<div dir="ltr"><div><div>I'm actually testing that out that today. Example prototype patch: <a href="https://github.com/lostgoat/linux/commit/c9d88d409d8655d63aa8386edc66687a2ba64a12">https://github.com/lostgoat/linux/commit/c9d88d409d8655d63aa8386edc66687a2ba64a12</a> </div><div>My goal is to first implement this approach, then slowly work my way towards the HW level optimizations. </div><div>The problem I expect to see with this approach is that there will still be unpredictably long latencies depending on what has been committed to the HW rings. </div><div>But it is definitely a good start. </div><div> </div>Regards, </div>Andres </div><div class="gmail_extra"> <div class="gmail_quote">On Fri, Dec 23, 2016 at 11:20 AM, Bridgman, John <<a href="mailto:John.Bridgman@amd.com" target="_blank">John.Bridgman@amd.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> <div dir="ltr"> <div id="m_2404866291218945997divtagdefaultwrapper" style="font-size:12pt;color:#000000;font-family:Calibri,Arial,Helvetica,sans-serif" dir="ltr"> One question I just remembered - the amdgpu driver includes some scheduler logic which maintains per-process queues and therefore avoids loading up the primary ring with a ton of work. Has there been any experimentation with injecting priorities at that level rather than jumping straight to HW-level changes ? <div style="color:rgb(0,0,0)"> <hr style="display:inline-block;width:98%"> <div id="m_2404866291218945997divRplyFwdMsg" dir="ltr">From: amd-gfx <<a href="mailto:amd-gfx-bounces@lists.freedesktop.org" target="_blank">amd-gfx-bounces@lists.freedesktop.org</a>> on behalf of Andres Rodriguez <<a href="mailto:andresx7@gmail.com" target="_blank">andresx7@gmail.com</a>> Sent: December 23, 2016 11:13 AM To: Koenig, Christian Cc: Zhou, David(ChunMing); Huan, Alvin; Mao, David; Sagalovitch, Serguei; <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank">amd-gfx@lists.freedesktop.org</a>; Andres Rodriguez; Pierre-Loup A. Griffais; Zhang, Hawking<div><div class="h5"> Subject: Re: [RFC] Mechanism for high priority scheduling in amdgpu</div></div> <div> </div> </div><div><div class="h5"> <div> <div dir="ltr"> <div>Hey Christian, <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> But yes, in general you don't want another compositor in the way, so we'll be acquiring the HMD display directly, separate from any desktop or display server.</blockquote> <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> </blockquote> Assuming that the the HMD is attached to the rendering device in some way you have the X server and the Compositor which both try to be DRM master at the same time. Please correct me if that was fixed in the meantime, but that sounds like it will simply not work. Or is this what Andres mention below Dave is working on ?. </blockquote> </div> You are correct on both statements. We can't have two DRM_MASTERs, so the current DRM+X does not support this use case. And this what Dave and Pierre-Loup are currently working on. <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> Additional to that a compositor in combination with X is a bit counter productive when you want to keep the latency low. </blockquote> <div> </div> <div>One thing I'd like to correct is that our main goal is to get latency _predictable_, secondary goal is to make it low. </div> <div>The high priority queue feature addresses our main source of unpredictability: the scheduling latency when the hardware is already full of work from the game engine. </div> <div>The DirectMode feature addresses one of the latency sources: multiple (unnecessary) context switches to submit a surface to the DRM driver. <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> Targeting something like Wayland and when you need X compatibility XWayland sounds like the much better idea. </blockquote> <div> </div> <div>We are pretty enthusiastic about Wayland (and really glad to see Fedora 25 use Wayland by default). Once we have everything working nicely under X (where most of the users are currently), I'm sure Pierre-Loup will be pushing us to get everything optimized under Wayland as well (which should be a lot simpler!). </div> <div>Ever since working with SurfaceFlinger on Android with explicit fencing I've been waiting for the day I can finally ditch X altogether :) </div> <div>Regards, </div> <div>Andres </div> </div> </div> <div class="gmail_extra"> <div class="gmail_quote">On Fri, Dec 23, 2016 at 5:54 AM, Christian König <<a href="mailto:christian.koenig@amd.com" target="_blank">christian.koenig@amd.