Hi, On Fri, 2019-05-17 at 16:43 -0400, Nicolas Dufresne wrote: > Le jeudi 16 mai 2019 à 20:45 +0200, Paul Kocialkowski a écrit : > > Hi, > > > > Le jeudi 16 mai 2019 à 14:24 -0400, Nicolas Dufresne a écrit : > > > Le mercredi 15 mai 2019 à 22:59 +0200, Paul Kocialkowski a écrit : > > > > Hi, > > > > > > > > Le mercredi 15 mai 2019 à 14:54 -0400, Nicolas Dufresne a écrit : > > > > > Le mercredi 15 mai 2019 à 19:42 +0200, Paul Kocialkowski a écrit : > > > > > > Hi, > > > > > > > > > > > > Le mercredi 15 mai 2019 à 10:42 -0400, Nicolas Dufresne a écrit : > > > > > > > Le mercredi 15 mai 2019 à 12:09 +0200, Paul Kocialkowski a écrit : > > > > > > > > Hi, > > > > > > > > > > > > > > > > With the Rockchip stateless VPU driver in the works, we now have a > > > > > > > > better idea of what the situation is like on platforms other than > > > > > > > > Allwinner. This email shares my conclusions about the situation and how > > > > > > > > we should update the MPEG-2, H.264 and H.265 controls accordingly. > > > > > > > > > > > > > > > > - Per-slice decoding > > > > > > > > > > > > > > > > We've discussed this one already[0] and Hans has submitted a patch[1] > > > > > > > > to implement the required core bits. When we agree it looks good, we > > > > > > > > should lift the restriction that all slices must be concatenated and > > > > > > > > have them submitted as individual requests. > > > > > > > > > > > > > > > > One question is what to do about other controls. I feel like it would > > > > > > > > make sense to always pass all the required controls for decoding the > > > > > > > > slice, including the ones that don't change across slices. But there > > > > > > > > may be no particular advantage to this and only downsides. Not doing it > > > > > > > > and relying on the "control cache" can work, but we need to specify > > > > > > > > that only a single stream can be decoded per opened instance of the > > > > > > > > v4l2 device. This is the assumption we're going with for handling > > > > > > > > multi-slice anyway, so it shouldn't be an issue. > > > > > > > > > > > > > > My opinion on this is that the m2m instance is a state, and the driver > > > > > > > should be responsible of doing time-division multiplexing across > > > > > > > multiple m2m instance jobs. Doing the time-division multiplexing in > > > > > > > userspace would require some sort of daemon to work properly across > > > > > > > processes. I also think the kernel is better place for doing resource > > > > > > > access scheduling in general. > > > > > > > > > > > > I agree with that yes. We always have a single m2m context and specific > > > > > > controls per opened device so keeping cached values works out well. > > > > > > > > > > > > So maybe we shall explicitly require that the request with the first > > > > > > slice for a frame also contains the per-frame controls. > > > > > > > > > > > > > > - Annex-B formats > > > > > > > > > > > > > > > > I don't think we have really reached a conclusion on the pixel formats > > > > > > > > we want to expose. The main issue is how to deal with codecs that need > > > > > > > > the full slice NALU with start code, where the slice_header is > > > > > > > > duplicated in raw bitstream, when others are fine with just the encoded > > > > > > > > slice data and the parsed slice header control. > > > > > > > > > > > > > > > > My initial thinking was that we'd need 3 formats: > > > > > > > > - One that only takes only the slice compressed data (without raw slice > > > > > > > > header and start code); > > > > > > > > - One that takes both the NALU data (including start code, raw header > > > > > > > > and compressed data) and slice header controls; > > > > > > > > - One that takes the NALU data but no slice header. > > > > > > > > > > > > > > > > But I no longer think the latter really makes sense in the context of > > > > > > > > stateless video decoding. > > > > > > > > > > > > > > > > A side-note: I think we should definitely have data offsets in every > > > > > > > > case, so that implementations can just push the whole NALU regardless > > > > > > > > of the format if they're lazy. > > > > > > > > > > > > > > I realize that I didn't share our latest research on the subject. So a > > > > > > > slice in the original bitstream is formed of the following blocks > > > > > > > (simplified): > > > > > > > > > > > > > > [nal_header][nal_type][slice_header][slice] > > > > > > > > > > > > Thanks for the details! > > > > > > > > > > > > > nal_header: > > > > > > > This one is a header used to locate the start and the end of the of a > > > > > > > NAL. There is two standard forms, the ANNEX B / start code, a sequence > > > > > > > of 3 bytes 0x00 0x00 0x01, you'll often see 4 bytes, the first byte > > > > > > > would be a leading 0 from the previous NAL padding, but this is also > > > > > > > totally valid start code. The second form is the AVC form, notably used > > > > > > > in ISOMP4 container. It simply is the size of the NAL. You must keep > > > > > > > your buffer aligned to NALs in this case as you cannot scan from random > > > > > > > location. > > > > > > > > > > > > > > nal_type: > > > > > > > It's a bit more then just the type, but it contains at least the > > > > > > > information of the nal type. This has different size on H.264 and HEVC > > > > > > > but I know it's size is in bytes. > > > > > > > > > > > > > > slice_header: > > > > > > > This contains per slice parameters, like the modification lists to > > > > > > > apply on the references. This one has a size in bits, not in bytes. > > > > > > > > > > > > > > slice: > > > > > > > I don't really know what is in it exactly, but this is the data used to > > > > > > > decode. This bit has a special coding called the anti-emulation, which > > > > > > > prevents a start-code from appearing in it. This coding is present in > > > > > > > both forms, ANNEX-B or AVC (in GStreamer and some reference manual they > > > > > > > call ANNEX-B the bytestream format). > > > > > > > > > > > > > > So, what we notice is that what is currently passed through Cedrus > > > > > > > driver: > > > > > > > [nal_type][slice_header][slice] > > > > > > > > > > > > > > This matches what is being passed through VA-API. We can understand > > > > > > > that stripping off the slice_header would be hard, since it's size is > > > > > > > in bits. Instead we pass size and header_bit_size in slice_params. > > > > > > > > > > > > True, there is that. > > > > > > > > > > > > > About Rockchip. RK3288 is a Hantro G1 and has a bit called > > > > > > > start_code_e, when you turn this off, you don't need start code. As a > > > > > > > side effect, the bitstream becomes identical. We do now know that it > > > > > > > works with the ffmpeg branch implement for cedrus. > > > > > > > > > > > > Oh great, that makes life easier in the short term, but I guess the > > > > > > issue could arise on another decoder sooner or later. > > > > > > > > > > > > > Now what's special about Hantro G1 (also found on IMX8M) is that it > > > > > > > take care for us of reading and executing the modification lists found > > > > > > > in the slice header. Mostly because I very disliked having to pass the > > > > > > > p/b0/b1 parameters, is that Boris implemented in the driver the > > > > > > > transformation from the DPB entries into this p/b0/b1 list. These list > > > > > > > a standard, it's basically implementing 8.2.4.1 and 8.2.4.2. the > > > > > > > following section is the execution of the modification list. As this > > > > > > > list is not modified, it only need to be calculated per frame. As a > > > > > > > result, we don't need these new lists, and we can work with the same > > > > > > > H264_SLICE format as Cedrus is using. > > > > > > > > > > > > Yes but I definitely think it makes more sense to pass the list > > > > > > modifications rather than reconstructing those in the driver from a > > > > > > full list. IMO controls should stick to the bitstream as close as > > > > > > possible. > > > > > > > > > > For Hantro and RKVDEC, the list of modification is parsed by the IP > > > > > from the slice header bits. Just to make sure, because I myself was > > > > > confused on this before, the slice header does not contain a list of > > > > > references, instead it contains a list modification to be applied to > > > > > the reference list. I need to check again, but to execute these > > > > > modification, you need to filter and sort the references in a specific > > > > > order. This should be what is defined in the spec as 8.2.4.1 and > > > > > 8.2.4.2. Then 8.2.4.3 is the process that creates the l0/l1. > > > > > > > > > > The list of references is deduced from the DPB. The DPB, which I thinks > > > > > should be rename as "references", seems more useful then p/b0/b1, since > > > > > this is the data that gives use the ability to implementing glue in the > > > > > driver to compensate some HW differences. > > > > > > > > > > In the case of Hantro / RKVDEC, we think it's natural to build the HW > > > > > specific lists (p/b0/b1) from the references rather then adding HW > > > > > specific list in the decode_params structure. The fact these lists are > > > > > standard intermediate step of the standard is not that important. > > > > > > > > Sorry I got confused (once more) about it. Boris just explained the > > > > same thing to me over IRC :) Anyway my point is that we want to pass > > > > what's in ffmpeg's short and long term ref lists, and name them that > > > > instead of dpb. > > > > > > > > > > > Now, this is just a start. For RK3399, we have a different CODEC > > > > > > > design. This one does not have the start_code_e bit. What the IP does, > > > > > > > is that you give it one or more slice per buffer, setup the params, > > > > > > > start decoding, but the decoder then return the location of the > > > > > > > following NAL. So basically you could offload the scanning of start > > > > > > > code to the HW. That being said, with the driver layer in between, that > > > > > > > would be amazingly inconvenient to use, and with Boyer-more algorithm, > > > > > > > it is pretty cheap to scan this type of start-code on CPU. But the > > > > > > > feature that this allows is to operate in frame mode. In this mode, you > > > > > > > have 1 interrupt per frame. > > > > > > > > > > > > I'm not sure there is any interest in exposing that from userspace and > > > > > > my current feeling is that we should just ditch support for per-frame > > > > > > decoding altogether. I think it mixes decoding with notions that are > > > > > > higher-level than decoding, but I agree it's a blurry line. > > > > > > > > > > I'm not worried about this either. We can already support that by > > > > > copying the bitstream internally to the driver, though zero-copy with > > > > > this would require a new format, the one we talked about, > > > > > SLICE_ANNEX_B. > > > > > > > > Right, but what I'm thinking about is making that the one and only > > > > format. The rationale is that it's always easier to just append a start > > > > code from userspace if needed. And we need a bit offset to the slice > > > > data part anyway, so it doesn't hurt to require a few extra bits to > > > > have the whole thing that will work in every situation. > > > > > > What I'd like is to eventually allow zero-copy (aka userptr) into the > > > driver. If you make the start code mandatory, any decoding from ISOMP4 > > > (.mp4, .mov) will require a full bitstream copy in userspace to add the > > > start code (unless you hack your allocation in your demuxer, but it's a > > > bit complicated since this code might come from two libraries). In > > > ISOMP4, you have an AVC header, which is just the size of the NAL that > > > follows. > > > > Well, I think we have to do a copy from system memory to the buffer > > allocated by v4l2 anyway. Our hardware pipelines can reasonably be > > expected not to have any MMU unit and not allow sg import anyway. > > The Rockchip has an mmu. You need one copy at least indeed, Is the MMU in use currently? That can make things troublesome if we run into a case where the VPU has MMU and deals with scatter-gather while the display part doesn't. As far as I know, there's no way for userspace to know whether a dma-buf-exported buffer is backed by CMA or by scatter-gather memory. This feels like a major issue for using dma- buf, since userspace can't predict whether a buffer exported on one device can be imported on another when building its pipeline. > e.g. file > to mem, or udpsocket to mem. But right now, let's say with ffmpeg/mpeg- > ts, first you need to copy the MPEG TS to mem, then to demux you copy > that H264 stream to another buffer, you then copy in the parser, > removing the start-code and finally copy in the accelerator, adding the > start code. If the driver would allow userptr, it would be unusable. > > GStreamer on the other side implement lazy conversion, so it would copy > the mpegts to mem, copy to demux, aggregate (with lazy merging) in the > parser (but stream format is negotiation, so it keeps the start-code). > If you request alignment=au, you have full frame of buffers, so if your > driver could do userptr, you can same that extra copy. > > Now, if we demux an MP4 it's the same, the parser will need do a full > copy instead of lazy aggregation in order to prepend the start code > (since it had an AVC header). But userptr could save a copy. > > If the driver requires no nal prefix, then we could just pass a > slightly forward