DSP Design Methodology (ASPI8-4) - Institut for Elektroniske Systemer

Download Report

Transcript DSP Design Methodology (ASPI8-4) - Institut for Elektroniske Systemer

ASPI8-4
DSP Design Methodology
ASPI8-4: DSP Design Methodologies, AAU
1/66
Practical issues
• Webpage: http://www.cs.aau.dk/~moullec/aspi8• Literature: available on the webpage and
in your mailbox (or ask Dorthe)
• YOUR feedback about the course is most welco
• during the lecture
• after the lecture ([email protected])
ASPI8-4: DSP Design Methodologies, AAU
2/66
Outline of the course
• MM1: Design meta models
Introduction/Y-chart model/Rugby meta model
• MM2: Computational models
FSMD/SDFG/CDFG/HCDFG/ASM-charts
• MM3: Design Domains
Function/Communication/Data/Timing
ASPI8-4: DSP Design Methodologies, AAU
3/66
MM1:
Design meta models
• Introduction
• Y-chart model
• Rugby meta model
• Conclusion
ASPI8-4: DSP Design Methodologies, AAU
4/66
Introduction: the need for design methodologies
• Pervasive computing with an always-on Internet infrastructure.
• Voice activated controls enabled by speech synthesis.
• Wireless communications that keep us in touch no matter where
or when we need it.
• A consumer industry dominated by a plethora of entertainment
devices.
Office
es
Gam
Music
S, …
MM
,
S
EM
oth
eto
Blu
ASPI8-4: DSP Design Methodologies, AAU
Intern
et
5/66
Introduction: the need for design methodologies
“ Digital signal processing (DSP) has become the technology
of focus with consensus expectations of exponential growth.
Everybody knows that DSP is the technology driver for the
semiconductor industry,”
Will Strauss, analyst, Forward Concepts Co., Tempe, AZ.
ASPI8-4: DSP Design Methodologies, AAU
6/66
Introduction: the need for design methodologies
But ???
Source: Xilinx
 NEW PLATFORMS (DSPs, FPGAs, SoC,…)
 NEW METHODOLOGIES
ASPI8-4: DSP Design Methodologies, AAU
7/66
Introduction: the need for design methodologies
But ???
{
nlSsBord = nl - n lCo re2;
ncSsBord = nc - ncCore2;
}
for (y=n lCore2; y<n lSs Bord; y++) {
for (x=ncCore2; x<ncSsBord; x++) {
{
xy Offset= (y * nc) + x;
erode=255;
}
for (k=0; k<nbPointIn Core; k++) {
erode= (erode && pImage[ xy Offset + offset[k] ]);O
ffice
}
if (erode) {
pImageErode[xyOffset]=255;
{
}
nlSsBord = nl - n lCo re2;
else {
ncSsBord = nc - ncCore2;
pImageErode[xyOffset]=0;
}
}
for (y=n
} lCore2; y<n lSs Bord; y++) {
for (x=ncCore2;
x<ncSsBord; x++) {
}
{
for (y=0; y<n lCore2; y++) {
xy Offset=
(y *x<(nc);
nc) + x;x++) {
for (x=0;
pImageErode[(y * nc) + x]=0;
erode=255;
}
}
}
{
for (k=0; k<nbPointIn Core; k++) {
nlSsBord = nl - n lCo re2;
erode= (erode && pImage[ xy Offset + offset[k] ]);
ncSsBord = nc - ncCore2;
}
}
if (erode) {
for (y=n lCore2; y<n lSs Bord; y++) {S,
pImageErode[xyOffset]=255;
M
for (x=ncCore2; x<ncSsBord;Ex++) {
}
{
else {
xy
Offset=
(y
*
nc)
+
x;
pImageErode[xyOffset]=0;
?
