Mantaflow: changing 'upres' value for noise destroys sim bake (using 'modular' cache) #78243

New Issue

Mark Spink · 2020-06-25T09:19:49+02:00

Mark Spink commented

2020-06-25 09:19:49 +02:00

System Information
Operating system: Windows-10-10.0.18362-SP0 64 Bits
Graphics card: GeForce GTX 1060/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 446.14

Blender Version
Broken: version: 2.83.0, branch: master, commit date: 2020-06-03 14:38, hash: 211b6c29f7
Worked: 2.82a and previous versions

Short description of error
[Any change from the default upres value in Domain settings kills/frees the sim bake]

Exact steps for others to reproduce the error
[Bake the attached, then check 'noise'; if you then change the upres setting at all you will lose the bake, and have to re-bake before you can bake the noise pass.

I don't know if this is a fault or a feature, but it certainly isn't an improvement to the 'modular' cache workflow (if anything it's back to the old 2.79/2.8 smoke sim 'do everything at once, and if it's not what you want- do the whole thing again' type of workflow). And it's not how the upres setting behaved in 2.82a or previous incarnations of Mantaflow in Blender.

I'd say this is a great shame if it is a feature, as the ability to bake the sim, then try different upres settings was, what I previously thought was fantastic, compared to the 'old' smoke sim.

NoiseUpRes_KillsBake_2.83.mp4

Here's the difference between 2.82a (first) and 2.83-

[{F8642311}]

I also tried this in 2.9 Alpha (24th June build) and it behaves the same as 2.83 release version, i.e. not like 2.82a or previous versions.

**System Information** Operating system: Windows-10-10.0.18362-SP0 64 Bits Graphics card: GeForce GTX 1060/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 446.14 **Blender Version** Broken: version: 2.83.0, branch: master, commit date: 2020-06-03 14:38, hash: `211b6c29f7` Worked: 2.82a and previous versions **Short description of error** [Any change from the default upres value in Domain settings kills/frees the sim bake] **Exact steps for others to reproduce the error** [Bake the attached, then check 'noise'; if you then change the upres setting at all you will lose the bake, and have to re-bake before you can bake the noise pass. I don't know if this is a fault or a feature, but it certainly isn't an improvement to the 'modular' cache workflow (*if anything it's back to the old 2.79/2.8 smoke sim 'do everything at once, and if it's not what you want- do the whole thing again' type of workflow*). And it's not how the upres setting behaved in 2.82a or previous incarnations of Mantaflow in Blender. I'd say this is a great shame if it *is* a feature, as the ability to bake the sim, then try different upres settings was, what I previously thought was fantastic, compared to the 'old' smoke sim. [NoiseUpRes_KillsBake_2.83.mp4](https://archive.blender.org/developer/F8642321/NoiseUpRes_KillsBake_2.83.mp4) Here's the difference between 2.82a (first) and 2.83- [{[F8642311](https://archive.blender.org/developer/F8642311/UpresFactorDestroysBake_2.83.blend)}] I also tried this in 2.9 Alpha (24th June build) and it behaves the same as 2.83 release version, i.e. not like 2.82a or previous versions.

Mark Spink commented

2020-06-25 09:19:49 +02:00

Added subscriber: @marks-4

blender-admin commented

2020-06-25 09:19:49 +02:00

#77713 was marked as duplicate of this issue

Philipp Oeser commented

2020-06-25 18:30:48 +02:00

Added subscribers: @sebbas, @lichtwerk

Philipp Oeser commented

2020-06-25 18:30:48 +02:00

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

Philipp Oeser commented

2020-06-25 18:30:48 +02:00

Not totally sure, but to me clearing the sim also seems unneccessary?

@sebbas: shouldnt the following actually be enough?
P1489: #78243 proposed solution



diff --git a/source/blender/makesrna/intern/rna_fluid.c b/source/blender/makesrna/intern/rna_fluid.c
index b4fa791362f..5bb727c739b 100644
--- a/source/blender/makesrna/intern/rna_fluid.c
+++ b/source/blender/makesrna/intern/rna_fluid.c
@@ -1547,7 +1547,7 @@ static void rna_def_fluid_domain_settings(BlenderRNA *brna)
                            "The noise simulation is scaled up by this factor (compared to the "
                            "base resolution of the domain)");
   RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
-  RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_Fluid_domain_reset");
+  RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_Fluid_noisecache_reset");
 
   prop = RNA_def_property(srna, "noise_type", PROP_ENUM, PROP_NONE);
   RNA_def_property_enum_sdna(prop, NULL, "noise_type");
@@ -1555,7 +1555,7 @@ static void rna_def_fluid_domain_settings(BlenderRNA *brna)
   RNA_def_property_ui_text(
       prop, "Noise Method", "Noise method which is used during the high-res simulation");
   RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
-  RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_Fluid_domain_reset");
+  RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_Fluid_noisecache_reset");
 
   prop = RNA_def_property(srna, "use_noise", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flags", FLUID_DOMAIN_USE_NOISE);

It is a bit hard to test atm, since after the VDB cache commits, the noise baking seems to be broken...?
Anyways, I tested this on a commit before 9fe64948ab, and it seems to work fine (I might be missing something though...)

