Attendance:

Others:

Discussion:

• NVIDIA GPU streaming:

•  

  • Mark: GPU compressor hasn’t been open sourced because of concern that the format might change. I’m not privy to details. Discussions are going on with a major client, but seem to be almost done. 

  • Mark: Would come with some CMake configuration settings, so dependency on CUDA would be optional. We recognize there are issues with distro builds.

  • Larry: With OSL builds, Intel came up with a way to build a variety of different compile configurations, then load the appropriate shared object at runtime depending on available hardware. Not sure that’s the best general solution.

  • Peter: Is it wise to have the library use the GPU without telling you? Would make it easier to manage the build. 

  • Mark: The big thing is GPU Direct Storage, direct to GPU storage.

• PR 174 Enabling SSE4 support:

•  

  • Larry: Need to look further, but probably a pretty good idea. Looks OK.

  • Peter: Builds if it can, but does runtime detection.

  • Larry: It’s just a CPU id.

  • Larry: Somebody should do the same for ARM. Some people package up the intrinsic into a function for convenience. Without that, changes will touch many places in the code, so a bit messy.

• Apple optimizations:

•  

  • Joseph: talked to Michael Johnson, he pointed some people to the to TSC invite, didn’t work apparently.

  • Joseph: Quote from Michael: “I’ve asked the people who did the rewrite to engage.” So whatever they’ve done might not merge easily with the new core changes.