es
Gam
Standards
S, …
, MM
EMS
…
MS ,
Blu
M
eto
oth
}
erode=255;
oth
}
eto
}
}
Blu
for{(k=0; k<nbPointIn Core; k++) {
for (y=0; y<n lCore2; y++)
erode=
(erode
&&
pImage[
xy
Offset
+
offset[k]
]);
for (x=0; x<(nc); x++) {
pImageErode[(y * nc) + }x]=0;
if (erode) {
}
pImageErode[xyOffset]=255;
}
}
else {
pImageErode[xyOffset]=0;
}
}
}
for (y=0; y<n lCore2; y++) {
for (x=0; x<(nc); x++) {
pImageErode[(y * nc) + x]=0;
}
}
{
nlSsBord = nl - n lCo re2;
ncSsBord = nc - ncCore2;
}
for (y=n lCore2; y<n lSs Bord; y++) {
for (x=ncCore2; x<ncSsBord; x++) {
{
xy Offset= (y * nc) + x;
erode=255;
}
for (k=0; k<nbPointIn Core; k++) {
erode= (erode && pImage[ xy Offset + offset[k] ]);
}
if (erode) {
pImageErode[xyOffset]=255;
}
else {
pImageErode[xyOffset]=0;
}
}
}
for (y=0; y<n lCore2; y++) {
for (x=0; x<(nc); x++) {
pImageErode[(y * nc) + x]=0;
}
}
{
nlSsBord = nl - n lCo re2;
ncSsBord = nc - ncCore2;
}
for (y=n lCore2; y<n lSs Bord; y++) {
for (x=ncCore2; x<ncSsBord; x++) {
{
xy Offset= (y * nc) + x;
erode=255;
}
{
for (k=0; k<nbPointIn Core; k++) {
nlSsBord = nl - n lCo re2;
erode= (erode && pImage[ xy Offset + offset[k] ]);
ncSsBord = nc - ncCore2;
}
}
if (erode) {
for (y=n lCore2; y<n lSs Bord; y++) {
pImageErode[xyOffset]=255;
for (x=ncCore2; x<ncSsBord; x++) {
}
else { {
xy Offset= (y * nc) + x;
pImageErode[xyOffset]=0;
}
erode=255;
}
}
}
fory<n
(k=0;
k<nbPointIn
for (y=0;
lCore2;
y++) { Core; k++) {
for (x=0;erode=
x<(nc);(erode
x++) &&
{ pImage[ xy Offset + offset[k] ]);
}
Music pImageErode[(y
* nc) + x]=0;
if (erode) {
}
pImageErode[xyOffset]=255;
}
{
}
nlSsBord = nl - n lCo re2;
else {
ncSsBord = nc - ncCore2;
pImageErode[xyOffset]=0;
}
}
for
} (y=n lCore2; y<n lSs Bord; y++) {
}for (x=ncCore2; x<ncSsBord; x++) {
{ (y=0; y<n lCore2; y++) {
Inte for
xy(x=0;
Offset=
(y * nc)
+ x;{
for
x<(nc);
x++)
rn e t
pImageErode[(y * nc) + x]=0;
}erode=255;
}}
for (k=0; k<nbPointIn Core; k++) {
erode= (erode && pImage[ xy Offset + offset[k] ]);
}
if (erode) {
pImageErode[xyOffset]=255;
}
else {
pImageErode[xyOffset]=0;
}
}
}
for (y=0; y<n lCore2; y++) {
for (x=0; x<(nc); x++) {
pImageErode[(y * nc) + x]=0;
}
}
 INCREASED COMPLEXITY (multiple standards, Quality of Service, …)
 NEW METHODOLOGIES
ASPI8-4: DSP Design Methodologies, AAU
8/66
Introduction: the need for design methodologies
What is a methodology ?
• “The way in which you find out information; a methodology describes
how something will be (or was) done. The methodology includes the
methods, procedures, and techniques used to collect and analyze
information”
• “A documented approach for performing activities in a coherent,
consistent, accountable, and repeatable manner”
• “A collection of methods, procedures, and standards that defines an
integrated synthesis of engineering approaches to the development of a
product”
• A method is not a methodology, it’s part of a methodology.
