Vijay Tatkar's Blog

New Look-and-feel to SunStudio Developers landing page
Check out the landing page for Sun Studio Developers. It has a new cool(er) look and better features than the old one; search facility in particular has been enhanced. There are quite a few new technical whitepapers added to this in addition to integrating Solaris development related material that used to reside at different sites earlier. In addition to this, the site features doc reference material, profiles of Sun Studio Heroes, Cool Tools (tools for CoolThreads development), Learning Materials, Support Services, and the popular Downloads service.
The URL is also a much more simpler one to remember:
http://developers.sun.com/sunstudio
Expect to see more enhancements, content additions, etc. Check it out from time to time.
Posted by tatkar ( Oct 30 2006, 10:14:36 AM PST ) Permalink Comments [0]

WebEx-based SunStudio Preview Video for Training Class
About 3 weeks, our TechPubs manager(called Docs outside of Sun lingo), David Lindt, and I did a short 11-minute take on SunStudio Web-based Training Class For Solaris Developers(offered here). The video that resulted from this has been used to promote this class. This was completely unscripted and unrehearsed and off-the-cuff with David leading the discussion and both of us sharing our views and opinions of the product. It was put up on YouTube to promote the class.
Take a look at it here: http://www.youtube.com/watch?v=ViM9P1Rk51s
Tell me what you think about it. I learned a few things myself: the next time around I will be more careful with diction and voice quality and tonal quality, but this time around it was with no preparation at all.
Hope you enjoy it and/or find it informative!
Posted by tatkar ( Oct 25 2006, 03:56:21 PM PDT ) Permalink Comments [0]

Sun Studio Compilers + Solaris have 15% advantage over MS Windows/XP
This interesting performance blog at Geek Patrol was being discussed on two aliases internally:
Sun Ultra 20 M2 Performance
It is an interesting comparison between Windows/XP Compiler and Sun Studio running on Solaris using a benchmark called geekbench. Geek Bench 2006 is available here for Solaris x86 (both 32bit and 64bit versions). Overall, the conclusion is that SunStudio does about 15% better. In particular, I like this pronouncement:
Solaris outperformed Windows in almost every benchmark category, even outperforming Windows dramatically in some specific tests (such as some of the floating point benchmarks). If you’re working with processor-intensive tasks, Solaris might be the operating system for you.
This is inspite of the Sun Studio numbers for STREAM being under-represented. John Poole, who wrote the blog, used the -fast flag, while adding -xvector=simd -xprefetch would have helped it enormously more (by about 50% or so, I'm guessing). My own references to STREAM here and here show these numbers to be compellingly higher.
Comments posted here on OSNews are also worth reading.
Posted by tatkar ( Oct 24 2006, 01:34:58 PM PDT ) Permalink Comments [3]

Compilers continue World Record Performance streak on new SunFire Rev F Opteron
Sun has made much of the Datacenter in a Trailer (aka Project Blackbox) concept, lately. It is clearly one of those ideas that are so compelling that they are also simple. The press has been quick to pick up on it. See here, here at IT Jungle  and here at Reuters, for example.
What didnt get as much press at this time was the other announcement from Sun: Sun has refreshed the entire x64 line of servers with the new AMD Opteron RevF (and whats more, at 2.8GHz, dual-core) chips.
Sun's compilers are at the fore-front once again in the SPEC CPU performance race. The current systems and Sun Studio 11 combine to give very impressive performance overall, nicely edging out systems based on Woodcrest, Itanium and Power5+ by some very healthy margins. In particular:

• The Sun Fire X4100 M2 is up to 40+% faster than Intel Xeon 5160-based servers from Dell, IBM and HP
• Sun Fire X4100 M2 4-core, 2p SPEC 2000 FPrate=119 (peak), 102(base)
• Woodcrest servers from IBM, HP, Dell 4-core, 2p SPEC 2000 FPrate= 82 to 85
  
• The Sun Fire X4200 M2 with  SPEC 2000 FP (peak) of 3515 beats the best Intel Itanium2 result  of 3045 by 15+%
• Sun Fire X4200 M2 with SPEC OMP result of 13222 beats the best 4-thread IBM Power5+ result by 60+%
  

Sun is on a roll here and Sun Studio 11 is helping differentiate. Even on the same system, the advantage with Sun Studio11 is about 20-25% over competitive compilers. For instance, the new Sun Fire X4100 with the score(see above) for SPECfp of 119 beats a similarly configured IBM box with SuSE9 and PathScale compilers with a SPECfp of 90.3 (If you dont like SPECrate numbers, how about SPECfp itself: 2857(base) with SunStudio 11 vs. 2218peak, 2029 base with SuSE9/PathScale on the IBM box... hmm, these IBM numbers are curiosly smaller than I'd have expected on this box.). The idea of using disclosed Base numbers is because these dont use autopar (automatic parallelization, so they are single-core base numbers).

