Threadwell

Recent Notes

DemetriSpanos
Oct 06, 2026
NOTimothyLottes
Oct 06, 2026
bonzajplc
Oct 06, 2026

nicbarkeragain

168 items with this tag.

Aug 28, 2026
nicbarkeragain
Aug 28, 2026
Similar to string tags, if your program has deferred actions e.g.
Aug 25, 2026
I think one of the biggest level ups I had in programming was shaking off the metaphor associated with data "moving".
Aug 20, 2026
I dev with the debugger attached 100% of the time, and unless I'm doing something visual like UI, I literally always put a breakpoint at the beginning of the new code I've just written and step through it before I ever let it "run".
Aug 05, 2026
An example of this - a classic way to construct a tree in code is with push / pop style calls.
Aug 03, 2026
IMO basically every "object creation API" you have should take a string label that is for internal use only.
May 10, 2026
PSA: if you've never done graphics programming before and are interested in learning the fundamentals, it is very important that you don't start with Vulkan.
Apr 06, 2026
I'm completely convinced at this point that the "Command Palette" is a fundamental UI concept, and should be in all applications.
Dec 18, 2025
When I'm making educational material, one of the techniques I almost always use is not just explaining why the correct idea works, but why the incorrect ideas don't work.
Jul 22, 2025
"Branchless Programming" is a term used to describe a programming style that improves performance by avoiding branches (e.g.
Jul 15, 2025
I can't help but have this feeling that we're missing something in programming language design that will make SIMD much easier for the average programmer to use.
Aug 28, 2026
@nicbarkeragain That's the kind of posts we want to see.
Aug 28, 2026
@nicbarkeragain It's also a good idea to add these to your mem allocs.
Aug 28, 2026
@nicbarkeragain I'm convinced you should never use observer pattern or callbacks.
Aug 26, 2026
@nicbarkeragain Something that CS suffers from is not being clear in what is true in the ideal model vs the actual implementation of a “computer.” For example, in relation to EE, ideal op-amps vs the actual transistor level circuitry of an op-amp allow for different approximations to be made.
Aug 26, 2026
some distinction between allocating new memory, copying src->dst, and then freeing src.
Aug 25, 2026
@nicbarkeragain I think it's actually *easier* to reason about your code when you pre-allocate and just "fill" buffers that never move.
Aug 25, 2026
@nicbarkeragain I still find myself sketching data flow diagrams when designing data processing oriented systems, but I agree that it often doesn't translate well to concrete interfaces in code, it misses any problems you might have with push/pull impedance mismatches, who "holds" memory etc.
Aug 08, 2026
@nicbarkeragain This just saved me from hours of debugging.
Aug 06, 2026
Rust's ownership + RAII makes such patterns really easy and nice to encode statically, no debug strings needed if the compiler guarantees you can't open another object at the same level if the previous one wasn't closed.
Aug 05, 2026
@nicbarkeragain Thanks a lot for the example
Aug 05, 2026
@nicbarkeragain 1.
Aug 05, 2026
@nicbarkeragain Ah, matching strings, just like xml 🙃
Aug 05, 2026
@nicbarkeragain DeferLoop is super useful for this
Aug 05, 2026
@nicbarkeragain this particular case though is easily solved with `defer` or even `closures`
Aug 05, 2026
I love everything about this.
Aug 05, 2026
@nicbarkeragain That's a great tip.
Aug 05, 2026
@nicbarkeragain Or you could use C++ with RAII and not have to tie yourself in knots
Aug 05, 2026
@nicbarkeragain tangently, this example illustates pretty good why I like syntax with `end` keyword and optional name of closee better than just `{ ...
Aug 04, 2026
@nicbarkeragain I accidentally discovered this myself.
Aug 04, 2026
@nicbarkeragain function "draw string, for debug, in my game's 3d graphical space"
Aug 04, 2026
@nicbarkeragain WebGPU has this and it’s a life saver for debugging
Aug 03, 2026
@nicbarkeragain I do and access it via 9p.
Aug 03, 2026
@nicbarkeragain brb adding this to my codebase
Aug 03, 2026
@nicbarkeragain You can also implicitly pass the source code location from creation so there's no need for explicit labeling a lot of the time.
Aug 03, 2026
@nicbarkeragain This is the way!
Aug 03, 2026
