NVIDIA's LLVM CUDA Compiler: Open-Source, 10% Faster

[phoronix]

Phoronix: NVIDIA's LLVM CUDA Compiler: Open-Source, 10% Faster

Back in December there was an announcement from NVIDIA that they would open-source their CUDA compiler based upon the LLVM back-end. NVIDIA today released their new CUDA implementation that's based upon LLVM. Besides being open-source, which will allow it to be ported to new (non-NVIDIA) architectures/hardware, there's also a measurable speed boost in the switch over to LLVM...

http://www.phoronix.com/vr.php?view=MTA0ODc

[cl333r]

It's certainly great, but it would be even better if it was OpenCL 1.2, cause we don't really need 2 technologies which are pretty much the exact same thing.

[agd5f]

Nothing seems to indicate that they released any source code at all much less under an open source license.

[elanthis]

It's certainly great, but it would be even better if it was OpenCL 1.2, cause we don't really need 2 technologies which are pretty much the exact same thing.

CUDA is much more than OpenCL. CUDA in particular allows a much wider range of source languages and features. If either were to disappear, I'd rather OpenCL bit it.

[smitty3268]

CUDA is much more than OpenCL. CUDA in particular allows a much wider range of source languages and features. If either were to disappear, I'd rather OpenCL bit it.

If OpenCL disappeared, CUDA would suddenly gain a lot of NVidia only features. That's why AMD will never support it, unless NVidia releases control of it to some 3rd party like Khronos.

[crazycheese]

CUDA is nvidia-only. This is its only, but huge disadvantage.
It is as if microsoft would opensource directx sdk, but still maintain monopoly, IP package and rights on directx.

[dhewg]

The press release from december reads "will provide the source code for the new NVIDIA® CUDA® LLVM-based compiler to academic researchers and software-tool vendors". That already smells like a non libre license. And there's no sign of any code accessible to the public.

Why does this article praise NV? Because "open source" was mentioned in a press release?
CUDA is just another vendor locked API. AMD should get the praise for pushing OpenCL.

[CornedBee]

LLVM is no longer an acronym. Please don't use the old "low-level virtual machine" anymore, as we've dropped this designation. It's not accurate and thus confusing. LLVM is now just a simple name.

[uid313]

Can this newly released open source code be used by the nouveau graphics device driver?

[FourDMusic]

According to an Anandtech article, which references the press release:

To actually add languages and architectures to CUDA LLVM you need the source code to it, and that’s where CUDA is becoming “open.” NVIDIA will not be releasing CUDA LLVM in a truly open source manner, but they will be releasing the source in a manner akin to Microsoft’s “shared source” initiative – eligible researchers and developers will be able to apply to NVIDIA for access to the source code. This allows NVIDIA to share CUDA LLVM with the necessary parties to expand its functionality without sharing it with everyone and having the inner workings of the Fermi code generator exposed, or having someone (i.e. AMD) add support for a new architecture and hurt NVIDIA’s hardware business in the process.

It seems that the CUDA compiler is being kind of opened up, really so that NVIDIA can reap the benefits of developers that extend CUDA functionality without releasing the full source. And having to apply for access to only part of the source which NVIDIA deems you need to program your extension, I'm not sure you can really call that open-source.