cycles hair (curve segments) problem #38304

New Issue

Ronny G · 2014-01-21T17:22:10+01:00

Ronny G commented

2014-01-21 17:22:10 +01:00

System Information
Operating system and graphics card

Blender Version
Broken: 2.69 an up

Short description of error

particle hair (curve segments) with children (simple/interpolated) and length < 100% produces black "artefacts"

Exact steps for others to reproduce the error

see blend file for details

add hair particle system
turn on children simple or interpolated
change the length with vertex group or random option
change hair primitive -> curve segments
adjust root and tip (cycles hair settings) if necessary
render :)

particle_black1.blend

**System Information** Operating system and graphics card **Blender Version** Broken: 2.69 an up **Short description of error** particle hair (curve segments) with children (simple/interpolated) and length < 100% produces black "artefacts" **Exact steps for others to reproduce the error** see blend file for details 1. add hair particle system 2. turn on children simple or interpolated 3. change the length with vertex group or random option 4. change hair primitive -> curve segments 5. adjust root and tip (cycles hair settings) if necessary 6. render :) ![hair_cull_on.png](https://archive.blender.org/developer/F71101/hair_cull_on.png) ![hair_cull_off.png](https://archive.blender.org/developer/F71102/hair_cull_off.png) ![hair_prob.png](https://archive.blender.org/developer/F71109/hair_prob.png) [particle_black1.blend](https://archive.blender.org/developer/F71105/particle_black1.blend)

Ronny G commented

2014-01-21 17:22:10 +01:00

Changed status to: 'Open'

Ronny G commented

2014-01-21 17:22:10 +01:00

Added subscriber: @nudelZ

Sv. Lockal commented

2014-01-21 17:28:48 +01:00

Added subscriber: @Lockal

Sv. Lockal commented

2014-01-21 17:29:47 +01:00

Added subscriber: @brecht

Sv. Lockal commented

2014-01-21 18:51:43 +01:00

Note for devs:
bvh_curve_refine is a source of NaNs, at least one comes from:

float3 tg = normalize_len(float4_to_float3(P2 - P1), &l);

when P2 - P1 => {0,0,0}; len({0,0,0}) => 0; {0,0,0}/0 => NaN.

This file probably demonstrates 2 different bugs: bad shading along the hair, when "Cull back faces" is disabled, and black dots on the tip of each hair.

NaNs were the reason, why I reverted b886c26d1f. Once hair functions generate bad normals, these nans may appear inside bvh_intersect_hair, where NaNs are no-candidate values for SSE functions (e. g. SSE min and max always return normal float values, if any). SSE code did not produce bad rays, it only exposed more generic problem.

Note for devs: `bvh_curve_refine` is a source of NaNs, at least one comes from: ``` float3 tg = normalize_len(float4_to_float3(P2 - P1), &l); ``` when [P2](https://archive.blender.org/developer/P2.txt) - [P1](https://archive.blender.org/developer/P1.txt) => {0,0,0}; len({0,0,0}) => 0; {0,0,0}/0 => NaN. This file probably demonstrates 2 different bugs: bad shading along the hair, when "Cull back faces" is disabled, and black dots on the tip of each hair. NaNs were the reason, why I reverted b886c26d1f70d512b4f68975142372e3bee81c89. Once hair functions generate bad normals, these nans may appear inside bvh_intersect_hair, where NaNs are no-candidate values for SSE functions (e. g. SSE min and max always return normal float values, if any). SSE code did not produce bad rays, it only exposed more generic problem.

Sergey Sharybin commented

2014-01-22 11:07:21 +01:00

Added subscriber: @Sergey

Sergey Sharybin commented

2014-01-22 11:07:21 +01:00

@brecht, @Lockal, anyone of you is looking into the issue? Would like to have all bugs which are being looked into to be assigned :)

Thomas Dinges commented

2014-01-27 11:50:19 +01:00

Added subscriber: @ThomasDinges

Thomas Dinges commented

2014-01-27 11:50:19 +01:00

Would be really nice to fix this, so we can also re-enable the SSE optimizations here again.

Sv. Lockal commented

2014-02-07 18:40:53 +01:00

@brecht, can you look at this? I'm not very confident in raytracing. One thing I can say is that nan's behavior depends on compiler. max(1.0, NaN) can return either NaN or 1.0, depending on how compiler allocates registers. GCC mostly eliminates nans in kernel_bvh_traversal.h (both in ray->P and ray->D), but MSVC exposes them. When ray has NaNs, msvc renders transparent pixels. That's how Koro's alpha map looks in GCC:

And that's how it looks with msvc 11:

@brecht, can you look at this? I'm not very confident in raytracing. One thing I can say is that nan's behavior depends on compiler. max(1.0, NaN) can return either NaN or 1.0, depending on how compiler allocates registers. GCC mostly eliminates nans in kernel_bvh_traversal.h (both in ray->P and ray->D), but MSVC exposes them. When ray has NaNs, msvc renders transparent pixels. That's how Koro's alpha map looks in GCC: ![linalpha.png](https://archive.blender.org/developer/F76377/linalpha.png) And that's how it looks with msvc 11: ![alpha.png](https://archive.blender.org/developer/F76379/alpha.png)

Brecht Van Lommel commented

2014-02-10 18:42:30 +01:00

I fixed the problem in this file now, see {a0c54d71d8a432e85fbcd9b01d05a52dee2659c3}.

I'm not sure that covers the entire issue, I didn't test with visual studio yet.

I fixed the problem in this file now, see {a0c54d71d8a432e85fbcd9b01d05a52dee2659c3}. I'm not sure that covers the entire issue, I didn't test with visual studio yet.

Thomas Dinges commented

2014-02-11 18:25:23 +01:00

Render of the attached file is fine in 3D View (Perspective Mode). Tested with a VC2008 x64 build.

Orographic mode still has some black artefacts, but that was already in 2.69.

Render of the attached file is fine in 3D View (Perspective Mode). Tested with a VC2008 x64 build. Orographic mode still has some black artefacts, but that was already in 2.69.

Brecht Van Lommel commented

2014-02-11 18:53:14 +01:00

Changed status from 'Open' to: 'Resolved'

Brecht Van Lommel closed this issue

2014-02-11 18:53:14 +01:00

Brecht Van Lommel self-assigned this 2014-02-11 18:53:14 +01:00

Brecht Van Lommel commented

2014-02-11 18:53:14 +01:00

Tested with Visual Studio 2013 and see to work fine. I committed b886c26d1f since that seems to work again.

Tested with Visual Studio 2013 and see to work fine. I committed b886c26d1f70d512b4f68975142372e3bee81c89 since that seems to work again.

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

cycles hair (curve segments) problem #38304