NVIDIA, Mentor Graphics May Harm GCC

[ldesnogu]

ARM:

http://infocenter.arm.com/help/index.jsp

http://infocenter.arm.com/help/topic/com.arm.doc.qrc0001l/QRC0001_UAL.pdf

Imagination:

http://www.imgtec.com/mips/developers/whitepapers.asp

Qualcomm:
You need to sign up with https://developer.qualcomm.com/ to access anything useful that is specific to Qualcomm products, but both Adreno and Snapdragon use the ARM ISA anyhow so it isn't a big deal in this case.

It took me less than 5 minutes to find all of this.

You're kidding right? We are talking GPU here...

[Nille]

Its not our problem if you ignore the point of concerns.

[johnc]

Where is the Point?

The point is that CUDA is something that is almost entirely used in a supercomputing context, especially when used with OpenACC. Having a dependency on a binary driver doesn't harm any of those customers.

Like I said, this whole thread is just more nvidia froth. Why not just post up the Linus pic and call it a day?

[Nille]

How many Oak Ridge-like Titan supercomputers are using the nouveau driver?

Where is the Point?

[johnc]

Round and round we go weeee.

The problem is obvious your mind is so closed you can not accept any of the reasoned responses myself and others have given. Licencing is not the issue here and never has been. This is a matter of principals.
Yes you can create non-free output but this is a side effect that is allowed rather than an explicit goal. Targeting PTX in gcc would have no other use than to produce non-free output as I have explained multiple times and so have others here:

Clearly we're doomed then. We need another article proclaiming the death of GCC at the hands of nvidia.

[johnc]

And not only a Nvidia Card it also requie the hole Nvidia BLOB.

How many Oak Ridge-like Titan supercomputers are using the nouveau driver?

[tuke81]

And this support is in GCC? Its not like that OpenACC is completely new. The New thing is now the support in the GCC and i see only that there is CUDA supported.

Well I don't really know. But gcc policy is not to accept any standard only partially. OpenCL1.2 is in OpenACC2.0(which Mentor talks). So if Mentor want's openACC to gcc it have to be whole thing not only part of it. It's like a situation with openMP, which have to use it's own dev-branch gomp before getting accepted to gcc:

http://gcc.gnu.org/ml/gcc/2013-09/msg00244.html

http://gcc.gnu.org/projects/gomp/

[Nille]

It's not only for nvidia hw, OpenACC has opencl support too, and it can be used with radeon gpus.http://www.openacc.org/SC2013

And this support is in GCC? Its not like that OpenACC is completely new. The New thing is now the support in the GCC and i see only that there is CUDA supported.

[tuke81]

In the hole discussion it look like some people forget something. The Topic is not only the Output. We have a API in GCC, Open ACC. The Hole API is only usable if you has a Nvidia Card and use the Nvidia BLOB. So i don't care if the GCC can create a Output for the CUDA Runtime, but i have a problem with it if they don't support the FOSS Way. This means the OpenCL Runtime of a other FOSS Runtime or a documented Instruction Set.

It's not only for nvidia hw, OpenACC has opencl support too, and it can be used with radeon gpus.

http://www.openacc.org/SC2013

PGI Booth #2509:

PGI showing OpenACC performance portability with one binary running on both AMD and NVIDIA GPUs

So one binary which uses nvidia/cuda and amd/opencl at the same time.

[Nille]

In the hole discussion it look like some people forget something. The Topic is not only the Output. We have a API in GCC, Open ACC. The Hole API is only usable if you has a Nvidia Card and use the Nvidia BLOB. So i don't care if the GCC can create a Output for the CUDA Runtime, but i have a problem with it if they don't support the FOSS Way. This means the OpenCL Runtime of a other FOSS Runtime or a documented Instruction Set.