Blender 2.78 can't render a plane with subdivision surface, displacement modifier & normal map on it #49523

New Issue

Syntex · 2016-10-01T16:56:32+02:00

Syntex commented

2016-10-01 16:56:32 +02:00

System Information
OS: Windows 10
GPU: GTX 1080
CPU: intel i7 5820k
RAM: 32gb
drivers: GeForce Game Ready Driver version 372.90
system-info.txt

Blender Version
Broken: 2.78
Worked: 2.78 release candidate v2

Short description of error
Blender cycles won't render a subdivided plane with displacement modifier & normal map on it. The program goes unresponsive or gets stuck when using either GPU or CPU to render it & won't close unless closed from task manager. I rendered a scene that had the same plane in it with 2.78 release candidate v2 and it worked fine.

Open the .blend file, press render and it won't render the image. If you use the preview render in viewport, it will say Blender is not responding at the top left corner. You can't cancel the render either.
BUG.blend

Exact steps for others to reproduce the error
Add a plane, scale it up > CTRL+A > appply scale. subdivide it 100 times in edit mode using w > subdivide. Add a subdivision surface with subdiv level 2 or 3. Then add a displacement modifier & a 4k displacement texture. It will still render it just fine. Then add a diffuse material. Add a normal map node & an image texture node with non-color data. Add a 4k normal map texture to the image texture node & connect it to the normal map's color input. Then connect the normal map node's output to the normal input of the diffuse shader. Try to render the scene > it gets stuck and won't render. If you change the viewport to rendered, blender stops responding & you have to close it from task manager.
reproduced_bug.blend

**System Information** OS: Windows 10 GPU: GTX 1080 CPU: intel i7 5820k RAM: 32gb drivers: GeForce Game Ready Driver version 372.90 [system-info.txt](https://archive.blender.org/developer/F372213/system-info.txt) **Blender Version** Broken: 2.78 Worked: 2.78 release candidate v2 **Short description of error** Blender cycles won't render a subdivided plane with displacement modifier & normal map on it. The program goes unresponsive or gets stuck when using either GPU or CPU to render it & won't close unless closed from task manager. I rendered a scene that had the same plane in it with 2.78 release candidate v2 and it worked fine. Open the .blend file, press render and it won't render the image. If you use the preview render in viewport, it will say Blender is not responding at the top left corner. You can't cancel the render either. [BUG.blend](https://archive.blender.org/developer/F372206/BUG.blend) **Exact steps for others to reproduce the error** Add a plane, scale it up > CTRL+A > appply scale. subdivide it 100 times in edit mode using w > subdivide. Add a subdivision surface with subdiv level 2 or 3. Then add a displacement modifier & a 4k displacement texture. It will still render it just fine. Then add a diffuse material. Add a normal map node & an image texture node with non-color data. Add a 4k normal map texture to the image texture node & connect it to the normal map's color input. Then connect the normal map node's output to the normal input of the diffuse shader. Try to render the scene > it gets stuck and won't render. If you change the viewport to rendered, blender stops responding & you have to close it from task manager. [reproduced_bug.blend](https://archive.blender.org/developer/F372220/reproduced_bug.blend)

Syntex commented

2016-10-01 16:56:32 +02:00

Changed status to: 'Open'

Syntex commented

2016-10-01 16:56:32 +02:00

Added subscriber: @Syntex3D

blender-admin commented

2016-10-01 16:56:32 +02:00

#49576 was marked as duplicate of this issue

Aaron Carlisle commented

2016-10-01 17:05:22 +02:00

Closed as duplicate of #49472

Aaron Carlisle closed this issue

2016-10-01 17:05:22 +02:00

Syntex commented

2016-10-01 17:08:03 +02:00

Changed status from 'Duplicate' to: 'Open'

Syntex reopened this issue

2016-10-01 17:08:03 +02:00

Syntex commented

2016-10-01 17:08:03 +02:00

I used normal subsurf modifier, not adaptive subsurf & the material is not black. This is not a duplicate of that bug

Sergey Sharybin commented

2016-10-05 10:56:36 +02:00

Added subscriber: @Sergey

Sergey Sharybin self-assigned this 2016-10-05 10:56:36 +02:00

Sergey Sharybin commented

2016-10-05 10:56:36 +02:00

Can reproduce the error on Windows. Doesn't happen on Linux tho. Indeed 2,78-rc2 synchronizes the scene quite fast while 2.78 release is getting stuck.

Something weird, will dig into!

Can reproduce the error on Windows. Doesn't happen on Linux tho. Indeed 2,78-rc2 synchronizes the scene quite fast while 2.78 release is getting stuck. Something weird, will dig into!

Sergey Sharybin commented

2016-10-05 17:14:13 +02:00

Added subscriber: @brecht

Sergey Sharybin commented

2016-10-05 17:14:13 +02:00

@brecht, this is caused by recent mikkspace fix from you. There seems to be some really suspecious code around "else" when updating min/max value. Leaving it out seemsto be more clear. But unfortunately, this particular bug seems to be a compiler bug of some sort. Here is a patch which works it arond:

