Sculpting with multiresolution modifier in 4.1 crashes #120187

New Issue

Mike-Roberts-2 · 2024-04-02T21:17:07+02:00

Mike-Roberts-2 commented

2024-04-02 21:17:07 +02:00

System Information
Operating system: Windows 11
Graphics card: Nvidia RTX 3060 with latest drivers

Blender Version
Broken: Blender 4.1
Worked: Blender 4

Whenever I try to sculpt with multiresolution modifier on an object at about 8 million polygons or above in 4.1 it crashes every time. Still works just fine in Blender 4.

Step 1: Add multiresolution modifier to object.
Step 2: Subdivide several times so you get to around 8 million or so polygons.
Step 3: Sculpt for about 30 seconds to a minute and it will crash.

Tried it on a second computer with Windows 10 and Nvidia RTX 3050 and had same issue in 4.1

**System Information** Operating system: Windows 11 Graphics card: Nvidia RTX 3060 with latest drivers **Blender Version** Broken: Blender 4.1 Worked: Blender 4 Whenever I try to sculpt with multiresolution modifier on an object at about 8 million polygons or above in 4.1 it crashes every time. Still works just fine in Blender 4. Step 1: Add multiresolution modifier to object. Step 2: Subdivide several times so you get to around 8 million or so polygons. Step 3: Sculpt for about 30 seconds to a minute and it will crash. Tried it on a second computer with Windows 10 and Nvidia RTX 3050 and had same issue in 4.1

Mike-Roberts-2 added the

labels 2024-04-02 21:17:08 +02:00

Jaimi McEntire referenced this issue

2024-04-02 21:21:58 +02:00

When sculpting with multires modifier, Randomize causes crash #120188

Germano Cavalcante commented

2024-04-02 22:31:36 +02:00

Can you check the latest official release, or the latest daily build?

Please try the latest daily build: https://builder.blender.org/download/

I was able to confirm some crashes in development versions of 4.1, but not in the last one.

Which object would be best to replicate the issue? (Cube, Monkey...)

How many subdivisions did you test? (6, 7)

Can you check the latest official release, or the latest daily build? Please try the latest daily build: https://builder.blender.org/download/ I was able to confirm some crashes in development versions of 4.1, but not in the last one. Which object would be best to replicate the issue? (Cube, Monkey...) How many subdivisions did you test? (6, 7)

Germano Cavalcante added

Status

Needs Information from User

and removed

Status

Needs Triage

labels 2024-04-02 22:31:45 +02:00

Mike-Roberts-2 commented

2024-04-02 23:50:58 +02:00

Tried today's latest build and still does it when I get to level 6. Tried it on sphere and multiple custom models on two different computers. If I go back to blender 4 I don't have the problem and everything works great.

Germano Cavalcante added

Status

Needs Triage

and removed

Status

Needs Information from User

labels 2024-04-03 00:04:25 +02:00

YimingWu commented

2024-04-03 08:16:07 +02:00

Hi @Mike-Roberts-2 Any chance you could get us the crash log?

On my machine (debug build) simply by subdividing the multires level to 6 will cause a crash in sculpt mode. For me it crashed in create_index_grids where args.grid_indices has a _size of 0. Not sure if that's the exact problem you are having

Hi @Mike-Roberts-2 Any chance you could get us the [crash log](https://docs.blender.org/manual/en/latest/troubleshooting/crash.html#windows)? On my machine (debug build) simply by subdividing the multires level to 6 will cause a crash in sculpt mode. For me it crashed in `create_index_grids` where `args.grid_indices` has a `_size` of 0. Not sure if that's the exact problem you are having

YimingWu commented

2024-04-03 08:43:08 +02:00

Will mark this as Confirmed, Looks to me that at least 40a5e739e2 is running correctly.

~~Will bisect.~~ Apparently I can't reproduce now on latest 8252208955 ... Could you guys try latest build and see again?

Will mark this as Confirmed, Looks to me that at least 40a5e739e270 is running correctly. ~~Will bisect.~~ Apparently I can't reproduce now on latest 82522089550c67be1abf ... Could you guys try latest build and see again?

