Enable UsdMtlx for next Gaffer release

[JeremyBooks]

Hello,

I am wondering if you guys could add the "PXR_ENABLE_MATERIALX_SUPPORT" CMAKE flag when compiling USD for the next release of Gaffer. This would be very beneficial for us, and would be great to avoid having to compile Gaffer ourselves just for this flag.

[johnhaddon]

Hi Jeremy, we're already trying to squeeze quite a lot into the upcoming 1.3 release, but in principle adding MaterialX support in USD sounds fine, so we'll see what we can do...

[JeremyBooks]

that's fair enough, sounds great. Thank you John

[johnhaddon]

I'm taking a look at this now, and hit some roadbumps. First, MaterialX hardcodes the C++ standard to 11, whereas OIIO requires 14 minimum - hacked around that. Second, MaterialX 1.38.7 (the latest) isn't compatible with USD 23.05 (also the latest) - looks like MaterialX broke API compatibility in a patch release. I got 1.38.6 to compile instead, but I don't seem to be able to get usdview to show anything useful. I'm testing with the OpenChessSet asset, and I just get this error over and over, and a black screen :

Warning: in _ValidateCompilation at line 227 of /home/john/dev/gafferDependencies/USD/working/USD-23.05/pxr/imaging/hdSt/glslProgram.cpp -- Failed to compile shader (FRAGMENT_SHADER): 0(4091) : error C1503: undefined variable "mx_surface_transmission"

Any advice? What do you need this for?

[JeremyBooks]

I believe that would work for us, definitely worth a shot! We can also look into usdView for this as well and let you know if we are able to come up with anything.

[johnhaddon]

OK, I think I have something working now. But it will go into Gaffer 1.3 which won't be released until July at the earliest. We could make some test builds in the interim though, if that helps you out.

[JeremyBooks]

Would it be a lot to ask for a test build? We are not in a big rush at this very moment, but as i'm sure you are aware that, can always change rather quickly, so it would be ideal to know if this will work for us sooner than later.

[johnhaddon]

Yep, that should be fine. At a minimum, you can always download a build artifact from one of our regular CI builds without us needing to do anything extra at all. And if you want something a little more official, we could make a beta release of 1.3.

[JeremyBooks]

yeah a build artifact from your CI should do just fine

[johnhaddon]

Once the CI is finished, there will be build artifacts available from this PR : #5345. My testing with it so far indicates that MaterialX shaders will load in usdview and be previewed (extremely slowly), and that they try to load into Gaffer, but that we might need to fix some stuff on our side until that's useful. Let us know how it goes for you...

[JeremyBooks]

great, thank you John! I will check back on the CI in a couple hours and let you know how it goes

[JeremyBooks]

Hey John, I was just testing the build from your PR and it looks like there is a couple library issues in this build? it seems it was updated to Imath 3.1 and Half library is missing completely. I am not entirely sure if there is other libraries missing/excluded from this build. I am a little confused by this since i would have assumed the CI tests would fail when trying to import the packages that are linked to these libraries.

My error:ERROR : startup/gui/menus.py : Error loading Arnold module - "libImath-2_4.so.24: cannot open shared object file: No such file or directory".

I completely removed our builds of GafferArnold, however i did compile them against Imath 3.1 with no luck, the so files in the build are still looking for IMath 2.4 and Half (2.4?)

[johnhaddon]

The Half library no longer exists with OpenEXR/Imath 3 - it has been subsumed into Imath. From https://github.com/AcademySoftwareFoundation/Imath/blob/main/docs/PortingGuide2-3.md :

To further simplify matters, the new Imath library includes the half data type directly, rather than maintaining it in a separate library.

I've checked and those artifacts run OK here, and I can't find anything in them linking to libHalf or Imath-2.4 - as you say, CI would fail if they did. I suspect you may not have completely removed your builds. For debugging this sort of thing, I usually set LD_DEBUG=files, which will spit out what is trying to be loaded and when, and for who.

[JeremyBooks]

oh man, sorry! I had a build a while ago with some incorrect configuration and it left a hidden build of GafferArnold in the same lib directory as my IECoreUSD build... sorry about that! glad nothing is wrong