Announcement

Collapse
No announcement yet.

AMDGPU DC Display Code Ported To GCN 1.0 "Southern Islands" GPUs

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #41
    Originally posted by nanonyme View Post
    Edit: engineers and managers aren't antiheroes but they aren't heroes either. Everyone occasionally makes bad judgement calls
    Except we as community don't made them our heroes because their engineering achievements or management prowess simply because very few people on this forum are kernel developers. People praise Bridgman and some developers on this forum because they spent a lot of their personal time on talking with community and sometimes solved issues that their employee wouldn't give a shit about.

    When it's come to large companies it's super rare to be able to communicate not just with some PR talking head, but with someone who actually care because you know they spent 10 years of their life working on it. At this point how good / bad their decisions were is pretty much secondary.

    Comment


    • #42
      Originally posted by pal666 View Post
      what vulkan-only game you are missing?
      It wasn't purchased for gaming :P Though I might do some light vulkan development myself, not sure what AMD tooling is like but it does seem to have better quality profiling for OpenCL vs Nvidia. If it did get proper amdgpu support, maybe it'd be a step closer to meeting ROCm, either or I'm not fussed, I just don't see 4 years as being that old. The card was the best value within budget available to me, plus I'd like to give AMD a try again(I think a 1030 could be had for a little bit more with better perf), as it's just for display purposes it'd be nice to have vs issues I get with nvidia sometimes, plus I can use it with wayland at a later date

      Comment


      • #43
        Originally posted by SXX⁣ View Post
        Man this is really not my business to tell who must post what on public forums, but here is fair feedback. I'm kind a super nerd who know more names of Linux graphics stack developers than movie actors and I track this forum, mailing lists and IRC for many years. On top of that I even know at least some publicly known part of RadeonHD story.

        Yet even for person like me tone of your posts is just off. Might be you just post it here for Bridgman / Airlie and then it's up to you, but for anyone else this kind of attitude just looks horrible. For someone who're never heard of RadeonHD it's would be just batshit crazy.

        Airlie wasn't the one behind KMS? Fine, tell it. They were guys who slowed down open source stack progress? Ok, that insigtful. But personal attacks over decade old events just look crazy.
        This is the specific bit of history you need to remember
        https://www.phoronix.com/scan.php?pa...item&px=NzEwNg
        RadeonHD was not just discontinued, people were laid off. This is very personal to libv. The entire matter matter has been so toxic all this time that you stopped hearing anything from the other side because they grew exhausted.

        As far as I've understood all boils down to Verhaegen considering Bridgman (and certain other people) personally accountable for the happenings around and related to OpenSUSE.
        Last edited by nanonyme; 10 October 2018, 03:03 AM.

        Comment


        • #44
          Originally posted by nanonyme View Post
          This is the specific bit of history you need to remember
          https://www.phoronix.com/scan.php?pa...item&px=NzEwNg
          RadeonHD was not just discontinued, people were laid off. This is very personal to libv. The entire matter matter has been so toxic all this time that you stopped hearing anything from the other side because they grew exhausted.

          As far as I've understood all boils down to Verhaegen considering Bridgman (and certain other people) personally accountable for the happenings around and related to OpenSUSE.
          I understand that you are trying to be fair here, but there are a few inaccuracies in there, some probably because you are trying to be too brief.

          First off, the lay-offs at SUSE had absolutely nothing to do with RadeonHD. This was a political move by Novell following the financial crisis kneejerk reaction of laying off a lot of people in novells older departments. Novell did not need to do this as it was sitting on a pile of cash, but when it did, it felt compelled to also lay off SUSE employees, and the lay-offs were almost all at SUSE Nuernberg. The selection of individuals laid off was on a system called "socialauswahl", and i was unlucky in being relatively young (back then) and a relatively recent employee of SuSE Nuernberg.

          Our mandate at SUSE with respect for the RadeonHD was as an antidote to a faltering ATI, and an attempt by AMD to get some order into that heap, specifically to stop bad ATI/fglrx press from ruining the AMD server market.

          Bridgman, an ATI veteran, was put in charge of getting us documentation, and was clearly acting from an ATI position from the getgo. Here is an old email i sent to SUSE management about him and his actions, a year into the project: https://libv.livejournal.com/27799.html

          After a driver was out in september 2007, and ATI had no way back, Mr Bridgman recruited Dave Airlie and Alex Deucher to create a competing (partially forked) driver, implementing everything the fglrx way. This driver became public in november 2007, "borrowing" a lot of radeonhd code. This was shored up by a campaign of shitsmearing by the likes of Daniel Stone and Adam Jackson. The latter culminated into the two of them hacking the by then dormant radeonhd repository, which was also revenge for Daniel Stone losing his job at nokia towards the person Daniel deemed responsible for him getting laid off. While i knew there was going to be a backlash by Dave, Daniel and Adam from the day we started thinking about writing the (now public) proposal to AMD for an open source driver for ATI, i never imagined that these guys would stoop to such depths.