Required Disclosure Statements:

SPEC, SPEComp, SPECCfp and SPECfp Rate are Registered Trademarks of Standard Performance Evaluation Sun's results were submitted for review. For SPEC comparisons, socket equates to chip.
Competitive results from www.spec.org as of Oct 15, 2006.

Sun Fire X4100 M2 (AMD Opteron model 2220SE, 4 cores, 2 chips): SPECfp_rate2000: 119, SPECfp_rate_base2000:  102;
IBM System X 3550 (Xeon 5160, 4 cores, 2 chips, Windows 2003 Server EE): SPECfp_rate_base2000 – 82.1
HP ProLiant DL360G5 (Xeon 5160, 4 cores, 2 chips, RHEL 4AS U3): SPECfp_rate2000 – 84.7
Dell PowerEdge 1950 (Xeon 5160, 4 cores, 2 chipsp, RHEL 4AS U3): SPECfp_rate2000 - 83.4
IBM System X 3455 (AMD Opteron model 2220SE, 4 cores, 2 chips, SLES 9 SP3): SPECfp_rate2000: 90.3

Sun Fire X4100 M2 (AMDOpteron model 2220SE, 4 cores, 2 chips): SPECfp2000 – 3515
Sun Fire X4100 M2 (AMD Opteron model 2220SE, 4 cores, 2 chips): SPECfp2000 Base– 2857
Bull NovaScale 3045 (Itanium2 9050, 8 cores, 4 chips, Bull Linux AS4): SPECfp2000 – 3017
IBM System X 3455 (AMD Opteron model 2220SE, 4 cores, 2 chips, SLES 9 SP3): SPECfp2000: 2218 Peak, 2029 Base

Sun Fire X4100 M2 (AMD Opteron model 2220SE, 4 cores, 2 chips, 4 thread): SPECompM2001 – 13222
IBM System p5 520 (POWER5+, 2 cores, 1 chip, 4 threads, AIX5L V5.3): SPECompM2001 - 8174

Posted by tatkar ( Oct 23 2006, 10:04:32 AM PDT ) Permalink Comments [0]

Performance Comparison: Sun Studio vs GCC on STREAM Benchmark I have previously described  the STREAM Benchmark and the results we were seeing with its OpenMP version and what we got by turning on Automatic Parallelization  in the compiler.
Here I'd like to put out comparative results with the GCC compiler

Function	Sun Studio 11(MB/s)	GCC4.1 (MB/s)
Copy	4658	2766
Scale	4614	2745
Add	4628	2970
Triad	4627	2969

This is roughly a 1.6x advantage with the Sun Studio compiler.
The comparisons were done on exactly the same box. The box was a SunFire V40z with 4 x 2.6GHz processors and PC3200 CL3 DDR SDRAM ECC Regd. memory.

The Optimization options used in these cases were:
Sun Studio: -fast -xarch=amd64a -xvector=simd -xprefetch -xprefetch_level=3
GCC 4.1: -O3 -funroll-all-loops -ffast-math -fpeephole -m64 -mtune=k8 -fprefetch-loop-arrays

Function	Sun Studio 11(MB/s)	Sun Studio 11(MB/s)  4proc  Autopar
Copy	4658	18120
Scale	4614	18108
Add	4628	17758
Triad	4627	17626

For a 4CPU machine, this is roughly a 3.9x scalability, which is incredible!
Of course, the GCC compiler isnt able to exploit such scalability because it does neither Automatic Parallelization nor OpenMP at this time. (Its working on at least OpenMP support, so at least this discrepancy will be addressed in a future release.).
Posted by tatkar ( Oct 18 2006, 02:08:06 PM PDT ) Permalink Comments [2]