Regression: Shading artefacts in sculpt mode #96810

New Issue

CHET · 2022-03-27T00:05:24+01:00

CHET commented

2022-03-27 00:05:24 +01:00

System Information
Operating system: Windows-10-10.0.19044-SP0 64 Bits
Graphics card: NVIDIA GeForce GTX 950M/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 496.13

Blender Version
Broken: version: 3.1.1 Release Candidate, branch: master, commit date: 2022-03-24 13:38, hash: fb2cb0324a
Worked: (newest version of Blender that worked as expected)

Short description of error
Shading artefacts in sculpt mode when use grab brush
blender_E1iUEGStqR.mp4

Exact steps for others to reproduce the error

Open attached file
Select the "Grab" brush
Click on random mesh positions
untitled.blend

**System Information** Operating system: Windows-10-10.0.19044-SP0 64 Bits Graphics card: NVIDIA GeForce GTX 950M/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 496.13 **Blender Version** Broken: version: 3.1.1 Release Candidate, branch: master, commit date: 2022-03-24 13:38, hash: `fb2cb0324a` Worked: (newest version of Blender that worked as expected) **Short description of error** Shading artefacts in sculpt mode when use grab brush [blender_E1iUEGStqR.mp4](https://archive.blender.org/developer/F12948645/blender_E1iUEGStqR.mp4) **Exact steps for others to reproduce the error** - Open attached file - Select the "Grab" brush - Click on random mesh positions [untitled.blend](https://archive.blender.org/developer/F12948649/untitled.blend)

CHET commented

2022-03-27 00:05:24 +01:00

Added subscriber: @cheteron

Philipp Oeser commented

2022-03-30 15:28:53 +02:00

Added subscriber: @lichtwerk

Philipp Oeser commented

2022-03-30 15:28:53 +02:00

Changed status from 'Needs Triage' to: 'Needs User Info'

Philipp Oeser commented

2022-03-30 15:28:53 +02:00

Cannot repro here

**System Information**
Operating system: Linux-5.13.0-0.rc6.45.fc35.x86_64-x86_64-with-glibc2.34.9000 64 Bits
Graphics card: NVIDIA GeForce GTX 970M/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 495.44
version: 3.1.1, branch: master, commit date: 2022-03-29 17:50, hash: `rBd91843f3b6c6`

Is this also an issue in previous version? https://download.blender.org/release/
Have you tried other GPU driver versions?

Cannot repro here ``` **System Information** Operating system: Linux-5.13.0-0.rc6.45.fc35.x86_64-x86_64-with-glibc2.34.9000 64 Bits Graphics card: NVIDIA GeForce GTX 970M/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 495.44 version: 3.1.1, branch: master, commit date: 2022-03-29 17:50, hash: `rBd91843f3b6c6` ``` Is this also an issue in previous version? https://download.blender.org/release/ Have you tried other GPU driver versions?

CHET commented

2022-03-30 15:33:38 +02:00

This comment was removed by @cheteron

*This comment was removed by @cheteron*

CHET commented

2022-04-01 23:28:01 +02:00

The problem is not only visual. Where artifacts appear, vertices seem to offset, or split, and when trying to smooth, the geometry is distorted. It's wire problem, and I don't know how to identify the problem, it appears randomly, sometimes when using dyntopo, and sometimes during the sculpting process or during the undo operation, in the same way, for unknown reasons, the problem itself disappears.
Sometimes I had to remesh the geometry to remove the artifact - it helped, but sometimes

The problem is not only visual. Where artifacts appear, vertices seem to offset, or split, and when trying to smooth, the geometry is distorted. It's wire problem, and I don't know how to identify the problem, it appears randomly, sometimes when using dyntopo, and sometimes during the sculpting process or during the undo operation, in the same way, for unknown reasons, the problem itself disappears. Sometimes I had to remesh the geometry to remove the artifact - it helped, but sometimes

CHET commented

2022-04-02 00:58:15 +02:00

Artifact appears after using mask and then draw brush
P.S. new nvidia drivers
blender_oPMoJ32PHQ.mp4

Artifact appears after using mask and then draw brush P.S. new nvidia drivers [blender_oPMoJ32PHQ.mp4](https://archive.blender.org/developer/F12962818/blender_oPMoJ32PHQ.mp4)

CHET commented

2022-04-06 17:22:47 +02:00

Blender 3.1.2 - same problem!

