Analysis of the WorstCase Execution Time WCET A survey

1
Analysisof theWorst-Case Execution Time
(WCET)-A survey
2
Introduction

• correctness P Q R
• logical correctness
• temporal requirements
• deadlines
• temporal correctness
• RTS .. real time system
• timeliness
• scheduling service time
• WCET .. worst case execution time
• design-aids for RTS

3
Why is WCET-analysis (WCETA) complicated ?

• abstraction
• mapping
• optimizations
• speculation
• unpredictable
• not aiming at worst-case
• instruction execution times
• ? problem-avoidance by simplification

4
WCET-analysis (WCETA)

• the real WCET
• by measurement
• measurement-based search-methods
• WCET-estimate
• approximation
• pessimism, optimism
• WCET-analysis
• statical analysis
• tightness

5
Classification ofWCETA-techniques

• Three orthogonal aspects
• Criteria for finding tools

6
WCETA-classification, classical

• high-level analysis
• low-level analysis
• Global-level analysis
• WCET calculation

7
High-level WCET-analysisSummation of basic
blocks

• Shaw
• pseudo-language
• execution time bounds
• timing formulae

8
Summation of basic blocks

• Puschner Koza
• practical example for WCET-orientation
• MAXT .. maximum execution time
• limitations in language constructs
• annotations markers scopes
• formulae

9
Graphs

• graphs widely used
• replaced/extended sum of block method
• advantages
• portability, abstraction
• graph-algorithms (i.e. path-problems)
• example timing tree (Puschner Schedl)
• cost function
• maximum cycle
• flow

10
(No Transcript)
11
WCETA and OOP-languages

• additional dynamic features
• encapsulation
• inheritance
• polymorphism
• features complicate static analysis
• improved pessimism
• example dynamic dispatching
• new annotations needed

12
class Triangle public double giveArea()
// herons formula
class RightTriangle extends
Triangle public double giveArea() // a
b / 2
class Program public static void
main( ) Triangle t new Triangle()
RightTriangle s new RightTriangle() t
s result t.giveArea() // right
class/method known only at run-time // ?
dispatching also at run-time dynamically
13
Automationavoidance of manual annotations

• manual annotations error-prone
• branch constraints by analysis
• iteration-based constraints
• effect-based constraints
• timing analysis WCET, BCET
• great effort needed, one example
• correlation between conditional branches
• ? O(C2) C .. of branches

14
Optimizing compilers

• data-dependencies
• transformations
• Enblom et al. co-transformation
• ODL
• HLA and LLA
• co-transformer transforms execution info
• notion of portability (ODL)

15
Optimizing compilers

• Puschner and Kirner
• flow facts from source code
• Whalley et al. optimize WCET
• ? optimizations for WC path
• approach reduce consuming instructions
• feedback to compiler
• possible drawback code explosion

16

• example superblock-formation

17
Low- global-level WCETACaching

• locality
• caches unconsidered in WCETA
• ? pessimism
• Whalley et al. extending summation of blocks
  method for considering instruction caching
• static cache simulator
• instruction-categorization always/first hit/miss
• timing tree

18
Caching

• Li et al. with different approach
• program path analysis used here
• to construct an ILP
• means path analysis gives constraints
• 2 types of constraints
• automatically derived constraints (structural)
• user-provided constraints (functional)
• cost-function total execution time
• where N of basic blocks
• xi execution count of the basic block i
• ci execution time of the basic block i

19
example
20
Caching

• Extension cache-hit and cache-miss times
• for instruction caching
• l-block (line-block)
• ? grouping of instructions
• total execution time
• where N basic blocks
• ni l-blocks of basic block i
• cache hit and cache miss counts of l-block
  Bi,j
• execution time of l-block Bi,j in case of
  hit/miss
• data-caching

21
Pipelining

• Whalley et al. significant reduction of
  pessimism considering pipelining
• pipeline-analysis, pipeline-simulation
• ? overlapping
• virtual pipeline
• many simulations
• ? hardware-dependent offset
• (can be positive !)

22
overlapping-example
23
Tools for practical use Tuning the WCET by
feedback

• many possible optimizations
• branch chaining, remove useless blocks,
• goal integrated development tool
• roll-back
• feedback
• Kirner et al. extensions for Matlab

24
tuning the WCET by feedback
25
JAVA and WCETA

• WCETA partitioned into two parts
• machine-independent flow analysis
• machine-dependent timing analysis
• abstract analysis WCEF
• annotations needed
• compiler-independent solution dummy-class
• portability

26
(No Transcript)
27
(No Transcript)
28
WCETA-friendly software design

• Puschner Burns
• temporally predictable code
• only one path
• avoids multiple paths and execution times
• new language-construct needed
• constant-time conditional
• evaluates both possible results of a conditional
  execution
• transformation rules for normal code
• main drawback efficiency

29
Conclusion