Texas Instruments Has New Open-Source Driver

[phoronix]

Phoronix: Texas Instruments Has New Open-Source Driver

While Texas Instruments released an open-source driver last year for the Linux kernel within the DRM area (the TILER driver), it didn't make it into the mainline tree for the lack of open-source user-space applications/drivers that could take advantage of it, i.e. the usual ARM graphics mess. Yesterday, however, Texas Instruments released a new open-source DRM driver for their OMAP platforms...

http://www.phoronix.com/vr.php?view=OTg3MA

[allquixotic]

These proprietary vultures (TI, Samsung, etc) are going to play the cat and mouse game with Linus and Dave, giving little bits and pieces of additional functionality over time, but without ever contributing a fully open source, complete 3d driver. It's frustrating to watch them try to get around the policies that have been set forth by the kernel developers and the license, knowing that once it is eventually accepted into mainline, we still won't have any open source 3d support on these chipsets.

They just won't do it. They probably aren't willing to give up built-in S3TC support in the 3d driver, or perhaps they have additional patent fears regarding their 3d code. Or maybe it's the pressure from the DRM cartels that keeps them from opening it all the way up. But so far, these mobile developers are unwilling to compromise and provide open hardware documentation and a mostly-working 3d driver, like AMD has.

I wish AMD would go into the business of making GLES SoC 3d acceleration engines. If they apply the same thought process to that as they have their radeon effort, we'd have excellent smartphone/embedded APUs with open source support. You just can't trust TI or Samsung to play by the spirit of open source, they are just concerned with complying with the letter of the license.

[FireBurn]

These proprietary vultures (TI, Samsung, etc) are going to play the cat and mouse game with Linus and Dave, giving little bits and pieces of additional functionality over time, but without ever contributing a fully open source, complete 3d driver. It's frustrating to watch them try to get around the policies that have been set forth by the kernel developers and the license, knowing that once it is eventually accepted into mainline, we still won't have any open source 3d support on these chipsets.

They just won't do it. They probably aren't willing to give up built-in S3TC support in the 3d driver, or perhaps they have additional patent fears regarding their 3d code. Or maybe it's the pressure from the DRM cartels that keeps them from opening it all the way up. But so far, these mobile developers are unwilling to compromise and provide open hardware documentation and a mostly-working 3d driver, like AMD has.

I wish AMD would go into the business of making GLES SoC 3d acceleration engines. If they apply the same thought process to that as they have their radeon effort, we'd have excellent smartphone/embedded APUs with open source support. You just can't trust TI or Samsung to play by the spirit of open source, they are just concerned with complying with the letter of the license.

They used to be in the business but they sold it of to Qualcomm

[agd5f]

These proprietary vultures (TI, Samsung, etc) are going to play the cat and mouse game with Linus and Dave, giving little bits and pieces of additional functionality over time, but without ever contributing a fully open source, complete 3d driver. It's frustrating to watch them try to get around the policies that have been set forth by the kernel developers and the license, knowing that once it is eventually accepted into mainline, we still won't have any open source 3d support on these chipsets.

There's not much they can do about it. They don't own the 3D IP.

[Adarion]

Exactly. The GPUs on the OMAP are from Imagination Technologies (same folks behind poulsbo!). I told TI people on LinuxTag in Berlin that all their nice and praised multimedia platforms are really constrained by this problem that they only provide a blob for some older Kernel versions.

[robclark]

These proprietary vultures (TI, Samsung, etc) are going to play the cat and mouse game with Linus and Dave, giving little bits and pieces of additional functionality over time, but without ever contributing a fully open source, complete 3d driver. It's frustrating to watch them try to get around the policies that have been set forth by the kernel developers and the license, knowing that once it is eventually accepted into mainline, we still won't have any open source 3d support on these chipsets.

They just won't do it. They probably aren't willing to give up built-in S3TC support in the 3d driver, or perhaps they have additional patent fears regarding their 3d code. Or maybe it's the pressure from the DRM cartels that keeps them from opening it all the way up. But so far, these mobile developers are unwilling to compromise and provide open hardware documentation and a mostly-working 3d driver, like AMD has.

I wish AMD would go into the business of making GLES SoC 3d acceleration engines. If they apply the same thought process to that as they have their radeon effort, we'd have excellent smartphone/embedded APUs with open source support. You just can't trust TI or Samsung to play by the spirit of open source, they are just concerned with complying with the letter of the license.

btw, it isn't about trying to play cat and mouse.. I would love nothing more to release 3d kernel and userspace, but unfortunately it is IMG's IP. We are just trying to do the best we can and get basic KMS support, and support for other acceleration blocks that we can provide open userspace for into the kernel.