Philipp Oeser commented

2022-04-06 17:29:53 +02:00

Have you checked 2.93 LTS? Is it the same problem there?

CHET commented

2022-04-06 17:43:55 +02:00

This problem appear only in 3.1+

Philipp Oeser commented

2022-04-06 17:50:34 +02:00

Changed status from 'Needs User Info' to: 'Needs Triage'

Germano Cavalcante commented

2022-04-06 23:09:03 +02:00

Added subscriber: @mano-wii

Germano Cavalcante commented

2022-04-06 23:09:03 +02:00

Changed status from 'Needs Triage' to: 'Confirmed'

Germano Cavalcante commented

2022-04-06 23:09:03 +02:00

I can reproduce the problem. But it's nothing as serious as what's shown in the video.

Is it only seen on Windows?
Anyway, I believe it deserves an investigation.

Operating system: Windows-10-10.0.22000-SP0 64 Bits
Graphics card: Radeon (TM) RX 480 Graphics ATI Technologies Inc. 4.5.14802 Core Profile/Debug Context 22.2.3 30.0.14029.5006

I can reproduce the problem. But it's nothing as serious as what's shown in the video. ![GIF.gif](https://archive.blender.org/developer/F12979295/GIF.gif) Is it only seen on Windows? Anyway, I believe it deserves an investigation. --- Operating system: Windows-10-10.0.22000-SP0 64 Bits Graphics card: Radeon (TM) RX 480 Graphics ATI Technologies Inc. 4.5.14802 Core Profile/Debug Context 22.2.3 30.0.14029.5006

Germano Cavalcante commented

2022-04-06 23:19:33 +02:00

It is difficult to find when the problem was introduced, but I noticed that in some versions the problem was as bad as the one shown in the video. So there has been an improvement in recent versions.

@cheteron, please try the latest daily build: https://builder.blender.org/download/

Go to File → Defaults → Load Factory Settings and then load your file to see if you still can reproduce this issue.

It is difficult to find when the problem was introduced, but I noticed that in some versions the problem was as bad as the one shown in the video. So there has been an improvement in recent versions. @cheteron, please try the latest daily build: https://builder.blender.org/download/ Go to File → Defaults → Load Factory Settings and then load your file to see if you still can reproduce this issue.

CHET commented

2022-04-06 23:21:56 +02:00

This is a problem because the mesh is not only visually distorted, it physically creates artifacts. And it need of a fix

Philipp Oeser commented

2022-04-27 14:21:09 +02:00

In #96810#1336836, @cheteron wrote:
This problem appear only in 3.1+

@mano-wii : can you bisect this? Seems like it could be high prio if you ask me...

> In #96810#1336836, @cheteron wrote: > This problem appear only in 3.1+ @mano-wii : can you bisect this? Seems like it could be high prio if you ask me...

Philipp Oeser changed title from ~~Shading artefacts in sculpt mode~~ to Regression: Shading artefacts in sculpt mode

2022-04-27 14:21:40 +02:00

Germano Cavalcante commented

2022-04-27 18:11:41 +02:00

Added subscriber: @HooglyBoogly

Germano Cavalcante commented

2022-04-27 18:11:41 +02:00

A bit difficult to bisect as the problem is random, subtle and behaves differently in different versions.
But I tried again and was more careful this time.
Apparently the problem was introduced in {7682d7de046185382985999f8f6b6e7dcf860582}
So @HooglyBoogly, any ideas what could be causing it?

