Sketching high-performance implementations of bitstream programs. - PowerPoint PPT Presentation

1 / 21
About This Presentation
Title:

Sketching high-performance implementations of bitstream programs.

Description:

an optimizer tailored to the algorithm. Challenges, and ... y. x. Na ve Compilation of BitStream Program. The example filter operates on 3- and 2-bit chunks. ... – PowerPoint PPT presentation

Number of Views:13
Avg rating:3.0/5.0
Slides: 22
Provided by: aso1
Category:

less

Transcript and Presenter's Notes

Title: Sketching high-performance implementations of bitstream programs.


1
Sketching high-performance implementations of
bitstream programs.
  • Armando Solar-Lezama, Rastislav Bodik
  • UC Berkeley

2
Bitstream programs
  • bitstream programs a growing domain
  • crypto, compression, NSA/BitTwiddle, coding in
    general.
  • bitstream algorithms easy to state
  • e.g., Drop every third bit in the bit stream.
  • but bitstream programs hard to implement
  • because efficient bit manipulations are hard to
    code
  • Can only work with word-size arrays of bits
  • Exponentially many ways of accomplishing the same
    task

FAST O(log n)
3
Current development process
  • A collaborative experience
  • domain expert writes a high-level algorithm, in
    C/Fortran,
  • system expert tunes its performance, often
    drastically turning the algorithm into ugly
    low-level code.
  • Now, if the original algorithm needs to be
    modified
  • introduce changes into optimized code
    (error-prone), or
  • rewrite the algorithm and repeat tuning
    (time-consuming).

4
Our development process
  • A (better) collaborative experience
  • domain expert writes a clean algorithm, in a
    clean DSL,
  • system expert optimizes the implementation by
    writing a (reusable) transformation specification
    in a TSL.
  • If the original algorithm is modified, then
  • simply reapply the transformation specification.
  • The transformation spec think of it as
  • sequence of program edits, or
  • high level description of the desired
    implementation, or
  • an optimizer tailored to the algorithm

5
Challenges, and our solution (overview)
  • How to make the transformation reusable?
  • Transformation should apply after the algorithm
    changes.
  • How to simplify transformation development?
  • Should be much easier than editing the program by
    hand.
  • Should not allow you to introduce bugs
  • Our solution a transformation sketch
  • system expert sketches the transformation,
  • details filled in automatically.

DSL StreamIt
TSL transf. spec
TSL with sketching
?
6
DSL StreamIt
  • High-level bitstream algorithms written in
    StreamIt.
  • Example drop every third bit.
  • filter dropThird
  • Work push 2 pop 3
  • for (int i0 ilt3 i)
  • x peek(i)
  • if (ilt2) push(x)
  • pop()

7
Naïve Compilation of BitStream Program
  • The example filter operates on 3- and 2-bit
    chunks.
  • unsuitable for the target machines instructions
  • So, how to generate code for the example filter?
  • transform the filter to operate on word-size
    chunks!
  • Decompose new filters into filters corresponding
    to machine instructions
  • Such low-level program can be expressed in
    StreamIt!

8
The development strategy
  • We offer several methods for lowering filters
    into low-level form (i.e., for implementing them
    with target code)
  • Manual-transformation M(P) For any filter P,
    the system expert could manually write a TSL spec
    to make P low level.
  • Auto-transformation N(P) For any filter P,
    the system knows how to produce a naïve lowering
    TSL spec N.
  • N is thus a simple code generator for P

9
Transforming into low-level form (1)
  • unroll 4x to make input/output a multiple of W4
    bits.

3
1 0 0 0 1 0
2
10
Transforming into low-level form (2)
  • decompose into filters operating on W4 bits of
    input.

12
1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 0 0 1 0
8
11
Transforming into low-level form (3)
  • decompose into filters producing W4 bits of
    output.

rrobin 4,4,4
1 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0
0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 1 0
or
12
Implementing a basic filter (4)
  • decompose word-size filter into available
    instructions.

1 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0
13
The development strategy
  • We offer several methods for lowering filters
    into low-level form (i.e., for implementing them
    with target code)
  • Manual-transformation M(P) For any filter P,
    the filter system expert could manually bring it
    to low level form.
  • Auto-transformation N(P) For any filter P,
    the system knows how to produce a naïve lowering
    TSL spec N.
  • N is thus a simple code generator for P
  • Half-way transformation. N(T(P)) System expert
    provides a spec T that transforms P so that N
    generates better code.
  • If written well, T brings P closer to low-level
    code (gives it a good structure) so that N can do
    a perfect job generating the code.

14
TSL Transformation spec language
  • TSL specification
  • describes how to transform a filter into a
    semantically equivalent one with a different
    structure.
  • so that you dont need to transform filters
    manually.
  • Example. TSL spec for the transformation you
    just saw
  • f Unroll4(dropThird)
  • f ColSplit4(f)
  • f.f_1 RowSplit4(f.f_1)
  • f.f_2 RowSplit4(f.f_2)
  • f.f_3 RowSplit4(f.f_3)
  • TSL Spec can also be thought of as an
    implementation specification.
  • Every statement in the above specification
    provides the system more details about the
    implementation.
  • This simple specification can be generated
    automatically
  • better TSL specs written by system expert, with
    help of sketches

15
Half way transformations
  • The expert guides lowering by imparting structure
    to the filter
  • If the system expert has an algorithm in mind to
    implement a particular filter, the expert can
    decompose the filter into a sequence of filters,
    each one implementing one step of the algorithm.
  • So, specifying an efficient bit manipulation
    often boils down to specifying a decomposition of
    the filters matrix.
  • filter filter1 x filter2
  • This can be done hierarchically,
  • Detail is only added where necessary.

