|
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
Vol. II, No. 2 BDTI's DSP Insider February 2002
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
This month:
*** TI's OMAP Lands in Palm
*** Lexra's IP Business: Another One Bites the Dust
*** Reconfigurable Hardware for DSP
*** "Impulse Response," a news analysis and opinion column written
by Jeff Bier, BDTI's General Manager, and featured in EE Times
*** Benchmarking C Compilers for DSP Applications
*** "Inside" Report on ARM
*** BDTImark2000(TM) for the TI 'C64xx
*** BDTIsimMark2000(TM) scores for ARM, SC110, SP-5, and DSP5685x
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
*** TI's OMAP Lands in Palm
On December 17, 2001 Palm and Texas Instruments announced that Palm
will base a set of next-generation handheld computers on TI's OMAP
processor platform, which combines an ARM ARM9 microprocessor core
with TI's 'C55xx DSP core. This announcement came as part of a
agreement between the two companies that covers everything from
collaboration on product development to joint marketing. Palm expects
to introduce OMAP-based handheld computers in about one year.
Palm had previously announced that it was porting its operating system
to ARM's family of ARM microprocessor cores; by choosing the OMAP
platform, Palm is clearly signaling that it considers high-performance
DSP applications a critical factor for the success of next-generation
handheld computers. Although ARM processors power many of today's
embedded DSP applications, e.g., portable digital audio players, these
cores offer limited signal processing speed. Consider the two cores
in the OMAP platform: according to BDTI's analysis, a 200 MHz 'C55xx
is over three times faster than a 200 MHz ARM9 on DSP tasks.
The agreement between Palm and TI demonstrates the effectiveness of
TI's savvy diversification strategy. In the early days, TI's DSP
efforts, like those of its competitors, focused almost exclusively on
telecommunication applications. Over time, TI broadened its scope to
markets such as hard disk drive servo controllers and digital cameras.
Many other DSP vendors have remained narrowly focused or have met
less success in new markets.
TI's approach to handheld computing is in keeping with a strategy of
diversification: in addition to its agreement with Palm, TI has
announced that its OMAP platform will support Microsoft's Windows CE
and the Linux operating system. As signal processing becomes ever
more pervasive, DSP vendors would do well to emulate TI's
diversification strategy.
*** Lexra's IP Business: Another One Bites the Dust
On December 31, 2001, MIPS announced that it was dropping its
patent-infringement lawsuit against Lexra as part of an agreement
between the two companies. As part of this agreement, Lexra will
assign its processor IP assets to MIPS and become a MIPS32 licensee.
This announcement spells the end of Lexra's DSP-enhanced licensable
core: Lexra is exiting the IP business and becoming a fabless
semiconductor company focused on network processors.
In its lawsuit, filed in November 1999, MIPS alleged that Lexra has
violated at least two and as many as eleven MIPS patents. One of the
patents covers a method for preserving precision in the
single-instruction multiple data (SIMD) operations in MIPS' MDMX
multimedia extensions. MIPS alleged that the SIMD operations in
Lexra's DSP-oriented Radiax instruction set extensions violated this
patent.
Ironically, while the MDMX extensions have been largely ignored, Lexra
claimed 11 licensees for its Radiax-enhanced cores. Now that MIPS
owns Lexra's processor IP, will MIPS adopt the Radiax DSP extensions
for its own use? MIPS certainly has strong motivation to revisit its
DSP strategy; competitors like ARM and Hitachi are gaining wide
acceptance of their DSP-enhanced processors, leaving MIPS to play
catch-up.
Lexra is not the only core licensor to exit the IP business recently.
Last February, Massana dropped its IP offerings and repositioned
itself as a fabless chip company. In December, picoTurbo announced
that it was exiting the IP business as part of a settlement with ARM.