@nicbarkeragain I don't suppose you could demonstrate this with a small code snippet/example that shows HOW one can label an allocation?
Aug 03, 2026
@nicbarkeragain Isn't this just a Python-style docstring with extra steps?
Aug 03, 2026
@nicbarkeragain https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/wdm/nf-wdm-exallocatepool2 🤘
Aug 03, 2026
@nicbarkeragain Can you post an example of how you would implement it?
Aug 03, 2026
@nicbarkeragain IIRC sokol_gfx does this.
Aug 03, 2026
@nicbarkeragain When I've seen this they usually require unique names and then you end up recapitulating the entire creation path in the string to make it unique but still readable, which I kinda hate.
Aug 03, 2026
@nicbarkeragain That's actually a smart idea.
Aug 03, 2026
@nicbarkeragain ya tagging allocations, and frames to actual assets in-game is absolutely necessary for development once your game stops being a small thing
May 11, 2026
@nicbarkeragain I've been doing some very intro stuff with SDL lately and it's been a real pleasure, it's pretty impressive what can be accomplished with a handful of functions and a lot of trial and error.
May 11, 2026
@nicbarkeragain I just started cranking out shaders, its helped alot in my understanding tbh
May 11, 2026
@nicbarkeragain @raysan5 raylib mentioned
May 10, 2026
@nicbarkeragain My only computer is an apple silicon macbook.
May 10, 2026
@nicbarkeragain Agreed.
May 10, 2026
@nicbarkeragain No, pretty sure you start with implementing boolean operators on solids yourself.
May 10, 2026
@nicbarkeragain Opengl 3.3 is fine for most learning demoing purposes, and if you are on windows probably d3d11 is the next step.
May 10, 2026
@nicbarkeragain I highly suggest sdl gpu because in addition to the shape being friendly to transition to vulkan from, it also comes with a set of constraints that make you think more creatively about how to work with data.
May 10, 2026
@nicbarkeragain Opengl is literally a waste of time.
May 10, 2026
@nicbarkeragain Planet Earth is far more locked-in to OpenGL than most people realize.
May 10, 2026
@nicbarkeragain What do you think about WebGPU.
May 10, 2026
@nicbarkeragain I used sokol with Odin bindings and it was the best decision I made
May 10, 2026
@nicbarkeragain I started with pset(x,y),color 🙃 and then just "writing pixes to memory" - I agree with the Vulkan sentiment though
May 10, 2026
@nicbarkeragain sokol is amazing, works at exactly the right level of abstraction to learn so much while still getting stuff done
May 10, 2026
@nicbarkeragain why are we recommending opengl in 2026? genuinely asking since it seems so old
May 10, 2026
@nicbarkeragain Yep, totally agree! I did this mistake.
May 10, 2026
@nicbarkeragain I started with https://learnopengl.com/ very good resource.
May 10, 2026
@nicbarkeragain I've been saying start with OpenGL 3.3 (or higher) instead of Vulkan forever.
Apr 07, 2026
@nicbarkeragain I see your “command palette” and raise you an “everything palette”
Apr 07, 2026
@nicbarkeragain that's why visual assist X and resharper find symbol stuff are the most powerful feature of all.
Apr 06, 2026
@nicbarkeragain macOS has this in all applications.
Dec 18, 2025
@nicbarkeragain This is probably why reading Rob Pike and Ken Thompson code back-to-back is so elucidating.
Dec 18, 2025
@nicbarkeragain When possible, I like to "make the mistakes" so we all learn from it together.
Jul 26, 2025
@nicbarkeragain If you're handling integers, just use a full bitmask and use boolean AND.
Jul 23, 2025
@nicbarkeragain People have been whipped for less than this.
Jul 23, 2025
@nicbarkeragain Seems like the problem is that both options are badly written and it's a matter of luck whether the compiler optimises it or not.
Jul 23, 2025
@nicbarkeragain Ternary operator in openmp GPU code is along the same lines.
Jul 23, 2025
@nicbarkeragain I love using arrays as jump tables for operations.
Jul 23, 2025
@nicbarkeragain remind me of rust
Jul 23, 2025
@nicbarkeragain This is what Apple does with the NPU and CoreML, isn’t it? Compiling the entire inference process into a single path, like the final boss of branchless programming?
Jul 23, 2025
The key here is that CPUs mostly only have two options:
Jul 23, 2025
@nicbarkeragain Google TCO .
Jul 23, 2025
@nicbarkeragain I still wonder if you can make compiler to auto-non-branchify code? Or transpiler.
Jul 23, 2025