ASPI8-4: DSP Design Methodologies, AAU
9/66
ASPI8-4: DSP Design Methodologies, AAU
10/66
Introduction: elements for a design methodology
• Domains: functional, structural, physical, …
• Abstraction levels: system, algorithmic, RTL, logic, circuit, …
• Activities: analysis, synthesis, refinement, optimization …
ASPI8-4: DSP Design Methodologies, AAU
11/66
Specification domains for digital systems
• Functional domain: emphasis is on behavior (functionality, I/O),
no reference to the implementation of this behavior
• Structural domain: interconnected functional components,
possibly hierarchical
• Physical domain: a.k.a geometric, physical placement in space and
physical properties, no direct relation to functionality
ASPI8-4: DSP Design Methodologies, AAU
12/66
Abstraction levels for digital systems
• System level: subsystems (modules,…) interacting with each other
(for example exchanging messages).
Structural elements: processsors, communication channels, memories,…
• Algorithmic level: a.k.a behavioral level, algorithm(s) describing
the functionality.
Structural elements: controller, netlist,…
• Register Transfer Level: a.k.a RTL level, operations described as
transfer of data between registers and functional units.
Structural elements: registers, ALUs, multiplexers, controller,…
ASPI8-4: DSP Design Methodologies, AAU
13/66
Abstraction levels for digital systems
• Logic level: operations described as Boolean equations.
Structural elements: gates and interconnections.
• Circuit level: differential equations describing the relation between
voltage, current, …
Structural elements: transistors, resistors, capacitors
ASPI8-4: DSP Design Methodologies, AAU
14/66
Meta models : Y-chart and Rugby
• Meta model: a model used to describe and analyze other models
• Higher level of abstraction
ASPI8-4: DSP Design Methodologies, AAU
15/66
Y-chart
• The Y-chart is a representation proposed by Gajski and Khan to capture
specifications domains, abstraction levels and their inter-relation.
• Specification domains are represented as three axes (hence the Y)
In each domain the specification can be at different abstraction levels.
• The Y-chart tries to capture the relation between design activities
(synthesis activities) such as synthesis, partitioning, …
ASPI8-4: DSP Design Methodologies, AAU
16/66
Domains and Levels of Modeling
Functional
Structural
high level of
abstraction
low level of
abstraction
Geometric
ASPI8-4: DSP Design Methodologies, AAU
“Y-chart” by
Gajski & Kahn
17/66
Domains and Levels of Modeling
Functional
Structural
Algorithm
(behavioral)
Register-Transfer
Language
Boolean Equation
Differential Equation
Geometric
ASPI8-4: DSP Design Methodologies, AAU
“Y-chart” by
Gajski & Kahn
18/66
Domains and Levels of Modeling
Functional
Structural
Processor-Memory
Switch
Register-Transfer
Gate
Transistor
Geometric
ASPI8-4: DSP Design Methodologies, AAU
“Y-chart” by
Gajski & Kahn
19/66
Domains and Levels of Modeling
Functional
Structural
Polygons
Sticks
Standard Cells
Floor Plan
Geometric
ASPI8-4: DSP Design Methodologies, AAU
“Y-chart” by
Gajski & Kahn
20/66
Inverted as compared to previous slides
ASPI8-4: DSP Design Methodologies, AAU
21/66
ASPI8-4: DSP Design Methodologies, AAU
22/66
Design activities
ASPI8-4: DSP Design Methodologies, AAU
23/66
Rugby meta model
Axel Jantsch, Shashi Kumar*, Ahmed Hemani
Royal Institute of Technology
Department of electronics
Electronic systems design laboratory
Electrum 229
S – 164 40 Kista, Sweden
* School of Engineering, Jönköping University, Jönköping,
Sweden
ASPI8-4: DSP Design Methodologies, AAU
24/66
Rugby meta model
“A conceptual framework, in which designs, design processes and
design tools can be studied.
