-/* Copyright (c) 2007-2018. The SimGrid Team. All rights reserved. */
+/* Copyright (c) 2007-2019. The SimGrid Team. All rights reserved. */
/* This program is free software; you can redistribute it and/or modify it
* under the terms of the license (GNU LGPL) which comes with this package. */
* "ssh://username@scm.gforge.inria.fr/svn/memo/people/pvelho/lagrange/ppf.ps".
*/
#include "src/kernel/lmm/maxmin.hpp"
+#include "src/surf/surf_interface.hpp"
#include "xbt/log.h"
#include "xbt/sysdep.h"
#include <algorithm>
-#include <cstdlib>
-#ifndef MATH
#include <cmath>
-#endif
+#include <cstdlib>
XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_lagrange, surf, "Logging specific to SURF (lagrange)");
XBT_LOG_NEW_SUBCATEGORY(surf_lagrange_dichotomy, surf_lagrange, "Logging specific to SURF (lagrange dichotomy)");
-#define SHOW_EXPR(expr) XBT_CDEBUG(surf_lagrange, #expr " = %g", expr);
-#define VEGAS_SCALING 1000.0
-#define RENO_SCALING 1.0
-#define RENO2_SCALING 1.0
+static constexpr double VEGAS_SCALING = 1000.0;
+static constexpr double RENO_SCALING = 1.0;
+static constexpr double RENO2_SCALING = 1.0;
namespace simgrid {
namespace kernel {
namespace lmm {
-double (*func_f_def)(const Variable&, double);
-double (*func_fp_def)(const Variable&, double);
-double (*func_fpi_def)(const Variable&, double);
-
System* make_new_lagrange_system(bool selective_update)
{
return new Lagrange(selective_update);
}
-/*
- * Local prototypes to implement the Lagrangian optimization with optimal step, also called dichotomy.
- */
-// computes the value of the dichotomy using a initial values, init, with a specific variable or constraint
-static double dichotomy(double init, double diff(double, const Constraint&), const Constraint& cnst, double min_error);
-// computes the value of the differential of constraint cnst applied to lambda
-static double partial_diff_lambda(double lambda, const Constraint& cnst);
-
-template <class CnstList, class VarList>
-static int __check_feasible(const CnstList& cnst_list, const VarList& var_list, int warn)
+bool Lagrange::check_feasible(bool warn)
{
- for (Constraint const& cnst : cnst_list) {
+ for (Constraint const& cnst : active_constraint_set) {
double tmp = 0;
for (Element const& elem : cnst.enabled_element_set) {
Variable* var = elem.variable;
if (double_positive(tmp - cnst.bound, sg_maxmin_precision)) {
if (warn)
XBT_WARN("The link (%p) is over-used. Expected less than %f and got %f", &cnst, cnst.bound, tmp);
- return 0;
+ return false;
}
XBT_DEBUG("Checking feasability for constraint (%p): sat = %f, lambda = %f ", &cnst, tmp - cnst.bound, cnst.lambda);
}
- for (Variable const& var : var_list) {
+ for (Variable const& var : variable_set) {
if (not var.sharing_weight)
break;
if (var.bound < 0)
if (double_positive(var.value - var.bound, sg_maxmin_precision)) {
if (warn)
XBT_WARN("The variable (%p) is too large. Expected less than %f and got %f", &var, var.bound, var.value);
- return 0;
+ return false;
}
}
- return 1;
+ return true;
}
-static double new_value(const Variable& var)
+double Lagrange::new_value(const Variable& var)
{
double tmp = 0;
tmp += var.mu;
XBT_DEBUG("\t Working on var (%p). cost = %e; Weight = %e", &var, tmp, var.sharing_weight);
// uses the partial differential inverse function
- return var.func_fpi(var, tmp);
+ return func_fpi(var, tmp);
}
-static double new_mu(const Variable& var)
+double Lagrange::new_mu(const Variable& var)
{
double mu_i = 0.0;
double sigma_i = 0.0;
for (Element const& elem : var.cnsts) {
sigma_i += elem.constraint->lambda;
}
- mu_i = var.func_fp(var, var.bound) - sigma_i;
+ mu_i = func_fp(var, var.bound) - sigma_i;
if (mu_i < 0.0)
return 0.0;
return mu_i;
}
-template <class VarList, class CnstList>
-static double dual_objective(const VarList& var_list, const CnstList& cnst_list)
+double Lagrange::dual_objective()
{
double obj = 0.0;
- for (Variable const& var : var_list) {
+ for (Variable const& var : variable_set) {
double sigma_i = 0.0;
if (not var.sharing_weight)
XBT_DEBUG("var %p : sigma_i = %1.20f", &var, sigma_i);
- obj += var.func_f(var, var.func_fpi(var, sigma_i)) - sigma_i * var.func_fpi(var, sigma_i);
+ obj += func_f(var, func_fpi(var, sigma_i)) - sigma_i * func_fpi(var, sigma_i);
if (var.bound > 0)
obj += var.mu * var.bound;
}
- for (Constraint const& cnst : cnst_list)
+ for (Constraint const& cnst : active_constraint_set)
obj += cnst.lambda * cnst.bound;
return obj;
print();
}
- if (not modified)
+ if (not modified_)
return;
/* Initialize lambda. */
- auto& cnst_list = active_constraint_set;
- for (Constraint& cnst : cnst_list) {
+ for (Constraint& cnst : active_constraint_set) {
cnst.lambda = 1.0;
cnst.new_lambda = 2.0;
XBT_DEBUG("#### cnst(%p)->lambda : %e", &cnst, cnst.lambda);
}
/*
- * Initialize the var_list variable with only the active variables. Initialize mu.