point to userptr and avoid ACV to ANNEX-B conversion, > which is a bit slower (even know it's nothing compare to the full > copies we already do. > > That was my argument in favour for no NAL prefix in term of efficiency, > and it does not prevent adding a control to enable start-code for cases > it make sense. I see, so the internal arcitecture of userspace software may not be a good fit for adding these bits and it could hurt performance a bit. That feels like a significant downside. > > So with that in mind, asking userspace to add a startcode it already > > knows doesn't seem to be asking too much. > > > > > On the other end, the data_offset thing is likely just a thing for the > > > RK3399 to handle, it does not affect RK3288, Cedrus or IMX8M. > > > > Well, I think it's best to be fool-proof here and just require that > > start code. We should also have per-slice bit offsets to the different > > parts anyway, so drivers that don't need it can just ignore it. > > > > In extreme cases where there is some interest in doing direct buffer > > import without doing a copy in userspace, userspace could trick the > > format and avoid a copy by not providing the start-code (assuming it > > knows it doesn't need it) and specifying the bit offsets accordingly. > > That'd be a hack for better performance, and it feels better to do > > things in this order rather than having to hack around in the drivers > > that need the start code in every other case. > > So basically, you and Tomas are both strongly in favour of adding > ANNEX-B start-code to the current uAPI. I have digged into Cedrus > registers, and it seems that it does have start-code scanning support. > I'm not sure it can do "full-frame" decoding, 1 interrupt per frame > like the RK do. That requires the IP to deal with the modifications > lists, which are per slices. Actually the bitstream parser won't reconfigure the pipeline configuration registers, it's only around for userspace to avoid implementing bitstream parsing, but it's a standalone thing. So if we want to do full-frame decoding we always need to reconfigure our pipeline (or do it like we do currently and just use one of the per-slice configuration and hope for the best). Do we have more information on the RK3399 and what it requires exactly? (Just to make sure it's not another issue altogether.) > My question is, are you willing to adapt the Cedrus driver to support > receiving start-code ? And will this have a performance impact or not ? > On RK side, it's really just about flipping 1 bit. > > On the Rockchip side, Tomas had concern about CPU wakeup and the fact > that we didn't aim at supporting passing multiple slices at once to the > IP (something RK supports). It's important to understand that multi- > slice streams are relatively rare and mostly used for low-latency / > video conferencing. So aggregating in these case defeats the purpose of > using slices. So I think RK feature is not very important. Agreed, let's aim for low-latency as a standard. > Of course, I do believe that long term we will want to expose bot > stream formats on RK (because the HW can do that), so then userspace > can just pick the best when available. So that boils down to our first > idea, shall we expose _SLICE_A and _SLICE_B or something like this ? > Now that we have progressed on the matter, I'm quite in favour of > having _SLICE in the first place, with the preferred format that > everyone should support, and allow for variants later. Now, if we make > one mandatory, we could also just have a menu control to allow other > formats. That seems fairly reasonable to me, and indeed, having one preferred format at first seems to be a good move. > > > > To me the breaking point was about having the slice header both in raw > > > > bitstream and parsed forms. Since we agree that's fine, we might as > > > > well push it to its logical conclusion and include all the bits that > > > > can be useful. > > > > > > To take your words, the bits that contain useful information starts > > > from the NAL type byte, exactly were the data was cut by VA-API and the > > > current uAPI. > > > > Agreed, but I think that the advantages of always requiring the start > > code outweigh the potential (yet quite unlikely) downsides. > > > > > > > > > But it also support slice mode, with an > > > > > > > interrupt per slice, which is what we decided to use. > > > > > > > > > > > > Easier for everyone and probably better for latency as well :) > > > > > > > > > > > > > So