Hacker News Reader

Comment by vetinari | Hacker News Reader

That pressupossition is not that unrealistic.

For example, see shader compilers for GPUs. How many open shader compilers do you see around? Is there any motivation to open them?

On the other side, during the 90's, we have seen many vendors come with new CPU ISAs, extensions of existing ISAs, new SOCs, etc. Many of them didn't have resources or will to write a new C compiler, so they wanted to use someting existing. They were willing to write new GCC backend - and GCC license basically forced them to be open. After this, there was no point in keeping ISA specification secret.

LLVM/clang does not have this effect. It pretty much rewards for being closed. So today, we have shitty (especially on ARM-SoCs) shader compilers for secret ISAs, and you aren't going to see their sources anytime soon.

View on HN

I am not very familiar with the world of shader compilers, but it seems like you have provided the most concrete example I've heard yet of how GCC benefited in the long term in a way that LLVM/clang didn't.

I would love to hear from anyone with more expertise in this realm who might be able to dispute this claim in any way... otherwise this seems to be a smoking gun in GCC's favor!

There are some llvm based shader compilers out there. Some even open (AMD), some not (Nvidia).

The situation with architectures in 90s sounds about right: Those that were kept closed (and where a gcc backend was thus no option for the vendor - mostly embedded stuff) have to this day shitty compilers with unpredictable optimizers.

I'm not sure about the exact conclusion you seem to draw but it seems to be something like "GPL projects tend to have better outcomes than non-GPL". The whole point of the thread doesn't support that conclusion coming directly from the GCC brain trust. It's only natural that niche GCC users are going to hang around the longest.

I'd also like to note LLVM/clang didn't gain large non-apple marketshare until GCC adopted GPLv3 which has more to with it's stagnation than anything IMO. v2 is palatable to many business needs, v3 is not.

Mostly I was looking for a specific/practical current example where using LLVM allowed a company to release something closed-source, where they would have been forced to open it if they used GCC.

The rubber apparently meets the road with the NVidia shader compiler.

> LLVM/clang does not have this effect. It pretty much rewards for being closed.

Only if you can convince people to use your architecture. The tools being closed is a minus in that, and must be taken into account with a lot of other stuff. A closed dev environment may well drown a brand-new ecosystem that you're trying to bootstrap.

Sony recently contributed back a ton of LLVM and clang stuff from their PS4 project, by the way. Why would they do that if keeping it closed was so rewarding?

Imagination just released their IR. LLVM has a backend for AMD GPUs. It's moving, slowly but it's moving...

Reply on news.ycombinator.com