Attendance:

Guests:

Discussion:

• Rod: new gdeflate thing in Microsoft DirectStorage.  They didn’t do the thing we asked for, which was straight up zip support.

• Kimball: not sure. Haven’t heard from Mark or Steve at NVIDIA yet. They’ve open sourced the decompression stuff, which has gdeflate and zstd.

• Rod: This may be more on the microsoft side. My preference is for standard exr’s to be read this way. 

• Kimball: Current decoding in the core library can do the decompression on the gpu, but you still have the swizzle cost. 

• Rod: But you have to know what you’re doing.

• Rod: 5.1 support for dealing with tiled mipmapped exr, pulling in only the parts that you need. But they’re stored uncompressed, tiles and mipmaps, so you have to be selective about how they’re viewed.

• Kimball: Aras’s blog. We’re doing swizzling of the 16 bits. It would be great to have a performance suite to measure this stuff. If we add gdeflate, zstd, blosc, we should have a performance suite.

• RGB: We have some images. CitySample, city rendered at 8K driving down the street. And forensic pictures to illustrate where the tiles are. We could contribute. 

• Kimball: Could go into the DPEL library.

• Larry: Image streaming, did you wedge things like tile size?

• Rod: We did a little with tile size. But we had a two-pronged approach: 1) Read the thing as is, with a projection map to see if you could see it, and 2) virtual texturing, occlusion testing so you don’t read things that are occluded.

• Rod: Can you send me who to contact?

• Kimball: Eric Enderton, TSC chair of DPEL. They have the budget.

• Kimball: What else for OpenEXR 4, a change in file format? 

•  

  • Larry: A few people have suggested if we repartition the file, reading could be optimized. 

  • Kimball: We could do that with the random tile order.

  • Larry: But make it standard, so you don’t have to know so much.

  • Larry: We’d also talked about metadata-only. Other data types.

  • Kimball: support for bfloat.

  • Peter: a flag to indicate that a value is not normalized. 

  • Kimball: We already have uint32.

  • Peter: a bitmask to tell you what range you’re using.

• Lutz: We’re still interested in gdefate, GPU-side decompression.

• Larry: David Morin has also called on the TAC to assemble a working group on virtual production.

• Rod: Regarding virtual production, OpenEXR, OCIO, OTIO are involved, it would be weird for it to be its own thing.