LWN: Comments on "Sun Microsystems Launches OpenSPARC Project"

This should help with porting also

zblaxell — Wed, 01 Feb 2006 05:44:51 +0000

I have this vision of a meeting room in Sun somewhere. A design engineer is getting chewed out for not responding to all these requests for porting specs.

"Porting specs? I'm a chip designer, not a technical writer! Just give 'em the damn VHDL files and tell 'em figure it out themselves..."

...or in the future...

"[PATCH] fixes mumblefoo error in L1 cache controller"

This vision may have been influenced by Microsoft's recent action, where after the time and expense of creating 12000 pages of utterly useless technical documentation (and being caught at it), they decided to try to get out of their EU obligations by releasing their source code.

Building masks for a modern CPU-sized chip is sufficiently expensive that when designers find errors they'll bend over backwards to correct the designs by editing them rather than generating a new set. Imagine software development where the linker for your target architecture costs a million dollars per .o file to run--developers do as much work as possible with a compiler for some other target, and if bugs are found after running the linker for the primary target, they try to patch the binaries as a first resort, and recompile as a last resort.

I'm not sure how anything like open source fits into that world--for example, the GPL wants the preferred form for editing the program, but in the chip design case there's edits to both the "source" and "binary". It's easy for smaller designs that can fit into a big FPGA (cheap compiler, no hand-editing afterwards), but that's a whole different technology.

Sun Microsystems Launches OpenSPARC Project

remijnj — Wed, 07 Dec 2005 17:29:44 +0000

I think you are right.

This article from 2001 tells the story.

Sun Microsystems Launches OpenSPARC Project

beoba — Wed, 07 Dec 2005 16:07:53 +0000

ESA? I'm picturing European Space Agency.

Sun Microsystems Launches OpenSPARC Project

khim — Tue, 06 Dec 2005 23:40:07 +0000

You can give pretty good approxymation for 99.999% cases - enough for GCC developers, for example (even if may be not for hand-assembly). But then you'll be forced to admit that your CPU can sometimes spend 50-80 CPU ticks as heater without when no actual work is done - and that's just embarassing...

Sun Microsystems Launches OpenSPARC Project

gnb — Tue, 06 Dec 2005 22:55:14 +0000

> Note that Intel, for instance, no longer even publish *instruction
>timings* for their CPUs (although AMD do).
Annoying though that is, I suspect it's almost unavoidable given the
complexity of modern CPUs ("well, it'll take this many cycles unless it's
a mis-predicted branch, or one of the registers it uses has a dependency
on a previous instruction, or ..."). In fact I'd be interested to know
which vendors have real instruction timings for their CPUs as opposed to
statistical averages based on simulating (hopefully representative)
benchmark kernels.

Sun Microsystems Launches OpenSPARC Project

pjones — Tue, 06 Dec 2005 21:56:45 +0000

Well, there are already two designs for 32-bit sparcs out in the wild -- microSPARC-IIep is mostly available*, and the ESA has their own which is available.

(* IIRC, the cpu design is available without the on-core pci controller, so you do have to build your own I/O devices.)

Sun Microsystems Launches OpenSPARC Project

jmorris42 — Tue, 06 Dec 2005 21:45:21 +0000

You just never know with Sun do ya.

Having details of the instruction set properly documented is great. Not sure what having the Verilog files available will accomplish but it can't be a bad thing.

Open Source hardware just isn't as useful since most people lack the ability to make use of it. Copyrights made perfect sense when only a few had a printing press, now many question it. The software world instantly chaffed under closed source because almost every computer is able to compile and vast numbers of users are also capable of some level of development activity. But since few have a chip fab.... yet. :)

Will be interesting to see if there is an uptake among the embedded world, but I'd think most of them would be more interested in the 32bit SPARC.

Sun Microsystems Launches OpenSPARC Project

asamardzic — Tue, 06 Dec 2005 21:31:06 +0000

Great stuff for all enthusiast... I remember enjoying examining LEON Sparc model in VDHL, can't wait to take a look into "real stuff". I hope only the release wont' be followed by "after-examining-this-model-you-are-not-allowed-to-write-a-line-of-Verilog-code" kind of license...

Sun Microsystems Launches OpenSPARC Project

nix — Tue, 06 Dec 2005 21:24:17 +0000

Specs at the level Sun are proposing are most unusual in and of themselves. Note that Intel, for instance, no longer even publish *instruction timings* for their CPUs (although AMD do).

Three cheers for Sun for this move.

Sun Microsystems Launches OpenSPARC Project

adulau — Tue, 06 Dec 2005 21:01:52 +0000

They should use the Open Cores (http://www.opencores.org/projects.cgi/web/opencores/mission) approach to release the specification as a real "free" hardware.

This should help with porting also

freeio — Tue, 06 Dec 2005 20:47:06 +0000

Considering how difficult it has been to get full data on the UltraSPARC III, (things required for accurate porting) this is an interesting change. While it is great for the hardware hacker, it is also a real benefit to the developers who do the direct interface to the silicon itself.

SUN is, at the very least, unpredictable.

Sun Microsystems Launches OpenSPARC Project

spot — Tue, 06 Dec 2005 20:46:02 +0000

Here's hoping we can get to the hardware specifications without having to look at CDDL code.