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> But yes, in general you don't want another compositor in the way, so we'll be acquiring the HMD display directly, separate from any desktop or display server. </blockquote> Assuming that the the HMD is attached to the rendering device in some way you have the X server and the Compositor which both try to be DRM master at the same time. Please correct me if that was fixed in the meantime, but that sounds like it will simply not work. Or is this what Andres mention below Dave is working on ?. Additional to that a compositor in combination with X is a bit counter productive when you want to keep the latency low. E.g. the "normal" flow of a GL or Vulkan surface filled with rendered data to be displayed is from the Application -> X server -> compositor -> X server. The extra step between X server and compositor just means extra latency and for this use case you probably don't want that. Targeting something like Wayland and when you need X compatibility XWayland sounds like the much better idea. Regards, Christian. <div class="m_2404866291218945997HOEnZb"> <div class="m_2404866291218945997h5"> Am 22.12.2016 um 20:54 schrieb Pierre-Loup A. Griffais: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Display concerns are a separate issue, and as Andres said we have other plans to address. But yes, in general you don't want another compositor in the way, so we'll be acquiring the HMD display directly, separate from any desktop or display server. Same with security, we can have a separate conversation about that when the time comes. On 12/22/2016 08:41 AM, Serguei Sagalovitch wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Andres, Did you measure latency, etc. impact of __any__ compositor? My understanding is that VR has pretty strict requirements related to QoS. Sincerely yours, Serguei Sagalovitch On 2016-12-22 11:35 AM, Andres Rodriguez wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Hey Christian, We are currently interested in X, but with some distros switching to other compositors by default, we also need to consider those. We agree, running the full vrcompositor in root isn't something that we want to do. Too many security concerns. Having a small root helper that does the privilege escalation for us is the initial idea. For a long term approach, Pierre-Loup and Dave are working on dealing with the "two compositors" scenario a little better in DRM+X. Fullscreen isn't really a sufficient approach, since we don't want the HMD to be used as part of the Desktop environment when a VR app is not in use (this is extremely annoying). When the above is settled, we should have an auth mechanism besides DRM_MASTER or DRM_AUTH that allows the vrcompositor to take over the HMD permanently away from X. Re-using that auth method to gate this IOCTL is probably going to be the final solution. I propose to start with ROOT_ONLY since it should allow us to respect kernel IOCTL compatibility guidelines with the most flexibility. Going from a restrictive to a more flexible permission model would be inclusive, but going from a general to a restrictive model may exclude some apps that used to work. Regards, Andres On 12/22/2016 6:42 AM, Christian König wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Hi Andres, well using root might cause stability and security problems as well. We worked quite hard to avoid exactly this for X. We could make this feature depend on the compositor being DRM master, but for example with X the X server is master (and e.g. can change resolutions etc..) and not the compositor. So another question is also what windowing system (if any) are you planning to use? X, Wayland, Flinger or something completely different ? Regards, Christian. Am 20.12.2016 um 16:51 schrieb Andres Rodriguez: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Hi Christian, That is definitely a concern. What we are currently thinking is to make the high priority queues accessible to root only. Therefore is a non-root user attempts to set the high priority flag on context allocation, we would fail the call and return ENOPERM. Regards, Andres On 12/20/2016 7:56 AM, Christian König wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> BTW: If there is non-VR application which will use high-priority h/w queue then VR application will suffer. Any ideas how to solve it? </blockquote> Yeah, that problem came to my mind as well. Basically we need to restrict those high priority submissions to the VR compositor or otherwise any malfunctioning application could use it. Just think about some WebGL suddenly taking all our rendering away and we won't get anything drawn any more. Alex or Michel any ideas on that? Regards, Christian. Am 19.12.2016 um 15:48 schrieb Serguei Sagalovitch: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> > If compute queue is occupied only by you, the efficiency > is equal with setting job queue to high priority I think. The only risk is the situation when graphics will take all needed CUs. But in any case it should be very good test. Andres/Pierre-Loup, Did you try to do it or it is a lot of work for you? BTW: If there is non-VR application which will use high-priority h/w queue then VR application will suffer. Any ideas how to solve it? Sincerely yours, Serguei Sagalovitch On 2016-12-19 12:50 AM, zhoucm1 wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Do you encounter the priority issue for compute queue with current driver? If compute queue is occupied only by you, the efficiency is equal with setting job queue to high priority I think. Regards, David Zhou On 2016年12月19日 13:29, Andres Rodriguez wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Yes, vulkan is available on all-open through the mesa radv UMD. I'm not sure if I'm asking for too much, but if we can coordinate a similar interface in radv and amdgpu-pro at the vulkan level that would be great. I'm not sure what that's going to be yet. - Andres On 12/19/2016 12:11 AM, zhoucm1 wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> On 2016年12月19日 11:33, Pierre-Loup A. Griffais wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> We're currently working with the open stack; I assume that a mechanism could be exposed by both open and Pro Vulkan userspace drivers and that the amdgpu kernel interface improvements we would pursue following this discussion would let both drivers take advantage of the feature, correct? </blockquote> Of course. Does open stack have Vulkan support? Regards, David Zhou <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> On 12/18/2016 07:26 PM, zhoucm1 wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> By the way, are you using all-open driver or amdgpu-pro driver? +David Mao, who is working on our Vulkan driver. Regards, David Zhou On 2016年12月18日 06:05, Pierre-Loup A. Griffais wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Hi Serguei, I'm also working on the bringing up our VR runtime on top of amgpu; see replies inline. On 12/16/2016 09:05 PM, Sagalovitch, Serguei wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Andres, <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> For current VR workloads we have 3 separate processes running actually: </blockquote> So we could have potential memory overcommit case or do you do partitioning on your own? I would think that there is need to avoid overcomit in VR case to prevent any BO migration. </blockquote> You're entirely correct; currently the VR runtime is setting up prioritized CPU scheduling for its VR compositor, we're working on prioritized GPU scheduling and pre-emption (eg. this thread), and in the future it will make sense to do work in order to make sure that its memory allocations do not get evicted, to prevent any unwelcome additional latency in the event of needing to perform just-in-time reprojection. <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> BTW: Do you mean __real__ processes or threads? Based on my understanding sharing BOs between different processes could introduce additional synchronization constrains. btw: I am not sure if we are able to share Vulkan sync. object cross-process boundary. </blockquote> They are different processes; it is important for the compositor that is responsible for quality-of-service features such as consistently presenting distorted frames with the right latency, reprojection, etc, to be separate from the main application. Currently we are using unreleased cross-process memory and semaphore extensions to fetch updated eye images from the client application, but the just-in-time reprojection discussed here does not actually have any direct interactions with cross-process resource sharing, since it's achieved by using whatever is the latest, most up-to-date eye images that have already been sent by the client application, which are already available to use without additional synchronization. <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> 3) System compositor (we are looking at approaches to remove this overhead) </blockquote> Yes, IMHO the best is to run in "full screen mode". </blockquote> Yes, we are working on mechanisms to present directly to the headset display without any intermediaries as a separate effort. <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> The latency is our main concern, </blockquote> I would assume that this is the known problem (at least for compute usage). It looks like that amdgpu / kernel submission is rather CPU intensive (at least in the default configuration). </blockquote> As long as it's a consistent cost, it shouldn't an issue. However, if there's high degrees of variance then that would be troublesome and we would need to account for the worst case. Hopefully the requirements and approach we described make sense, we're looking forward to your feedback and suggestions. Thanks! - Pierre-Loup <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> Sincerely yours, Serguei Sagalovitch From: Andres Rodriguez <<a href="mailto:andresr@valvesoftware.com" target="_blank">andresr@valvesoftware.com</a>> Sent: December 16, 2016 10:00 PM To: Sagalovitch, Serguei; <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank"> amd-gfx@lists.freedesktop.org</a> Subject: RE: [RFC] Mechanism for high priority scheduling in amdgpu Hey Serguei, <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> [Serguei] No. I mean pipe :-) as MEC define it. As far as I understand (by simplifying) some scheduling is per pipe. I know about the current allocation scheme but I do not think that it is ideal. I would assume that we need to switch to dynamical partition of resources based on the workload otherwise we will have resource conflict between Vulkan compute and OpenCL. </blockquote> I agree the partitioning isn't ideal. I'm hoping we can start with a solution that assumes that only pipe0 has any work and the other pipes are idle (no HSA/ROCm running on the system). This should be more or less the use case we expect from VR users. I agree the split is currently not ideal, but I'd like to consider that a separate task, because making it dynamic is not straight forward :P <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> [Serguei] Vulkan works via amdgpu (kernel submissions) so amdkfd will be not involved. I would assume that in the case of VR we will have one main application ("console" mode(?)) so we could temporally "ignore" OpenCL/ROCm needs when VR is running. </blockquote> Correct, this is why we want to enable the high priority compute queue through libdrm-amdgpu, so that we can expose it through Vulkan later. For current VR workloads we have 3 separate processes running actually: 1) Game process 2) VR Compositor (this is the process that will require high priority queue) 3) System compositor (we are looking at approaches to remove this overhead) For now I think it is okay to assume no OpenCL/ROCm running simultaneously, but I would also like to be able to address this case in the future (cross-pipe priorities). <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> [Serguei] The problem with pre-emption of graphics task: (a) it may take time so latency may suffer </blockquote> The latency is our main concern, we want something that is predictable. A good illustration of what the reprojection scheduling looks like can be found here: <a href="https://community.amd.com/servlet/JiveServlet/showImage/38-1310-104754/pastedImage_3.png" rel="noreferrer" target="_blank">https://community.amd.com/servlet/JiveServlet/showImage/38-1310-104754/pastedImage_3.png</a> <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> (b) to preempt we need to have different "context" - we want to guarantee that submissions from the same context will be executed in order. </blockquote> This is okay, as the reprojection work doesn't have dependencies on the game context, and it even happens in a separate process. <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> BTW: (a) Do you want "preempt" and later resume or do you want "preempt" and "cancel/abort" </blockquote> Preempt the game with the compositor task and then resume it. <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> (b) Vulkan is generic API and could be used for graphics as well as for plain compute tasks (VK_QUEUE_COMPUTE_BIT). </blockquote> Yeah, the plan is to use vulkan compute. But if you figure out a way for us to get a guaranteed execution time using vulkan graphics, then I'll take you out for a beer :) Regards, Andres ________________________________________ From: Sagalovitch, Serguei [<a href="mailto:Serguei.Sagalovitch@amd.com" target="_blank">Serguei.Sagalovitch@amd.com</a>] Sent: Friday, December 16, 2016 9:13 PM To: Andres Rodriguez; <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank"> amd-gfx@lists.freedesktop.org</a> Subject: Re: [RFC] Mechanism for high priority scheduling in amdgpu Hi Andres, Please see inline (as [Serguei]) Sincerely yours, Serguei Sagalovitch From: Andres Rodriguez <<a href="mailto:andresr@valvesoftware.com" target="_blank">andresr@valvesoftware.com</a>> Sent: December 16, 2016 8:29 PM To: Sagalovitch, Serguei; <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank"> amd-gfx@lists.freedesktop.org</a> Subject: RE: [RFC] Mechanism for high priority scheduling in amdgpu Hi Serguei, Thanks for the feedback. Answers inline as [AR]. Regards, Andres ________________________________________ From: Sagalovitch, Serguei [<a href="mailto:Serguei.Sagalovitch@amd.com" target="_blank">Serguei.Sagalovitch@amd.com</a>] Sent: Friday, December 16, 2016 8:15 PM To: Andres Rodriguez; <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank"> amd-gfx@lists.freedesktop.org</a> Subject: Re: [RFC] Mechanism for high priority scheduling in amdgpu Andres, Quick comments: 1) To minimize "bubbles", etc. we need to "force" CU assignments/binding to high-priority queue when it will be in use and "free" them later (we do not want forever take CUs from e.g. graphic task to degrade graphics performance). Otherwise we could have scenario when long graphics task (or low-priority