As in the Lexra/MIPS settlement, picoTurbo will cease business as an
IP vendor and relinquish its IP assets to ARM (unlike Lexra and
Massana, it appears that picoTurbo does not have a fall-back position
and is closing its doors). The number of core vendors has exploded in
recent years; the Lexra and picoTurbo settlements may signal the start
of a significant contraction in the ranks of core vendors. It is
quite possible that other core vendors will go the way of Lexra,
Massana, and picoTurbo before this cycle ends.
*** BDTI Case Study
This Month: Reconfigurable Hardware for DSP
Reconfigurable hardware includes a broad spectrum of architectures
ranging from FPGAs to processors with reconfigurable data paths.
Between these extremes are FGPAs with hard-wired functions like
multipliers and embedded processor cores. Ideally, reconfigurable
architectures combine some of the best attributes of ASICs (e.g.,
hardware structures highly tuned for the application) with some of the
best attributes of traditional processors (e.g., solid development
tools, flexibility to make last-minute changes). However, these
benefits may be difficult to attain, and it is equally possible for
reconfigurable architectures to combine some of the worst attributes
of other technologies, such as long, painful development cycles and
poor cost-effectiveness.
In the past few years, vendors of reconfigurable hardware have
increasingly emphasized the DSP capabilities of their products and
have introduced new architectures, development tools, and IP building
blocks targeting DSP applications. One reason for this increasing
activity is the increasing computational workload of DSP applications
— reconfigurable hardware enables system developers to create highly
parallel hardware structures that match the needs of key algorithms.
This can be particularly powerful for complex algorithms that do not
map efficiently to traditional processors and for performance-hungry
applications with rapidly changing specifications.
Comparing reconfigurable devices to traditional processors requires
careful thought because the underlying architectures are fundamentally
different. For example, reconfigurable devices often can perform an
enormous number of operations in parallel, but harnessing that
potential parallelism to quickly create an efficient implementation of
a particular algorithm can be extremely challenging.
Beginning several years ago, BDTI has been involved in evaluating
reconfigurable architectures for DSP applications. For example, in a
contract consulting engagement, BDTI evaluated a novel reconfigurable
processor for use in telecom applications. BDTI compared the
performance of this reconfigurable processor to that of a popular DSP
processor using both the BDTI Benchmarks(TM) and custom benchmarks
derived from the target application. Going forward, BDTI will expand
its analysis activities focused on reconfigurable hardware. For
example, BDTI will publish a report exploring the strengths and
weaknesses of FPGAs for DSP applications later this year.
To learn how BDTI's DSP expertise can help you deploy or use
reconfigurable hardware in DSP applications, or to learn more about
BDTI's upcoming FPGA report, please contact Jeremy Giddings at BDTI
(giddings@BDTI.com).
*** Impulse Response, by Jeff Bier
C-ing the Future of Hardware Design
At one time, most DSP applications were powered either by
instruction-set processors or by application-specific fixed-function
chips. There was a clear distinction between these two types of
solutions and clear differences in the associated design
methodologies: processors required software development, while ASICs
required hardware design. Today, the line between hardware and
software has been blurred, particularly with flexible architectures
that combine instruction-set processors with custom, customizable, or
reconfigurable logic.
One key advantage of such flexible architectures is the ability to
move workloads between the predefined processor and specialized logic
incrementally; e.g., by first implementing the application on the
processor and then creating just enough custom logic to meet
performance requirements. However, doing so can be made difficult by
the differences between the software and hardware design languages.
One way to reduce the burden of application development for this kind
of a flexible architecture is to design the software and the hardware
with similar languages. Given its popularity, C is an obvious
candidate for a common design language; companies like Adelante and
Celoxica offer tools that allow implementation from C-like code into
custom logic. This approach seems to be gaining momentum: in
November, Celoxica announced agreements with both Altera and Xilinx to
adapt its C-based language for use on next-generation FPGAs from these
two companies.