YimingWu added

Module

Sculpt, Paint & Texture

and removed

labels 2024-04-03 08:43:32 +02:00

YimingWu added

Priority

Normal

Status

Needs Information from User

and removed

labels 2024-04-03 08:52:01 +02:00

Philipp Oeser commented

2024-04-03 09:34:48 +02:00

Just noting something similar has been reported in #120080

@ChengduLittleA : can you repro tat as well?

Just noting something similar has been reported in #120080 @ChengduLittleA : can you repro tat as well?

YimingWu commented

2024-04-03 12:10:40 +02:00

@lichtwerk Nope in latest main I can't make sculpt+multires it to really crash even at ~8 million polys. Are you able to reproduce?

Philipp Oeser commented

2024-04-03 12:23:50 +02:00

@lichtwerk Nope in latest main I can't make sculpt+multires it to really crash even at ~8 million polys. Are you able to reproduce?

#120080 (comment)

> @lichtwerk Nope in latest main I can't make sculpt+multires it to really crash even at ~8 million polys. Are you able to reproduce? https://projects.blender.org/blender/blender/issues/120080#issuecomment-1158692

YimingWu commented

2024-04-03 12:35:27 +02:00

Still unable to crash it in windows using 4.1 or 4.2... Tried extensive sculpting and also the randomize option. 🤔

Here's the system info I have on windows:

Still unable to crash it in windows using 4.1 or 4.2... Tried extensive sculpting and also the randomize option. 🤔 Here's the system info I have on windows:

system-info.txt

21 KiB

Germano Cavalcante commented

2024-04-03 16:49:41 +02:00

After absentmindedly sculpting a sphere with multiresolution subdivided 6 times for a few minutes, I witnessed the crash in Blender 4.1

I tried to replicate with a 4.1 debug build, but as soon as I entered sculpt mode with the sphere, I came across 5 asserts launched simultaneously:

BLI_assert failed: D:\BlenderDev\blender\source\blender\blenlib\BLI_span.hh:318, first(), at 'size_ > 0'

(Reminds me of #120179 which was recently fixed... cc. @HooglyBoogly ).

So I tested it with a debug build of main (assert resolved) and after 6 minutes of sculpting I had the crash:

Visual Studio ended up crashing too, so I couldn't copy the call stack, but I remember it was something in sculpt's UNDO system.

It appears to be a problem where Blender runs out of memory.

It seems like something really difficult to bisect...

After absentmindedly sculpting a sphere with multiresolution subdivided 6 times for a few minutes, I witnessed the crash in Blender 4.1 I tried to replicate with a 4.1 debug build, but as soon as I entered sculpt mode with the sphere, I came across 5 asserts launched simultaneously: ``` BLI_assert failed: D:\BlenderDev\blender\source\blender\blenlib\BLI_span.hh:318, first(), at 'size_ > 0' ``` (Reminds me of #120179 which was recently fixed... cc. @HooglyBoogly ). So I tested it with a debug build of main (assert resolved) and after 6 minutes of sculpting I had the crash: ![image](/attachments/3b821dbd-580d-4a1b-978d-215a3480c2d0) Visual Studio ended up crashing too, so I couldn't copy the call stack, but I remember it was something in sculpt's UNDO system. It appears to be a problem where Blender runs out of memory. It seems like something really difficult to bisect...

image.png

15 KiB

Germano Cavalcante commented

2024-04-03 18:37:35 +02:00

While bisecting, I was able to identify a pattern.
In bad commits, performance is greatly affected.
So, it was easy to identify which commits were bad and which were good, just checking the lag.

