tiling.c File Reference

#include <stdlib.h>
#include <stdio.h>
#include <limits.h>
#include "genC.h"
#include "misc.h"
#include "linear.h"
#include "ri.h"
#include "effects.h"
#include "ri-util.h"
#include "effects-util.h"
#include "matrice.h"
#include "tiling.h"

Include dependency graph for tiling.c:

Go to the source code of this file.

Functions
tiling	loop_nest_to_tile (Psysteme sc, int ls, Pbase base_index, int first_parallel_level, int last_parallel_level, int perfect_nested_level)
	Code Generation for Distributed Memory Machines. More...

Function Documentation

loop_nest_to_tile()

tiling loop_nest_to_tile	(	Psysteme	sc,
		int	ls,
		Pbase	base_index,
		int	first_parallel_level,
		int	last_parallel_level,
		int	perfect_nested_level
	)

Code Generation for Distributed Memory Machines.

tiling.c

Functions to decide a specific tiling for a loop nest

File: tiling.c

PUMA, ESPRIT contract 2701

Francois Irigoin, Corinne Ancourt, Lei Zhou 1991 loop_nest_to_tile():

Because the number of elements per bank ligne is usefull to optimize the tiling, parameter ls is used temporarily to build it. Because ls is the number of bytes of each bank ligne, it is divided by the assumed number of bytes needed for the
location of one element (4 for int and real)

build the diagonal matrix: ls x I

build origin tile vector as the first iteration to minimize the number of partial tiles (heuristics)

vect_chg_coeff(&porigin,var,min);

Parameters

sc	c
ls	s
base_index	ase_index
first_parallel_level	irst_parallel_level
last_parallel_level	ast_parallel_level
perfect_nested_level	erfect_nested_level

Definition at line 58 of file tiling.c.

{
 
    tiling tile;
    matrice M;
    Pvecteur porigin = VECTEUR_NUL;
    Pvecteur pv;
    int vs = vect_size((Pvecteur)(base_index));
    Psysteme sc_proj;
    Value min,max;
    bool faisable;
    int i;
 
    /* Because the number of elements per bank ligne is usefull to
       optimize the tiling,  parameter ls is used temporarily to build it.
       Because ls is the number of bytes of each bank ligne, it is 
       divided by the assumed number of bytes needed for the  
       location of one element (4 for int and real) */
 
    int lsd = ls/4;
 
    /* build the diagonal matrix: ls x I */
 
    M = matrice_new(vs,vs);
    matrice_nulle(M,vs,vs); 
    for (i =1; i<= vs && i < first_parallel_level; i++)
        ACCESS(M,vs,i,i) = VALUE_ONE;
    for (; i<= vs && i <= last_parallel_level; i++)
        ACCESS(M,vs,i,i) = int_to_value(lsd);
    for (i = MIN(perfect_nested_level+1,last_parallel_level+1); i<= vs; i++)
        ACCESS(M,vs,i,i) = VALUE_CONST(999);
 
  
    /* build origin tile vector as the first iteration to minimize
       the number of partial tiles (heuristics) */
 
    for (pv = base_index; !VECTEUR_NUL_P(pv); pv = pv->succ) {
        Variable var = vecteur_var(pv);
        sc_proj = sc_dup(sc);
        faisable =  sc_minmax_of_variable(sc_proj,var, &min, &max);
        if (faisable) 
            /*      vect_chg_coeff(&porigin,var,min);*/
            vect_chg_coeff(&porigin,var,VALUE_ONE);
        else
            pips_internal_error("illegal empty iteration domain");
 
    }
  
    tile = make_tiling(M, porigin);
    return tile;
}

References ACCESS, int_to_value, make_tiling(), matrice_new, matrice_nulle(), max, min, MIN, pips_internal_error, sc_dup(), sc_minmax_of_variable(), Svecteur::succ, VALUE_CONST, VALUE_ONE, vect_chg_coeff(), vect_size(), VECTEUR_NUL, VECTEUR_NUL_P, and vecteur_var.

Referenced by loop_nest_to_wp65_code().

Here is the call graph for this function:

Here is the caller graph for this function: