AMD Open-Source S.I. Botched, Hope For The Future

[glisse]

Ya, you were stupid enough to buy our hardware but you can't use it so stop whining. Is not an acceptable position for any company to take. As for MESA it will never be any where near up to par as long as the card manufacturers refuse to let people know how the hardware is designed to work. This isn't like looking at an analogue circuit. Unless you know how the chips are designed and work together the amount of work required to reverse engineer how a modern graphics card works is beyond the resources that any one could reasonably be expected to expend. Documentation is the only viable way to have what should be considered a reasonable level of performance in this day and age. But both NVidia and AMD have decided that they are never going to provide what is needed so the open source drivers for these companies will always suck.

AMD official support is through the closed source driver. They support the community effort to make an open source driver but i don't think it's considered as what they recommand to use to their customer. You better get your fact right before making claim. One can totaly reverse engineer GPU without knowing deeply how things works. nouveau is a brilliant example of that. On the other end AMD is providing documentations that cover all the acceleration bits you need to make a driver. Yes there is lacking area like power management or video decoding but it was for long time considered as secondary.

[jrch2k8]

AMD official support is through the closed source driver. They support the community effort to make an open source driver but i don't think it's considered as what they recommand to use to their customer. You better get your fact right before making claim. One can totaly reverse engineer GPU without knowing deeply how things works. nouveau is a brilliant example of that. On the other end AMD is providing documentations that cover all the acceleration bits you need to make a driver. Yes there is lacking area like power management or video decoding but it was for long time considered as secondary.

well make LOT of sense that features like video decoding get consireded as secondary when there are still many important bits to be finished like get parity with opengl specs or improve shaders compilers, etc but either way it seems you can't stop conspiracy junkies and nVidia fanboys for now since they seem just to bypass this kind of post and go on trolling sad but true

[jvillain]

On the other end AMD is providing documentations that cover all the acceleration bits you need to make a driver.

More like several years ago AMD made some documentation available for the cards back then and have done next to nothing since. I understand people with older cards being happy. I didn't have any big beefs with how my HD4000 worked until it failed. There is all the difference in the world between reading about these problems and living through them.

Yes there is lacking area like power management or video decoding but it was for long time considered as secondary.

Every thing is lacking from the stand point of an end user because all these pieces were so late coming. Yes there is experimental code available in non main stream repositories but hunting them down building them and then trying to get them to work with out breaking other parts of the system is more than a little problematic. Especially since the only documentation on how it is supposed to fit together is what Michael is posting. If they aren't even including the required parts in Rawhide as even a proof of concept then end users aren't going to see this stuff until at least F19 or 20. It is guaranteed not to be in F18 so we are what a year away or so from support for end users in even the most bleeding edge distros? Reinstalling with Fedora 13 or what ever the last version that the blob works with isn't an option at al, for me. Not because of pig headedness but because other things I need aren't there. Ignoring these problems won't make them go away. What it will do however is fool people into thinking they can use these SI cards which will then leave them with a shit experience that will turn them off of AMD products all together like what has happened to me. Or worse they will just say fuck it and load windows or go buy a mac.

[jrch2k8]

More like several years ago AMD made some documentation available for the cards back then and have done next to nothing since. I understand people with older cards being happy. I didn't have any big beefs with how my HD4000 worked until it failed. There is all the difference in the world between reading about these problems and living through them.



Every thing is lacking from the stand point of an end user because all these pieces were so late coming. Yes there is experimental code available in non main stream repositories but hunting them down building them and then trying to get them to work with out breaking other parts of the system is more than a little problematic. Especially since the only documentation on how it is supposed to fit together is what Michael is posting. If they aren't even including the required parts in Rawhide as even a proof of concept then end users aren't going to see this stuff until at least F19 or 20. It is guaranteed not to be in F18 so we are what a year away or so from support for end users in even the most bleeding edge distros? Reinstalling with Fedora 13 or what ever the last version that the blob works with isn't an option at al, for me. Not because of pig headedness but because other things I need aren't there. Ignoring these problems won't make them go away. What it will do however is fool people into thinking they can use these SI cards which will then leave them with a shit experience that will turn them off of AMD products all together like what has happened to me. Or worse they will just say fuck it and load windows or go buy a mac.

1. the documentation is available up to 7000 series [just missing some bits here and there like UVD] ask bridgman or glisse for a link since i don't remember it right now but you need to understand this documentation is not about how to write a driver but a pure GPU ASM and electronic registers like the CPU documentation from intel/AMD or any other vendor(take a look and you will see what pure asm register looks like) note that this is not a mistake documentation should be like that. so all opengl call end being a bunch of those ASM commands depending of the GPU so is mesa/gallium problem how it use those ASM bits and that require extensive analize and test.[trust me when i say is anything but trivial in fact extremely hard don't even gets close LOL]

2. OSS driver are not meant for ppl that can't properly use git for now since this drivers and most of the linux graphics is in alpha state[aka under heavy development], so you should stick to fglrx [even if it means downgrade the distro] wich is the officially supported driver[i know it sucks]. so if you are willing to be an alpha tester like many of us you should [maybe having a second partition so you don't mess your main distro] build regularly these components[i do it daily | once you get used to it is fast]

* kernel[stick to drm-next if possible until the next kernel rc start coming[may kill kittens so keep an stable kernel around]]
* libdrm with libkms git
* libxcb git
* mesa git
* cairo in case you wanna test weston
* DDX git
* glamor git[broken for xserver 1.13]
* xorg-server git[glamor you need to stick with 1.12 for now]
* put in a bookmark bar mesa bugzilla
* enjoy

another good tip is read daily mesa/ddx cgit logs so you can pinpoint when a nice patch lands to test it
another one could be install apitrace 3.0 it does wonders and even save kittens/puppies

i understand this seems overkill but no one ever has said that OSS graphic stack is ready for production or even worse recommended for regular user as official driver, ofc it will get there since everyday keep getting closer and remember that 2 or 3 years ago linux graphic support was near 0[like bsd or solaris] so the graphic team basically has to rethink everything from scratch to be where we are today with an competitive modern graphic stack[technically speaking since is still work in progress but technically speaking the new stack is more capable than windows/mac put togheter they just need to end putting all the bits togheter and optimize but they are like 8 ppl due to the massive level of knowledge you need to program a GPU]

[agd5f]

Or worse they will just say fuck it and load windows or go buy a mac.

Linux has a much lower desktop market share and as such a much lower level of investment across the board. This is true across the board (hw vendors, oems, desktop software stacks, etc.) Yes, I know, chicken and egg. Life is hard on the fringe, but can also be more interesting.

The closed source driver has full support for all currently available cards (largely because it is able to leverage code shared with windows) and supports a wide range of distro versions.

[PsynoKhi0]

Ya, you were stupid enough to buy our hardware but you can't use it so stop whining. Is not an acceptable position for any company to take.

Neither is this an acceptable position for a user to take.
"Can't use it"??? Eh? HD7770 owner here. Working fine. Perfect? Nope, but tell you what, when I find a bug, my first reaction is NOT to come here waving arms around, screaming bloody murder and making doomsday claims. Instead I GATHER INFORMATION TO GET THE BUG FIXED.
So maybe, just maybe, being more specific instead of whining would make you more valuable?

[jvillain]

jrch2k8 thank you for your well thought out post. My training was in hardware design and I have read many spec sheets for chips and devices over the years. I went and looked at what AMD had available on the developer site a couple of weeks ago and know what these are. Again thank you for your post.

But I have hit the limit of what I am willing to put up with. I bit the bullet and went and bought the parts to build an Intel based machine. Tomorrow I will take my HD7700 card out of my machine and enjoyed smashing it with a hammer. Yes I might be one customer but I also have considerable input in the purchasing decisions of a number of companies. I also push the companies I work with to help support open source in various ways. I'm not an arm waver as some momo just described me I have been a specialist in Linux since the early days of Slackware. If I am as frustrated as I am there is a real problem. I was also called a lousy customer. But there is an old saying in business. "If the customer is mad they have a right to be. Either because you did some thing wrong or because you didn't manage expectations." Where AMD is running into problems is in trying to have their cake and eat it to. You can't be the saviour of open source and at the same time not support open source at all. Pick a message and stick with it. But as AMD continues to shed market share I am sure they will continue to sit around and scratch their heads as to why.

Feel free to close my account Michael.

[bridgman]

jrch2k8 thank you for your well thought out post. My training was in hardware design and I have read many spec sheets for chips and devices over the years. I went and looked at what AMD had available on the developer site a couple of weeks ago and know what these are. Again thank you for your post.

You mentioned "developer site" so I'm just checking... you were looking at the documentation section of RadeonFeature, right ? If you look anywhere else you're only going to see a subset...

http://www.x.org/wiki/RadeonFeature#Documentation

You can't be the saviour of open source and at the same time not support open source at all. Pick a message and stick with it.

With respect, I believe we are doing just that. I don't know where either of your statements came from.

I don't believe we have ever even hinted at being anything like the saviour of open source. We said that we would start supporting open source driver development again while continuing to work on the fglrx binary driver in parallel.

You say "not support open source at all", and I don't understand how you can think that. We have four full time developers working on the open source drivers and supporting other community developers, plus others working part time and behind the scenes. Is there some terribly evil thing we're doing which somehow invalidates all that for you ?

[fuzz]

What the hell... honestly it's practically amazing what the open source AMD team has accomplished with their resources.

I don't know how some of this stuff gets so twisted.

[enrico.tagliavini]

I would posit just the opposite. In that case, the user probably can't tell whether he's using vesa or a native driver. Both will put an image on the screen a perform quite well for non-3D tasks which is what the vast majority of users do. Even for basic 3D, llvmpipe should do a pretty decent job in the interim. For that, the driver already works quite well.

Try with a laptop, it will became warm in a second, the battery will just yell for the discharge rate. It is not only a 3D performance problem. I tried llvmpipe (by accident to be honest) on my brand new quad core ivy bridge. It is not bad, the first minute I didn't saw the difference, but then I started moving windows around.... KDE works only with xrender backend with llvmpipe and you can see the difference. Now this is with an high end CPU.... Really you can't pretend users are ok with this. Some user might be, but really you simply can't ship it as ready for the mass. I would not say it is botched, it is not. It is simply not ready. As John said the community didn't worked a lot on the driver -> you can't expect it will autocomplete. I'm not pointing the finger to someone here, nobody is doing a bad work here, I want this to be clear. But radeon still has major missing features for the masses.