This web log is the product of the collaborative, innovative, virtual minds of the editors of the IBM developerWorks Multicore acceleration (Cell/B.E. SDK) zone.

Faster than a speeding bullet: LANL Roadrunner makes TIME Top Ten list ...

... of technology for 2008: Yes, the LANL Roadrunner (see disclaimer for name usage) is number 10 on TIME magazine's "Best Inventions of 2008" list:

On May 26, at 3:30 in the morning, a $133 million supercomputer nicknamed Roadrunner broke the long-sought-after petaflop barrier: 1 quadrillion calculations per second. Built by IBM for Los Alamos National Laboratory, Roadrunner will be used primarily to simulate the effects of aging on nuclear weapons. Next up: the exaflop barrier.

The other nine before LANL Roadrunner:

1. The Retail DNA Test.
2. The Tesla Roadster.
3. The Lunar Reconnaissance Orbiter.
4. Hulu.com.
5. The Large Hadron Collider.
6. The Global Seed Vault.
7. The Chevy Volt.
8. Bullets That Shoot Bullets.
9. The Orbital Internet.

LANL Roadrunner takes a Tekne Award
A win for superior computing in Minnesota: IBM-Rochester received a Tekne Award for the LANL Roadrunner supercomputer (see disclaimer for name usage). Specifically, the company won the IT-Software & Hardware, Communications and Infrastructure award for the hybrid that is the first computer to break the petaflop barrier. (A Tekne Award acknowledges companies and individuals who have demonstrated superior technology advancement and leadership in Minnesota. There is no requirement that this superiority has to also manifest itself outside of MN, but it usually does.)

Catching up with LANL Roadrunner?
Is Cray closing in?: Oak Ridge National Labs just tested its newly installed AMD-based Cray XT5 supercomputer ("Jaguar") and it peaked out at 1.64pflops, just under the 1.7pflops peak for the Cell/B.E.-/Opteron hybrid LANL Roadrunner (see disclaimer for name usage), in essence making it the world's fastest computer for (non-military) science research.

Design on a dime: Folding@home awarded for its good design

PS3 at heart of distributed, sharing system: The StanfordU Folding@home project has just grabbed the Japanese Ministry of International Trade and Industry's Good Design Award for outstanding design in industrial and consumer products. (Folding@home is a distributed PS3 network, chock full of Cell/B.E. processors and visualizer software, designed to let users share idle PS3 time with scientists studying protein folding.) According to the jury: "Analysis of proteins for the purpose of shedding light on diseases is just one example of solution design for social issues, a stance that indicates the direction that design should take in the future. Motivating the people who will be involved in these studies will be the key to success, but the program functions well as an idea for making participation in this project visible on a global scale."

Games people play: Two perspectives on simulation performance

Per cost and raw performance: In the excellent ars technica thinkpiece "Putting the PS3's brain to work," Matt Ford details an article from the Journal of Physical Review E: Statistical, Nonlinear, and Soft Matter Physics ("Accuracy of the lattice-Boltzmann method using the Cell processor") that discusses the computer simulation transition from faster and faster single processors to large multiprocessing systems. The article compares how two Cell/B.E. systems -- the PS3 (best performance per cost) and an IBM QS20/QS21 blade server (best for raw performance) -- handles simulations. The group was trying to determine whether the Cell/B.E. processor's inability to accurately represent floating-point numbers caused a significant loss of precision in the simulations.

(Lattice Boltzmann is a relatively new simulation technique for complex fluid systems. Traditional methods solve the conservation equations of macroscopic properties -- mass, momentum, and energy -- numerically; LB models the fluid consisting of fictive particles and such particles perform consecutive propagation and collision processes over a discrete lattice mesh. Because of its particulate nature and local dynamics, LB has advantages in dealing with complex boundaries, incorporating of microscopic interactions, and parallelization of the algorithm.)

Trends and tradeoffs: YDL says hello to Fixstars Solutions

Terra Soft becoming part of Fixstars: Terra Soft Solutions (Yellow Dog Linux) and Fixstars (a company specializing in Cell/B.E. software) announced that Terra Soft has been bought by Fixstars and will be known as Fixstars Solutions. Many YDL users are acquainted (at least peripherally) with TSS's CEO Kai Staats; Staats will continue on as Fixstars Solutions's COO. The original Fixstars provided Cell/B.E.-based solutions in Japan using YDL, so the company thought it might be a good match to bring YDL development into the fold. The new company will continue work on Linux for the Power architecture with a primary focus on Cell/B.E. systems.

Product watch: Toshiba's upscaling TV

Cell/B.E. speed makes HD TV prettier: Why are we talking about television? Because the heart of the Toshiba Regza 42ZV555D and 46ZV555D TVs -- the first "upscaling" TVs according to the company -- is the Cell Broadband Engine processor.