The commit causing the issue is 4b427e1b57
@HooglyBoogly ^

This is the call_stack:

>	blender.exe!blender::ed::sculpt_paint::undo::store_coords(Object * ob, blender::ed::sculpt_paint::undo::Node * unode) Line 1297	C++
 	blender.exe!tbb::interface7::internal::delegated_function<`blender::ed::sculpt_paint::undo::push_node'::`2'::<lambda_1> const ,void>::operator()() Line 97	C++
 	[External Code]	
 	blender.exe!tbb::interface7::internal::isolate_impl<void,`blender::ed::sculpt_paint::undo::push_node'::`2'::<lambda_1> const>(const blender::ed::sculpt_paint::undo::push_node::__l2::<lambda_1> & f) Line 216	C++
 	blender.exe!blender::ed::sculpt_paint::undo::push_node(Object * ob, PBVHNode * node, blender::ed::sculpt_paint::undo::Type type) Line 1524	C++
 	blender.exe!do_brush_action_task(Object * ob, const Brush * brush, PBVHNode * node) Line 3290	C++
 	blender.exe!blender::FunctionRef<void __cdecl(blender::IndexRange)>::callback_fn<`do_brush_action'::`55'::<lambda_1> const>(__int64 callable, blender::IndexRange <params_0>) Line 96	C++
 	blender.exe!tbb::interface9::internal::dynamic_grainsize_mode<tbb::interface9::internal::adaptive_mode<tbb::interface9::internal::auto_partition_type> >::work_balance<tbb::interface9::internal::start_for<tbb::blocked_range<__int64>,`blender::threading::detail::parallel_for_impl'::`5'::<lambda_1>,tbb::auto_partitioner const >,tbb::blocked_range<__int64> >() Line 442	C++
 	blender.exe!tbb::interface9::internal::start_for<tbb::blocked_range<__int64>,`blender::threading::detail::parallel_for_impl'::`5'::<lambda_1>,tbb::auto_partitioner const >::execute() Line 143	C++
 	[External Code]

This is the line when it crashed:

unode->normal[index] = float3(CCG_elem_offset_no(&key, elem, i));

While bisecting, I was able to identify a pattern. In bad commits, performance is greatly affected. So, it was easy to identify which commits were bad and which were good, just checking the lag. The commit causing the issue is 4b427e1b57 @HooglyBoogly ^ This is the call_stack: ``` > blender.exe!blender::ed::sculpt_paint::undo::store_coords(Object * ob, blender::ed::sculpt_paint::undo::Node * unode) Line 1297 C++ blender.exe!tbb::interface7::internal::delegated_function<`blender::ed::sculpt_paint::undo::push_node'::`2'::<lambda_1> const ,void>::operator()() Line 97 C++ [External Code] blender.exe!tbb::interface7::internal::isolate_impl<void,`blender::ed::sculpt_paint::undo::push_node'::`2'::<lambda_1> const>(const blender::ed::sculpt_paint::undo::push_node::__l2::<lambda_1> & f) Line 216 C++ blender.exe!blender::ed::sculpt_paint::undo::push_node(Object * ob, PBVHNode * node, blender::ed::sculpt_paint::undo::Type type) Line 1524 C++ blender.exe!do_brush_action_task(Object * ob, const Brush * brush, PBVHNode * node) Line 3290 C++ blender.exe!blender::FunctionRef<void __cdecl(blender::IndexRange)>::callback_fn<`do_brush_action'::`55'::<lambda_1> const>(__int64 callable, blender::IndexRange <params_0>) Line 96 C++ blender.exe!tbb::interface9::internal::dynamic_grainsize_mode<tbb::interface9::internal::adaptive_mode<tbb::interface9::internal::auto_partition_type> >::work_balance<tbb::interface9::internal::start_for<tbb::blocked_range<__int64>,`blender::threading::detail::parallel_for_impl'::`5'::<lambda_1>,tbb::auto_partitioner const >,tbb::blocked_range<__int64> >() Line 442 C++ blender.exe!tbb::interface9::internal::start_for<tbb::blocked_range<__int64>,`blender::threading::detail::parallel_for_impl'::`5'::<lambda_1>,tbb::auto_partitioner const >::execute() Line 143 C++ [External Code] ``` This is the line when it crashed: ``` unode->normal[index] = float3(CCG_elem_offset_no(&key, elem, i)); ```

Germano Cavalcante closed this issue

2024-04-03 18:37:38 +02:00

Germano Cavalcante reopened this issue

2024-04-03 18:38:18 +02:00

Germano Cavalcante added

and removed