+ * Initialize the active variables. Initialize mu.
*/
- auto& var_list = variable_set;
- for (Variable& var : var_list) {
+ for (Variable& var : variable_set) {
if (not var.sharing_weight)
var.value = 0.0;
else {
}
/* Compute dual objective. */
- double obj = dual_objective(var_list, cnst_list);
+ double obj = dual_objective();
/* While doesn't reach a minimum error or a number maximum of iterations. */
int iteration = 0;
XBT_DEBUG("-------------- Gradient Descent ----------");
/* Improve the value of mu_i */
- for (Variable& var : var_list) {
+ for (Variable& var : variable_set) {
if (var.sharing_weight && var.bound >= 0) {
XBT_DEBUG("Working on var (%p)", &var);
var.new_mu = new_mu(var);
XBT_DEBUG("Updating mu : var->mu (%p) : %1.20f -> %1.20f", &var, var.mu, var.new_mu);
var.mu = var.new_mu;
- double new_obj = dual_objective(var_list, cnst_list);
+ double new_obj = dual_objective();
XBT_DEBUG("Improvement for Objective (%g -> %g) : %g", obj, new_obj, obj - new_obj);
xbt_assert(obj - new_obj >= -epsilon_min_error, "Our gradient sucks! (%1.20f)", obj - new_obj);
obj = new_obj;
}
/* Improve the value of lambda_i */
- for (Constraint& cnst : cnst_list) {
+ for (Constraint& cnst : active_constraint_set) {
XBT_DEBUG("Working on cnst (%p)", &cnst);
cnst.new_lambda = dichotomy(cnst.lambda, partial_diff_lambda, cnst, dichotomy_min_error);
XBT_DEBUG("Updating lambda : cnst->lambda (%p) : %1.20f -> %1.20f", &cnst, cnst.lambda, cnst.new_lambda);
cnst.lambda = cnst.new_lambda;
- double new_obj = dual_objective(var_list, cnst_list);
+ double new_obj = dual_objective();
XBT_DEBUG("Improvement for Objective (%g -> %g) : %g", obj, new_obj, obj - new_obj);
xbt_assert(obj - new_obj >= -epsilon_min_error, "Our gradient sucks! (%1.20f)", obj - new_obj);
obj = new_obj;
}
- /* Now computes the values of each variable (\rho) based on the values of \lambda and \mu. */
+ /* Now computes the values of each variable (@rho) based on the values of @lambda and @mu. */
XBT_DEBUG("-------------- Check convergence ----------");
overall_modification = 0;
- for (Variable& var : var_list) {
+ for (Variable& var : variable_set) {
if (var.sharing_weight <= 0)
var.value = 0.0;
else {
}
XBT_DEBUG("-------------- Check feasability ----------");
- if (not __check_feasible(cnst_list, var_list, 0))
+ if (not check_feasible(false))
overall_modification = 1.0;
XBT_DEBUG("Iteration %d: overall_modification : %f", iteration, overall_modification);
}
- __check_feasible(cnst_list, var_list, 1);
+ check_feasible(true);
if (overall_modification <= epsilon_min_error) {
XBT_DEBUG("The method converges in %d iterations.", iteration);
/*
* Returns a double value corresponding to the result of a dichotomy process with respect to a given
- * variable/constraint (\mu in the case of a variable or \lambda in case of a constraint) and a initial value init.
+ * variable/constraint (@mu in the case of a variable or @lambda in case of a constraint) and a initial value init.
