Blender.org

Are there any plans for implementation of HLSL shaders in Blender?

Somebody correct me if I'm mistaken - HLSL is for Windows and Direct X, and Blender runs on many platforms that MS does not support with Direct X.

In tuhopuu gameengine you can find support for GLSL shaders.

(http://projects.blender.org/viewcvs/vie ... ot=tuhopuu)

/jesterKing

Shaders may be implemented i openGL also using
OpenGL® Shading Language

Here is a quote from http://www.theregister.co.uk/2003/07/01 ... ces_opengl

The HLSL was approved by the OpenGL Architecture Review Board, the standard's governing body, on 11 June and is likely to form the basis for OpenGL 2.0, the next major revision of the open graphics API.

More ...

http://www.opengl.org/documentation/oglsl.html

http://www.theregister.co.uk/2003/07/01 ... ces_opengl

You can use GLSL in tuhopuu2.

Check out the PyDoc stuff for how to load your shader, and how to set uniforms, etc.

Many thank's JesterKing and alien-xmp, I will check that.

You can use GLSL in tuhopuu2.

Excellent work Kester... this could be very promising indeed !!!

Quick question though...

Won't this kill / hinder the performance of the game engine to some degree ???

For example, the game engine isn't known for its speed ( no offence intended ), so wouldn't GLSL bring it down even more ???

On a side note...

Great news about the 'Shadow' support in the game engine, all of your work in this field is greatly appreciated !!!

Take care,

Panther

Won't this kill / hinder the performance of the game engine to some degree ???

It really depends on the graphics card. I test with a Radeon 9800XT and a GeForce FXGo5200. The Radeon loves it, it barely misses a step. The FX5200 does get killed. The FX5200 just doesn't have the fill rate, the speed is proportional to the size of the object being shaded.

Also, vertex buffer objects and display lists are in Tuhopuu2.

Also also, gcc generates faster code than MSVC (~33%)

Put these together, and I'm getting around 60fps with 100k onscreen quads. This is of course static geometry, deformed objects such as armatures and zsorted objects and the rest of the game will slow you down.

alien-xmp wrote: It really depends on the graphics card. I test with a Radeon 9800XT and a GeForce FXGo5200. The Radeon loves it, it barely misses a step. The FX5200 does get killed. The FX5200 just doesn't have the fill rate, the speed is proportional to the size of the object being shaded.

There is a BIG difference between those two cards, especially in price. I hope you didn't mean it as a direct comparison of nvidia vs ati.

Which driver versions were you using for each as well?

alien-xmp wrote: Also also, gcc generates faster code than MSVC (~33%)

Do you have proof of this? I've read nothing but article after article stating the opposite.

It really depends on the graphics card. I test with a Radeon 9800XT and a GeForce FXGo5200. The Radeon loves it, it barely misses a step. The FX5200 does get killed.

ATI giving better OpenGL performance than NVIDIA... I never thought I'd see the day !!!

I'm using a standard ( not 'Go' ) 128MB GeForceFX 5200, what do you think my chances are of getting 60FPS performing a similar test to you - Good / bad ???

gcc generates faster code than MSVC (~33%)

Interesting... is that documented somewhere / do all the developers know that ???

I know we're speaking about Tuhopuu here, but does anyone know which compiler is used to generate all the 'Official' BF releases ???

I'm pretty sure everything execpt for Windows is compiled with gcc currently for the offical builds.

Thanks for the fast response SirDude !!!

IoN_PuLse wrote:

alien-xmp wrote: It really depends on the graphics card. I test with a Radeon 9800XT and a GeForce FXGo5200. The Radeon loves it, it barely misses a step. The FX5200 does get killed. The FX5200 just doesn't have the fill rate, the speed is proportional to the size of the object being shaded.

There is a BIG difference between those two cards, especially in price. I hope you didn't mean it as a direct comparison of nvidia vs ati.

Which driver versions were you using for each as well?

Yeah I know. One of the nvidia papers showed a similar performance jump between the 5200 and their high end models. The FX cards also have a nicer shader engine (longer shaders, dFdx works.) The FX is also running at 1600x1200, while the ATI is at 1024x768, which is 240% more pixels to shade.

I mostly test driver 61.06 on Linux, and occaisionally use 61.76 on Windows. The ATI is on Catalyst 4.6. (Not the latest, but it will take me all day to download the next revision.)

IoN_PuLse wrote:

alien-xmp wrote: Also also, gcc generates faster code than MSVC (~33%)

Do you have proof of this? I've read nothing but article after article stating the opposite.

I know, I was quite surprised when I saw it, and I only compiled with gcc because I was helping someone on IRC. I'm basing it on game framerates:
Ballercoaster: MSVC: ~120fps, gcc: ~160fps
55Wheels: MSVC: ~30fps, gcc: ~40fps

MSVC extra compiler flags: /Og /Oip /GR
GCC extra compiler flags: -O2 -ffast-math

alien-xmp wrote:

IoN_PuLse wrote:

alien-xmp wrote: Also also, gcc generates faster code than MSVC (~33%)

Do you have proof of this? I've read nothing but article after article stating the opposite.

I know, I was quite surprised when I saw it, and I only compiled with gcc because I was helping someone on IRC. I'm basing it on game framerates:
Ballercoaster: MSVC: ~120fps, gcc: ~160fps
55Wheels: MSVC: ~30fps, gcc: ~40fps

MSVC extra compiler flags: /Og /Oip /GR
GCC extra compiler flags: -O2 -ffast-math

Since the game engine is mostly C++, could it be that this speed increase is limited to C++ code for whatever reason? Because for the rest of blender, I've heard quite the opposite.

Martin

So this is really a comparison of GCC for windows and MSVC?