Viewport gets stuck when two group instances are emitted from one face using dupli-verts #48913

New Issue

GiantCowFIlms · 2016-07-21T04:37:14+02:00

GiantCowFIlms commented

2016-07-21 04:37:14 +02:00

System Information
Windows 8.1 Enterprise 64-bit
GPU: Nvidia GTX 670

Blender Version
Broken: 2.77a abf6f08

Short description of error
view port render gets stuck in loop of restarting, either during caching the BVH or on the first sample, depending on how large the scene is. Note that F12 render still works.

Gif:

Exact steps for others to reproduce the error

Full reproduction steps (from beginning):
Start with a clean .blend
follow the steps in this post:
http://blender.stackexchange.com/a/58304/3127
then grab the group instance empty, shift-D copy it, and move it over.
Switch to view port render mode, the bug will appear.

Blend file based reproduction steps:
Switch to view port render mode, the bug will appear.

Blend file
Stuck Viewport Demo.blend

**System Information** Windows 8.1 Enterprise 64-bit GPU: Nvidia GTX 670 **Blender Version** Broken: 2.77a abf6f08 **Short description of error** view port render gets stuck in loop of restarting, either during caching the BVH or on the first sample, depending on how large the scene is. Note that F12 render still works. Gif: ![Chat22.gif](https://archive.blender.org/developer/F325695/Chat22.gif) **Exact steps for others to reproduce the error** Full reproduction steps (from beginning): Start with a clean .blend follow the steps in this post: http://blender.stackexchange.com/a/58304/3127 then grab the group instance empty, shift-D copy it, and move it over. Switch to view port render mode, the bug will appear. Blend file based reproduction steps: Switch to view port render mode, the bug will appear. Blend file [Stuck Viewport Demo.blend](https://archive.blender.org/developer/F325699/Stuck_Viewport_Demo.blend)

GiantCowFIlms commented

2016-07-21 04:37:14 +02:00

Changed status to: 'Open'

GiantCowFIlms commented

2016-07-21 04:37:14 +02:00

Added subscriber: @GiantCowFIlms

gandalf3 commented

2016-07-21 04:38:38 +02:00

Added subscriber: @gandalf3

Bastien Montagne commented

2016-07-21 20:08:27 +02:00

Added subscribers: @Sergey, @mont29

Sergey Sharybin was assigned by Bastien Montagne

2016-07-21 20:08:27 +02:00

Bastien Montagne commented

2016-07-21 20:08:27 +02:00

Can confirm the issue…

Cycles keeps aborting its update (synchronize) call back because either session->ready_to_reset() or session->scene->mutex.try_lock() fail (line 813 of blender_session.cpp).

Will let Cycles/depsgraph guy dig further here, @sergey. ;)

Can confirm the issue… Cycles keeps aborting its update (synchronize) call back because either `session->ready_to_reset()` or `session->scene->mutex.try_lock()` fail (line 813 of blender_session.cpp). Will let Cycles/depsgraph guy dig further here, @sergey. ;)

Sergey Sharybin commented

2016-07-28 11:43:56 +02:00

Developer note.

it's nothing really to do with dependency graph or the locking/notification mechanism mentioned here. It is to do with how dupli_object.c generates persistent ID: there are two object parented to the plane, yeah of them duplicates two suzannes. This makes the persistent IDs be the following:

Via Monkey:

0 0 2147483647 2147483647 2147483647 2147483647 2147483647 2147483647
1 0 2147483647 2147483647 2147483647 2147483647 2147483647 2147483647

Via Monkey.001

0 0 2147483647 2147483647 2147483647 2147483647 2147483647 2147483647
1 0 2147483647 2147483647 2147483647 2147483647 2147483647 2147483647

This makes two different instances to have different transformation matrix but same persistent ID, which confuses Cycles.

So this issue is to be solved on object_dupli,c level with some careful thoughts how to avoid unnecessary persistent ID changes when geometry changes over the time.