P408: Fix for #49523

diff --git a/intern/mikktspace/mikktspace.c b/intern/mikktspace/mikktspace.c
index 8d51816..8c74c68 100644
--- a/intern/mikktspace/mikktspace.c
+++ b/intern/mikktspace/mikktspace.c
@@ -579,13 +579,18 @@ static void MergeVertsFast(int piTriList_in_and_out- [ ], STmpVert pTmpVert- [ ], cons
 {
 	// make bbox
 	int c=0, l=0, channel=0;
-	float fvMin- [x] = {INFINITY, INFINITY, INFINITY};
-	float fvMax- [x] = {-INFINITY, -INFINITY, -INFINITY};
+	float fvMin- [x], fvMax[3];
 	float dx=0, dy=0, dz=0, fSep=0;
-	for (l=iL_in; l<=iR_in; l++)
-		for (c=0; c<3; c++)
+	for (c=0; c<3; c++) {
+		fvMin- [x] = min(pTmpVert[iL_in].vert- [x], INFINITY);
+		fvMax- [x] = fvMin[c];
+	}
+	for (l=(iL_in+1); l<=iR_in; l++) {
+		for (c=0; c<3; c++) {
 			if (fvMin- [x]>pTmpVert- [x].vert- [x]) fvMin- [x]=pTmpVert- [x].vert[c];
-			else if (fvMax- [x]<pTmpVert- [x].vert- [x]) fvMax- [x]=pTmpVert- [x].vert[c];
+			if (fvMax- [x]<pTmpVert- [x].vert- [x]) fvMax- [x]=pTmpVert- [x].vert[c];
+		}
+	}
 
 	dx = fvMax- [x]-fvMin[0];
 	dy = fvMax- [x]-fvMin[1];

But can't doublecheck whether the initial report you've been fixing here is still fixed (for some reason reverting the fix does not cause any crashes or anytthing).

Mind having a second glance perhaps?

@brecht, this is caused by recent mikkspace fix from you. There seems to be some really suspecious code around "else" when updating min/max value. Leaving it out seemsto be more clear. But unfortunately, this particular bug seems to be a compiler bug of some sort. Here is a patch which works it arond: [P408: Fix for #49523](https://archive.blender.org/developer/P408.txt) ``` diff --git a/intern/mikktspace/mikktspace.c b/intern/mikktspace/mikktspace.c index 8d51816..8c74c68 100644 --- a/intern/mikktspace/mikktspace.c +++ b/intern/mikktspace/mikktspace.c @@ -579,13 +579,18 @@ static void MergeVertsFast(int piTriList_in_and_out- [ ], STmpVert pTmpVert- [ ], cons { // make bbox int c=0, l=0, channel=0; - float fvMin- [x] = {INFINITY, INFINITY, INFINITY}; - float fvMax- [x] = {-INFINITY, -INFINITY, -INFINITY}; + float fvMin- [x], fvMax[3]; float dx=0, dy=0, dz=0, fSep=0; - for (l=iL_in; l<=iR_in; l++) - for (c=0; c<3; c++) + for (c=0; c<3; c++) { + fvMin- [x] = min(pTmpVert[iL_in].vert- [x], INFINITY); + fvMax- [x] = fvMin[c]; + } + for (l=(iL_in+1); l<=iR_in; l++) { + for (c=0; c<3; c++) { if (fvMin- [x]>pTmpVert- [x].vert- [x]) fvMin- [x]=pTmpVert- [x].vert[c]; - else if (fvMax- [x]<pTmpVert- [x].vert- [x]) fvMax- [x]=pTmpVert- [x].vert[c]; + if (fvMax- [x]<pTmpVert- [x].vert- [x]) fvMax- [x]=pTmpVert- [x].vert[c]; + } + } dx = fvMax- [x]-fvMin[0]; dy = fvMax- [x]-fvMin[1]; ``` But can't doublecheck whether the initial report you've been fixing here is still fixed (for some reason reverting the fix does not cause any crashes or anytthing). Mind having a second glance perhaps?

Sergey Sharybin commented

2016-10-05 17:16:15 +02:00

Added subscriber: @jkmonsun

blender-admin commented

2016-10-06 03:09:46 +02:00

This issue was referenced by 5a0f397eaa

This issue was referenced by 5a0f397eaa6b5c05a6d312617eeebd96e977fa80

Brecht Van Lommel commented

2016-10-06 03:14:40 +02:00

It's odd, the else indeed shouldn't have been there but not sure why this is happening. Restoring it to work more like the old code seems to fix it, so I committed that. I left out the min() since I don't think it's needed.

It's odd, the `else` indeed shouldn't have been there but not sure why this is happening. Restoring it to work more like the old code seems to fix it, so I committed that. I left out the `min()` since I don't think it's needed.

Brecht Van Lommel commented

2016-10-06 03:15:05 +02:00

Changed status from 'Open' to: 'Resolved'

Brecht Van Lommel closed this issue

2016-10-06 03:15:05 +02:00

Sergey Sharybin commented

2016-10-06 09:34:20 +02:00

@brecht, seems to be some sort of compiler issue. I remember having similar crazy thing happening, but that was caused by code area not having SSE2 flag and which tried to use SSE2 function.. But nowadays SSE2 is enabled globally. Also didn't see anything odd with CFLAGS or anything. Would be interesting to see if the issue happens with MSVC2015, but it'll take forever for me to test that on my laptop, so can't be bothered right now. It also didn't happen in debug builds and in release builds on Linux.

Thanks for giving an extra help here!

@brecht, seems to be some sort of compiler issue. I remember having similar crazy thing happening, but that was caused by code area not having SSE2 flag and which tried to use SSE2 function.. But nowadays SSE2 is enabled globally. Also didn't see anything odd with CFLAGS or anything. Would be interesting to see if the issue happens with MSVC2015, but it'll take forever for me to test that on my laptop, so can't be bothered right now. It also didn't happen in debug builds and in release builds on Linux. Thanks for giving an extra help here!

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

Blender 2.78 can't render a plane with subdivision surface, displacement modifier & normal map on it #49523