What's "upscaling"? HD TVs have something called an upscaler built in -- what it does is convert video that is shot at a different resolution to the screen to fit the screen (if you didn't, you'd get a small picture in the center of a big screen). The upscaling hardware:

• Stretches the picture to fit.
• Applies video processing to make the picture look as good as possible.
• Applies processing to convert standard-definition, interlaced pictures into progressive ones because LCD and plasma TVs can only display a progressive image. (With an interlaced image, you see the odd lines then the even ones; with a progressive image, you see all the lines sweeping left to right at the same time.)

Where does the Cell/B.E. processor come in? Faster processing, better quality.

From the Cell/B.E. Architecture forum

This blog-based column looks at some of the more interesting problems and challenges posed recently in the Cell Broadband Engine Architecture forum.

jlabellan84 wants to know if SDK 3.0 apps will play nice with SDK 3.1: I have two questions about SDK3.1:

• Are there problems with backward compatibility? That is, will my applications developed in SDK3.0 run in SDK3.1? and my ALF applications?
• Is it possible to update my PS3 with SDK3.1?

[IBM SDK Service Administrator]: Yes, 3.0 applications will run on 3.1. While we have not done testing on PS3s, it should work the same as 3.0.

redMoon started this in September 2008 when he asked for help to install SDK 3.0 on a PS3 with Fedora Core 8. There are now three forum pages of troubleshooting advice, including whether all the ISOs have been mounted (and how you tell); how to do some manual installation; device file name changes; why installing numactl and blas is important; the whys of using ppu-embedspu instead of embedspu; and more.

People still want to know more about how to install simulator.sdk on Ubuntu: Back in June, iamrohitbanga wanted to know a definitive solution to converting rpm files to debian so he could install the SDK simulator. jgrossm1 suggested IBM Cell SDK/SDK for Multicore Acceleration on Debian/Ubuntu HOWTO. mkistler suggested that "In my experience, the message error while loading shared libraries: libtcl8.4.so: cannot open shared object file: No such file or directory means that the loader could not find libtcl8.4.so when it is trying to load the shared libraries needed by the simulator executable. Is the directory containing libtcl8.4.so in your LD_LIBRARY_PATH?"

The latest in the saga from kauboyhere: I get the same error as mentioned by iamrohitbanga. I installed systemsim-cell 3.0 from its .rpm package using Alien, on Ubuntu 8.04; when I tried ../run_gui from systemsim-cell/run/cell/linux in the Terminal, I got the /opt/ibm/systemsim-cell/bin/systemsim-cell: error while loading shared libraries: libtcl8.4.so: cannot open shared object file: No such file or directory error, as mentioned earlier. I was able to locate the libtcl8.4.so file in my /usr/lib directory. Any solution? tcl8.4 and tk8.4 are installed already.

[choosyg]: I also had serious troubles converting the rpms to deb packages on Hardy, it seems to be a problem with alien/dpkg. Finaly i installed Gutsy and the very same script worked out well. Then i was able to install the simulator deb packages on Hardy and it works fine.

Now the simulatro works, but using spu-gcc failes with the following Message:

/opt/cell/toolchain/bin/spu-gcc -W -Wall -Winline -Wno-main -I. -I 
/opt/cell/sysroot/usr/spu/include -I 
/opt/cell/sysroot/opt/cell/sdk/usr/spu/include -O3 -c wavelet_spu.c
/tmp/ccxZAQnm.s: Assembler messages:
/tmp/ccxZAQnm.s:7: Error: alignment not a power of 2
/tmp/ccxZAQnm.s:11: Error: no such instruction: `hbrr .L8,printf'
/tmp/ccxZAQnm.s:12: Error: no such instruction: `stqd $80,-16($sp)'
/tmp/ccxZAQnm.s:13: Error: no such instruction: `il $80,0'
...

The same program/makefile workes fine on my Fedora 7 System with Simulator.

Panajev2001a wants to know what environmental variable SPE_START_DEBUG does (in reference to gdb debugging).

[IBM SDK Service Administrator]: Information I found:

SPU_DEBUG_START and SPU_INFO were used to set the newly created SPU threads in a sleep state after creation. libspe2 then displayed a message how to connect to the SPU thread with spu-gdb via stderr. Today's combined debugger has different ways to attach to newly created SPU threads. For example GDB has a new command set spu stop-on-load [on|off]. With this set to on the newly created SPU thread is also stopped by GDB itself.

The code for SPU_DEBUG_START and SPU_INFO is still in libspe2 to support the libspe1-to-2 wrapper. So if you set these variables it should show the original behavior of stopping then SPU threads. But this is something that has not been tested (or supported) since the combined debugger was released.

Cell/B.E. Security SDK: APIs: PPE-assisted functions

printf(“This is a %s call with %d parameters.\n”, “PPE-assisted call”,
     (int) 3);

size of auxiliary buffer = 16 + 3*16 + (48 + 32) + 32 = 176 bytes

#include <libisolation.h>

int change_ppuassist_buf_len(unsigned int new_len)

Product watch: SDK 3.1 documentation available

Version 3.1 of the SDK for Multicore Acceleration is here: With new documentation and links to the exciting new IBM Systems Information Center, a single, searchable Web place where the documentation is "live."

