Shaders using halfs are usually faster even on GeForce 7 class hw because of dedicated normalization unit and increased latency hiding capabilities.Originally Posted by Korval
Shaders using halfs are usually faster even on GeForce 7 class hw because of dedicated normalization unit and increased latency hiding capabilities.
Nonsense.Don't know its origin but a half float is in most conditions alot more than enough, and half memory cost
Half-floats are a phantom. An illusion. For half of the hardware out there (more than that, now that nVidia has largely abandoned them), half-floats don't even exist.
There's no memory savings with half-floats because ATi doesn't support them. They never did. On no ATi hardware will you see a benefit when you use them. And since it isn't something that is cross-platform, it isn't something you can rely on.
Now, maybe it's something you want to test for and swap in different shaders. But it certainly isn't something that ATi or anyone else should be forced to support in any way just because nVidia decided to save some transistors in their NV3x/4x hardware.
You are greatly underestimating features of the D3DX. There is more in it beyond simple helpers. I would suggest that you read its documentation.
Even the HLSL compiler is part of the D3DX and not of the core runtime. Or the D3DX contains mathematics operations optimized by individual CPU vendors (as far as I know). Also the effect framework which appears to be used by some professional developers is part of the D3DX.
Some features of D3DX are clearly meant for use in build pipeline (e.g. the unwrapper of uv coordinates, functions for calculating coefficients for precomputed radiance transfer). Yes, you can write your own version however why do that if someone else already created usable one.
DX SDK is about nice package in which you get documentation, headers, latest version of D3DX library and examples.Actually, you don't need an "SDK" at all (actually I never understood what this SDK thing is about?
If they really choose to abandon halfs, I wish they would at least add a color-renderable GL_RG32F_ARB texture format to get 64 bits per pixel (mainly to save bandwidth in PBO operations).
Right now, the only non-RECT color-renderable tex format with 64bpp is GL_RGBA16F_ARB. It's supported natively up till NV44 that I know of, don't know if it's supported natively on more recent hardware though.
Nvidia's CUDA language still has support for halfs through the CUDA driver API and CUDA is only supported on GF8 and up so I guess it's still supported natively on GF8 hardware...
Please correct me if I'm wrong
cross-platform what is that exactly.... OpenGL isnt even supported directly under most of the worlds computers, i.e sure XP did support OpenGL 1.1, the word cross platform is an illusion..
Halfs aint the hardware has it, or you havent used 16 bit floating point blend? Never tried a fast normalization using half4 ?
Knowledge is Power. Power Corrupts. Study Hard. Be Evil
You can get all the OpenGL functionality you need just by downloading the appropriate graphics card driver.
About cross-platform... I just can't use Direct3D in Linux
Korval is talking about the half type within the shader (which still has advantages on GF7 and older hw, I do not know how it is with the G80), not about the 16F format of textures/render buffers.
Ah, I see. Guess I was thrown off by the 'half memory cost'-quote in his post
And, i was talking about the 16F textures/render buffers ><
"Not needed at all" was the phrase I replied to; this implies there is no requirement AT ALL for it, however the fact that it could be helpful (and indeed, one the reasons I'd recommend learning D3D over OGL right now is that having D3DX and DXUT takes away some of the pain) proves there is a need for beginners.
It ok for people who have been using 3D APIs for a while to say 'oh, you can code this in a day or two' or 'people in industry won't use it' and I'm sure it's true however everyone has to start some where, and at the point of learning the API to get an intrest not everyone can write the code in a day or two or wants to use some great abstraction over the API; they want to use the API and it's helpful to have support code there which is 'offical' and known to work properly.
Yes, a cleaner API will help and that should be a priority ofcourse but you can't sit there and write something off as 'not needed' because professionals wont use it.. there is more to the OpenGL world than professionals and people who have been using it for years and, like it or not, these people sometimes need something to help them in.
So, no, no problems with the english language here thanks.
Posting Permissions
