User Tools

Site Tools


Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
ffnamespace:tutorial [2014/09/14 18:59]
aldinuc
ffnamespace:tutorial [2015/09/08 16:52] (current)
torquati
Line 2: Line 2:
  
  
-===== Full 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 September 2015) 
 +  * [[http://​calvados.di.unipi.it/​storage/​tutorial/​fftutorial.pdf|PDF file]] (version ​September 2015 
 +  * [[http://​calvados.di.unipi.it/​storage/​tutorial/​fftutorial_source_code.tgz | Tests and examples - source code tarball]] (version September 2015)
  
 ===== Very short Tutorial ===== ===== Very short Tutorial =====
Line 25: Line 27:
 <code c++> <code c++>
 /* this is a 3-stage pipeline example */ /* this is a 3-stage pipeline example */
 +#include <​iostream>​
 #include <​ff/​pipeline.hpp>​ #include <​ff/​pipeline.hpp>​
 using namespace ff; using namespace ff;
 typedef long fftask_t; typedef long fftask_t;
  
-struct firstStage: ff_node_t<​task_t> {+struct firstStage: ff_node_t<​fftask_t> {
     fftask_t *svc(fftask_t *t) {     fftask_t *svc(fftask_t *t) {
  for(long i=0;​i<​10;​++i) ff_send_out(new fftask_t(i));​  for(long i=0;​i<​10;​++i) ff_send_out(new fftask_t(i));​
Line 39: Line 42:
     return t;     return t;
 } }
-struct thirdStage: ff_node_t<​task_t> {+struct thirdStage: ff_node_t<​fftask_t> {
     fftask_t *svc(fftask_t *t) {     fftask_t *svc(fftask_t *t) {
  std::cout << "​stage"​ << get_my_id() << " received " << *t << "​\n";​  std::cout << "​stage"​ << get_my_id() << " received " << *t << "​\n";​
Line 47: Line 50:
 }; };
 int main() { int main() {
-    ​ff_pipe<fftask_t> pipe(new firstStage, secondStage, ​new thirdStage); +    ​ff_Pipe<> pipe(make_unique<​firstStage>(), 
-    ​pipe.cleanup_nodes(); // cleanup at exit+                   ​make_unique<​ff_node_F<​fftask_t>​ >(secondStage) 
 +                   ​make_unique<​thirdStage>(
 +                   ​);
     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 68: Line 73:
  
 int main() { int main() {
-    std::​vector<​ff_node*> W = {new thirdStage, new thirdStage}; // the farm has 2 workers +    std::vector<​std::​unique_ptr<ff_node> W; 
-    ​ff_pipe<fftask_t> pipe(new firstStage, secondStage, ​new ff_farm<>​(W)); +    ​// the farm has 2 workers 
-    ​pipe.cleanup_nodes();+    ​W.push_back( make_unique<​thirdStage>​());​ 
 +    W.push_back( make_unique<​thirdStage>​());​ 
 +     
 +    ff_Pipe<> pipe(make_unique<​firstStage>(), 
 +                   ​make_unique<​ff_node_F<​fftask_t>​ >(secondStage), 
 +                   ​make_unique<ff_Farm<​fftask_t> ​>(std::move(W)) 
 +                   ​);
     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 103: Line 114:
 === 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 functionsare 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;
  
 +the resulting DAG is sketched in the following figure:
 +{{:​ffnamespace:​strassen.png?​300|}}
 +
 +The DAG's instructions can be executed in parallel simply respecting data dependencies.
 +===== Some valid combinations of pipeline and farm (and feedback) =====
 +
 +{{:​ffnamespace:​composition2.png?​400|}}
ffnamespace/tutorial.1410713974.txt.gz · Last modified: 2014/09/14 18:59 by aldinuc