The model has similar objectives as the well known Y chart but its
scope is extended to handle designs and design processes required
for complex systems requiring
concurrent processes and mixed HW/SW implementation.
The Rugby model has four domains, namely,
Computation, Communication, Data and Time. “
Axel Jantsch, Shashi Kumar, Ahmed Hemani
ASPI8-4: DSP Design Methodologies, AAU
25/66
Rugby meta model
“The behavioral domain of the Y chart is replaced with more
restricted computation domain.
The structural and physical domain of the Y chart are merged into
a more generic domain called Communication.
The new domains Data and Time have become necessary to model
data abstractions used at various levels of design, and to explicitly
model timing constraints at various levels in the design process,
respectively”.
Axel Jantsch, Shashi Kumar, Ahmed Hemani
ASPI8-4: DSP Design Methodologies, AAU
26/66
Rugby meta model
“The Rugby model is able to represent mixed HW/SW designs and
design processes for HW/SW codesign at various levels of
abstraction.
It not only can represent state-of-the-art of current electronic
systems and electronic system design automation,
but it also points to gaps in the availability of tools and
methodologies for designing complex system”.
Axel Jantsch, Shashi Kumar, Ahmed Hemani
ASPI8-4: DSP Design Methodologies, AAU
27/66
Rugby meta model
Domain, Hierarchy, and Abstraction:
“Abstraction and hierarchy are two different means to handle
complexity.
Hierarchy partitions a system into smaller parts; abstraction replaces
one model with another model which contains significantly less
detail and information.
Both reduce the amount of information and details that must be
considered for a particular purpose.
A third way to tackle complexity is the analytical slicing of models
into domains.
Unlike hierarchy and abstraction it does not lead to physically
separate models or parts of a model, but it is an analytical means to
study different aspects of a model separately”. Axel Jantsch, Shashi Kumar, Ahmed Hemani
ASPI8-4: DSP Design Methodologies, AAU
28/66
Rugby meta model
Hierarchy: “A hierarchy is a, possibly recursive, partitioning of a
design model such, that the details of each part is hidden into a
lower hierarchical level.
Hierarchy defines the amount of information presented and visible
at a particular hierarchical level of a model.
At all hierarchy levels the same modelling concepts are used. The
motivation for hierarchy is to hide information when it is not
needed and to display details when they are useful”.
Axel Jantsch, Shashi Kumar, Ahmed Hemani
ASPI8-4: DSP Design Methodologies, AAU
29/66
Rugby meta model
Abstraction: “An abstraction level defines the modelling concepts
and their semantics for representing a system.
The type of information available at different levels is different. A
higher level ignores some irrelevant information at a lower level
and encodes it using different concepts.
Abstraction defines the type of information present in a model.
Unlike hierarchy, abstraction is not concerned with the amount of
information visible, but with the semantic principles of a model. In
general, the movement from high to low abstraction levels includes
a decision making process”.
Axel Jantsch, Shashi Kumar, Ahmed Hemani
ASPI8-4: DSP Design Methodologies, AAU
30/66
Rugby meta model
Domain: “A domain is an aspect of a model which can logically be
analysed independently from other
aspects.
A domain focuses on one design aspect. Real models always contain
several aspects or domains
but different models may emphasize one domain more than another.
Models, which focus on one particular
domain use modelling notations and constructs to model the design
aspect of concern explicitly.
Other design aspects may be implicitly part of the models. Whereas
hierarchy and abstraction simplifies the design, domain partitioning helps the developers of tools and
methodologies to cope with the
complexity. The domains considered in this article are computation,
communication, data, and time.”
Axel Jantsch, Shashi Kumar, Ahmed Hemani
ASPI8-4: DSP Design Methodologies, AAU
31/66
Rugby meta model
“While hierarchical partitioning is mostly a manual endeavour, the
definition of abstraction levels and transformations between them, is
behind most of the advances in design automation.
