3 problems here

I gave win10 in a vm a spin and there's definitely different behavior between win7 and 10, before we land this i'd like to see if i cat atleast explain whats going on.

I currently don't have an nvidia card so can't check, but can you see if switching this option to maximum performance (for just blender, second tab in that screenshot, not global) solves the issue when blender is running alone?

Yeah i did the same thing, edit the scene, minimize, right click on the button and select close, with this patch blender pops up, asks if i want to save, (and then proceeds not to quit, but thats a different bug)

I don't have win10 to test on but this seems to work on win7

That's ok, i'll close it for you

Just pulled, it's fixed now.

Please follow our submission template and guidelines and make a complete, valid bug report, with the required info, precise description of the issue, precise steps to reproduce it, small and simple .blend and/or other files to do so if needed, etc.

Spend some more time on this, and i either end up with something half-assed or start re-writing the whole thing. I'll shelf the idea for now until i have a good amount of time to do it properly

Looking good now, just turn it on by default if someone doesn't like it, they can turn it off on their own.

The rename from WINDOWS_USE_VISUAL_STUDIO_FOLDERS to WINDOWS_USE_VISUAL_STUDIO_PROJECT_FOLDERS missed the one n build_files\cmake\platform\platform_win32.cmake and one in intern\cycles\cmake\macros.cmake breaking the feature completly (back to a giant list of projects in the root now)

BF_Freestyle looks a little nicer now, I'm not minding it as much anymore, @Harley Acheson (harley) you have any opinion here having this on/off by default?

The screenshot looked great, once applied I found myself struggling to find files in BF_FreeStyle though (it's a maze in there) so there is something to be said for a nice flat layout.

This is not the bug tracker for UPBGE please report issues to their tracker available here

• Fix capitalization of some of the includes.

Meh, don't worry about it.

Brecht got me curious about cycles preprocessing, I ran the build twice, once with Preprocess to file and once regular just to get an idea how much time is spend in the preprocessor, there was no option for not writing out the result so the preprocessor segment has some extra cost for IO and the actual savings would be lower than what is shown here.

In D2606#114565, @Sergey Sharybin (sergey) wrote:

2 year ago when I made this diff for freestyle, I actually also looked at ceres as a candidate for PCH, the results were rather inconclusive and disappointing so I didn't pursue it further

Indeed inconclusive =\

Yeah but given the release and debug build times are so far apart, it doesn't seem likely a whole lot of time is being spent in the front end of the compiler where PCH's would help.

In D2606#114548, @Sergey Sharybin (sergey) wrote:

That brings it down to about 11% of the total build time, it's without a doubt an improvement, and if I can't have it in svn I'll happily compromise on this option.

Well, good we are moving towards compromise :) Still feels weird. I think i've got different percentage here, but i'm also on Linux.
Wack glog into a PCH? :D Could also help Libmv and, possibly, some areas of Cycles? Eigen could also slow down compilation.

I setup the blender_precompile_headers macro generic enough that other projects can use it if needed, but it's better to look at those in a case by case basis.

In D2606#114532, @Sergey Sharybin (sergey) wrote:

Better and more acceptable solution is to disable WITH_LIBMV_SCHUR_SPECIALIZATIONS from blender_full.cmake, and have it only enabled for final releases. having those disabled during development is a good idea anyway, no matter if you're working directly on Libmv/Ceres, or if you don't even plan to touch those.

Cycles won't benefit from precompiled headers all that much, the heaviest items in cycles (kernels) spend most of the build time the optimizer stages due to the crazy amount of inlining we do, making the front end (parsing) cost almost neglectable in comparison.

That is something i wouldn't accept. You are trying to optimize work which is done automatically by a compiler and move more work which is to be done manually by those who use and maintain the library.

Also when looking at this in the past, PCH speeds things up if lots and lots of compile units all include the same files

when excluding the intern/ headers

Actually, this *IS* still worth doing, build time in ms

You can't statically link glibc like the other deps, because some services on linux like nss require dynamic linking to operate properly.

It does we ship 30+ dependencies, the thing is for technical reasons you can't ship glibc.

• Emit proper formatted VC error messages with source information.

This caused the cycles_denoise test denoise_passes to fail not sure if it needs a reference image update or something else is wrong since all differences are in the dark part of the image, can you take a look @Lukas Stockner (lukasstockner97) ?

I opted to just direct people to use output pane/build since the actual errors coming out of a single off by one padding error doing it that way are... yeahhhh....

Please do not triage your own ticket.

It's really weird that Windows users who don't have glibc installed don't have to change OS to run Blender,

unsure, but lets ping @Brecht Van Lommel (brecht) and @Aaron Carlisle (Blendify) , one of them will have an idea on how to handle this.

Please also think of having a patch ready for the manual for when this lands

I looked a little further into this, the tests are failing because the geometry generated is actually different between x86 and x64 (errors out during CD_MPOLY compare in customdata_compare) the result doesn't look wrong, up to you if you want to keep it at medium.

Not sure who to assign here @Antonio Vazquez (antoniov) is the goto GP guy, but @Brecht Van Lommel (brecht) added the test.

found it, the culprit seems to be the ccl_try_align(16) on float3 is a no-op on x86/windows

The actual tests are failing, the memleak is probably just a symptom.

@Campbell Barton (campbellbarton) mind taking a peek here?