A bit difficult to bisect as the problem is random, subtle and behaves differently in different versions. But I tried again and was more careful this time. Apparently the problem was introduced in {7682d7de046185382985999f8f6b6e7dcf860582} So @HooglyBoogly, any ideas what could be causing it? ![GIF.gif](https://archive.blender.org/developer/F12979295/GIF.gif)

Hans Goudey commented

2022-04-28 20:18:08 +02:00

No ideas at the moment. But I'll look through that commit in more depth when I get a chance, maybe there's some incorrect part.

Julien Kaspar commented

2022-05-12 10:22:20 +02:00

Added subscriber: @JulienKaspar

Julien Kaspar commented

2022-05-12 10:22:20 +02:00

@HooglyBoogly Are there any updates on this issue?

Jeroen Bakker commented

2022-05-18 18:55:01 +02:00

Added subscriber: @Jeroen-Bakker

Jeroen Bakker commented

2022-05-18 18:55:01 +02:00

Seems like they occur at pbvhnode edges...

Hans Goudey commented

2022-05-18 19:25:08 +02:00

Sorry, I don't have any updates at the moment. The last time I looked into this I didn't get anywhere. Errors at node edges makes sense to me.

Brecht Van Lommel commented

2022-05-18 20:52:07 +02:00

Added subscriber: @brecht

Brecht Van Lommel commented

2022-05-18 20:52:07 +02:00

pbvh_update_normals_store_task_cb does not look thread safe. Two threads calling BLI_BITMAP_DISABLE to write bits in the same integer will not get synchronized correctly.

`pbvh_update_normals_store_task_cb` does not look thread safe. Two threads calling `BLI_BITMAP_DISABLE` to write bits in the same integer will not get synchronized correctly.

Hans Goudey commented

2022-05-18 21:17:29 +02:00

Ah, that makes sense to me. I wonder if it's better to use atomics for this or to just switch from a bitmap to a boolean array.

Brecht Van Lommel commented

2022-05-18 21:20:53 +02:00

Or just zero the entire bitmap afterwards. It's not ideal in that it involves all vertices when ideally we only work on affected nodes, but this being a bitmap array it's relatively small.

Hans Goudey self-assigned this 2022-05-19 23:33:48 +02:00

WMROssi commented

2022-05-20 00:59:33 +02:00

Added subscriber: @Wesley-Rossi

blender-admin commented

2022-05-20 09:49:29 +02:00

This issue was referenced by b8bd20d7e0

This issue was referenced by b8bd20d7e0238df9a7f558892377f6b304d5ca75

Hans Goudey commented

2022-05-20 09:49:51 +02:00

Changed status from 'Confirmed' to: 'Resolved'

Hans Goudey closed this issue

2022-05-20 09:49:51 +02:00

CHET commented

2022-07-06 21:49:30 +02:00

Why was the topic closed without closing the bug?

Why was the topic closed without closing the bug? ![image.png](https://archive.blender.org/developer/F13263613/image.png) ![blender_7ekQc4IQeY.jpg](https://archive.blender.org/developer/F13263609/blender_7ekQc4IQeY.jpg)

CHET commented

2022-07-06 21:53:19 +02:00

blender_FoqZoWEGst.mp4

[blender_FoqZoWEGst.mp4](https://archive.blender.org/developer/F13263627/blender_FoqZoWEGst.mp4)

CHET commented

2022-07-11 17:55:54 +02:00

Guys, open topic! Problem not resolved!

Julien Kaspar commented

2022-07-12 10:16:41 +02:00

Changed status from 'Resolved' to: 'Confirmed'

Julien Kaspar reopened this issue

2022-07-12 10:16:41 +02:00

Julien Kaspar commented

2022-07-12 10:16:41 +02:00

I can still reproduce this. Both with the included example file and with factory settings. On linux.
@HooglyBoogly Can you investigate?

I can still reproduce this. Both with the included example file and with factory settings. On linux. @HooglyBoogly Can you investigate?