compute) will took all (extra) CUs and high--priority will wait for needed resources. It will not be visible on "NOP " but only when you submit "real" compute task so I would recommend not to use "NOP" packets at all for testing. It (CU assignment) could be relatively easy done when everything is going via kernel (e.g. as part of frame submission) but I must admit that I am not sure about the best way for user level submissions (amdkfd). [AR] I wasn't aware of this part of the programming sequence. Thanks for the heads up! Is this similar to the CU masking programming? [Serguei] Yes. To simplify: the problem is that "scheduler" when deciding which queue to run will check if there is enough resources and if not then it will begin to check other queues with lower priority. 2) I would recommend to dedicate the whole pipe to high-priority queue and have nothing their except it. [AR] I'm guessing in this context you mean pipe = queue? (as opposed to the MEC definition of pipe, which is a grouping of queues). I say this because amdgpu only has access to 1 pipe, and the rest are statically partitioned for amdkfd usage. [Serguei] No. I mean pipe :-) as MEC define it. As far as I understand (by simplifying) some scheduling is per pipe. I know about the current allocation scheme but I do not think that it is ideal. I would assume that we need to switch to dynamical partition of resources based on the workload otherwise we will have resource conflict between Vulkan compute and OpenCL. BTW: Which user level API do you want to use for compute: Vulkan or OpenCL? [AR] Vulkan [Serguei] Vulkan works via amdgpu (kernel submissions) so amdkfd will be not involved. I would assume that in the case of VR we will have one main application ("console" mode(?)) so we could temporally "ignore" OpenCL/ROCm needs when VR is running. <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> we will not be able to provide a solution compatible with GFX worloads. </blockquote> I assume that you are talking about graphics? Am I right? [AR] Yeah, my understanding is that pre-empting the currently running graphics job and scheduling in something else using mid-buffer pre-emption has some cases where it doesn't work well. But if with polaris10 it starts working well, it might be a better solution for us (because the whole reprojection work uses the vulkan graphics stack at the moment, and porting it to compute is not trivial). [Serguei] The problem with pre-emption of graphics task: (a) it may take time so latency may suffer (b) to preempt we need to have different "context" - we want to guarantee that submissions from the same context will be executed in order. BTW: (a) Do you want "preempt" and later resume or do you want "preempt" and "cancel/abort"? (b) Vulkan is generic API and could be used for graphics as well as for plain compute tasks (VK_QUEUE_COMPUTE_BIT). Sincerely yours, Serguei Sagalovitch From: amd-gfx <<a href="mailto:amd-gfx-bounces@lists.freedesktop.org" target="_blank">amd-gfx-bounces@lists.freedesktop.org</a>> on behalf of Andres Rodriguez <<a href="mailto:andresr@valvesoftware.com" target="_blank">andresr@valvesoftware.com</a>> Sent: December 16, 2016 6:15 PM To: <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank">amd-gfx@lists.freedesktop.org</a> Subject: [RFC] Mechanism for high priority scheduling in amdgpu Hi Everyone, This RFC is also available as a gist here: <a href="https://gist.github.com/lostgoat/7000432cd6864265dbc2c3ab93204249" rel="noreferrer" target="_blank">https://gist.github.com/lostgoat/7000432cd6864265dbc2c3ab93204249</a> [RFC] Mechanism for high priority scheduling in amdgpu <a href="http://gist.github.com" rel="noreferrer" target="_blank">gist.github.com</a> [RFC] Mechanism for high priority scheduling in amdgpu [RFC] Mechanism for high priority scheduling in amdgpu <a href="http://gist.github.com" rel="noreferrer" target="_blank">gist.github.com</a> [RFC] Mechanism for high priority scheduling in amdgpu [RFC] Mechanism for high priority scheduling in amdgpu <a href="http://gist.github.com" rel="noreferrer" target="_blank">gist.github.com</a> [RFC] Mechanism for high priority scheduling in amdgpu We are interested in feedback for a mechanism to effectively schedule high priority VR reprojection tasks (also referred to as time-warping) for Polaris10 running on the amdgpu kernel driver. Brief context: -------------- The main objective of reprojection is to avoid motion sickness for VR users in scenarios where the game or application would fail to finish rendering a new frame in time for the next VBLANK. When this happens, the user's head movements are not reflected on the Head Mounted Display (HMD) for the duration of an extra frame. This extended mismatch between the inner ear and the eyes may cause the user to experience motion sickness. The VR compositor deals with this problem by fabricating a new frame using the