have!

2026-05-07 05:45:32 GMT

I am yet again getting horrifically confused by what "column-major" and "row-major" actually mean. Stop it. Just show me what gets multiplied by what to produce what. Then I will know if your library is righteous or abhorrent, and can adjust accordingly.

2026-05-05 07:40:58 GMT

- reply

@nprofile…26t9 That's exactly what I did.

2026-05-04 23:20:57 GMT

- reply

@nprofile…26t9 It's just a compile-time constant, and a perfect power of two at that. What scale would you prefer? Is 233 nanometers not fine enough?

2026-05-04 20:42:29 GMT

- reply

@nprofile…26t9 Ah, I see the confusion. Maybe you believe my entire engine is fixed point? No, that would be mad. This is 2026 not 1996.
My engine needs to know about real units. It uses floats just like usual, and in regular SI units.
This thing here is just a specific representation of worldspace positions - it is fixed point, so that we have uniform precision everywhere in the world. You need to know how to turn one into the other. That's the scale.

2026-05-04 19:40:15 GMT

- reply

@nprofile…26t9 You need to do physics and gameplay and so on. Miles per hour. Humans that are 6 feet tall. Real-world stuff.
This conversation continues to be bizarre. I wrote games on the Spectrum, ST and 8086 machines. You read my blog post. You know I understand how fixed point works.

2026-05-04 19:03:57 GMT

- reply

@nprofile…26t9 Fixed point has a scale. The size of the LSB. You need to pick that. I chose 233nm for the reasons given.
I feel like we're missing something in our communication - this is just normal standard ordinary fixed point.

2026-05-04 18:35:12 GMT

- reply

@nprofile…26t9 I'm confused. You're arguing I don't need a scale, but then arguing that my scale is wrong. You should decide!
Anyway, not sure what your gripe is. 32 bits is not enough, so the next step is 64 bits.
I wanted something that covers a reasonable range with reasonable precision. Plenty of things want to represent planets in a solar system.
I wanted a scale that you could "read" in a debugger. "Upper DWORD is kilometers" is a useful property.

2026-05-04 17:24:32 GMT

- reply

@nprofile…26t9 Do not speak to me of the fixed-point magic - I was there when it was written :-)
But the correct answer is that if you find yourself doing any sort of fixed-point maths that isn't simple addition and subtraction, you've done it wrong. Float32 is perfectly fine for many many thing, just not world positions or world times:
https://tomforsyth1000.github.io/blog.wiki.html#%5B%5BA%20matter%20of%20precision%5D%5D

2026-05-04 05:04:37 GMT

- reply

Incidentally this means the smallest distance is 233 nanometers, which should be sufficient for most gaming purposes?

2026-05-04 04:59:17 GMT

Writing a fixed-point position class, and Neptune is 4.4 billion km away. So a reasonable scale for the upper 32bits would be (1<<32)km. But converting to meters that's (1<<32)*1000, which is not an exact float32. How about I use (1<<32)*1024 instead? So not kilometers, but... kibimeters?

2026-05-03 23:47:04 GMT

- reply

@nprofile…gd2y I remember that game blowing my mind, yes! And as you say, they got the controls right.

2026-05-03 18:49:00 GMT

- reply

@nprofile…9whj Infestation and Cybercon 3 specifically. I still remember seeing curves shapes in Cybercon 3 and being amazed. It's a simple trick, but very effective.

2026-05-03 18:21:07 GMT

Was watching a bunch of ST and Amiga games that we would today call "First Person Shooters". The graphics are simple, but acceptable, and usually around 10fps.
The surprising thing is that almost none of them use the mouse - they're all keyboard or joystick controls. And these aren't DOS ports - they were all first on either ST or Amiga. I guess we just never thought of it?
Even years later, Doom's mouse controls were pretty funky, and lots of people didn't use them.

2026-05-02 15:53:34 GMT

- reply

@nprofile…wgz8 Yeah, I resigned myself to monitor glasses about a year ago. I also have another set specifically for the phone. They're not that different, but enough to give me greater comfort for longer use.
I have a set of varifocals (which you can get at the supermarket now!) for traveling that can do both distances, but I don't like the peripheral distortion for longer-term use.

2026-05-01 08:30:54 GMT

Anyone know a simple source-level Windows+DX11 app with all the boring message pump etc stuff done. Bonus points for a simple 3D scene with flyable camera.
I have a rendering idea I want to try out, but all my standard 3D rendering harnesses are OLD, or already have a chonky game inside them.

2026-04-30 20:22:27 GMT

- reply

@nprofile…y977 Is the Tactical Cello canon?

2026-04-29 05:27:05 GMT

- reply

@nprofile…yk06 Yeah I'm always torn about recommending some sports, because I know my fellow nerds - we always take everything too far. Climbing and lifting have obvious dangers. That's what I like about rowing - it's super low-impact, so old rowers keep right on doing it.
On the other hand, weighting the risks of someone getting too fit and too strong, versus doing nothing - I think it's worth it :-)

2026-04-28 19:16:05 GMT

- reply

Don't go nuts of course - plenty of climbers get tendon problems because they climb and climb and climb. But a little really helps. When I start to feel "tight wrists" it's usually because for one reason or another I haven't had a row in a while. So that's my advice - LHTAPTDA.

