Didn't test all formats but, saving the default cube at 1080p
Someone will have to take a closer look, gut feeling says It's likely a latency/iops related issue, I suspect the fast formats likely do fewer large writes while the slower formats do frequent…
- No idea, this could be diagnosed by running
msiexec /i blender-4.2.0-windows-x64.msi /l* blender.logand looking at the generated blender.log
2+3. The store version will use a sandboxed…
I'd probably place just the add_definitions(-DNOMINMAX) back where they went missing but use the new target for everything else, as to not upset things too much and deal with a min/max cleanup…
intern/mikktspaceseems fine to changeintern/mantaflowdoesn't use TBB just does the work so the headers from extern/mantaflow don't throw a fitextern/mantaflowuses-DTBB=1…
-DNOMINMAX went missing here, which may come back to haunt you, my plan was to add this define to the platform flags so all files would have it, but have not done so since some warns popped up about double defines in places that also tried dealing with this problem themselves, which i had no time to resolve just yet.
can't this just depend on bf::dependencies::optional::tbb ? ideally there should be no more instances of ${TBB_INCLUDE_DIRS} and ${TBB_LIBRARIES} through out our cmakefiles, except for the file that defines the ``bf::dependencies::optional::tbb` target
the PUGIXML_HEADER_ONLY doesn't work according to sebastian, we'll need #define PUGIXML_API __attribute__((visibility("hidden"))) but be sure to put it in a guard as not to trip up MSVC
with BITCODE=OFF it appears to build fine with no further python issues, it warns about our cuda version
-- CUDA SDK install folder found: C:\Program Files\NVIDIA GPU…I'm afraid not, given we also build with BITCODE=ON it'll still call bakeKernel and run into python issues there.
Windows and linux are fundamentally not all that different, except for the make helper, couple of pointers here though:
- we build the deps on vs 2019 - v16.11.26 , it has seen no other more…
Projects not building for Debug is generally handled in the main CMakeLists.txt
The marking of the script as executable can be handled using the PATCH_COMMAND as it only needs to be executed once.
by setting just CONFIGURE_COMMAND and not BUILD_COMMAND cmake will just call make to build the project which is problematic since the PLATFORM_ALT_GENERATOR is set to ninja.
Only a single after_install is allowed, these two will need to be merge or cmake will error out.
This doesn't work on windows, it requires set PYTHON_BIN=${PYTHON_BINARY} this probably be fixed by sticking it in the HIPRT_WORKAROUND section above however since that has been moved to a patch step and....