*
- * @param init initial value for \mu or \lambda
- * @param diff a function that computes the differential of with respect a \mu or \lambda
+ * @param init initial value for @mu or @lambda
+ * @param diff a function that computes the differential of with respect a @mu or @lambda
* @param var_cnst a pointer to a variable or constraint
* @param min_erro a minimum error tolerated
*
* @return a double corresponding to the result of the dichotomy process
*/
-static double dichotomy(double init, double diff(double, const Constraint&), const Constraint& cnst, double min_error)
+double Lagrange::dichotomy(double init, double diff(double, const Constraint&), const Constraint& cnst,
+ double min_error)
{
double min = init;
double max = init;
return ((min + max) / 2.0);
}
-static double partial_diff_lambda(double lambda, const Constraint& cnst)
+double Lagrange::partial_diff_lambda(double lambda, const Constraint& cnst)
{
double diff = 0.0;
// replace value of cnst.lambda by the value of parameter lambda
sigma_i = (sigma_i - cnst.lambda) + lambda;
- diff += -var.func_fpi(var, sigma_i);
+ diff += -func_fpi(var, sigma_i);
}
diff += cnst.bound;
return diff;
}
-/** \brief Attribute the value bound to var->bound.
+/** @brief Attribute the value bound to var->bound.
*
- * \param func_fpi inverse of the partial differential of f (f prime inverse, (f')^{-1})
- *
- * Set default functions to the ones passed as parameters. This is a polymorphism in C pure, enjoy the roots of
- * programming.
+ * @param func_f function (f)
+ * @param func_fp partial differential of f (f prime, (f'))
+ * @param func_fpi inverse of the partial differential of f (f prime inverse, (f')^{-1})
*
+ * Set default functions to the ones passed as parameters.
*/
-void set_default_protocol_function(double (*func_f)(const Variable& var, double x),
- double (*func_fp)(const Variable& var, double x),
- double (*func_fpi)(const Variable& var, double x))
+void Lagrange::set_default_protocol_function(double (*func_f)(const Variable& var, double x),
+ double (*func_fp)(const Variable& var, double x),
+ double (*func_fpi)(const Variable& var, double x))
{
- func_f_def = func_f;
- func_fp_def = func_fp;
- func_fpi_def = func_fpi;
+ Lagrange::func_f = func_f;
+ Lagrange::func_fp = func_fp;
+ Lagrange::func_fpi = func_fpi;
}
+double (*Lagrange::func_f)(const Variable&, double);
+double (*Lagrange::func_fp)(const Variable&, double);
+double (*Lagrange::func_fpi)(const Variable&, double);
+
/**************** Vegas and Reno functions *************************/
/* NOTE for Reno: all functions consider the network coefficient (alpha) equal to 1. */
/*
- * For Vegas: $f(x) = \alpha D_f\ln(x)$
- * Therefore: $fp(x) = \frac{\alpha D_f}{x}$
- * Therefore: $fpi(x) = \frac{\alpha D_f}{x}$
+ * For Vegas: $f(x) = @alpha D_f@ln(x)$
+ * Therefore: $fp(x) = @frac{@alpha D_f}{x}$
+ * Therefore: $fpi(x) = @frac{@alpha D_f}{x}$
*/
double func_vegas_f(const Variable& var, double x)
{
}
/*
- * For Reno: $f(x) = \frac{\sqrt{3/2}}{D_f} atan(\sqrt{3/2}D_f x)$
- * Therefore: $fp(x) = \frac{3}{3 D_f^2 x^2+2}$
- * Therefore: $fpi(x) = \sqrt{\frac{1}{{D_f}^2 x} - \frac{2}{3{D_f}^2}}$
+ * For Reno: $f(x) = @frac{@sqrt{3/2}}{D_f} atan(@sqrt{3/2}D_f x)$
+ * Therefore: $fp(x) = @frac{3}{3 D_f^2 x^2+2}$
+ * Therefore: $fpi(x) = @sqrt{@frac{1}{{D_f}^2 x} - @frac{2}{3{D_f}^2}}$
*/
double func_reno_f(const Variable& var, double x)
{
}
/* Implementing new Reno-2
- * For Reno-2: $f(x) = U_f(x_f) = \frac{{2}{D_f}}*ln(2+x*D_f)$
+ * For Reno-2: $f(x) = U_f(x_f) = @frac{{2}{D_f}}*ln(2+x*D_f)$
* Therefore: $fp(x) = 2/(Weight*x + 2)
* Therefore: $fpi(x) = (2*Weight)/x - 4
*/
