Details

Reviewers

Campbell Barton (campbellbarton)
Sergey Sharybin (sergey)

Maniphest Tasks

T41092: Very high memory usage for Cycles baking compared to ordinary renders

Commits

rBS394c5318c6fe: Bake-API: reduce memory footprint when baking more than one object (Fix T41092)
rC2015345f93a3: Bake-API: reduce memory footprint when baking more than one object (Fix T41092)
rB394c5318c6fe: Bake-API: reduce memory footprint when baking more than one object (Fix T41092)
rB087ee57dbaa2: Bake-API: reduce memory footprint when baking more than one object [test build]

Summary

combine all the highpoly pixel arrays into a single array. I use a global to deliver the correct data to the renderer. Since baking/rendering is a blocking operation no one can bake more than one objcet at the same time anyways.

Diff Detail

Repository: rB Blender

Event Timeline

Dalai Felinto (dfelinto) updated this revision to Diff 2518.Sep 1 2014, 10:04 AM

Dalai Felinto (dfelinto) retitled this revision from to Bake-API: reduce memory footprint when baking more than one object (Fix T41092).

Dalai Felinto (dfelinto) updated this object.

Dalai Felinto (dfelinto) added reviewers: Sergey Sharybin (sergey), Campbell Barton (campbellbarton).

Dalai Felinto (dfelinto) added subscribers: Brecht Van Lommel (brecht), Andy Davies (metalliandy).

For the records, a final solution via this approach should use R instead of G to store baking_object_id

Dalai Felinto (dfelinto) added a task: T41092: Very high memory usage for Cycles baking compared to ordinary renders.Sep 3 2014, 2:42 PM

Bartosz Moniewski (monio) added a subscriber: Bartosz Moniewski (monio).Sep 3 2014, 3:58 PM

marc dion (marcclintdion) added a subscriber: marc dion (marcclintdion).Sep 20 2014, 4:41 AM

Bartosz Moniewski (monio) mentioned this in T41092: Very high memory usage for Cycles baking compared to ordinary renders.Oct 9 2014, 6:09 PM

The biggest thing i don't like about is the usage of Global for such a task. I wouldn't want this G to become worrying about some specific jobs.

source/blender/blenkernel/BKE_global.h
100 ↗	(On Diff #2518)	Don't do this. It's not what G is intended to handle.

This revision now requires changes to proceed.Feb 4 2015, 7:26 PM

Merge remote-tracking branch 'origin/master' into arcpatch-D772

only updating the patch, I may resume working on it soon

Different approach, no longer using G (using Render though)

@Sergey Sharybin (sergey) how about now?

Dalai Felinto (dfelinto) mentioned this in T43553: Cycles dupli object baking issues.Feb 10 2015, 5:18 PM

Closed by commit rB087ee57dbaa2: Bake-API: reduce memory footprint when baking more than one object [test build] (authored by Dalai Felinto (dfelinto)). · Explain WhyFeb 10 2015, 6:07 PM

This revision was automatically updated to reflect the committed changes.

aff, Phabricator and its automatic closing of tasks

Ok, question: is it possible to pass the object_id via python API rather than requesting it from the Cycles side?

Ok, question: is it possible to pass the object_id via python API rather than requesting it from the Cycles side?

I always assumed the object id is not consistent between Blender and Cycles, but I can look at it. I can see how that would be beneficial here.
An option is to have a RE_engine_object_id_get(re, object). Wouldn't that be more safe-proof?

Just an update against latest master, still waiting to hear from @Sergey Sharybin (sergey) (or @Campbell Barton (campbellbarton) perhaps?)

Merge remote-tracking branch 'origin/master' into fix-bake-dupli

Tested this patch for the last day,
works fine and reduces memory usage drastically in many cases, thus avoids many crashes with baking.

@Dalai Felinto (dfelinto), i'm still not sure why it's a function of RenderEngine instead and not an argument of bake() function.

Passing object_id (id of the lookup table) to the renderer (cycles)

@Sergey Sharybin (sergey) you were right, there was no reason not to pass the id for the renderer.

Sergey Sharybin (sergey) added inline comments.Apr 17 2015, 4:59 PM

intern/cycles/blender/blender_session.cpp
501	Is `b_engine` still needed here?
source/blender/editors/object/object_bake_api.c
683	Is it correct that both hi and lo poly arrays are the same size?

From review: b_engine() is no longer necessary
Update bake API doc

LGTM

This revision is now accepted and ready to land.Apr 17 2015, 5:17 PM

Closed by commit rB394c5318c6fe: Bake-API: reduce memory footprint when baking more than one object (Fix T41092) (authored by Dalai Felinto (dfelinto)). · Explain WhyApr 17 2015, 5:29 PM

This revision was automatically updated to reflect the committed changes.

Bake-API: reduce memory footprint when baking more than one object (Fix T41092)
ClosedPublic
Actions

Details

Diff Detail

Event Timeline

Revision Contents
Changeset List

Diff 3456

intern/cycles/blender/blender_session.cpp

source/blender/editors/object/object_bake_api.c

source/blender/makesrna/intern/rna_render.c

source/blender/render/extern/include/RE_bake.h

source/blender/render/intern/include/render_types.h

source/blender/render/intern/source/bake_api.c

source/blender/render/intern/source/external_engine.c

source/blenderplayer/bad_level_call_stubs/stubs.c

Bake-API: reduce memory footprint when baking more than one object (Fix T41092)ClosedPublicActions

Details

Diff Detail

Event Timeline

Revision ContentsChangeset List

Diff 3456

intern/cycles/blender/blender_session.cpp

source/blender/editors/object/object_bake_api.c

source/blender/makesrna/intern/rna_render.c

source/blender/render/extern/include/RE_bake.h

source/blender/render/intern/include/render_types.h

source/blender/render/intern/source/bake_api.c

source/blender/render/intern/source/external_engine.c

source/blenderplayer/bad_level_call_stubs/stubs.c

Bake-API: reduce memory footprint when baking more than one object (Fix T41092)
ClosedPublic
Actions

Revision Contents
Changeset List