RFD 1137 details: What the encode measured
Throughput
Measured on this desk, 16-thread CPU, one job:
| stage | quantity |
|---|---|
| input | 96 frames, 1024 by 1024, packed 8-bit RGBA |
| raw size | 402.7 MB, which is 96 x 1024 x 1024 x 4 |
| output | 23.4 MiB, CineForm RGBA_4444, quality 3 |
| ratio | 16.8 to 1 |
| encoder time | 554 ms |
| rate | about 173 frames per second |
| playback | 8 fps, so 12 seconds of clip |
The clip is 12 seconds and took 0.554 s to make, which is 21 times realtime. interactor-cineform publishes 112 fps at 1920 by 1080 for RGB; this is a smaller frame with an alpha channel, and the two figures sit either side of the pixel-count scaling as expected.
Against the render the encode disappears. The same 96 frames took 32.3 s on the card and would take 2 h 5 min at llvm’s 78 s a frame.
What the clip does to a frame hash
Every render writes a sidecar carrying a sha256 of its PNG. CineForm is visually lossless and is not bit-exact, so a frame decoded from the clip will not reproduce that digest. Nothing is wrong with either, and the .cff says which artefact the hash names.
Two ways to keep a check, and neither is chosen here:
- Hash the container. One digest for the sweep, which answers “is this the clip we made” and not “is this frame the frame we rendered”.
- Keep the PNGs as the archival form and treat the clip as the viewing copy. Costs the storage the clip was meant to save.
The build, and what it cost to find
Four traps, each of which produced a wrong answer rather than an error.
repo init inside 3-interactor/interactor-cineform re-pointed the WORKSPACE manifest at that project. It reports only “repo has been initialized in C:-keypoint”. Recovery: the manifests git keeps its history, git -C .repo/manifests reflog names the branch, and repo init -u .../weftspun-keypoint -b <branch> -m default.xml restores it. repo list | wc -l is the check.
up.py under nohup builds the library and then starts the encoder as its own child, so both exit together. The next command then waits on a service that is not there and prints nothing at all.
iceoryx2’s Windows PAL prints FindNextFileA ... there are no more files and RemoveDirectoryA ... the directory is not empty during normal startup. They look like failures and are not.
The composing manifest was incomplete. service-cineform/default.xml listed the two programs and the bus, but a manifest root’s own list is not read when the project is one of ninety-four, so a workspace sync gave the programs and nothing they link. The root manifest now carries iceoryx2, cineform-sdk, libwebm and FTXUI under the service, and both builds take them through the cache paths the CMakeLists expose.