Not totally sure, but to me clearing the sim also seems unneccessary? @sebbas: shouldnt the following actually be enough? [P1489: #78243 proposed solution](https://archive.blender.org/developer/P1489.txt) ``` diff --git a/source/blender/makesrna/intern/rna_fluid.c b/source/blender/makesrna/intern/rna_fluid.c index b4fa791362f..5bb727c739b 100644 --- a/source/blender/makesrna/intern/rna_fluid.c +++ b/source/blender/makesrna/intern/rna_fluid.c @@ -1547,7 +1547,7 @@ static void rna_def_fluid_domain_settings(BlenderRNA *brna) "The noise simulation is scaled up by this factor (compared to the " "base resolution of the domain)"); RNA_def_property_clear_flag(prop, PROP_ANIMATABLE); - RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_Fluid_domain_reset"); + RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_Fluid_noisecache_reset"); prop = RNA_def_property(srna, "noise_type", PROP_ENUM, PROP_NONE); RNA_def_property_enum_sdna(prop, NULL, "noise_type"); @@ -1555,7 +1555,7 @@ static void rna_def_fluid_domain_settings(BlenderRNA *brna) RNA_def_property_ui_text( prop, "Noise Method", "Noise method which is used during the high-res simulation"); RNA_def_property_clear_flag(prop, PROP_ANIMATABLE); - RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_Fluid_domain_reset"); + RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_Fluid_noisecache_reset"); prop = RNA_def_property(srna, "use_noise", PROP_BOOLEAN, PROP_NONE); RNA_def_property_boolean_sdna(prop, NULL, "flags", FLUID_DOMAIN_USE_NOISE); ``` It is a bit hard to test atm, since after the VDB cache commits, the noise baking seems to be broken...? Anyways, I tested this on a commit before 9fe64948ab, and it seems to work fine (I might be missing something though...)

Philipp Oeser commented

2020-06-25 22:59:15 +02:00

In #78243#965562, @lichtwerk wrote:
It is a bit hard to test atm, since after the VDB cache commits, the noise baking seems to be broken...?

see #78272 (Mantaflow - Bake not showing (when noise is used))

> In #78243#965562, @lichtwerk wrote: > It is a bit hard to test atm, since after the VDB cache commits, the noise baking seems to be broken...? see #78272 (Mantaflow - Bake not showing (when noise is used))

Mark Spink commented

2020-07-02 03:05:20 +02:00

Changed status from 'Confirmed' to: 'Resolved'

Mark Spink closed this issue

2020-07-02 03:05:20 +02:00

Mark Spink self-assigned this 2020-07-02 03:05:20 +02:00

Mark Spink commented

2020-07-02 03:05:20 +02:00

This comment was removed by @marks-4

*This comment was removed by @marks-4*

Mark Spink commented

2020-07-02 03:11:07 +02:00

Marked as resolved by mistake! Sorry! Can't change it back to 'open'...

Philipp Oeser commented

2020-07-02 07:20:37 +02:00

Changed status from 'Resolved' to: 'Confirmed'

Philipp Oeser reopened this issue

2020-07-02 07:20:37 +02:00

Iago Mota commented

2020-07-08 15:30:26 +02:00

Added subscriber: @Iago-Diogo-de-Vasconcelos-Mota

Jacques Lucke commented

2020-07-13 15:14:01 +02:00

Added subscribers: @YuraKharitonow, @Arken, @JacquesLucke

Stefano Lazzaroni commented

2020-07-15 12:35:41 +02:00

Added subscriber: @Lazza

Stefano Lazzaroni commented

2020-07-15 12:35:41 +02:00

I've tried today to play with resolutions of particles in liquid and noise in gas and after the commit by @sebbas in https://developer.blender.org/rBdbcc74f8010f685156cf6416454a772f66fb6dfa he seems to have fixed every cache problem in mantaflow. Go take a look and mark as Resolved.

Sebastián Barschkis commented

2020-07-15 12:41:11 +02:00

Changed status from 'Confirmed' to: 'Resolved'

Sebastián Barschkis closed this issue

2020-07-15 12:41:11 +02:00

Sebastián Barschkis commented

2020-07-15 12:41:11 +02:00

Ok good, closing now!

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

Mantaflow: changing 'upres' value for noise destroys sim bake (using 'modular' cache) #78243