[Shaderhacker]

All:

I'm about to delve into Renderman at work and I'm quite excited. I have some initial notes about Renderman's architecture that is pretty interesting when comparing it to Mental Ray's API for writing shaders:

1) Shader Types supported: Displacement, Surface, Light, Volume, Imager.

Mental Ray supports quite a few more shader types: Geometry, Displacement, Surface, Light, Volume, Utility, Texture, Post-process (Imager).

2) PRMan technically doesn't support Post-process of images.

3) PRMan supports only 2 data classes: uniform and varying. This is sufficient given the Reyes architecture.

4) Triangle renderer doesn't lend itself to SIMD (although a GPU could be made to do this, i.e. RenderAnts). Perhaps this is a big disadvantage since Reyes can make more use of parallel execution in GPUs.

5) No numeric types except float in PRMan. That means no INTS, BOOLEANS, DOUBLES, etc..

6) No pointers in RSL.

7) No multi-dimensional arrays.

8) Shaders operate on micropolygon grids or each vertex in the micropolygon grid.

9) Only 2 math matrix operators ( dot, and cross ). This may be sufficient.

10) No operator over loading (i.e. I can't change *,/,+,- for different data types. Basically, color * vector is not valid).

11) All variables must be initialized. This might snag me quite a few times when developing.

12) PRMan has an array of global variables that basically are a representation of the renderer state in Mental Ray. Some are values automatically written to by the renderer and other variables, you can write to (i.e Oi, Ci).

13) All shader parameters are uniform by default.

14) diffuse() + ambient() seem to be overloaded.

More to come as I delve into it.

-M

[sijen]

2) PRMan technically doesn't support Post-process of images.
what do you mean?

[Shaderhacker]

Originally Posted by sijen

2) PRMan technically doesn't support Post-process of images.
what do you mean?

After delving more into RMan, it does have some ways to post-process and image. In order to do any kind of real Tone mapping though, I'd have to make a DSO shadeop to build my class and functions to upscale and downscale the image.

-M

[yamily]

Originally Posted by Shaderhacker

After delving more into RMan, it does have some ways to post-process and image. In order to do any kind of real Tone mapping though, I'd have to make a DSO shadeop to build my class and functions to upscale and downscale the image.

-M

mark!
Thanks for sharing.