GCC Prepares To Drop Support For CompactRISC CR16

[phoronix]

Phoronix: GCC Prepares To Drop Support For CompactRISC CR16

After being marked as obsolete in GCC 12, GNU Compiler Collection developers are now preparing to remove compiler support for the CompactRISC CR16 architecture in GCC 13...

https://www.phoronix.com/news/GCC-Dropping-CompactRISC-CR16

[uid313]

Good, get rid of all legacy proprietary instruction set architectures in favor of RISC-V!

[carewolf]

I wonder what other 16-bit architectures remain? Looking around I spoted PDP-11 still being default 16-bit integers.

[ssokolow]

I wonder what other 16-bit architectures remain? Looking around I spoted PDP-11 still being default 16-bit integers.

I don't know about GCC, but MSP430 is still supported by rustc last I checked, and, depending on whether you mean == or <=, AVR (which is 8-bit) could count.

[Vistaus]

And all Phoronix people cried, as they either still use that arch or don't like dropping old cruft.

[skeevy420]

I'm no expert in what CompactRISC CR16 is even used for, but, regardless of the architecture or feature, is depreciating something with 12 and removing it with 13 simply moving too fast when the release time frame of GCC seems to be about every six months? Some LTS distributions haven't even made it to 10 let alone 12.

The six month-ish window from 12 to 13 seems like it could be moving so fast that the people using this architecture may not even find out that they've been depreciated until a year or two from now whenever they update whatever OS they cross-compile their code from.

[andrei_me]

release time frame of GCC seems to be about every six months?

That's LLVM, GCC as far as I remember is on a yearly cadence

[skeevy420]

That's LLVM, GCC as far as I remember is on a yearly cadence

I looked at this table wrong.

6 or 12 months, my point on that still stands. Unless the end-user is on Arch, Fedora, or something else bleeding edge or really on top of the news then they may not even know about a GCC depreciation or dropping of a feature before it is too late.

[PerformanceExpert]

I looked at this table wrong.

6 or 12 months, my point on that still stands. Unless the end-user is on Arch, Fedora, or something else bleeding edge or really on top of the news then they may not even know about a GCC depreciation or dropping of a feature before it is too late.

You suggest there will be many end users that might be affected. I bet there are none. I don't think the distros even support ancient ISAs unless they get paid to do so.

GLIBC still supports m68k, IA64, Sparc, PA and even Alpha! Plus there are a bunch of others which I have never heard of, like CSKY, or1k, NIOS2 - perhaps they are still used somewhere by someone, but nobody seems to actively maintain any of these.

I don't see the point of keeping old cruft if there is neither active development nor a large user base. It's not like they are immediately lost forever once support in latest GCC/GLIBC is removed (since anyone can build older GCC/GLIBC for many years). And at some point ISAs simply die, and the world moves on to something better.

[brucehoult]

I have never heard of, like CSKY, or1k, NIOS2 - perhaps they are still used somewhere by someone

CSKY is a proprietary ISA by Alibaba's T-Head and it supported by Linux. T-Head have gone all-in on RISC-V for new stuff but they still sell a lot of CSKY in China. Exact same deal with Andes NDS32, which you don't mention -- lots of it still in use, but they company is 100% RISC-V going forward. MIPS too.

NIOS2 is the soft core supplied by an obscure company called "Intel" for use in their FPGAs. Or, at least, it's the old core/ISA. The new one is NIOS V which is ... you guessed it ... RISC-V.

or1k is OpenRISC, which has actually been used in commercial stuff. The Berkeley people looked it using it for their new ISA back in 2010 but at the time it was 32 bit only and it also had very little unused encoding space to add new stuff. However it's a pretty good point-of-time-in-technology ISA pretty much on a par with MIPS. It runs Linux, has SIMD. It's probably going to die quite quickly now that RISC-V has vector and SIMD extensions.