Hans Goudey commented

2022-07-29 17:39:26 +02:00

I've looked into this for a while and found that a PBVH node calculating its vertices' normals can write into the normals of another node. I think ideally each node would affect the normals of its own unique vertices (vertices that are not part of any other node).

~~This change that checks if a vertex is in the node before writing to the normal fixes the problem for me, but adds exponential complexity within each node. I'll work on optimizing it next.~~

diff --git a/source/blender/blenkernel/intern/pbvh.c b/source/blender/blenkernel/intern/pbvh.c
index 6cebcdfea4e..6ff20ad30c1 100644
--- a/source/blender/blenkernel/intern/pbvh.c
+++ b/source/blender/blenkernel/intern/pbvh.c
@@ -1017,71 +1017,84 @@ static void pbvh_update_normals_clear_task_cb(void *__restrict userdata,
 {
   PBVHUpdateData *data = userdata;
   PBVH *pbvh = data->pbvh;
   PBVHNode *node = data->nodes[n];
   float(*vnors)[3] = data->vnors;
 
   if (node->flag & PBVH_UpdateNormals) {
     const int *verts = node->vert_indices;
     const int totvert = node->uniq_verts;
     for (int i = 0; i < totvert; i++) {
       const int v = verts[i];
       if (pbvh->vert_bitmap[v]) {
         zero_v3(vnors[v]);
       }
     }
   }
 }
 
+static bool span_contains(const int *data, const int data_num, const int value)
+{
+  for (int i = 0; i < data_num; i++) {
+    if (data[i] == value) {
+      return true;
+    }
+  }
+  return false;
+}
+
 static void pbvh_update_normals_accum_task_cb(void *__restrict userdata,
                                               const int n,
                                               const TaskParallelTLS *__restrict UNUSED(tls))
 {
   PBVHUpdateData *data = userdata;
 
   PBVH *pbvh = data->pbvh;
   PBVHNode *node = data->nodes[n];
   float(*vnors)[3] = data->vnors;
 
   if (node->flag & PBVH_UpdateNormals) {
     unsigned int mpoly_prev = UINT_MAX;
     float fn[3];
 
     const int *faces = node->prim_indices;
     const int totface = node->totprim;
 
     for (int i = 0; i < totface; i++) {
       const MLoopTri *lt = &pbvh->looptri[faces[i]];
       const unsigned int vtri[3] = {
           pbvh->mloop[lt->tri[0]].v,
           pbvh->mloop[lt->tri[1]].v,
           pbvh->mloop[lt->tri[2]].v,
       };
       const int sides = 3;
 
       /* Face normal and mask */
       if (lt->poly != mpoly_prev) {
         const MPoly *mp = &pbvh->mpoly[lt->poly];
         BKE_mesh_calc_poly_normal(mp, &pbvh->mloop[mp->loopstart], pbvh->verts, fn);
         mpoly_prev = lt->poly;
       }
 
       for (int j = sides; j--;) {
         const int v = vtri[j];
+        if (!span_contains(node->vert_indices, node->uniq_verts, v)) {
+          continue;
+        }
 
         if (pbvh->vert_bitmap[v]) {
           /* NOTE: This avoids `lock, add_v3_v3, unlock`
            * and is five to ten times quicker than a spin-lock.
            * Not exact equivalent though, since atomicity is only ensured for one component
            * of the vector at a time, but here it shall not make any sensible difference. */
           for (int k = 3; k--;) {
             atomic_add_and_fetch_fl(&vnors[v][k], fn[k]);
           }
         }
       }
     }
   }
 }
 
 static void pbvh_update_normals_store_task_cb(void *__restrict userdata,
                                               const int n,
                                               const TaskParallelTLS *__restrict UNUSED(tls))

I'm not sure why this wasn't observed before, which makes me a bit skeptical. Another possible explanation is that not all necessary PBVH nodes are tagged for an update, but situation described above sounds a bit more intuitive, so I'll go with that for now.