Developer note. it's nothing really to do with dependency graph or the locking/notification mechanism mentioned here. It is to do with how dupli_object.c generates persistent ID: there are two object parented to the plane, yeah of them duplicates two suzannes. This makes the persistent IDs be the following: ``` Via Monkey: ``` 0 0 2147483647 2147483647 2147483647 2147483647 2147483647 2147483647 1 0 2147483647 2147483647 2147483647 2147483647 2147483647 2147483647 ``` Via Monkey.001 ``` 0 0 2147483647 2147483647 2147483647 2147483647 2147483647 2147483647 1 0 2147483647 2147483647 2147483647 2147483647 2147483647 2147483647 This makes two different instances to have different transformation matrix but same persistent ID, which confuses Cycles. So this issue is to be solved on `object_dupli,c` level with some careful thoughts how to avoid unnecessary persistent ID changes when geometry changes over the time.

blender-admin commented

2016-08-14 15:33:21 +02:00

This issue was referenced by blender/cycles@29de8bb2a8

This issue was referenced by blender/cycles@29de8bb2a8a8fcc6c2138cc8564bbf1bda970e40

blender-admin commented

2016-08-14 15:33:21 +02:00

This issue was referenced by c0161a1bab

This issue was referenced by c0161a1bab71331f518bd5c8c1b091df7ee12074

Brecht Van Lommel commented

2016-08-15 15:27:29 +02:00

Added subscriber: @brecht

Brecht Van Lommel commented

2016-08-15 15:27:30 +02:00

Here's a simple patch that solves it, making the persistent ID longer to take into account the group/base index.
P385: (An Untitled Masterwork)

diff --git a/intern/cycles/blender/blender_util.h b/intern/cycles/blender/blender_util.h
index d5dbaba..e79f2bb 100644
--- a/intern/cycles/blender/blender_util.h
+++ b/intern/cycles/blender/blender_util.h
@@ -698,7 +698,7 @@ protected:
 
 /* Object Key */
 
-enum { OBJECT_PERSISTENT_ID_SIZE = 8 };
+enum { OBJECT_PERSISTENT_ID_SIZE = 16 };
 
 struct ObjectKey {
 	void *parent;
diff --git a/source/blender/blenkernel/intern/object_dupli.c b/source/blender/blenkernel/intern/object_dupli.c
index 21a0246..3a0222b 100644
--- a/source/blender/blenkernel/intern/object_dupli.c
+++ b/source/blender/blenkernel/intern/object_dupli.c
@@ -221,31 +221,39 @@ static void make_child_duplis(const DupliContext *ctx, void *userdata, MakeChild
 
 	if (ctx->group) {
 		unsigned int lay = ctx->group->layer;
+		unsigned int groupid = 0;
 		GroupObject *go;
-		for (go = ctx->group->gobject.first; go; go = go->next) {
+		for (go = ctx->group->gobject.first; go; go = go->next, groupid++) {
 			Object *ob = go->ob;
 
 			if ((ob->lay & lay) && ob != obedit && is_child(ob, parent)) {
+				DupliContext pctx;
+				copy_dupli_context(&pctx, ctx, ctx->object, NULL, groupid, false);
+
 				/* mballs have a different dupli handling */
 				if (ob->type != OB_MBALL)
 					ob->flag |= OB_DONE;  /* doesnt render */
 
-				make_child_duplis_cb(ctx, userdata, ob);
+				make_child_duplis_cb(&pctx, userdata, ob);
 			}
 		}
 	}
 	else {
 		unsigned int lay = ctx->scene->lay;
+		unsigned int baseid = 0;
 		Base *base;
-		for (base = ctx->scene->base.first; base; base = base->next) {
+		for (base = ctx->scene->base.first; base; base = base->next, baseid++) {
 			Object *ob = base->object;
 
 			if ((base->lay & lay) && ob != obedit && is_child(ob, parent)) {
+				DupliContext pctx;
+				copy_dupli_context(&pctx, ctx, ctx->object, NULL, baseid, false);
+
 				/* mballs have a different dupli handling */
 				if (ob->type != OB_MBALL)
 					ob->flag |= OB_DONE;  /* doesnt render */
 
-				make_child_duplis_cb(ctx, userdata, ob);
+				make_child_duplis_cb(&pctx, userdata, ob);
 			}
 		}
 	}
diff --git a/source/blender/makesdna/DNA_object_types.h b/source/blender/makesdna/DNA_object_types.h
index d53c836..d24c7fa 100644
--- a/source/blender/makesdna/DNA_object_types.h
+++ b/source/blender/makesdna/DNA_object_types.h
@@ -331,7 +331,7 @@ typedef struct DupliObject {
 
 	/* persistent identifier for a dupli object, for inter-frame matching of
 	 * objects with motion blur, or inter-update matching for syncing */
-	int persistent_id- [x]; /* MAX_DUPLI_RECUR */
+	int persistent_id[16]; /* 2*MAX_DUPLI_RECUR */
 
 	/* particle this dupli was generated from */
 	struct ParticleSystem *particle_system;

In 2.8 we might consider changing dupliverts and duplifaces so that duplicators use a group or object specified in a property, rather than through parenting. That dependency has always been in the wrong direction I think, and then as a side effect we could shorten the persistent ID again.