[allquixotic]

btw, it isn't about trying to play cat and mouse.. I would love nothing more to release 3d kernel and userspace, but unfortunately it is IMG's IP. We are just trying to do the best we can and get basic KMS support, and support for other acceleration blocks that we can provide open userspace for into the kernel.

Hey, certainly understandable. As far as legal and technical recourse, you really don't have much of either when you are licensing someone else's technology and they say "no". What more can you do?

I sympathize (much as I sympathize with those who use other proprietary software with restrictive licensing terms), but I still wish that things could be better.

Certainly, as a significant business partner that can largely affect Imagination Technologies' bottom line, TI has more clout with IMG than consumers, right? If you guys pull out from your deals with IMG, they'll lose millions of dollars. If a couple thousand consumers make a big gripe, they can still turn a blind eye, because it has no tangible quantifiable impact on their bottom line.

TI isn't the first vendor that would like to be open-source friendly, but can't be because they based their graphics core on IMG tech. Hey, it's ok. But can't you try to do something about it? When you talk to IMG, they listen. When Intel talks to IMG, they listen. Both TI and Intel would like to release fully open drivers for their graphics cores based on IMG IP. Until your post I didn't know that TI was in that boat too, but it's good to know that you are.

All I'm saying is, use your company's clout (and maybe team up with Intel to bring a unified message) to say, "we want the ability to create open source graphics drivers for your graphics core". If they flat-out say "no way it'll ever happen", you could say, "we'll start looking around for another graphics core to use for our next product cycle, then". Their eyes will get big and a bunch of corporate lawyers at IMG will start stirring in their seats.

This just has to stop. I know that there are probably good technical reasons (i.e. patents owned by IMG) that make it very compelling to want to license IMG's graphics cores for your SoC chipsets. They seem very low-power while providing a nice performance/cost ratio. Heck, I have two Android 2.2 smartphones that both have IMG graphics cores, with two different ARM SoC manufacturers having licensed the same core. Even Apple's iPads use IMG cores, from what I understand (and Apple would be the most likely candidate for rolling their own, if they were able).

It sounds like almost the entire industry is sold hook-line-and-sinker on IMG graphics cores for any 3d-accelerated device smaller than an ultraportable laptop. Do they really have that broad of a patent, or is it just that they have a huge technological advantage over the competition for that form factor?

BTW, I did a little bit of digging, and I couldn't find any evidence that Nvidia's Tegra platform is based on IMG IP. If you guys could somehow license Nvidia's graphics core for your next OMAP chip, that would be extremely awesome. Nvidia isn't exactly the most friendly vendor towards open source, but they are at least more reasonable than IMG. And at least threatening to move away from IMG would certainly make them think twice if you state lack of open-source 3d accel as one of the reasons why you are dissatisfied with your business with them.

I don't need to recite to someone with your industry knowledge all the virtues of open source acceleration, and how it's not only about freedom but also practical advantages like stability and compatibility. I'm sure you realize that a binary driver blob shoehorned into a rapidly-changing platform is an abomination that quickly experiences bitrot and can't be fixed when issues arise. Anyone who is developing Linux-based operating systems (and right now in the embedded space, that's a whole lot of people) is sure to value the practical advantages of open source graphics drivers. Since that is the case, it's time to stand up for what you value, and tell your business partners at Imagination Technologies that there are other viable options out there that are more likely to offer open source graphics drivers, and you may start to work with them if IMG won't budge on the issue.

Thanks for reading, and thanks for at least caring about open source drivers, even if there is nothing that you personally can do to improve the issue. But those of us who do value software freedom in and of itself have been maligning the proprietary driver policy of IMG for many years; this community's outrage at IMG can be traced back to the original Asus netbooks with Poulsbo graphics. And that was in 2007, wasn't it? Long time ago.

[daniels]

All I'm saying is, use your company's clout (and maybe team up with Intel to bring a unified message) to say, "we want the ability to create open source graphics drivers for your graphics core". If they flat-out say "no way it'll ever happen", you could say, "we'll start looking around for another graphics core to use for our next product cycle, then". Their eyes will get big and a bunch of corporate lawyers at IMG will start stirring in their seats.

I think it's fair to say that quite a companies have tried quite doggedly for some time. There's really not much option in the mobile GPU space though: you've got ARM's Mali with equally closed drivers, or Samsung's cores which even if they would license to other SoC vendors, still have closed drivers I believe.

*shrug*

[robclark]

Hey, certainly understandable. As far as legal and technical recourse, you really don't have much of either when you are licensing someone else's technology and they say "no". What more can you do?

I sympathize (much as I sympathize with those who use other proprietary software with restrictive licensing terms), but I still wish that things could be better.