user's updated head position in combination with the previous frames. This avoids a prolonged mismatch between the HMD output and the inner ear. Because of the adverse effects on the user, we require high confidence that the reprojection task will complete before the VBLANK interval. Even if the GFX pipe is currently full of work from the game/application (which is most likely the case). For more details and illustrations, please refer to the following document: <a href="https://community.amd.com/community/gaming/blog/2016/03/28/asynchronous-shaders-evolved" rel="noreferrer" target="_blank">https://community.amd.com/community/gaming/blog/2016/03/28/asynchronous-shaders-evolved</a> Gaming: Asynchronous Shaders Evolved | Community <a href="http://community.amd.com" rel="noreferrer" target="_blank">community.amd.com</a> One of the most exciting new developments in GPU technology over the past year has been the adoption of asynchronous shaders, which can make more efficient use of ... Gaming: Asynchronous Shaders Evolved | Community <a href="http://community.amd.com" rel="noreferrer" target="_blank">community.amd.com</a> One of the most exciting new developments in GPU technology over the past year has been the adoption of asynchronous shaders, which can make more efficient use of ... Gaming: Asynchronous Shaders Evolved | Community <a href="http://community.amd.com" rel="noreferrer" target="_blank">community.amd.com</a> One of the most exciting new developments in GPU technology over the past year has been the adoption of asynchronous shaders, which can make more efficient use of ... Requirements: ------------- The mechanism must expose the following functionaility: * Job round trip time must be predictable, from submission to fence signal * The mechanism must support compute workloads. Goals: ------ * The mechanism should provide low submission latencies Test: submitting a NOP packet through the mechanism on busy hardware should be equivalent to submitting a NOP on idle hardware. Nice to have: ------------- * The mechanism should also support GFX workloads. My understanding is that with the current hardware capabilities in Polaris10 we will not be able to provide a solution compatible with GFX worloads. But I would love to hear otherwise. So if anyone has an idea, approach or suggestion that will also be compatible with the GFX ring, please let us know about it. * The above guarantees should also be respected by amdkfd workloads Would be good to have for consistency, but not strictly necessary as users running games are not traditionally running HPC workloads in the background. Proposed approach: ------------------ Similar to the windows driver, we could expose a high priority compute queue to userspace. Submissions to this compute queue will be scheduled with high priority, and may acquire hardware resources previously in use by other queues. This can be achieved by taking advantage of the 'priority' field in the HQDs and could be programmed by amdgpu or the amdgpu scheduler. The relevant register fields are: * mmCP_HQD_PIPE_PRIORITY * mmCP_HQD_QUEUE_PRIORITY Implementation approach 1 - static partitioning: ------------------------------------------------ The amdgpu driver currently controls 8 compute queues from pipe0. We can statically partition these as follows: * 7x regular * 1x high priority The relevant priorities can be set so that submissions to the high priority ring will starve the other compute rings and the GFX ring. The amdgpu scheduler will only place jobs into the high priority rings if the context is marked as high priority. And a corresponding priority should be added to keep track of this information: * AMD_SCHED_PRIORITY_KERNEL * -> AMD_SCHED_PRIORITY_HIGH * AMD_SCHED_PRIORITY_NORMAL The user will request a high priority context by setting an appropriate flag in drm_amdgpu_ctx_in (AMDGPU_CTX_HIGH_PRIORITY or similar): <a href="https://github.com/torvalds/linux/blob/master/include/uapi/drm/amdgpu_drm.h#L163" rel="noreferrer" target="_blank">https://github.com/torvalds/linux/blob/master/include/uapi/drm/amdgpu_drm.h#L163</a> The setting is in a per context level so that we can: * Maintain a consistent FIFO ordering of all submissions to a context * Create high priority and non-high priority contexts in the same process Implementation approach 2 - dynamic priority programming: --------------------------------------------------------- Similar to the above, but instead of programming the priorities and amdgpu_init() time, the SW scheduler will reprogram the queue priorities dynamically when scheduling a task. This would involve having a hardware specific callback from the scheduler to set the appropriate queue priority: set_priority(int ring, int index, int priority) During this callback we would have to grab the SRBM mutex to perform the appropriate HW programming, and I'm not really sure if that is something we should be