It came from the Lab: Don't eat so much CO2

Cell to reduce greenhouse gases via fast sims: IBM and USaskatchewan are collaborating to understand better how to limit carbon dioxide emissions from coal-fired electricity plants -- they're doing this by setting up a simulation system consisting of BladeCenter QS22 Cell/B.E. blades. USaskatchewan research lead Raymond Spiteri noted that such simulations -- including determining how the plants can be modified to harness pre-released CO2 and change it into benign by-products or other fuels like methanol -- "used to take 10 days to complete ..." because the university rented time on a system to run the results. Now the results "will now be completed in under one day ... getting results back faster enables us to be more ambitious in our simulations and try more ideas and different scenarios." Spiteri has already started working with IBM scientists to build a model for relevant chemical reactions and to craft algorithms and code for the Cell/B.E. processors.

Product watch: TotalView supports Sony's BCU-100

Adds support for the ZEGO platform: TotalView Technologies has added support for Sony's BCU-100 Cell Platform (ZEGO) to its TotalView 8.6 source code debugging, analysis, and memory-error-detection tool. The Cell/B.E.- and RSX-based BCU-100 is designed (through its multicore architecture) to improve the efficiency of workflows, especially in the area of digital video production. With TotalView 8.6's Cell/B.E. support, ZEGO users can now examine and troubleshoot digital video program execution. (For more on ZEGO, see "Sony debuts new hybrid multicore Cell/B.E. platform." For more on TotalView, see "TotalView Debugger gets Cell/B.E. support.")

Oddments: What the world will look like ...

when robots take your job: Electrical engineer Marshall Brain (also founder of How Stuff Works) thinks robots may put us out of work in the next 20 to 30 years. At the recent Singularity Summit, Brain claimed that the coming industrial revolution will probably realize the fear that the original Industrial Revolution didn't -- that a "second intelligent species" may be a disruptive technology, at least to the human economy. But he's not trying to hold back the coming tide. Brain suggests we restructure the economy before this happens to make it beneficial for all, humans and robots. For example, distribute the benefits of the revolution to every person via a US$25K-year stipend. (He's also proposed some other interesting ideas to remove "dysfunctional elements from the capitalistic system.")

when robots take your dog's job too!: GATech researcher hope to reduce the cost and availability problems in the service dog industry by designing a robot to handle those tasks. The researchers claim that there are never enough service dogs to go around, they cost about US$16K a piece, and it takes about two years to train them. Phase One of their research yielded a point-and-click laser the user can use to point out the object to be fetched. The robot dogs already respond to a range of verbal commands, so the initial prototype was built with off-the-shelf robotic parts with a custom manipulator that mimics a dog's mouth.

Cell/B.E. Security SDK: APIs: Data transfer

#include <libisolation.h>

int copyin(uint64_t ea, void *ls, uint32_t size)

#include <libisolation.h>

int copyout(uint64_t ea, void *ls, uint32_t size)

#include <libisolation.h>

int decrypt_in(void *dst_ls, const uint64_t ea, const uint32_t message_size,
     const vector unsigned char shared_key)

#include <libisolation.h>

int encrypt_out(const uint64_t ea, void *src_ls, const uint32_t
     message_size, const vector unsigned char shared_key)

Cell/B.E. Security SDK: APIs: Secure File Storage

#include <libisolation.h>

SFS_FILE *sfs_file_var;

#include <libisolation.h>

SFS_FILE *sfs_fopen(char *filename, char *mode,
     const vector unsigned char key);

#include <libisolation.h>

int sfs_fclose(SFS_FILE *stream);

#include <libisolation.h>

int sfs_fread(void *buffer, size_t element_size, size_t count,
     SFS_FILE *stream);

#include <libisolation.h>

int sfs_fwrite(void *buffer, size_t element_size, size_t count,
     SFS_FILE *stream);

#include <libisolation.h>

int sfs_fseek(SFS_FILE *stream, long int offset, int wherefrom);

#include <libisolation.h>

int sfs_ftell(SFS_FILE *stream);

Product watch: SDK version 3.1 available

Version 3.1 of the SDK for Multicore Acceleration is here: With new documentation and links to the exciting new IBM Systems Information Center, a single, searchable Web place where the documentation is "live."

It came from the Lab: Plasmon lens create denser chips

Plasmonic lithography: UC-Berkeley engineers have found a way to breathe new life into optical lithography called plasmonic lithography -- by using metal lenses that focus light by exciting electrons (plasmons), they've been able to create line patterns 80nm wide or about 10 times smaller. The resolution and defraction you can get with regular light is a limiting factor; to overcome this, they leveraged a property of metals -- the presence of free electrons at the surface that oscillate when exposed to light (evanescent waves, smaller than the wavelength of light).