@nicbarkeragain Close enough, the for loop is a branch, however the code can be compiled so that it has a branch that is likely to be taken.
Jul 23, 2025
@nicbarkeragain İt's pretty hard to tell when removing a branch is actually going to help your performance.
Jul 22, 2025
@nicbarkeragain I sometimes use that principle in SQL so there's only one if statement instead of like 10
Jul 22, 2025
@nicbarkeragain What would be an actual real life use case?
Jul 22, 2025
@nicbarkeragain It’s a beautiful thing but also pointless and unreadable unless in an environment that can’t afford an if statement.
Jul 22, 2025
@nicbarkeragain I thought this kind of masking was standard regardless of branch clobbering.
Jul 22, 2025
@nicbarkeragain Cool.
Jul 22, 2025
@nicbarkeragain It's cool, but unless you unroll the loop fully, it's still branching.
Jul 22, 2025
@nicbarkeragain One additional read instead of using the result of "==".
Jul 22, 2025
@nicbarkeragain of course, it's important to be careful.
Jul 22, 2025
@nicbarkeragain ```[<int>] = <val>```
Jul 22, 2025
@nicbarkeragain That's cool, but also compilers know how to do it as well, sometimes the assembly it generates when it spots a branchless version by itself is better then the one generated by compiling your implementation.
Jul 22, 2025
@nicbarkeragain I don't think you even need a multiplication.
Jul 22, 2025
@nicbarkeragain What colour scheme is that btw?
Jul 22, 2025
@nicbarkeragain If you are doing a for loop you are still branching
Jul 22, 2025
A really long time ago, programming a simple game in basic on a 6502 based Commodore PET, I changed the logic from:
Jul 22, 2025
@nicbarkeragain Why should you do something that your compiler do for you?
Jul 22, 2025
@nicbarkeragain Why can't compilers implement this optimization themselves? At least in this example, it seems like a fairly mechanical transformation
Jul 22, 2025
i'm branch-free by default.
Jul 22, 2025
@nicbarkeragain Wait I am cooked.
Jul 22, 2025
@nicbarkeragain I think it's a cool field, like only using stack memory (e.g.
Jul 22, 2025
@nicbarkeragain It's good inside hot loops and shaders.
Jul 22, 2025
@nicbarkeragain A more compact example is using a bitmask integer.
Jul 22, 2025
@nicbarkeragain It's also a key component in developing good implementations of cryptography to prevent side channel and timing attacks.
Jul 22, 2025
@nicbarkeragain Shouldn't a good optimizer already do this?
Jul 22, 2025
@nicbarkeragain It's funny when some people say "Branches are essentially free these days" Yeah, not so much :)
Jul 22, 2025
@nicbarkeragain Godbolt example of branchless vs branchfull if-statement
Jul 22, 2025
@nicbarkeragain So...
Jul 22, 2025
@nicbarkeragain Looks pretty useless to me in this case.
Jul 22, 2025
@nicbarkeragain Here is a real-life example (excuse my French):
Jul 22, 2025
@nicbarkeragain Clang failed to recognize it should use a select op in the IR and using a ternary helped.
Jul 22, 2025
@nicbarkeragain I don't mind a useful if statement but when I see an else I'm ripping that crap out.
Jul 22, 2025
@nicbarkeragain I have no idea if it improves performance (in the grand scheme of a legacy canvas app) but I use tricks like this when rendering e.g.
Jul 22, 2025
@nicbarkeragain It can be done by tools like Terser or by compiler if they are really equal
Jul 22, 2025
@nicbarkeragain Australia represent!
Jul 22, 2025
@nicbarkeragain You nerd-sniped me to actually try the sth out (the Creel's example) .
Jul 22, 2025
@nicbarkeragain You mean functional programming?
Jul 22, 2025
@nicbarkeragain What is the measured performance impact?
Jul 22, 2025
@nicbarkeragain Is it for performance reasons? I've only seen it in security contexts, e.g.
Jul 22, 2025
@nicbarkeragain 10x better? I thought branch prediction wouldn't have SO much weight, crazy
Jul 22, 2025
@nicbarkeragain Aka Erlang, Prolog etc.
Jul 22, 2025
@nicbarkeragain I'd usually use & or >> but same idea
Jul 22, 2025
@nicbarkeragain It also matters how random the data is.
Jul 22, 2025
@nicbarkeragain Gross just use an if.
Jul 22, 2025
@nicbarkeragain I’d also maintain state machines are a kind of branchless programming.
Jul 22, 2025
@nicbarkeragain Great post! Which Color scheme / font is used in the picture?
Jul 22, 2025
@nicbarkeragain * 1 if x else 0
Jul 22, 2025
@nicbarkeragain this will be slower unless your compiler is really bad.
Jul 22, 2025
@nicbarkeragain Compilers already inline them.
Jul 22, 2025
