08:42 tomeu: pH5: btw, this script has been quite useful in checking my understanding of the bitstream format: https://paste.debian.net/1308739/
09:46 pH5: tomeu: oh nice, thank you. I should take that and make one for v8 as well. so far, my process is running the same data through both my encoder and the blob (with VIV_VX_ENABLE_SAVE_NETWORK_BINARY=1) and aligning the resulting bitstreams.
09:47 pH5: hence the spaghettification in the areas where I'm still unsure.
09:47 tomeu: that's what I did as well, being able to parse what they generate is kind of another way of validating our knowledge/hypotheses
09:48 tomeu: and for big compressed buffers, it can help by summarizing, or by generating a full dump that can be compared
19:13 tomeu: uff, finally got all tests passing with the improved caching
19:13 tomeu: and also improves performance in each model by 0.5ms, compared to the last measuring
19:22 tomeu: I'm quite curious now about what the blob does to be faster
19:23 f_: haha
19:27 f_: always be curious lol
19:27 f_: That said, nice performance improvement!