2026-04-28 19:15:57 GMT

- reply

My (amateur) theory is those tendons were designed for climbing trees. They're massively strong - when I row, the power of the five largest muscles in my body goes through them, and they easily handle it. But when you type or mouse or play piano, they soften up and "stick". Reset them by LHTAPTDA.

2026-04-28 19:15:48 GMT

- reply

Anything where you put big, static stress on those wrist tendons seems to help. Lift Heavy Things And Put Them Down Again of course. Or I like to row (I used to row on the water, now on a machine). Climbing and bouldering is good too. You want big "hook hands" with slow strong strain.

2026-04-28 19:15:36 GMT

Lots of friends with sore wrists - endemic for coders. I'm not a doctor, but anecdataly lifting heavy things and putting them down again is really good for this. This is very counterintuitive - most recommend rest. But rest is not a long-term solution.

2026-04-28 07:05:02 GMT

- reply

@nprofile…ftfu @nprofile…vzdm @nprofile…valu They're half the size if the array is < 65k entries.

2026-04-28 04:00:53 GMT

- reply

@nprofile…xlrs I have survived many years literally writing raw HTML. Which does work! Better than Wordpress :-)

2026-04-28 01:28:18 GMT

What's your favourite static web page builder? Something that takes some form of markup language and splits out some HTML. Nothing fancy, I just really really need to get my (read-only) web page/blog updated. Currently using https://pandoc.org/ which is great, just wondering if there's other options.

2026-04-27 08:32:52 GMT

- reply

@nprofile…j05z The free memory pools is the one place I still use LLs, yeah. I mean - by definition you can't move them, so...

2026-04-27 08:26:53 GMT

- reply

This popped into my head because I was writing some 68000 code, and I found myself thinking that hey - I could use a linked list here.
And then my brain caught up and said THE HELL YOU WILL, SINNER!
But they actually make sense in that world. Just... not any more.

2026-04-27 08:26:32 GMT

It's funny that algo folks still like to talk about linked lists. But these days we avoid them like the plague because they thrash caches and prefetchers.
In the 16-bit era, we didn't have either of those things, so linked lists made a decent amount of sense.
But it's been 30 years! Re-evaluate!

2026-04-25 22:56:11 GMT

Wait wait WHAT. Look, there are many many ways to deal with coordinates starting at the top-left, but this is crazypants.
https://pico-8.fandom.com/wiki/Sin

2026-04-25 20:02:02 GMT

I've been teaching my nephews programming with code.org and Scratch/Blockly. They're doing great, and at some point I'd like to get them on to TIC80 or PICO8, but I'm a little apprehensive about the sliding blocks->writing text transition. Has anybody done this with kids, and got any tips?

2026-04-25 05:05:30 GMT

- reply

@nprofile…yk06 This was fascinating and... puzzling. Many odd odd choices that you called out. I am sure there must be some good reasons for them - these are smart people - but like you they do leave me with many questions.
Great job reverse-engineering all these!

2026-04-25 01:16:32 GMT

- reply

@nprofile…fa5z Well that's a whole other discussion, on which I and many others are perfectly comfortable with Knuth being wrong.
It is the act of thinking one level down, that ensures that people don't end up with a mad panic scramble to optimise inherently slow code at the end of the project.

2026-04-25 00:23:37 GMT

- reply

@nprofile…fa5z OK, well you're entitled to your opinion. But I find it reasonable, practical, and extremely useful to do these things. Just writing code in a language and blindly assuming "the compiler will figure it all out" is, in my experience, in multiple languages, how you end up with slow ugly code.

2026-04-24 23:30:29 GMT

- reply

@nprofile…fa5z You maybe misunderstood me - I am not (always) talking about assembly language. Check out the rest of the thread.

2026-04-24 19:24:44 GMT

- reply

Writing C++ code? You should be able to mentally translate this into C. Where are the creator/destructors happening, and what do they turn into? Is that function call virtual? How does that work in practice, and what are the perf implications?
If you're writing C, you really should be able to read assembly, and know what instructions are available, how flow control turns into branches, what a cache line is, what happens when you run out of registers.

2026-04-24 19:24:35 GMT

- reply

So if you're writing Python, you do need to understand what will become a dictionary, and roughly how duck-typing works.
If you're writing a JITted language, you should roughly understand how the JITter works.
If you're writing C# or other managed language, be aware of how the GC works, and you should be able to write the equivalent C++ code without too much trouble.

2026-04-24 19:24:25 GMT

- reply

But these two things are related - language choice, and how far down you need to know. Because I do think that whatever language you use, you should know the "next level down". When you're writing code, you should be able to fairly simply translate in your head to the language below it.

2026-04-24 19:24:04 GMT

- reply

I can't find it in the video, but there was a question about how far down the stack you need to know. I mean Fabian and I like to know all the way down to the metal. But that's because we're strange. Most people do not - it's absolutely not necessary to get stuff done.