          In the end, AMD CPUs stopped being competitive, and the financial crisis happened, and AMD ran out of cash and had to sell off its fabs and the ATI Imageon (now qualcomms adreno). The old AMD had gradually lost ground internally in the internal AMD vs ATI battle, with ATI gaining the upper hand over its new corporate overlord (i can dig out quotes from Daniel Stone labelling the actual open source driver from SUSE as a "corporate" project -- this was plainly parroting ATI sentiments with respect to how AMD was trying to gain control of the failing ATI internal structures). The strongest datapoint for how this all fit together from an open source driver point of view is that the second AMD stopped partnering with SUSE in writing the RadeonHD driver, the flow of documentation (for anything but the shader ISA) stopped.

          From the start, we from the RadeonHD project knew we had to battle ATI and i knew we would have to face crap from the usual suspects. We just could not have imagined how badly AMD would lose control over ATI internally, or the extent to which the usual suspects lacked scruples. They were clearly more concerned with killing a project from SUSE and yours truly than they were with creating technically superior software or creating a free driver.

          And that, briefly, is why someone like Dave Airlie should not be credited with things he did not do, and things he negatively influenced. He should also not be praised for RADV, as RADV would not have been needed if AMD had been able to continue its path towards an open source graphics driver stack, with or without the involvement of yours truly or SUSE. The same applies to the powerstruggle and shitsmearing around the amd_gpu driver a few years ago. By supporting ATI, and by countering AMD, Dave Airlie directly helped create the same circumstances that he would later try to gain fame over.

          Comment


          • #45
            Originally posted by agd5f View Post
            This is already the default behavior. DC is only enabled by default on asics that are fully supported. You can set dc=0 or dc=1 to override the default behavior (-1).
            This is not what I wrote. DC could be automatically enabled depending on whether unsupported outputs actually exist on the board, not whether the ASIC can in principle support them.
            So a Kaveri Notebook or a CIK GPU which has no VGA output would enjoy DC, while those who have would automatically fall back to non-DC.

            Comment


            • #46
              Originally posted by chithanh View Post
              This is not what I wrote. DC could be automatically enabled depending on whether unsupported outputs actually exist on the board, not whether the ASIC can in principle support them.
              So a Kaveri Notebook or a CIK GPU which has no VGA output would enjoy DC, while those who have would automatically fall back to non-DC.
              Ah, I misunderstood then. That is somewhat complicated to implement due to the way the driver is structured and how atomic is handled in the drm core.

              Comment


              • #47
                Originally posted by schmidtbag View Post
                That's because those cards are hardware incompatible with FP64. So, I'm not sure what else you can be asking for.
                The answer for those hardware incompatible GPUs is soft FP64. And as I already mentioned, patches exists but were never accepted upstream, so the vast majority of cards are locked out from from OpenGL 4+.
                We cant say they offer OpenGL 4.4 for the r600 cards if that means there are actually just 4-6 cards where its really enabled by default. Yeah I know you can override but that's no proper solution for the average Joe.

                Originally posted by schmidtbag View Post
                The HD 7000 series was released in 2012. Not sure how that equates to 4 years.
                The last GPU based on GCN 1.0 was released just last year. https://www.techpowerup.com/gpu-spec...&sort=released
                Even though they just release slow and low end cards, they continue to use this technology. So we cant say GCN1.0 is abandoned thus they dont need to care for those anymore.
                Of course everyone understands that they should focus on newer Gens and cards but really, this has taken now some years now again so the time has certainly come where they should start moving.
                The good thing is that's old technology. Must bugs should be ironed out. So they really dont need to invent the wheel again.
                That google is desperate enough to try themselves implementing proper AMDGPU support is really poor for AMD.

                Comment


                • #48
                  Originally posted by libv View Post
                  Our mandate at SUSE with respect for the RadeonHD was as an antidote to a faltering ATI, and an attempt by AMD to get some order into that heap, specifically to stop bad ATI/fglrx press from ruining the AMD server market.
                  Not exactly. We had been supporting open source driver development between ~1998 and ~2002 with Tungsten Graphics / VA Linux, but when we acquired FireGL there was an attempt to standardize on the fglrx driver (written for IBM graphics and ported to our HW starting with R200) for all uses, not just workstation.