in this case, indeed we strictly require on start-code. Though, to > > > > > > > me this is not a great reason to make a new fourcc, so we will try and > > > > > > > use (data_offset = 3) in order to make some space for that start code, > > > > > > > and write it down in the driver. This is to be continued, we will > > > > > > > report back on this later. This could have some side effect in the > > > > > > > ability to import buffers. But most userspace don't try to do zero-copy > > > > > > > on the encoded size and just copy anyway. > > > > > > > > > > > > > > To my opinion, having a single format is a big deal, since userspace > > > > > > > will generally be developed for one specific HW and we would endup with > > > > > > > fragmented support. What we really want to achieve is having a driver > > > > > > > interface which works across multiple HW, and I think this is quite > > > > > > > possible. > > > > > > > > > > > > I agree with that. The more I think about it, the more I believe we > > > > > > should just pass the whole [nal_header][nal_type][slice_header][slice] > > > > > > and the parsed list in every scenario. > > > > > > > > > > What I like of the cut at nal_type, is that there is only format. If we > > > > > cut at nal_header, then we need to expose 2 formats. And it makes our > > > > > API similar to other accelerator API, so it's easy to "convert" > > > > > existing userspace. > > > > > > > > Unless we make that cut the single one and only true cut that shall > > > > supersed all other cuts :) > > > > > > That's basically what I've been trying to do, kill this _RAW/ANNEX_B > > > thing and go back to our first idea. > > > > Right, in the end I think we should go with: > > V4L2_PIX_FMT_MPEG2_SLICE > > V4L2_PIX_FMT_H264_SLICE > > V4L2_PIX_FMT_HEVC_SLICE > > > > And just require raw bitstream for the slice with emulation-prevention > > bits included. > > That's should be the set of format we start with indeed. The single > format for which software gets written and tested, making sure software > support is not fragmented, and other variants should be something to > opt-in. Cheers for that! Paul > > Cheers, > > > > Paul > > > > > > > > For H.265, our decoder needs some information from the NAL type too. > > > > > > We currently extract that in userspace and stick it to the > > > > > > slice_header, but maybe it would make more sense to have drivers parse > > > > > > that info from the buffer if they need it. On the other hand, it seems > > > > > > quite common to pass information from the NAL type, so maybe we should > > > > > > either make a new control for it or have all the fields in the > > > > > > slice_header (which would still be wrong in terms of matching bitstream > > > > > > description). > > > > > > > > > > Even in userspace, it's common to just parse this in place, it's a > > > > > simple mask. But yes, if we don't have it yet, we should expose the NAL > > > > > type, it would be cleaner. > > > > > > > > Right, works for me. > > > > > > Ack. > > > > > > > Cheers, > > > > > > > > Paul > > > > > > > > > > > > - Dropping the DPB concept in H.264/H.265 > > > > > > > > > > > > > > > > As far as I could understand, the decoded picture buffer (DPB) is a > > > > > > > > concept that only makes sense relative to a decoder implementation. The > > > > > > > > spec mentions how to manage it with the Hypothetical reference decoder > > > > > > > > (Annex C), but that's about it. > > > > > > > > > > > > > > > > What's really in the bitstream is the list of modified short-term and > > > > > > > > long-term references, which is enough for every decoder. > > > > > > > > > > > > > > > > For this reason, I strongly believe we should stop talking about DPB in > > > > > > > > the controls and just pass these lists agremented with relevant > > > > > > > > information for userspace. > > > > > > > > > > > > > > > > I think it should be up to the driver to maintain a DPB and we could > > > > > > > > have helpers for common cases. For instance, the rockchip decoder needs > > > > > > > > to keep unused entries around[2] and cedrus has the same requirement > > > > > > > > for H.264. However for cedrus/H.265, we don't need to do any book- > > > > > > > > keeping in particular and can manage with the lists from the bitstream > > > > > > > > directly. > > > > > > > > > > > > > > As discusses today, we still need to pass that list. It's being index > > > > > > > by the HW to retrieve the extra information we have collected about the > > > > > > > status