Regression: Multires wireframes are displayed incomplete #104915

New Issue

Julien Kaspar · 2023-02-18T12:38:15+01:00

Julien Kaspar commented

2023-02-18 12:38:15 +01:00

System Information
Operating system: Linux-6.1.12-gentoo-x86_64-AMD_Ryzen_Threadripper_1950X_16-Core_Processor-with-glibc2.36 64 Bits
Graphics card: Quadro GP100/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 525.89.02

Blender Version
Broken: version: 3.6.0 Alpha, branch: HEAD (modified), commit date: 2023-02-17 18:54, hash: 8806d7551db3
Worked: (newest version of Blender that worked as expected)

This is apparently more random than expected (once a bisect gave 10354b043f, another run gave 77963ff778 as the culprit)

Short description of error
When adding a multires modifier to an object and subdividing it, the wireframe overlay is not displayed completely. Random edges are missing from the overlay.

Exact steps for others to reproduce the error

Add a cube
Add a multires modifier and subdivide a few times
Enable the wireframe overlay or switch to the wireframe shading mode

**System Information** Operating system: Linux-6.1.12-gentoo-x86_64-AMD_Ryzen_Threadripper_1950X_16-Core_Processor-with-glibc2.36 64 Bits Graphics card: Quadro GP100/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 525.89.02 **Blender Version** Broken: version: 3.6.0 Alpha, branch: HEAD (modified), commit date: 2023-02-17 18:54, hash: `8806d7551db3` Worked: (newest version of Blender that worked as expected) This is apparently more random than expected (once a bisect gave 10354b043f2a6bee589bb85dd67cc51d7218595a, another run gave 77963ff7783c7a97de590fde677321b64d261a7a as the culprit) **Short description of error** When adding a multires modifier to an object and subdividing it, the wireframe overlay is not displayed completely. Random edges are missing from the overlay. ![image](/attachments/8aa5b7f1-dfaf-410f-8ea6-36e028a25928) **Exact steps for others to reproduce the error** - Add a cube - Add a multires modifier and subdivide a few times - Enable the wireframe overlay or switch to the wireframe shading mode

image.png

224 KiB

Julien Kaspar added the

labels 2023-02-18 12:38:16 +01:00

Pratik Borhade commented

2023-02-20 12:09:58 +01:00

@JulienKaspar hi, could you check again in current master?
I'm able to redo this in buildbot build from 15Feb but not in current master, 3.5.

@JulienKaspar hi, could you check again in current master? I'm able to redo this in buildbot build from 15Feb but not in current master, 3.5.

Pratik Borhade added

Status

Needs Information from User

and removed

Status

Needs Triage

labels 2023-02-20 12:10:08 +01:00

Julien Kaspar commented

2023-02-20 13:01:02 +01:00

I'm still able to reproduce it.

Pratik Borhade added

Status

Needs Triage

and removed

Status

Needs Information from User

labels 2023-02-20 13:09:56 +01:00

Pratik Borhade commented

2023-02-20 13:24:12 +01:00

Managed to trigger the issue in latest daily build. Though unable to repro in own build.
@lichtwerk , can you check? This seems a recent regression: c2c62c3618 (good) - c3b53abc10 (bad)

Managed to trigger the issue in latest daily build. Though unable to repro in own build. @lichtwerk , can you check? This seems a recent regression: c2c62c3618ee (good) - c3b53abc109d (bad)

Pratik Borhade added

and removed

labels 2023-02-20 13:24:42 +01:00

Philipp Oeser commented

2023-02-20 14:07:08 +01:00

Will bisect.

Philipp Oeser commented

2023-02-20 14:34:23 +01:00

Seems like this was caused by 77963ff778

CC @pragma37 ^^

Seems like this was caused by 77963ff7783c7a97de590fde677321b64d261a7a CC @pragma37 ^^

Philipp Oeser changed title from ~~Multires wireframes are displayed incomplete~~ to Regression: Multires wireframes are displayed incomplete

2023-02-20 14:35:23 +01:00

Philipp Oeser added

and removed

labels 2023-02-20 14:35:30 +01:00

Miguel Pozo commented

2023-02-20 15:41:46 +01:00

@lichtwerk Could you double-check?
The missing lines are random every time I open Blender.

I can reproduce it after reverting that commit (and even the previous commit referenced by it).
Also, that commit is not related to the Workbench or the Overlay engines, so I really don't see how it could be the cause.