16
Half way transformations-example
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
F.F_1
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1
Half way transformation to specify FAST bit
shifting algorithm
F.F_1
F
F.F_2
F.F_3
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
F.F_2
  • System Expert provides high level decomposition
  • System Takes care of Lowering F.F_1, F.F_2 and
    F.F_3
  • Correctness is guaranteed as long as
  • F F.F_3xF.F_2xF.F_1
  • Fully Specifying F.F._1, F.F_2 and F.F_3 is still
    too difficult. We would like to be able to sketch
    them

1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0
F.F_3
17
The development strategy
  • We offer several methods for lowering filters
    into low-level form (i.e., for implementing them
    with target code)
  • Manual-transformation M(P) For any filter P,
    the filter system expert could manually bring it
    to low level form.
  • Auto-transformation N(P) For any filter P,
    the system knows how to produce a naïve lowering
    TSL spec N.
  • N is thus a simple code generator for P
  • Half-way transformation. N(T(P)) System expert
    provides a spec T that transforms P so that N
    generates better code.
  • If written well, T brings P closer to low-level
    code (gives it a good structure) so that N can do
    a perfect job generating the code.
  • Half-specified transformation. N(sT(P)) System
    expert provides a sketch of a transformation T.
  • The compiler completes the sketch by requiring
    that the completed transformation (sT) produces
    a filter semantically equivalent to P.

18
Sketching a transformation
  • A transformation sketch
  • Specifies the number of stages in the
    decomposition
  • Gives constraints on terms of the decomposition
  • System derives a decomposition satisfying the
    constraints and semantically equivalent to
    original filter
  • Example. A fragment of TSL for FAST compaction

filter shift(116 by 0 1) x
shift(116 by 0 2) x shift(116
by 0 4)
19
Sketching a transformation
  • The PermutFactor function specifies each step as
    a list of constraints
  • PermutFactor constrList constrList
  • Constraints of 3 types
  • Type 1 specific shift amount
  • shift( bitList by x)
  • Type 2 undetermined shift amount
  • shift( bitList by ?)
  • Type 3 limited choice of shift amount
  • shift( bitList by a b )
  • System ignores constraints on discarded bits

20
Sketching a transformation-example
  • First we must unroll to get a multiple of the
    word size
  • Unroll16(filter)
  • We want to use the fast algorithm for compacting
    the bits in each word
  • The second word can only pack half the bits at a
    time, since half will go to one word and half to
    another

21
Sketching a transformation-example
  • If we first pack all bits within a word, we can
    better exploit the parallelism afforded by the
    fast algorithm.
  • We want to sketch this transformation.

22
Sketching a transformation-example
  • Sketch of the transformation
  • Sketch of the transformation
  • PermutFactor(filter)
  • Sketch of the transformation
  • PermutFactor
  • shift(116 by ?),
    shift(1732 by ?), shift(3348 by ?)
  • (filter)
  • Sketch of the transformation
  • PermutFactorshift(12 by 0), shift(1718 by 0),
    shift(3334 by 0)
  • shift(116 by ?),
    shift(1732 by ?), shift(3348 by ?)
  • (filter)
  • After this is done, we can proceed hierarchically

23
Sketching a transformation-example
  • We can select a specific part of the algorithm
    and add more detail
  • Specification of fast bit packing algorithm
    within each word
  • PermutFactor shift(116 by 0 1),
  • shift(116 by 0
    2),
  • shift(116 by 0
    4)
  • ( F_i )

24
Complete TSL spec for FAST
  • WSIZE16
  • subsequence UnrollWSIZE(subsequence)
  • subsequence PermutFactor shift(12 by 0),
    shift(1718 by 0), shift(3334 by 0),

  • shift(116 by ?), shift(1732 by ?), shift(3348
    by ?)
  • ( subsequence )
  • subsequence.subsequence_1DiagSplitWSIZE(subsequ
    ence)
  • for(i0 ilt3 i)
  • bsequence.subsequence_1.filter(i)
  • PermutFactor shift(116 by 0 1),
  • shift(116 by 0
    2),
  • shift(116 by 0
    4)
  • ( subsequence.subsequen
    ce_1.filter(i) )
  • Size 13 lines

25
Reusability of the TSL specs
  • A TSL spec is reusable under a given change to
    the program if the changed program can still be
    profitably transformed with the TSL spec.
  • Experiment
  • Start with the program P drop every third
    bit,
  • Write the FAST spec for it,
  • Modify P to drop the second bit in each three
    bits.
  • no change in the original TSL spec needed.
  • Modify P to drop every fourth bit.
  • Only minor change in the TSL spec needed.

26
Performance Gain bit compaction
  • On a pentium III processor running at 1.5 GHz,
    the unoptimized code took 0.068 seconds to
    process .127 Mb of data 100 times.
  • On the same machine, optimized version took 0.041
    seconds, a performance gain of 39.

27
Performance Gain Permutation from DES
  • On the same machine the unoptimized code took
    .102 seconds to process the same amount of data.
  • After a simple TSL specification
  • DESIPPermutFactor
  • shift(1231 by -33),
  • shift(2232 by 0),
  •               shift(33263 by 0),
  • shift(34264 by 33)
  • ,(DESIP)
  • Optimized version took .073 seconds, a speedup
    of 28

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
28
Conclusions
  • Our system allows the separation of the algorithm
    specification from performance tuning through the
    use of TSL
  • System expert only needs to specify high level
    transformations, system can take care of the
    details
  • Sketching the transformations makes them reusable
    and easier to write.
Write a Comment
User Comments (0)
About PowerShow.com