I think I can finally say I've successfully implemented the Radiance Cascades algorithm! :D
Running super fast now thanks to OpenCL compute kernels.

Time to integrate it into a game :)

僕のOpenCLどこ？なにもしないから出てきてよ

Vivo X100s With Mediatek Dimensity 9300+

Benchmarking ⬇️

AnTuTu V10:
2305268 Points

Geekbench 6:
SCS: 2313 Points
MCS: 7743 Points
OpenCL: 13786 Points

#Dimensity9300Plus #Mediatek

Pythonの勉強に初めて全文読んだプログラムもGPU計算(当時はOpenCLだった)だったし、どこかしらGPU計算と縁がある

✅SECURITY TOOLS Benefit from Hashcat’s CUDA/OpenCL capabilities for password recovery

کار کردن تو دنیای c/cpp تمومی نداره
به چیزایی مثل
Qt
Opencl
Opengl
Unreal engine
و هزار تا کوفت و زهرمار دیگه هم هست:)
که خب وابسته به هدف و پول ادم میره دنبالش:)

In QCOM docs:

> OpenCL is easy to program, debug, and profile thanks to all the resources available for Adreno, including Adreno OpenCL programming guide, Snapdragon profiler/debugger, and Adreno SDK examples.

Grimm 🌱/acc vittorio These are 4 competing GPU APIs: CUDA, OpenCL, WebGPU, GLSL shaders. Its been useful to me as a many-dimensional shape artist as u see here, a shape that appears to vibrate, bend, bounce, & fit itself to balls rolling on it. If ppl play they will learn math youtube.com/watch?v=eWNZk0…

OpenCL didn't have promise. It's fucking dog crap and I and many others refused to touch it because it promised to turn your code into an unmaintainable nightmare.

Im not locked into GLSL. I rarely write GLSL code directly anymore. Its a compile target that my GPU language calls. dagball/lib/Ap.js calls dagball/lib/TinyGLSL.js calls WebGL2_GLSL in browser calls GPU hardware. Ap.js could in theory compile to WebGPU or OpenCL with small mods.

trust me, gamers & normies had nothing to do with the success of CUDA or the failure of OpenCL. OpenCL failed as a production API because it was fucking dog crap. You had to write your compute kernels as strings, which IDEs simply have no ability to parse and assist in debugging.

Despite its somewhat ugly & unconventional interface, @Nvidia CUB has been my favorite library over the years. Such a gem! Even in the most trivial operations, consistently 50% faster than the Thrust wrapper on H100 - 3 TB/s vs 2.1 TB/s

github.com/ashvardanian/P…

One of AMD's first big fuck-ups when it came to compute (and by extension, getting their foot in the AI door) was the same reason Apple fucked it up: they ignored the gaming/normie space when hyping OpenCL up

MΛX (mxcop) Yaazarai Xor I've worked at both AMD and NVIDIA. I was there at AMD (and sent a scathing email) the day the CPU back-end for OpenCL was dropped. I was also at NVIDIA when this happened: developer.nvidia.com/blog/nvidia-is…

AMD doesn't give a damn about OpenCL anymore, but both NVIDIA and Intel have…

GeekBench 6 FreeのOpenGL（OpenCLだったかも）ベンチも回したけど、スコア80000くらいで公式サイトに上がってるスコアは240000くらいだったから、大きく乖離してた…つまり1/3程度しか性能が出てないことになる

Actors are better for continuous time ABM simulations with heterogeneous agents

GPU/CUDA/OpenCL are better suited for discrete time and more regular spaces with a single agent type, E.g. Cellular automata on grids or particles in continuous space

Intel Arc A770をUbuntuやFedoraで試してみたかったが

Linux Kernel 6.8.x ではGPUに影響を及ぼす計算バグが2個あってOpenCLが壊れているとのこと。

6.5.x や 6.6.x では大丈夫らしい。

Exynos 2200のベンチスコア漁ってみたけど
やっぱRDNAじゃ無理よな…

Adrenoに勝てたのはOpenCLだけとか…
そりゃまぁWindowsDirectXベース設計のアーキテクチャを
androidに持っていったって
無駄にダイサイズ大きくなるだけで
殆使われない機能だらけだろうに…

FFmpeg Dmitriy Kovalenko especially VAAPI, Intel QSV, AMD Accel, AI things, OpenCL things, etc.