of the reference frames. In the case of Hantro, which process > > > > > > > the modification list from the slice header for us, we also need that > > > > > > > list to construct the unmodified list. > > > > > > > > > > > > > > So the problem here is just a naming problem. That list is not really a > > > > > > > DPB. It is just the list of long-term/short-term references with the > > > > > > > status of these references. So maybe we could just rename as > > > > > > > references/reference_entry ? > > > > > > > > > > > > What I'd like to pass is the diff to the references list, as ffmpeg > > > > > > currently provides for v4l2 request and vaapi (probably vdpau too). No > > > > > > functional change here, only that we should stop calling it a DPB, > > > > > > which confuses everyone. > > > > > > > > > > Yes. > > > > > > > > > > > > > - Using flags > > > > > > > > > > > > > > > > The current MPEG-2 controls have lots of u8 values that can be > > > > > > > > represented as flags. Using flags also helps with padding. > > > > > > > > It's unlikely that we'll get more than 64 flags, so using a u64 by > > > > > > > > default for that sounds fine (we definitely do want to keep some room > > > > > > > > available and I don't think using 32 bits as a default is good enough). > > > > > > > > > > > > > > > > I think H.264/HEVC per-control flags should also be moved to u64. > > > > > > > > > > > > > > Make sense, I guess bits (member : 1) are not allowed in uAPI right ? > > > > > > > > > > > > Mhh, even if they are, it makes it much harder to verify 32/64 bit > > > > > > alignment constraints (we're dealing with 64-bit platforms that need to > > > > > > have 32-bit userspace and compat_ioctl). > > > > > > > > > > I see, thanks. > > > > > > > > > > > > > - Clear split of controls and terminology > > > > > > > > > > > > > > > > Some codecs have explicit NAL units that are good fits to match as > > > > > > > > controls: e.g. slice header, pps, sps. I think we should stick to the > > > > > > > > bitstream element names for those. > > > > > > > > > > > > > > > > For H.264, that would suggest the following changes: > > > > > > > > - renaming v4l2_ctrl_h264_decode_param to v4l2_ctrl_h264_slice_header; > > > > > > > > > > > > > > Oops, I think you meant slice_prams ? decode_params matches the > > > > > > > information found in SPS/PPS (combined?), while slice_params matches > > > > > > > the information extracted (and executed in case of l0/l1) from the > > > > > > > slice headers. > > > > > > > > > > > > Yes you're right, I mixed them up. > > > > > > > > > > > > > That being said, to me this name wasn't confusing, since > > > > > > > it's not just the slice header, and it's per slice. > > > > > > > > > > > > Mhh, what exactly remains in there and where does it originate in the > > > > > > bitstream? Maybe it wouldn't be too bad to have one control per actual > > > > > > group of bitstream elements. > > > > > > > > > > > > > > - killing v4l2_ctrl_h264_decode_param and having the reference lists > > > > > > > > where they belong, which seems to be slice_header; > > > > > > > > > > > > > > There reference list is only updated by userspace (through it's DPB) > > > > > > > base on the result of the last decoding step. I was very confused for a > > > > > > > moment until I realize that the lists in the slice_header are just a > > > > > > > list of modification to apply to the reference list in order to produce > > > > > > > l0 and l1. > > > > > > > > > > > > Indeed, and I'm suggesting that we pass the modifications only, which > > > > > > would fit a slice_header control. > > > > > > > > > > I think I made my point why we want the dpb -> references. I'm going to > > > > > validate with the VA driver now, to see if the references list there is > > > > > usable with our code. > > > > > > > > > > > Cheers, > > > > > > > > > > > > Paul > > > > > > > > > > > > > > I'm up for preparing and submitting these control changes and updating > > > > > > > > cedrus if they seem agreeable. > > > > > > > > > > > > > > > > What do you think? > > > > > > > > > > > > > > > > Cheers, > > > > > > > > > > > > > > > > Paul > > > > > > > > > > > > > > > > [0]: https://lkml.org/lkml/2019/3/6/82 > > > > > > > > [1]: https://patchwork.linuxtv.org/patch/55947/ > > > > > > > > [2]: https://chromium.googlesource.com/chromiumos/third_party/kernel/+/4d7cb46539a93bb6acc802f5a46acddb5aaab378 > > > > > > > > -- Paul Kocialkowski, Bootlin Embedded Linux and kernel engineering https://bootlin.com