@lichtwerk Could you double-check? The missing lines are random every time I open Blender. I can reproduce it after reverting that commit (and even the previous commit referenced by it). Also, that commit is not related to the Workbench or the Overlay engines, so I really don't see how it could be the cause.

Philipp Oeser commented

2023-02-21 09:54:46 +01:00

@pragma37 : hm, made another run, now bisect points me to 10354b043f

Doesnt really make too much sense either?

I assume this pops up more randomly than expected...

Just to narrow it down @pragma37 : can you also repro in 8703db393bfb?

If not, maybe you can run a bisect?

@pragma37 : hm, made another run, now bisect points me to 10354b043f2a6bee589bb85dd67cc51d7218595a Doesnt really make too much sense either? I assume this pops up more randomly than expected... Just to narrow it down @pragma37 : can you also repro in 8703db393bfb? If not, maybe you can run a bisect?

Miguel Pozo commented

2023-02-21 20:07:26 +01:00

@lichtwerk I actually can't reproduce it in 10354b043f. 🤔

8703db393b crashes on startup for me, but I doubt it's related to this issue.

If not, maybe you can run a bisect?

I fear my processor is not up to the task.
A debug build after a checkout like these ones takes me around 10-15 minutes.

My only tip is to toggle the modifier visibility several times to ensure you didn't just get lucky.

@lichtwerk I actually can't reproduce it in 10354b043f. 🤔 8703db393bfb crashes on startup for me, but I doubt it's related to this issue. > If not, maybe you can run a bisect? I fear my processor is not up to the task. A debug build after a checkout like these ones takes me around 10-15 minutes. My only tip is to toggle the modifier visibility several times to ensure you didn't just get lucky.

Jeroen Bakker commented

2023-02-23 10:49:28 +01:00

Feels like a threading race condition. On my system it is hard to detect as one segment might just be missing. I also noticed that the behavior isn't consistent between each commit. This makes bisecting not consistent:

first bisect try pointer me towards: 32149f8d7a Reverting this change on blender-v3.5-release still shows the issue.
Second time pointer to 284cdbb6cf what makes more sense. Reverting this change on blender-v3.5-release still shows the issue.

Different approach
I added the number of segments that would be drawn. And printed that out in the drawing code. This way I could just click for a minute and after wards detect if there were some issues with inside a batch.

Using a cube with 2 subdivisions have 384 edges, where 96 should be visible.

diff --git a/source/blender/draw/intern/mesh_extractors/extract_mesh_vbo_edge_fac.cc b/source/blender/draw/intern/mesh_extractors/extract_mesh_vbo_edge_fac.cc
index 4360fd16b9d..72ed3e664aa 100644
--- a/source/blender/draw/intern/mesh_extractors/extract_mesh_vbo_edge_fac.cc
+++ b/source/blender/draw/intern/mesh_extractors/extract_mesh_vbo_edge_fac.cc
@@ -165,6 +165,18 @@ static void extract_edge_fac_finish(const MeshRenderData *mr,
 {
   GPUVertBuf *vbo = static_cast<GPUVertBuf *>(buf);
   MeshExtract_EdgeFac_Data *data = static_cast<MeshExtract_EdgeFac_Data *>(_data);
+  {
+    int buf_len = mr->loop_len + mr->loop_loose_len;
+    int c = 0;
+    for (int ml_index = 0; ml_index < buf_len; ml_index++) {
+      if (data->vbo_data[ml_index] != 0) {
+        c++;
+      }
+    }
+    if (c != 384 && c != 96) {
+      printf("%s: %d\n", __func__, c);
+    }
+  }
 
   if (GPU_crappy_amd_driver() || GPU_minimum_per_vertex_stride() > 1) {
     /* Some AMD drivers strangely crash with VBO's with a one byte format.

When switching optimum display I get now the next results.

extract_edge_fac_finish: 94
extract_edge_fac_finish: 92
extract_edge_fac_finish: 94
extract_edge_fac_finish: 94
extract_edge_fac_finish: 94

This VBO extractor is single threaded; what makes me wonder if it is an input data issue, and not a rendering issue.

With this change the bisect points to 7ca651d182, what makes a lot more sense. I am not able to revert this patch to validate correctness.

@HooglyBoogly mind looking into this issue?

