Slashdot Mirror
Software Rendering Engine GPU-Accelerated By WebCL
Phopojijo writes "OpenGL and DirectX have been the dominant real-time graphics APIs for decades. Both are catalogs of functions which convert geometry into images using predetermined mathematical algorithms (scanline rendering, triangles, etc.). Software rendering engines calculate colour values directly from the fundamental math. Reliance on OpenGL and DirectX could diminish when GPUs are utilized as general 'large batches of math' solvers which software rendering engines offload to. Developers would then be able to choose their algorithms for best suits their project, even native to web browsers with the upcoming WebCL."
...is something i'd really like to see -especially one that does Glide and the dither+undither characteristic of the first three Voodoo cards.
I know there's MAME/MESS but I don't think they do the infamous filtering.
Developers would then be able to choose their algorithms for best suits their project, even native to web browsers with the upcoming WebCL."
If web browsers were people, that statement would have caused a mass suicide of them. Guys, stop trying to turn the browser into a platform. It introduces so many layers of complexity and security issues that it's a miracle anyone has any trust or faith in the internet at all. It's getting to the point where the only way to safely browse the net is to shove the entire browser into a virtual machine... and even that only manages to protect your own computer, say nothing of your online activities, credentials, life, etc.
We need to be making browsers simpler, not more complex. Feature bloat is making these things a leper's colony inside your computer... a cesspool of malware and vulnerability. Don't add to it by coming up with some new way for developers to directly access the hardware of your computer because you're too fucking lazy to write an app to do whatever it is, and want to cram it into the browser instead. You're just encouraging them.
Seriously, we need a 12 step program for these "web 2.0" people.
#fuckbeta #iamslashdot #dicemustdie
Great. I'd sure like my GPU, with its mad low level optimizations and surely ugly code to be used by unsigned code from random sources.
Java applets are far too secure, let's get to the lowest level!
Stupidity is the root of all evil.
Some want to use the same algorithms OpenGL and DirectX does... and those APIs are still for them.
Some do not. A good example is Epic Games who, in 2008, predicted "100% of the rendering code" for Unreal Engine 4 would be programmed directly for the GPUs. The next year they found the cost prohibitive so they kept with DirectX and OpenGL at least for a while longer. Especially for big production houses, if there is a bug or a quirk in the rendering code, it would be nice to be able to fix the problem directly rather than hack in a workaround.
The terminology in the summary is confusing and wrong.
First of all, software rendering vs. hardware rendering isn't the same as scanline rendering vs. "rendering from the underlying math", which I assume is a bad attempt at a layman's description of raytracing. You can have a scanline triangle renderer in software, and you can have a raytracer in hardware. It is true that most GPUs are built for scanline rendering and not raytracing, but plenty of raytracers have been written that run on GPUs.
Second, if your renderer runs on the GPU using OpenCL, then it is not a software renderer, it is a hardware renderer, perhaps with a little more of the work done in programmable shaders and a little less done on the fixed function hardware.
What they meant to say was that you can program your own hardware renderer using CL kernels, rather than rely on the ever-decreasing fixed-function hardware that the triangle pipeline normally uses.
The only fixed-function capabilities in a modern GPU are texture filtering and rasterization. The vertex processing, lighting, and shading are all programmable.
There is something interesting and new here, which is that maybe sometime in the future the programmable hardware will be good enough that the fixed-function stuff can be done away with completely.
Hamsters are at least as feathery as penguins. HamLix
Many 3D engines are carefully tuned to the limited bandwidth to the GPU cards that provides them just enough bandwidth per frame to transfer the necessary geometry/textures/etc for that frame. The results, of course, stay on the GPU card and are just outputted to the frame buffer. Now, in addition to that existing overhead, the engine writer would now have to transfer back the results/frame buffer back to the CPU to process, generate an image, that is then passed back to the GPU to be displayed as an image? Or am I missing something?
While I'm sure it would allow customized algorithms, they would have to be rather unique to not be handled by the current state of geometry/vertex/fragment shaders. Are they thinking some of non-triangular geometry?
Maybe there is a way to send the result of the maths directly to the frame buffer while it's on the GPU?
Things you think are in the Constitution, but are not.