While hierarchy is a general and important concept, it is not explicit in
the Rugby model.
We assume, hierarchy is possible at all abstraction levels in any
domain.
The Rugby model is based on these definitions and identifies four
domains, computation, communication, time and data, and several
abstraction levels for each domain.
Hierarchy is not further discussed but we assume, that a model at any
abstraction level can have an arbitrary number of hierarchy levels.”
Axel Jantsch, Shashi Kumar, Ahmed Hemani
ASPI8-4: DSP Design Methodologies, AAU
32/66
“The model derives its name from the
similarity of its visual representation (see
figure 4) to the shape of a Rugby, with the
domain lines forming the seams.”
Axel Jantsch, Shashi Kumar, Ahmed Hemani
ASPI8-4: DSP Design Methodologies, AAU
33/66
Axel Jantsch, Shashi Kumar, Ahmed Hemani
ASPI8-4: DSP Design Methodologies, AAU
34/66
ASPI8-4: DSP Design Methodologies, AAU
35/66
ASPI8-4: DSP Design Methodologies, AAU
36/66
ASPI8-4: DSP Design Methodologies, AAU
37/66
ASPI8-4: DSP Design Methodologies, AAU
38/66
ASPI8-4: DSP Design Methodologies, AAU
39/66
ASPI8-4: DSP Design Methodologies, AAU
40/66
ASPI8-4: DSP Design Methodologies, AAU
41/66
ASPI8-4: DSP Design Methodologies, AAU
42/66
ASPI8-4: DSP Design Methodologies, AAU
43/66
ASPI8-4: DSP Design Methodologies, AAU
44/66
ASPI8-4: DSP Design Methodologies, AAU
45/66
ASPI8-4: DSP Design Methodologies, AAU
46/66
ASPI8-4: DSP Design Methodologies, AAU
47/66
ASPI8-4: DSP Design Methodologies, AAU
48/66
ASPI8-4: DSP Design Methodologies, AAU
49/66
ASPI8-4: DSP Design Methodologies, AAU
50/66
ASPI8-4: DSP Design Methodologies, AAU
51/66
ASPI8-4: DSP Design Methodologies, AAU
52/66
ASPI8-4: DSP Design Methodologies, AAU
53/66
ASPI8-4: DSP Design Methodologies, AAU
54/66
ASPI8-4: DSP Design Methodologies, AAU
55/66
ASPI8-4: DSP Design Methodologies, AAU
56/66
ASPI8-4: DSP Design Methodologies, AAU
57/66
ASPI8-4: DSP Design Methodologies, AAU
58/66
ASPI8-4: DSP Design Methodologies, AAU
59/66
ASPI8-4: DSP Design Methodologies, AAU
60/66
ASPI8-4: DSP Design Methodologies, AAU
61/66
ASPI8-4: DSP Design Methodologies, AAU
62/66
ASPI8-4: DSP Design Methodologies, AAU
63/66
Rugby summary
• Separates the modeling issues computation,
communication, time and data;
• Defines abstraction levels in these four domains
independently;
• Allows to organize the design phases with respect
to the abstraction levels;
ASPI8-4: DSP Design Methodologies, AAU
64/66
Conclusions
What have we seen in the lecture ?
• motivation for DSP design methodologies
• some elements for DSP design methodologies
 Domains, abstraction level and design activities
• the Y-chart model
• the rugby meta-model
• illustration of the rugby meta-model
• Meta-models can be used to define the domains of design
• they can be used to analyze and describe other methods (SPU, OO…)
• they make the design domains explicit
• they can be used to map a particular design method
• they can be used to find room for improvement in existing methods
ASPI8-4: DSP Design Methodologies, AAU
65/66
Conclusions
Exercices:
1) define the new terms encountered in the articles:
• A Model of Design Representation and Synthesis
• The Rugby Meta-Model
2) For those who know the SPU method, try to put in line with the
rugby model
ASPI8-4: DSP Design Methodologies, AAU
66/66