                  When AMD bought ATI in 2006 one of the concerns that bubbled up was that fglrx was not a good fit for some of the AMD customer use cases, and that we really needed to either get back to supporting open source driver development or come up with another solution that could provide comparable ease of use.

                  As I learned later there were two main planning exercises resulting from this - one in the GPU SW team and another in the server CPU team working with SUSE. I was part of the GPU SW planning effort and wrote the final proposal with a lot of input from others.

                  Along the way two things happened:

                  1. While I was catching up with the current state of open source GPU drivers I noticed a couple of forum posts saying essentially "if AMD was serious about open source drivers they would call Dave Airlie and ask him what they should do" - so I did just that. Dave told me about the KMS idea that he and a couple of people were working on and suggested that I also talk to Alex Deucher - this would have been around May 2007 - and we all worked together on the proposal. Luc, I gather you still don't believe any of this ?

                  2. The AMD server folks working with SUSE suggested that we include SUSE in the initial driver effort. That seemed reasonable since the SUSE working relationship was already in place so we tweaked the proposal (I had been thinking about Tungsten Graphics based on previous ATI projects), took it to the executives, and received approval.

                  Originally posted by libv View Post
                  Bridgman, an ATI veteran, was put in charge of getting us documentation, and was clearly acting from an ATI position from the getgo. Here is an old email i sent to SUSE management about him and his actions, a year into the project: https://libv.livejournal.com/27799.html
                  "Bridgman, an ATI veteran, was put in charge of the project under the GPU SW VP" would an equally accurate statement, at least from our perspective. The problem (which we didn't discover until the following year) was that we had somehow ended up with a disconnect re: which of the proposals had been accepted, and so which company was responsible for setting the priorities. As a consequence you felt that we were meddling in "your" project rather than doing our jobs, although that was not actually the case.

                  There were honest differences in terms of technical priorities - we wanted a quick/simple user modesetting X driver using AtomBIOS so we could focus on 3D and KMS and built our plan around getting to acceleration as quickly as possible, while you felt that getting a really polished user mode X driver was the top priority, that VBIOS had no place in a graphics driver, and that 3D/drm and KMS were less important.

                  At the end of the day though, the key difference seemed to be that we thought we were working with you to implement our plan, while you felt that you were saving AMD from us and implementing the SUSE proposal you had helped to write. If you think about both sides of the story and try to focus on actual events rather than only remembering your interpretations at the time then the history will make a lot more sense... and maybe we can go back to drinking beer at conferences and arguing about the future instead of this.

                  Originally posted by libv View Post
                  After a driver was out in september 2007, and ATI had no way back, Mr Bridgman recruited Dave Airlie and Alex Deucher to create a competing (partially forked) driver, implementing everything the fglrx way.
                  A more accurate statement would be "months before a driver was out, when AMD was still planning, Mr Bridgman contacted Dave Airlie and Alex Deucher and worked with them on an open source driver plan, implementing nothing the fglrx way other than using AtomBIOS to speed up the initial HW bringup".

                  IIRC your "competing driver" comment refers to using radeon to support acceleration work since drm support was not yet available in RadeonHD - does that ring a bell ? I don't remember the details (it was 11 years ago after all) but it was something like that.
                  Last edited by bridgman; 14 October 2018, 04:43 AM.
                  Test signature

                  Comment


                  • #49
                    Originally posted by Strunkenbold View Post
                    That google is desperate enough to try themselves implementing proper AMDGPU support is really poor for AMD.
                    Perhaps I misunderstood your post, but you seem to be saying that the fact other community developers are contributing to the open source driver is a bad thing, and that the measure of success for an open source driver is AMD writing all the code themselves. Do I have that right ?
                    Last edited by bridgman; 13 October 2018, 10:10 AM.
                    Test signature

                    Comment


                    • #50
                      Originally posted by bridgman View Post

                      Perhaps I misunderstood your post, but you seem to be saying that the fact other community developers are contributing to the open source driver is a bad thing, and that the measure of success for an open source driver is AMD writing all the code themselves. Do I have that right ?
                      Common, you know it is not a problem who contribute. Its a good thing for the users.
                      Its good that google, valve, feral and the wine guys fix the AMD drivers.

                      The only problem with this is that they all doing this because AMD is either years too late (see radv, now DC for SI) or just dont have the manpower to fix bugs (see current state of Mesa RadeonSI, https://www.gamingonlinux.com/wiki/Games_broken_on_Mesa).
                      Whats the current status of the control panel actually? (next year or after next year?)

                      Comment

                      Working...
                      X