@nicbarkeragain What kind of applications is this practically useful for? Shaders / GPU stuff? Ultra low latency fintech stuff? Embedded?
Jul 22, 2025
@nicbarkeragain you can find a lot of examples of this method in assembly books.
Jul 16, 2025
@nicbarkeragain What popped into my head is languages like Uiua
Jul 16, 2025
@nicbarkeragain But is it really that hard to use though? Or what scenario do you mean? If we are to make use of SIMD we will be doing operations on collections of numbers.
Jul 16, 2025
@nicbarkeragain Hmmn.
Jul 16, 2025
@nicbarkeragain I’ve been wondering about this for over twenty years, but never found anything I felt was truly accessible.
Jul 16, 2025
@nicbarkeragain Honestly it will definitely be a random idea someone has at 3 am and everyone will just go "how did we not think about that earlier?" and then a revolution begins
Jul 15, 2025
@nicbarkeragain The answer exists.
Jul 15, 2025
@nicbarkeragain Reduced instruction sets.
Jul 15, 2025
@nicbarkeragain ISPC
Jul 15, 2025
@nicbarkeragain Mojo lang has SIMD aware data types and makes it a first class citizen.
Jul 15, 2025
My gamedev head just wants a sort of game mechanic lego for conveying a simd operation frame.
Jul 15, 2025
@nicbarkeragain I have hope that this is true.
Jul 15, 2025
@nicbarkeragain FFmpeg thinks otherwise.
Jul 15, 2025
@nicbarkeragain Have you ever studied APL?
Jul 15, 2025
@nicbarkeragain 128-bit seems to be pretty widespread, so it might be a good baseline? even in armv7 cores, not sure about risc-v and powerpc though.
Jul 15, 2025
@nicbarkeragain C# makes it fairly straightforward.
Jul 15, 2025
@nicbarkeragain Intel's ISPC
Jul 15, 2025
@nicbarkeragain Llm baked into gcc who’s only job is making things simd where possible.
Jul 15, 2025
@nicbarkeragain Perhaps something like ISPC / how we author shaders in GPU could help ?
Jul 15, 2025
@nicbarkeragain Relational model + columnar data.
Jul 15, 2025
@nicbarkeragain languages that have comptime also have wonderful support for SIMD: mojo and zig
Jul 15, 2025
@nicbarkeragain Its called array programming.
Jul 15, 2025
@nicbarkeragain Maybe Futhark, or array programming.
Jul 15, 2025
@nicbarkeragain https://github.com/HigherOrderCO/Bend
Jul 15, 2025
@nicbarkeragain It hasn't been stabilized yet but Rust's portable SIMD was pretty easy to work with https://doc.rust-lang.org/std/simd/index.html
Jul 15, 2025
@nicbarkeragain more arrays
Jul 15, 2025
@nicbarkeragain openmp does good enough job with compiler flags like march=avx2, i think the focus should be making compilers smarter, for the cases where parallelism is the primary concern one should use cuda
Jul 15, 2025
@nicbarkeragain I think ISPC more or less solved this and in a C-compatible way.
Jul 15, 2025
@nicbarkeragain Might be an oversimplification but I think the focus might need to be aimed more at grouping data efficiently.
Jul 15, 2025
@nicbarkeragain So, FORTRAN?
Jul 15, 2025
@nicbarkeragain the industry is trying to do that with vector-length-agnostic architectures, like SVE, RVV, and others.
Jul 15, 2025
@nicbarkeragain The "average programmer" you mentioned uses OOP and other things that make their programs so slower and fatter that SIMD would make literally zero impact within that scale of inefficiency through bad programming practices.
Jul 15, 2025
@nicbarkeragain @Modular Mojo has a first class SIMD support AFAIK but I've never used it https://docs.modular.com/mojo/stdlib/builtin/simd/SIMD#implemented-traits
Jul 15, 2025
@nicbarkeragain I agree, there is something weird with SIMD accessibility today, looks like there is room for innovation
Jul 15, 2025
@nicbarkeragain I think we need a system visualizer where devs can view their implementation running through the processor/gpus/memory/network.
Jul 15, 2025
@nicbarkeragain have you spent time writing SIMD algorithms for general-purpose tasks? it does not seem possible to make this stuff accessible, because the set of available doodads and their costs influences your algorithm design
Jul 15, 2025
@nicbarkeragain have you looked at Odin’s SIMD?
Jul 15, 2025
@nicbarkeragain Whats wrong with using inline ASM?
Jul 15, 2025
@nicbarkeragain Just enable arithmetic operations on fixed sized arrays and optimise this to use SIMD where possible
Jul 15, 2025
@nicbarkeragain Have you met the average programmer?
Jul 15, 2025
@nicbarkeragain

Created with Quartz v5.0.0 © 2026

GitHub