Feels like a threading race condition. On my system it is hard to detect as one segment might just be missing. I also noticed that the behavior isn't consistent between each commit. This makes bisecting not consistent: - first bisect try pointer me towards: 32149f8d7a1c277b1816a42a0ee99adb4a253a62 Reverting this change on blender-v3.5-release still shows the issue. - Second time pointer to 284cdbb6cf5061e2cb17ea6ca9f277e94b2bb62f what makes more sense. Reverting this change on blender-v3.5-release still shows the issue. **Different approach** I added the number of segments that would be drawn. And printed that out in the drawing code. This way I could just click for a minute and after wards detect if there were some issues with inside a batch. Using a cube with 2 subdivisions have 384 edges, where 96 should be visible. ``` diff --git a/source/blender/draw/intern/mesh_extractors/extract_mesh_vbo_edge_fac.cc b/source/blender/draw/intern/mesh_extractors/extract_mesh_vbo_edge_fac.cc index 4360fd16b9d..72ed3e664aa 100644 --- a/source/blender/draw/intern/mesh_extractors/extract_mesh_vbo_edge_fac.cc +++ b/source/blender/draw/intern/mesh_extractors/extract_mesh_vbo_edge_fac.cc @@ -165,6 +165,18 @@ static void extract_edge_fac_finish(const MeshRenderData *mr, { GPUVertBuf *vbo = static_cast<GPUVertBuf *>(buf); MeshExtract_EdgeFac_Data *data = static_cast<MeshExtract_EdgeFac_Data *>(_data); + { + int buf_len = mr->loop_len + mr->loop_loose_len; + int c = 0; + for (int ml_index = 0; ml_index < buf_len; ml_index++) { + if (data->vbo_data[ml_index] != 0) { + c++; + } + } + if (c != 384 && c != 96) { + printf("%s: %d\n", __func__, c); + } + } if (GPU_crappy_amd_driver() || GPU_minimum_per_vertex_stride() > 1) { /* Some AMD drivers strangely crash with VBO's with a one byte format. ``` When switching optimum display I get now the next results. ``` extract_edge_fac_finish: 94 extract_edge_fac_finish: 92 extract_edge_fac_finish: 94 extract_edge_fac_finish: 94 extract_edge_fac_finish: 94 ``` This VBO extractor is single threaded; what makes me wonder if it is an input data issue, and not a rendering issue. With this change the bisect points to 7ca651d18274acacfe8e0ecb7bc6fdd3399ea01b, what makes a lot more sense. I am not able to revert this patch to validate correctness. @HooglyBoogly mind looking into this issue?

Hans Goudey was assigned by Jeroen Bakker

2023-02-23 11:01:01 +01:00

Jeroen Bakker added

and removed

labels 2023-02-23 11:01:20 +01:00

Jeroen Bakker added

and removed

labels 2023-02-23 11:04:03 +01:00

Philipp Oeser commented

2023-02-23 11:52:40 +01:00

Could #103706 or #103700 (b247588dc0) be related, too?

In any case, will change back to Modeling (dont think Nodes & Physics fits here)

Could #103706 or #103700 (b247588dc0) be related, too? In any case, will change back to Modeling (dont think Nodes & Physics fits here)

Philipp Oeser added

and removed

labels 2023-02-23 11:52:55 +01:00

Hans Goudey commented

2023-02-23 13:52:14 +01:00

Yeah, a race condition with a bitmap isn't that surprising. I thought I checked that but I might have missed the multires modifier.

Yeah, a race condition with a bitmap isn't _that_ surprising. I thought I checked that but I might have missed the multires modifier.

Hans Goudey added

and removed

labels 2023-02-23 13:52:32 +01:00

Hans Goudey added this to the 3.5 milestone 2023-02-23 22:02:16 +01:00

Hans Goudey referenced a pull request that will close this issue

2023-02-23 22:51:49 +01:00

Fix #104915: Race condition writing subsurf optimal display edges #105156

Hans Goudey commented

2023-02-23 22:52:43 +01:00

Thanks for the report and the investigation everyone. I should have caught this earlier since I've run into a similar issue before. Anyway, it's fixed by #105156.

Hans Goudey referenced this issue from a commit

2023-02-26 23:59:13 +01:00

Fix #104915: Race condition writing subsurf optimal display edges

Hans Goudey closed this issue

2023-02-26 23:59:15 +01:00

Blender Bot added

and removed

labels 2023-02-26 23:59:16 +01:00