A couple weeks I added to FMXutils repository a conversion of the “Compute Boids” flocking demo.
This builds upon the WebGPU driver for FireMonkey, but uses a dedicated class where the Compute & rendering shaders are hosted.
Two ping-pong buffers are used to store and updated particle data, which are used to mimic the flocking behavior of birds.
There are many patterns, principles, and development methodologies out there that are evangelized.
They can have fancy names, some are old, some are new. But they all boil down to just three “Unities”, which have nothing in particular to do with computing or development, and all to do with those that are doing development and design: humans.
You find those three Unities in practically all crafts.
The FMXUtils repository now implements support for GPU Buffers for Delphi FMX (with DX11 or WebGPU).
Adding GPU buffers required some trickery and hackery, but the speedups can be massive: 2x to 3x speedups on low-end integrated GPUs, with reports of 20x (twenty times) speedups on discrete GPUs!
I tried to make the support as easy and backward compatible as possible, while dancing around missing private fields and non-virtual methods.
Just a quick post to announce the Delphi-WebGPU repository.
What ?
Why ?
To take the words from webgpu-headers:
While WebGPU is a JavaScript API made for the Web, it is a good tradeoff of ergonomic, efficient and portable graphics API.
In the previous article on 3D point cloud for FMX, systematic memory copy when drawing primitives came out as #1 bottleneck in FMX performance when rendering millions of points.
But was it just a false positive ?
I hacked FMXU.Context.DX11 to eliminate the memory copies… in the most hacky way, don’t ask for code it’s not re-useable 🙂 (more…)
With Delphi 12.Community Edition released not too long ago, it was an opportunity to test run what’s possible with FMX on the 3D side.
And while Delphi FireMonkey has 3D support, it’s not really a core feature. For a test run, the simplest of all 3D entities, a point cloud can serve as a decent stress test.
Point Clouds are now ubiquitous with the advent of 3D capture devices, which go from LIDAR to RGB-D cameras and all kinds of AI augmented approaches. So there is a plethora of 3D point cloud datasets out there.
In another comment of the Curiouser Case, Arnaud Bouchez pointed to an interesting optimization that is use in mORMot‘s UpperCopy255Buf function: a branchless parallel upper case conversion.
At the core of that implementation are the following lines of code
c := src[i]; d := c or _80; dest[i] := c - ((d - _61) and not (d - _7b)) and ((not c) and _80) shr 2;
Ok, it may not be too obvious what happens at first sight. Let’s break it down.
(more…)
In a comment to the previous article on a case insensitive hash code, Stefan Glienke pointed to an approach used in Spring4D’s comparers, which is a delightful hack.
Rather than converting the string to a “proper lower case”, it converts the string to an “approximate lower case” using an “or $20”, which happens to be good enough for a hash on string identifiers.
To figure the trick, one needs to check the ASCII Table. (more…)
Recent commits to the DWScript repository doubled the compiler performance when compiling many small scripts, like happens in the unit tests suites.
This started from a first profiling run where the memory allocations around the UnicodeLowerCase function came out as top bottlenecks.
Thing is, Pascal being a case-insensitive language, there are lots of case-insensitive comparisons, lookups, searches and hashes, and turns out a key hash code was computed with code like