Right-o.
From the perspective of the scene BSP, the hair object is just an opaque
box. When a ray/scanline hits that box, the hair sub-renderer is called,
which uses its own BSP, the settings of which are found on the individual
hair primitive's ppg.
A BSP tree is only used to accelerate ray/triangle intersection. It does
this by collecting triangles into a recursive set of axis-aligned boxes and
eliminating those triangles whose boxes the ray doesn't hit. That's all.
In scanline/raytrace mode, the hair is rendered directly from the hair
object, using the original hair data. In this case the hair's BSP settings
are used.
In rasterizer mode, the hair is converted directly to triangles and added to
the global pool of renderable triangles. In this case the hair's BSP
settings are ignored.
- ½
________________________________
From: owner-xsi(at)Softimage.COM [mailto:owner-xsi(at)Softimage.COM] On Behalf Of
Bernard Lebel
Sent: 30-May-2007 09:50
To: XSI(at)Softimage.COM
Subject: Re: massive hair instance render
Gods of mental ray, correct me if I'm wrong.........
Just to be pedantically clear about what does the hair BSP:
The geometry generated by hair is part of the "main" BSP tree and changing
the hair generation settings like number of hairs and strand multiplier will
affect the "main" BSP tree, as this will affect the number of triangles in
the scene. You can verify that by doing diagnostic renders:
Application.CreatePrim("Sphere", "MeshSurface", "", "")
Application.DeleteObj("sphere")
Application.CreatePrim("Cylinder", "MeshSurface", "", "")
Application.SetValue("cylinder.cylinder.radius", 2.39, "")
Application.SetValue("cylinder.cylinder.height", 40, "")
Application.SetValue("cylinder.polymsh.geom.subdivu", 24, "")
Application.SetValue("cylinder.polymsh.geom.subdivv ", 24, "")
Application.SetValue("cylinder.polymsh.geom.subdivbase",
24, "")
Application.SetValue("cylinder.polymsh.geom.subdivbase",
24, "")
Application.SetValue("cylinder.polymsh.geom.subdivv", 24, "")
Application.SetValue("cylinder.polymsh.geom.subdivv", 48, "")
Application.FreezeObj("", "", "")
Application.Rotate("", 0, 0, 90, "siRelative", "siLocal", "siObj", "siXYZ",
"", "", "", "", "", "", "", 0, "")
Application.FreezeObj("", "", "")
Application.ActivateObjectSelTool("")
Application.DeselectAll ()
Application.ActivateObjectSelTool("")
Application.CreatePrim("Sphere", "MeshSurface", "", "")
Application.Translate ("", 0, 0, -7.84688268062173, "siRelative", "siLocal",
"siObj", "siXYZ", "", "", "", "", "", "", "", "", "", 0, "")
Application.Translate("", 0, 0, 0.721777065527067, "siRelative", "siLocal",
"siObj", "siXYZ", "", "", "", "", "", "", "", "", "", 0, "")
Application.ApplyHairOp("sphere", "")
Application.InspectObj("ViewRenderOptions", "", "", 1, "")
Application.SetValue ("ViewRenderOptions.DiagnosticPhoton",
0, "")
Application.SetValue("ViewRenderOptions.DiagnosticGrid ",
0, "")
Application.SetValue("ViewRenderOptions.DiagnosticGrid", 0,
"")
Application.SetValue("ViewRenderOptions.DiagnosticBspTreeDepth",
1, "")
Draw a render region
Set the strand multiplier to 0, render
Set the strand multiplier to 2, render
Set the total hairs to 13,000, render
Unless there are shaders being used that modify the hair geometry on-the-fly
or other characteristics of hair, changing the hair BSP should not affect
actual render time (when buckets are being rendered).
The hair BSP deals with the pre-processing phase, that is, the generation of
hair geometry. The aim at the hair BSP tree is not to speed up ray
intersection, but to speed up things like surface collisions and similar
things that require the renderer to know where the geometry is in space to
generate accurately the hair geometry. So changing the hair BSP should
affect the hair generation time only (if, like I said, there is nothing that
modifies that geometry or else during rendering).
Ultimately the hair BSP will affect the "overall" render time, because the
rendering doesn't happen without the pre-processing of hair.
Bernard
On 5/30/07, Kim Aldis < XSI(at)kim-aldis.co.uk> wrote:
> I'm not sure if this also applies to instancing but hair has its own BSP
> settings. Normal hair renders that can make a huge difference.
>
> > -----Original Message-----
> > From: owner-xsi(at)Softimage.COM [mailto:owner-xsi(at)Softimage.COM] On
> > Behalf Of Bernard Lebel
> > Sent: 30 May 2007 01:27
> > To: XSI(at)Softimage.COM
> > Subject: Re: massive hair instance render
> >
> > Definitely consider using dynamic bsp and lower the mesh splitting
> > factor (try 0.5, then 0.2, then 0.1...). Will make rendering slower,
> > but will render. Also, definitely consider using the rasterizer, if it
> > doesn't change too much your lighting.
> >
> >
> > Cheers
> > Bernard
> >
> >
> >
> >
> > On 5/29/07, Kris Rivel <krisrivel(at)gmail.com> wrote:
> > > I'm building a tree with many extremely photo-realistic leaves on it.
> > > I'm using some SSS shading which looks great but together with
> > massive
> > > leaf instance count, my renders are hitting the roof. I'm trying to
> > > stick to a 2GB memory limit too which is all thats available on the
> > farm
> > > I will be using. Just curious if anyone has any tips for handling
> > such
> > > massive hair instance renders. I still have a lot of tweaking and
> > > optimization to fiddle with but it would be great to know of any
> > other
> > > nice tricks up front that I may not have thought about. Thanks.
> > >
> > > Kris
> > > ---
> > > Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in
> > body:
> > > unsubscribe xsi
> > >
> > ---
> > Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in
> > body:
> > unsubscribe xsi
>
> ---
> Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in body:
> unsubscribe xsi
>
