This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision Next revision Both sides next revision | ||
ffnamespace:tutorial [2014/09/14 18:09] aldinuc |
ffnamespace:tutorial [2015/09/08 14:53] torquati |
||
---|---|---|---|
Line 1: | Line 1: | ||
{{description>Fastflow tutorial: basic programming concepts and related technologies}} | {{description>Fastflow tutorial: basic programming concepts and related technologies}} | ||
+ | |||
+ | |||
===== Tutorial ===== | ===== Tutorial ===== | ||
- | The FastFlow tutorial can be dowloaded [[http://calvados.di.unipi.it/storage/tutorial/fftutorial.pdf|here]] (version August 2014). The simple tests and examples contained in the tutorial are available as tgz tarball [[http://calvados.di.unipi.it/storage/tutorial/fftutorial_source_code.tgz | here]]. | + | * [[http://calvados.di.unipi.it/storage/tutorial/html/tutorial.html|Single HTML file]] (version August 2014) |
+ | * [[http://calvados.di.unipi.it/storage/tutorial/fftutorial.pdf|PDF file]] (version September 2014) | ||
+ | * [[http://calvados.di.unipi.it/storage/tutorial/fftutorial_source_code.tgz | Tests and examples - source code tarball]] (version September 2014) | ||
+ | ===== Very short Tutorial ===== | ||
==== The FastFlow programming model ==== | ==== The FastFlow programming model ==== | ||
Line 10: | Line 15: | ||
- | + | ==== The basic ==== | |
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | + | ||
- | ===== The basic ===== | + | |
=== Pipeline === | === Pipeline === | ||
Line 60: | Line 49: | ||
}; | }; | ||
int main() { | int main() { | ||
- | ff_pipe<fftask_t> pipe(new firstStage, secondStage, new thirdStage); | + | ff_Pipe<> pipe(make_unique<firstStage>(), make_unique<ff_node_F<fftask> >(secondStage), make_unique<thirdStage>()); |
- | pipe.cleanup_nodes(); // cleanup at exit | + | |
if (pipe.run_and_wait_end()<0) error("running pipe"); | if (pipe.run_and_wait_end()<0) error("running pipe"); | ||
return 0; | return 0; | ||
Line 116: | Line 104: | ||
=== Data Dependency Tasks Executor (aka MDF) === | === Data Dependency Tasks Executor (aka MDF) === | ||
- | ===== Some valid combinations of pipeline and farm (and feedback) ===== | + | The data-flow programming model is a general approach to parallelization |
+ | based upon data dependencies among a program's operations. The computations is expressed | ||
+ | by the data-flow graph, i.e. a DAG whose nodes are instructions and arcs are pure data dependencies. | ||
+ | If instead of simple instructions, portions of code (sets of instructions or functions) are used as graph's nodes, then it is called the macro data-flow model (MDF). It is worth noting that, the data-flow programming model is able to work both on stream of values and on a single value. | ||
- | {{:ffnamespace:composition2.png?400|}} | + | As an example, considering the [[http://en.wikipedia.org/wiki/Strassen_algorithm |Strassen's algorithm]] described by the following sequence of instructions operating on (sub-)matrices : |
+ | S1 = A11 + A22; S2 = B11 + B22; S3 = A21 + A22; S4 = B12 - B22; S5 = B21 - B11; | ||
+ | S6 = A11 + A12; S7 = A21 - A11; S8 = B11 + B12; S9 = A12 - A22; S10 = B21 + B22; | ||
+ | P1 = S1 * S2; P2 = S3 * B11; P3 = A11 * S4; P4 = A22 * S5; P5 = S6 * B22; P6 = S7 * S8; P7 = S9*S10 | ||
+ | C11 = P1 + P4 - P5 + P7; C12 = P3 + P5; C21 = P2 + P4; C22 = P1 - P2 + P3 + P6; | ||
- | <html> | + | the resulting DAG is sketched in the following figure: |
- | <div class="maketitle"> | + | {{:ffnamespace:strassen.png?300|}} |