02:55 DemiMarie: robclark: is there a timeline on virtio-GPU native contexts for Intel and AMD being upstreamed?
03:36 DemiMarie: Is it safe to not load the HuC/VCN firmware? Will doing so have any side effects beyond rendering the respective engines unusable?
04:02 bcheng: Lynne: the 8 refs assert happens because you are sending providing 8 references in pReferenceSlots when there is only 1 reference
04:05 bcheng: One the first inter frame, I got referenceSlotsCount = 1, and all the pReferenceSlots enteries have slotIndex = 0
04:05 bcheng: s/One/on
04:07 bcheng: I'm pretty sure I didn't see this problem in the v4 patch you sent on ffmpeg-devel previously
04:19 Lynne: pretty sure you tested an older radv version
04:20 Lynne: that assert was addeed later, and if I remove it, decoding works
04:21 Lynne: like 1647: "assert(used_slots == 0x1ff && i == STD_VIDEO_AV1_NUM_REF_FRAMES);"
04:28 bcheng: ah yeah it was an older radv version
04:30 bcheng: in any case, I don't think multiple entries of pReferenceSlots with the same slotIndex and picture resource is valid?
04:30 bcheng: VUID-vkCmdDecodeVideoKHR-pDecodeInfo-07264: Each video picture resource corresponding to the pPictureResource member specified in the elements of pDecodeInfo->pReferenceSlots must be unique within pDecodeInfo->pReferenceSlots
04:39 bcheng: Lynne: the assert reads fine to me. the only time it should be triggered is if slotIndex is reused in pReferenceSlots, which would be illegal for AV1
06:16 Lynne: doesn't that assert the total number of refs used?
08:23 yshui`: has MESA_LOADER_DRIVER_OVERRIDE stopped working? i set it to swrast yet mesa still loads zink
08:25 yshui`: looks like i have to set both LIBGL_ALWAYS_SOFTWARE and MESA_LOADER_DRIVER_OVERRIDE to get it working
13:01 zmike: is this egl or glx
19:20 Lynne: bcheng: btw pretty sure it's the other way around, lr size == real unit size, not the log2 for the spec fix
19:21 Lynne: ah, nevermind
19:21 Lynne: I could have sworn during the meeting that it was the other way around
20:05 sakuska: karolherbst: you are sniffing mwk's fart again and playing big boss big gun as operator? But gotta be honest you are such a big gun like 50line java trace in open source development, no critics with 10years you found a way to fry eggs with graphics cards. This for a steroid munching cripple is big heroic act no doubt.
20:22 sakuska: there is a better method at storing the arrays as data wire format. however that compiler based packer works on data too, you could access elements through the transition graph too, so in that case the transition is just adding some prefix or base to access results.
20:34 sakuska: So yeah i am doing some baby steps towards modern compute paradigm. The plan starts to get into place where to begin the coding cycle, but i was investigating scala methods too, those are also pretty interesting, as to how they make bitsets.
20:35 sakuska: i had not realised such methods before , i do differently, but yeah i see that this might work in fact what martin odersky did.
20:38 sakuska: i only recently did start to think that such modulo and division iteration based thing might work, but i never implemented the code , and bang scala does something that i hypothesized only on.
21:55 sakuska: I myself have not tried any new kernels on toshiba laptop, only on macbook that just has decently stronger cpu it first time installed nicely on mxlinux with a warning, i got a new battery to it, and generally have lots of hw to work with. But i am risking my ass off, so it's my final year, i hope to start coding this month and will get another work to do besides tournaments.
22:06 sakuska: gst467 is being at a prototype/test polygon, a new non-volatile memory.
22:41 sakuska: Anyways transitions are done based of alus per arithmetic ops when the compression has no respective mapping like the integer was not naturally small int and that is what dictionary based compression needs to do, it needs to convert the results into correct ones per every operation.
22:51 sakuska: So that is known and some routines i developed that are known give some starting success, and i mentioned that compiler calculates the combination based of addition, where operand 1 is placed as odd small ints to map powers, and second operand is mapped with even integers or vice versa
22:52 sakuska: that would understand which operand was bigger as needed for all bunch of ops, like div since u/v is not the same as v/u when operands are not same valued.
22:53 sakuska: but yes technically they may have the compiler available through signal collect RDF, it supports some of the sparql
22:54 sakuska: and has integer compression routines, so they might or may have the best code around w3 consortium stuff.
22:55 sakuska: generally we look at the future with quite optimistic tones.
22:57 sakuska: I may fail, as failure i have been, but that is not due to technical limitations, it's just that myself i have been wreck of nerves, always back off to find a calmer situation and tbh. that never comes alike, its some form of guilt based procrastination maybe indeed