Although the idea of a unified design language is tantalizing, C may
not be up to the challenge. Because C lacks support for many
fundamental DSP concepts like fractional data types, it is difficult
for compilers to generate efficient DSP code from C. In addition, a
"C-based" hardware design language may in fact bear only superficial
resemblance to C; hence, moving application tasks between the
instruction-set processor and the specialized hardware may require
much more effort than simply using a different compiler.
Of course, the potential benefits and drawbacks of more-uniform design
methodologies are not limited to customizable or reconfigurable
processors. DSP applications increasingly depend on heterogeneous
architectures such as a combination of a general-purpose processor and
a DSP processor. Those who are able to adapt their design
methodologies to make effective use of such heterogeneous designs will
have a significant advantage in fielding competitive products.
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
*** Benchmarking C Compilers for DSP Applications
Some months ago, the "DSP Insider" announced the commercial roll-out
of BDTI's C compiler benchmarking methodology. BDTI is about to kick
off the inaugural application of the methodology in a multi-client
study. The methodology incorporates quantitative analysis of
compiler efficiency using a variety of benchmarks as well as detailed
qualitative analysis of usability, features, strengths, and
weaknesses.
If you are interested in participating in the study, please contact
Jeremy Giddings (giddings@BDTI.com).
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
*** "Inside ARM"
BDTI's newest "Inside" report, "Inside the ARM ARM7, ARM9, and
ARM9E," provides benchmarking data and analysis of the DSP
capabilities of the most widely used family of processor cores. This
report provides insight into how the newer ARM cores, the ARM9 and
DSP-enhanced ARM9E, stack up against the venerable ARM7, as well as
against comparable competitors from Hitachi, Motorola, and TI.
For information, follow the link from the BDTI home page at
www.BDTI.com. And for information on all of BDTI's technical reports,
go to www.BDTI.com/products/services_overview.htm#publications.
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
*** BDTImark2000(TM) for the TI 'C64xx
BDTI has released a BDTImark2000 score for the Texas Instruments
TMS320C64xx.
For this score, as well as BDTImark2000 and BDTIsimMark2000 scores for
other processors, go to www.BDTI.com/bdtimark/BDTImark2000.htm.
The BDTImark2000 is a summary measure of DSP speed distilled from a
suite of DSP benchmarks developed and independently verified by BDTI.
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
*** BDTIsimMark2000(TM) scores for ARM, SC110, SP-5, and DSP5685x
BDTI has published the first set of BDTIsimMark2000 scores on its web
site. Processors included are the 3DSP SP-5, the ARM ARM7, ARM9, and
ARM9E, the Motorola DSP5685x, and the StarCore SC110.
For these scores, go to www.BDTI.com/bdtimark/BDTImark2000.htm.
The BDTIsimMark2000 is a summary measure of simulated DSP speed
distilled from a suite of DSP benchmarks developed and independently
verified by BDTI.
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
*** About BDTI
BDTI is an independent source for DSP technology analysis and
optimized DSP software. From rigorous technical analyses of
processors for DSP, such as the "Inside" series of processor reports,
to highly regarded technology training classes, BDTI is the trusted
independent source for reliable information on DSP technology.
For more information, visit our Web site at www.BDTI.com.
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
The next issue of BDTI's "DSP Insider" is coming in March. An
archive of previous issues of the BDTI "DSP Insider" is now available
on BDTI's Web site. Follow the link from
www.BDTI.com/dspinsider/dspinsider.htm.
If you have comments, suggestions, or other feedback about the "DSP
Insider," please send email to dspinsider@bdti.com.
BDTI's "DSP Insider" is a free monthly electronic newsletter published
by Berkeley Design Technology, Inc. If our newsletter was forwarded
to you and you would like to receive it regularly, please register at
www.BDTI.com/dspinsider.htm.
If you no longer wish to receive the "DSP Insider," send an email
message to dspinsider@bdti.com with the words "Remove me" in the
subject line.
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
BDTI's DSP Insider (c) 2002 Berkeley Design Technology, Inc.
<< previous issue - next issue >>
|
