6 ms·

ARM has plenty of docs, sources, etc. for its Mali GPU: https://developer.arm.com/ip-products/graphics-and-multimedia/mali-gpus https://developer.arm.com/ip-pro

ARM has plenty of docs, sources, etc. for its Mali GPU: https://developer.arm.com/ip-products/graphics-and-multimedia/mali-gpus https://developer.arm.com/ip-products/graphics-and-multimedi... Their GPU drivers are even distributed by Debian: https://wiki.debian.org/MaliGraphics https://wiki.debian.org/MaliGraphics You are probably conflating vendors like Broadcom who license the ARM ISA and processor IP but add their own (closed) GPU to SoCs they produce, like those used in the Raspberry Pi. Remember ARM is just a company that license and sell an ISA and chip design. It's up to system integrators to turn those designs into physical chips, and sadly we have seen many of them don't value making their designs available to open source, etc. The solution to this is to not support those vendors, not to complain at ARM.

stefan_ 5y ago
Mali is every bit as proprietary as Nvidia.
- hda2 5y ago
  > Mali is every bit as proprietary as Nvidia. I don't think anyone was claiming it isn't. I think GP's (and his GP) point is that ARM /usually/ makes the information needed to develop high quality open source drivers available where as Nvidia tends to do the opposite. Whether these technologies are proprietary or not is orthogonal to this issue.
  - stefan_ 5y ago
    But it absolutely doesn't, that is the point. There is no register level documentation or documentation for the instruction set the chips use. There is no pertinent architecture documentation at all. Having a list of performance counters does not equal documentation.
rektide 5y ago
I've spent about 10 minutes clicking through looking for anything even mildly technical. The first link you sent me, "plenty of docs", is a 2 page PDF product brief & 5 web pages describing some features at a high level. There's a ton of docs... for game developers, application developers, &c. Talking up features & capabilities. But for making drivers? For describing how to use the hardware? There's some links to embedded development / DDK development sites for arm, but none of them are open accecss. ARM wont help. The Debian drivers you cite are the Panfrost Lima drivers, which are the reverse engineering works everyone else did in the dark with no help from ARM, until very very recently. I am not conflating anything. ARM has provided no documentation no support to help folks make drivers. I stand by my original post entirely, and disagree deeply. ARM's gpu chip design is stamped out into countless chips, verbatim or close to verbatim. Even if individual chip makers did want to support their chips, I suspect it is in fact ARM who would block them from releasing open source drivers, from documenting the chips they make: ARM's desire has been to protect it's IP, and to help the various embedded support partners that do know how to work with their chips and their IP in a secure fashion. ARM has been a bad player in the Linux world, has worked against allowing Linux support. The problem is absolutely ARM, not those vendors.
- 2526702934 5y ago
  Can someone explain to me how ARM expect to protect its IP if it doesn't publish anything about it? I was under the impression that in order to protect IP, one has to either patent it or copyright it, thus some form of publication must be made. If ARM does publish technical documentation (1), ARM has a way to prove infrigement or prior art whenever someone else tries to use their IP without their approval. If ARM doesn't, how can they prove anything? (1) Obviously, if they publish the tech. docs, they might as well help driver implementors, might they not?

stefan_ 5y ago

Mali is every bit as proprietary as Nvidia.

hda2 5y ago
> Mali is every bit as proprietary as Nvidia. I don't think anyone was claiming it isn't. I think GP's (and his GP) point is that ARM /usually/ makes the information needed to develop high quality open source drivers available where as Nvidia tends to do the opposite. Whether these technologies are proprietary or not is orthogonal to this issue.
- stefan_ 5y ago
  But it absolutely doesn't, that is the point. There is no register level documentation or documentation for the instruction set the chips use. There is no pertinent architecture documentation at all. Having a list of performance counters does not equal documentation.

rektide 5y ago

I've spent about 10 minutes clicking through looking for anything even mildly technical. The first link you sent me, "plenty of docs", is a 2 page PDF product brief & 5 web pages describing some features at a high level. There's a ton of docs... for game developers, application developers, &c. Talking up features & capabilities. But for making drivers? For describing how to use the hardware? There's some links to embedded development / DDK development sites for arm, but none of them are open accecss. ARM wont help. The Debian drivers you cite are the Panfrost Lima drivers, which are the reverse engineering works everyone else did in the dark with no help from ARM, until very very recently. I am not conflating anything. ARM has provided no documentation no support to help folks make drivers. I stand by my original post entirely, and disagree deeply. ARM's gpu chip design is stamped out into countless chips, verbatim or close to verbatim. Even if individual chip makers did want to support their chips, I suspect it is in fact ARM who would block them from releasing open source drivers, from documenting the chips they make: ARM's desire has been to protect it's IP, and to help the various embedded support partners that do know how to work with their chips and their IP in a secure fashion. ARM has been a bad player in the Linux world, has worked against allowing Linux support. The problem is absolutely ARM, not those vendors.

2526702934 5y ago
Can someone explain to me how ARM expect to protect its IP if it doesn't publish anything about it? I was under the impression that in order to protect IP, one has to either patent it or copyright it, thus some form of publication must be made. If ARM does publish technical documentation (1), ARM has a way to prove infrigement or prior art whenever someone else tries to use their IP without their approval. If ARM doesn't, how can they prove anything? (1) Obviously, if they publish the tech. docs, they might as well help driver implementors, might they not?