2026-04-24 19:23:51 GMT

- reply

Fabian talked some about how languages are fundamentally tools, and they're all flawed, and no one language will solve all problems. And I absolutely agree. Some people spend a lot of time creating new languages, and I'm just not sure it's productive.

2026-04-24 17:25:21 GMT

- reply

@nprofile…9wsp @nprofile…hcxx @nprofile…h6ty @nprofile…22w0 The other tricky part is when you want to mix'n'match between the "kernel" model where everything stays strictly in its lane, and other SIMD models that can be friendlier to memory access patterns etc. Such as... big matrix multiplies! Which are very trendy right now.
So picking a method that is flexible and can change execution models is really useful. Which encourages the explicit model, or something close, maybe with some HW assist.

2026-04-24 17:22:12 GMT

- reply

@nprofile…9wsp @nprofile…hcxx @nprofile…h6ty @nprofile…22w0 They can all do simple if/else/endif flow control, the tricky stuff comes when you want to encourage reconvergence (something Nvidia seems to have significant trouble with!) or need to support things like arbitrary gotos or "exception" handling. Lots of fun details. This is why I spent a lot of my second time at Intel looking at this stuff. I came to some conclusions, but not very firm ones. There are many ways to shave this yak!

2026-04-24 17:19:09 GMT

- reply

@nprofile…9wsp @nprofile…hcxx @nprofile…h6ty @nprofile…22w0 The three methods I know:
1. Each lane literally has a IP counter, compared against the real IP.
2. Each lane has a counter that is inc/dec by branches, and the instruction only executes if the counter == 0. There's a stack of IPs behind each.
3. A hardware stack of IPs & execution masks - basically the automatic version of explicit predication.
There's lots of hybrids between the above, and which is better depends on so many things.

2026-04-24 07:56:10 GMT

It is spooky how much @nprofile…22w0 and I agree on so many many things. So in lieu of me making a video of my ramblings, watch his instead.
https://www.youtube.com/watch?v=UBhT7nbWpMg

2026-04-24 06:06:00 GMT

- reply

@nprofile…9wsp @nprofile…hcxx @nprofile…h6ty @nprofile…22w0 Oh right - you folks would have been real frontrunners in this stuff. The rest of us were still maybe doing 2x2 processing at the time.

2026-04-21 21:51:14 GMT

- reply

@nprofile…hcxx @nprofile…h6ty @nprofile…22w0 No. I am saying "all this was known well before 2008"

2026-04-21 21:33:46 GMT

- reply

@nprofile…hcxx @nprofile…h6ty @nprofile…22w0 Yes, I very completely understand the system. I refer the honourable gentleman to my previous answer.

2026-04-21 21:06:37 GMT

- reply

@nprofile…hcxx @nprofile…h6ty @nprofile…22w0 SIMD machines with predication have been around far longer than GPUs, and I personally guarantee all this was known well before 2008, because that's when we published the Siggraph paper - we'd known about it for years before (and I'm pretty sure I didn't invent it!)

2026-04-10 21:28:30 GMT

- reply

@nprofile…w2cy Writing OSes is like writing your own language - it's a trap. You will spend a gigantic amount of time on it, and then... it doesn't run anything interesting, and nobody else wants to use it.
Look - I understand hobbies - I really do. But people put way more work in than they would a simple hobby because they dream that these will be useful to someone else. And then they never are.

2026-04-07 04:04:39 GMT

- reply

@nprofile…valu @nprofile…w2cy The pupil dilation is effect IS linked between eyes - it's a diagnostic test for brain damage.
The pupil response is almost instant, but is only about a 10x difference. It is not "night vision".
Night vision ("scotopic") is a much slower chemical response of the retina. Not only is it not shared between the eyes, it is local to each part of the retina. It takes much longer - typically about ten minutes. It can adapt to a much larger range of vision (about 1000x).

2026-04-06 21:21:54 GMT

- reply

@nprofile…40t9 Or you could render chunks of the background as texture-mapped triangles at the correct time.

2026-04-06 21:13:59 GMT

- reply

@nprofile…w2cy Also, pilots flying over big cities at night. Shirley at least one of them is a pilot?
In the army, if you're firing up a flare at night, you shut one eye while the flare is up. The eye that stays open will lose its night vision. Then when the flare goes out, you open the other eye, and it still has night vision.

2026-04-04 17:23:10 GMT

- reply

@nprofile…ncqk My other tip would be - don't hoard too much early on. I wasted a lot of time building and managing those annoying floating chests to hold stuff for later - but then later I either didn't need it, or it became easier to find.
Once you can built a proper base - THAT'S when you can hoard :-)

2026-04-04 07:45:44 GMT

- reply

@nprofile…j05z Ah yeah was just looking at an old old video of some digging. But they removed it - no digging now.

2026-04-04 07:37:33 GMT

- reply

@nprofile…j05z Erm... I don't remember any dynamic landscape. Are you thinking of a different game?