Certainly, as a significant business partner that can largely affect Imagination Technologies' bottom line, TI has more clout with IMG than consumers, right? If you guys pull out from your deals with IMG, they'll lose millions of dollars. If a couple thousand consumers make a big gripe, they can still turn a blind eye, because it has no tangible quantifiable impact on their bottom line.

TI isn't the first vendor that would like to be open-source friendly, but can't be because they based their graphics core on IMG tech. Hey, it's ok. But can't you try to do something about it? When you talk to IMG, they listen. When Intel talks to IMG, they listen. Both TI and Intel would like to release fully open drivers for their graphics cores based on IMG IP. Until your post I didn't know that TI was in that boat too, but it's good to know that you are.

All I'm saying is, use your company's clout (and maybe team up with Intel to bring a unified message) to say, "we want the ability to create open source graphics drivers for your graphics core". If they flat-out say "no way it'll ever happen", you could say, "we'll start looking around for another graphics core to use for our next product cycle, then". Their eyes will get big and a bunch of corporate lawyers at IMG will start stirring in their seats.

Well, it would be a bit easier if there was some other alternative IP choice that was open. (And, well, it would still also have to be competitive on a performance standpoint too).

It could also perhaps help if there was a coordinated effort between Intel, TI, and other vendors who use IMG. I really don't think it will happen with just TI alone.

This just has to stop. I know that there are probably good technical reasons (i.e. patents owned by IMG) that make it very compelling to want to license IMG's graphics cores for your SoC chipsets. They seem very low-power while providing a nice performance/cost ratio. Heck, I have two Android 2.2 smartphones that both have IMG graphics cores, with two different ARM SoC manufacturers having licensed the same core. Even Apple's iPads use IMG cores, from what I understand (and Apple would be the most likely candidate for rolling their own, if they were able).

It sounds like almost the entire industry is sold hook-line-and-sinker on IMG graphics cores for any 3d-accelerated device smaller than an ultraportable laptop. Do they really have that broad of a patent, or is it just that they have a huge technological advantage over the competition for that form factor?

BTW, I did a little bit of digging, and I couldn't find any evidence that Nvidia's Tegra platform is based on IMG IP. If you guys could somehow license Nvidia's graphics core for your next OMAP chip, that would be extremely awesome. Nvidia isn't exactly the most friendly vendor towards open source, but they are at least more reasonable than IMG. And at least threatening to move away from IMG would certainly make them think twice if you state lack of open-source 3d accel as one of the reasons why you are dissatisfied with your business with them.

No, tegra is based on Nvidia's own IP.. and I'm not really sure that Nvidia is better. At least IMG does have a GPL kernel module (not just binary blob w/ wrapper).

I don't need to recite to someone with your industry knowledge all the virtues of open source acceleration, and how it's not only about freedom but also practical advantages like stability and compatibility. I'm sure you realize that a binary driver blob shoehorned into a rapidly-changing platform is an abomination that quickly experiences bitrot and can't be fixed when issues arise. Anyone who is developing Linux-based operating systems (and right now in the embedded space, that's a whole lot of people) is sure to value the practical advantages of open source graphics drivers. Since that is the case, it's time to stand up for what you value, and tell your business partners at Imagination Technologies that there are other viable options out there that are more likely to offer open source graphics drivers, and you may start to work with them if IMG won't budge on the issue.

Thanks for reading, and thanks for at least caring about open source drivers, even if there is nothing that you personally can do to improve the issue. But those of us who do value software freedom in and of itself have been maligning the proprietary driver policy of IMG for many years; this community's outrage at IMG can be traced back to the original Asus netbooks with Poulsbo graphics. And that was in 2007, wasn't it? Long time ago.

One other thought.. I know everyone wants a full open userspace, but I suspect a compromise solution is something with at least a chance of happening.. ie. all the stuff about buffer management, MMU programming, and that sort of thing (where the kernel<->user interface is involved) is not really anything particularly magic. But this is mostly what the DRM kernel part is dealing with. I can't imagine that IMG or any of the other 3d IP vendors feel that this part of the stack is sensitive IP. (And btw, I'm not speaking for IMG or TI or anyone other than myself here.) The shader compiler, firmware, and anything exposing the instruction set and data flow within the 3d core, this I think they feel more protective of. So maybe some sort of partial open userspace part with a few binary blob parts would be a practical step in the right direction.

[_ONH_]

At the time AMD drops down with the Wattage TDP of their Brazos SoC to a comparable level/by the same Performance (even when it is nor ARM), and device maker that care to oss shift that way. I would like to see the IMG people to allow fully oss drivers. But I think we are at least 1.5 year away from this point.