~~I've looked into this for a while and found that a PBVH node calculating its vertices' normals can write into the normals of another node. I think ideally each node would affect the normals of its own unique vertices (vertices that are not part of any other node).~~ ~~This change that checks if a vertex is in the node before writing to the normal fixes the problem for me, but adds exponential complexity within each node. I'll work on optimizing it next.~~ ```lines=5 diff --git a/source/blender/blenkernel/intern/pbvh.c b/source/blender/blenkernel/intern/pbvh.c index 6cebcdfea4e..6ff20ad30c1 100644 --- a/source/blender/blenkernel/intern/pbvh.c +++ b/source/blender/blenkernel/intern/pbvh.c @@ -1017,71 +1017,84 @@ static void pbvh_update_normals_clear_task_cb(void *__restrict userdata, { PBVHUpdateData *data = userdata; PBVH *pbvh = data->pbvh; PBVHNode *node = data->nodes[n]; float(*vnors)[3] = data->vnors; if (node->flag & PBVH_UpdateNormals) { const int *verts = node->vert_indices; const int totvert = node->uniq_verts; for (int i = 0; i < totvert; i++) { const int v = verts[i]; if (pbvh->vert_bitmap[v]) { zero_v3(vnors[v]); } } } } +static bool span_contains(const int *data, const int data_num, const int value) +{ + for (int i = 0; i < data_num; i++) { + if (data[i] == value) { + return true; + } + } + return false; +} + static void pbvh_update_normals_accum_task_cb(void *__restrict userdata, const int n, const TaskParallelTLS *__restrict UNUSED(tls)) { PBVHUpdateData *data = userdata; PBVH *pbvh = data->pbvh; PBVHNode *node = data->nodes[n]; float(*vnors)[3] = data->vnors; if (node->flag & PBVH_UpdateNormals) { unsigned int mpoly_prev = UINT_MAX; float fn[3]; const int *faces = node->prim_indices; const int totface = node->totprim; for (int i = 0; i < totface; i++) { const MLoopTri *lt = &pbvh->looptri[faces[i]]; const unsigned int vtri[3] = { pbvh->mloop[lt->tri[0]].v, pbvh->mloop[lt->tri[1]].v, pbvh->mloop[lt->tri[2]].v, }; const int sides = 3; /* Face normal and mask */ if (lt->poly != mpoly_prev) { const MPoly *mp = &pbvh->mpoly[lt->poly]; BKE_mesh_calc_poly_normal(mp, &pbvh->mloop[mp->loopstart], pbvh->verts, fn); mpoly_prev = lt->poly; } for (int j = sides; j--;) { const int v = vtri[j]; + if (!span_contains(node->vert_indices, node->uniq_verts, v)) { + continue; + } if (pbvh->vert_bitmap[v]) { /* NOTE: This avoids `lock, add_v3_v3, unlock` * and is five to ten times quicker than a spin-lock. * Not exact equivalent though, since atomicity is only ensured for one component * of the vector at a time, but here it shall not make any sensible difference. */ for (int k = 3; k--;) { atomic_add_and_fetch_fl(&vnors[v][k], fn[k]); } } } } } } static void pbvh_update_normals_store_task_cb(void *__restrict userdata, const int n, const TaskParallelTLS *__restrict UNUSED(tls)) ``` ~~I'm not sure why this wasn't observed before, which makes me a bit skeptical. Another possible explanation is that not all necessary PBVH nodes are tagged for an update, but situation described above sounds a bit more intuitive, so I'll go with that for now.~~

blender-admin commented

2022-08-02 18:00:38 +02:00

This issue was referenced by efe0e2b183

This issue was referenced by efe0e2b18370972c14383e809719c2b606abc414

Hans Goudey commented

2022-08-02 18:01:35 +02:00

Changed status from 'Confirmed' to: 'Resolved'

Hans Goudey closed this issue

2022-08-02 18:01:35 +02:00

Thomas Dinges added this to the 3.3 LTS milestone 2023-02-08 15:37:46 +01:00

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

Regression: Shading artefacts in sculpt mode #96810