Here's a simple patch that solves it, making the persistent ID longer to take into account the group/base index. [P385: (An Untitled Masterwork)](https://archive.blender.org/developer/P385.txt) ```diff diff --git a/intern/cycles/blender/blender_util.h b/intern/cycles/blender/blender_util.h index d5dbaba..e79f2bb 100644 --- a/intern/cycles/blender/blender_util.h +++ b/intern/cycles/blender/blender_util.h @@ -698,7 +698,7 @@ protected: /* Object Key */ -enum { OBJECT_PERSISTENT_ID_SIZE = 8 }; +enum { OBJECT_PERSISTENT_ID_SIZE = 16 }; struct ObjectKey { void *parent; diff --git a/source/blender/blenkernel/intern/object_dupli.c b/source/blender/blenkernel/intern/object_dupli.c index 21a0246..3a0222b 100644 --- a/source/blender/blenkernel/intern/object_dupli.c +++ b/source/blender/blenkernel/intern/object_dupli.c @@ -221,31 +221,39 @@ static void make_child_duplis(const DupliContext *ctx, void *userdata, MakeChild if (ctx->group) { unsigned int lay = ctx->group->layer; + unsigned int groupid = 0; GroupObject *go; - for (go = ctx->group->gobject.first; go; go = go->next) { + for (go = ctx->group->gobject.first; go; go = go->next, groupid++) { Object *ob = go->ob; if ((ob->lay & lay) && ob != obedit && is_child(ob, parent)) { + DupliContext pctx; + copy_dupli_context(&pctx, ctx, ctx->object, NULL, groupid, false); + /* mballs have a different dupli handling */ if (ob->type != OB_MBALL) ob->flag |= OB_DONE; /* doesnt render */ - make_child_duplis_cb(ctx, userdata, ob); + make_child_duplis_cb(&pctx, userdata, ob); } } } else { unsigned int lay = ctx->scene->lay; + unsigned int baseid = 0; Base *base; - for (base = ctx->scene->base.first; base; base = base->next) { + for (base = ctx->scene->base.first; base; base = base->next, baseid++) { Object *ob = base->object; if ((base->lay & lay) && ob != obedit && is_child(ob, parent)) { + DupliContext pctx; + copy_dupli_context(&pctx, ctx, ctx->object, NULL, baseid, false); + /* mballs have a different dupli handling */ if (ob->type != OB_MBALL) ob->flag |= OB_DONE; /* doesnt render */ - make_child_duplis_cb(ctx, userdata, ob); + make_child_duplis_cb(&pctx, userdata, ob); } } } diff --git a/source/blender/makesdna/DNA_object_types.h b/source/blender/makesdna/DNA_object_types.h index d53c836..d24c7fa 100644 --- a/source/blender/makesdna/DNA_object_types.h +++ b/source/blender/makesdna/DNA_object_types.h @@ -331,7 +331,7 @@ typedef struct DupliObject { /* persistent identifier for a dupli object, for inter-frame matching of * objects with motion blur, or inter-update matching for syncing */ - int persistent_id- [x]; /* MAX_DUPLI_RECUR */ + int persistent_id[16]; /* 2*MAX_DUPLI_RECUR */ /* particle this dupli was generated from */ struct ParticleSystem *particle_system; ``` In 2.8 we might consider changing dupliverts and duplifaces so that duplicators use a group or object specified in a property, rather than through parenting. That dependency has always been in the wrong direction I think, and then as a side effect we could shorten the persistent ID again.

Sergey Sharybin commented

2016-08-17 14:52:30 +02:00

@brecht, any reason why not to commit this?

Brecht Van Lommel commented

2016-08-17 22:50:00 +02:00

I haven't had the time to test this properly yet, just wanted to post something so no one wastes time trying to fix this.

Brecht Van Lommel commented

2016-08-18 00:00:10 +02:00

Changed status from 'Open' to: 'Resolved'

Brecht Van Lommel closed this issue

2016-08-18 00:00:10 +02:00

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

Viewport gets stuck when two group instances are emitted from one face using dupli-verts #48913