doing from the scheduler. On the positive side, this approach would allow us to program a range of priorities for jobs instead of a single "high priority" value", achieving something similar to the niceness API available for CPU scheduling. I'm not sure if this flexibility is something that we would need for our use case, but it might be useful in other scenarios (multiple users sharing compute time on a server). This approach would require a new int field in drm_amdgpu_ctx_in, or repurposing of the flags field. Known current obstacles: ------------------------ The SQ is currently programmed to disregard the HQD priorities, and instead it picks jobs at random. Settings from the shader itself are also disregarded as this is considered a privileged field. Effectively we can get our compute wavefront launched ASAP, but we might not get the time we need on the SQ. The current programming would have to be changed to allow priority propagation from the HQD into the SQ. Generic approach for all HW IPs: -------------------------------- For consistency purposes, the high priority context can be enabled for all HW IPs with support of the SW scheduler. This will function similarly to the current AMD_SCHED_PRIORITY_KERNEL priority, where the job can jump ahead of anything not commited to the HW queue. The benefits of requesting a high priority context for a non-compute queue will be lesser (e.g. up to 10s of wait time if a GFX command is stuck in front of you), but having the API in place will allow us to easily improve the implementation in the future as new features become available in new hardware. Future steps: ------------- Once we have an approach settled, I can take care of the implementation. Also, once the interface is mostly decided, we can start thinking about exposing the high priority queue through radv. Request for feedback: --------------------- We aren't married to any of the approaches outlined above. Our goal is to obtain a mechanism that will allow us to complete the reprojection job within a predictable amount of time. So if anyone anyone has any suggestions for improvements or alternative strategies we are more than happy to hear them. If any of the technical information above is also incorrect, feel free to point out my misunderstandings. Looking forward to hearing from you. Regards, Andres _______________________________________________ amd-gfx mailing list <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank">amd-gfx@lists.freedesktop.org</a> <a href="https://lists.freedesktop.org/mailman/listinfo/amd-gfx" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/amd-gfx</a> amd-gfx Info Page - <a href="http://lists.freedesktop.org" rel="noreferrer" target="_blank"> lists.freedesktop.org</a> <a href="http://lists.freedesktop.org" rel="noreferrer" target="_blank">lists.freedesktop.org</a> To see the collection of prior postings to the list, visit the amd-gfx Archives. Using amd-gfx: To post a message to all the list members, send email ... amd-gfx Info Page - <a href="http://lists.freedesktop.org" rel="noreferrer" target="_blank"> lists.freedesktop.org</a> <a href="http://lists.freedesktop.org" rel="noreferrer" target="_blank">lists.freedesktop.org</a> To see the collection of prior postings to the list, visit the amd-gfx Archives. Using amd-gfx: To post a message to all the list members, send email ... _______________________________________________ amd-gfx mailing list <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank">amd-gfx@lists.freedesktop.org</a> <a href="https://lists.freedesktop.org/mailman/listinfo/amd-gfx" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/amd-gfx</a> </blockquote> _______________________________________________ amd-gfx mailing list <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank">amd-gfx@lists.freedesktop.org</a> <a href="https://lists.freedesktop.org/mailman/listinfo/amd-gfx" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/amd-gfx</a> </blockquote> </blockquote> </blockquote> _______________________________________________ amd-gfx mailing list <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank">amd-gfx@lists.freedesktop.org</a> <a href="https://lists.freedesktop.org/mailman/listinfo/amd-gfx" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/amd-gfx</a> </blockquote> </blockquote> </blockquote> Sincerely yours, Serguei Sagalovitch _______________________________________________ amd-gfx mailing list <a href="mailto:amd-gfx@lists.freedesktop.org" target="_blank">amd-gfx@lists.freedesktop.org</a> <a href="https://lists.freedesktop.org/mailman/listinfo/amd-gfx" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/amd-gfx</a> </blockquote> </blockquote> </blockquote> </blockquote> </blockquote> Sincerely yours, Serguei Sagalovitch </blockquote> </blockquote> </div> </div> </blockquote> </div> </div> </div> </div></div></div> </div> </div> </blockquote></div> </div>