]> www.ginac.de Git - ginac.git/blobdiff - ginac/mul.cpp
- Partially solves performance regression in expand(), gcd(), etc [Sheplyakov].
[ginac.git] / ginac / mul.cpp
index 962947898baf696c245ab816866fb286a39c5fdf..db8b9f14acd308d76344ed57bfc44857f63e4d24 100644 (file)
@@ -3,7 +3,7 @@
  *  Implementation of GiNaC's products of expressions. */
 
 /*
- *  GiNaC Copyright (C) 1999-2006 Johannes Gutenberg University Mainz, Germany
+ *  GiNaC Copyright (C) 1999-2007 Johannes Gutenberg University Mainz, Germany
  *
  *  This program is free software; you can redistribute it and/or modify
  *  it under the terms of the GNU General Public License as published by
@@ -27,8 +27,6 @@
 
 #include "mul.h"
 #include "add.h"
-#include "color.h"
-#include "clifford.h"
 #include "power.h"
 #include "operators.h"
 #include "matrix.h"
@@ -36,6 +34,7 @@
 #include "lst.h"
 #include "archive.h"
 #include "utils.h"
+#include "compiler.h"
 
 namespace GiNaC {
 
@@ -86,20 +85,20 @@ mul::mul(const epvector & v)
        GINAC_ASSERT(is_canonical());
 }
 
-mul::mul(const epvector & v, const ex & oc)
+mul::mul(const epvector & v, const ex & oc, bool do_index_renaming)
 {
        tinfo_key = &mul::tinfo_static;
        overall_coeff = oc;
-       construct_from_epvector(v);
+       construct_from_epvector(v, do_index_renaming);
        GINAC_ASSERT(is_canonical());
 }
 
-mul::mul(std::auto_ptr<epvector> vp, const ex & oc)
+mul::mul(std::auto_ptr<epvector> vp, const ex & oc, bool do_index_renaming)
 {
        tinfo_key = &mul::tinfo_static;
        GINAC_ASSERT(vp.get()!=0);
        overall_coeff = oc;
-       construct_from_epvector(*vp);
+       construct_from_epvector(*vp, do_index_renaming);
        GINAC_ASSERT(is_canonical());
 }
 
@@ -220,8 +219,12 @@ void mul::do_print_csrc(const print_csrc & c, unsigned level) const
                c.s << "(";
 
        if (!overall_coeff.is_equal(_ex1)) {
-               overall_coeff.print(c, precedence());
-               c.s << "*";
+               if (overall_coeff.is_equal(_ex_1))
+                       c.s << "-";
+               else {
+                       overall_coeff.print(c, precedence());
+                       c.s << "*";
+               }
        }
 
        // Print arguments, separated by "*" or "/"
@@ -313,7 +316,11 @@ int mul::degree(const ex & s) const
        epvector::const_iterator i = seq.begin(), end = seq.end();
        while (i != end) {
                if (ex_to<numeric>(i->coeff).is_integer())
-                       deg_sum += i->rest.degree(s) * ex_to<numeric>(i->coeff).to_int();
+                       deg_sum += recombine_pair_to_ex(*i).degree(s);
+               else {
+                       if (i->rest.has(s))
+                               throw std::runtime_error("mul::degree() undefined degree because of non-integer exponent");
+               }
                ++i;
        }
        return deg_sum;
@@ -326,7 +333,11 @@ int mul::ldegree(const ex & s) const
        epvector::const_iterator i = seq.begin(), end = seq.end();
        while (i != end) {
                if (ex_to<numeric>(i->coeff).is_integer())
-                       deg_sum += i->rest.ldegree(s) * ex_to<numeric>(i->coeff).to_int();
+                       deg_sum += recombine_pair_to_ex(*i).ldegree(s);
+               else {
+                       if (i->rest.has(s))
+                               throw std::runtime_error("mul::ldegree() undefined degree because of non-integer exponent");
+               }
                ++i;
        }
        return deg_sum;
@@ -412,7 +423,7 @@ ex mul::eval(int level) const
                return *this;
        }
        
-       int seq_size = seq.size();
+       size_t seq_size = seq.size();
        if (overall_coeff.is_zero()) {
                // *(...,x;0) -> 0
                return _ex0;
@@ -436,9 +447,81 @@ ex mul::eval(int level) const
                }
                return (new add(distrseq,
                                ex_to<numeric>(addref.overall_coeff).
-                               mul_dyn(ex_to<numeric>(overall_coeff))))
-                     ->setflag(status_flags::dynallocated | status_flags::evaluated);
+                               mul_dyn(ex_to<numeric>(overall_coeff)))
+                      )->setflag(status_flags::dynallocated | status_flags::evaluated);
+       } else if ((seq_size >= 2) && (! (flags & status_flags::expanded))) {
+               // Strip the content and the unit part from each term. Thus
+               // things like (-x+a)*(3*x-3*a) automagically turn into - 3*(x-a)2
+
+               epvector::const_iterator last = seq.end();
+               epvector::const_iterator i = seq.begin();
+               epvector::const_iterator j = seq.begin();
+               std::auto_ptr<epvector> s(new epvector);
+               numeric oc = *_num1_p;
+               bool something_changed = false;
+               while (i!=last) {
+                       if (likely(! (is_a<add>(i->rest) && i->coeff.is_equal(_ex1)))) {
+                               // power::eval has such a rule, no need to handle powers here
+                               ++i;
+                               continue;
+                       }
+
+                       // XXX: What is the best way to check if the polynomial is a primitive? 
+                       numeric c = i->rest.integer_content();
+                       const numeric& lead_coeff =
+                               ex_to<numeric>(ex_to<add>(i->rest).seq.begin()->coeff).div_dyn(c);
+                       const bool canonicalizable = lead_coeff.is_integer();
+
+                       // XXX: The main variable is chosen in a random way, so this code 
+                       // does NOT transform the term into the canonical form (thus, in some
+                       // very unlucky event it can even loop forever). Hopefully the main
+                       // variable will be the same for all terms in *this
+                       const bool unit_normal = lead_coeff.is_pos_integer();
+                       if (likely((c == *_num1_p) && ((! canonicalizable) || unit_normal))) {
+                               ++i;
+                               continue;
+                       }
+
+                       if (! something_changed) {
+                               s->reserve(seq_size);
+                               something_changed = true;
+                       }
+
+                       while ((j!=i) && (j!=last)) {
+                               s->push_back(*j);
+                               ++j;
+                       }
+
+                       if (! unit_normal)
+                               c = c.mul(*_num_1_p);
+
+                       oc = oc.mul(c);
+
+                       // divide add by the number in place to save at least 2 .eval() calls
+                       const add& addref = ex_to<add>(i->rest);
+                       add* primitive = new add(addref);
+                       primitive->setflag(status_flags::dynallocated);
+                       primitive->clearflag(status_flags::hash_calculated);
+                       primitive->overall_coeff = ex_to<numeric>(primitive->overall_coeff).div_dyn(c);
+                       for (epvector::iterator ai = primitive->seq.begin();
+                                       ai != primitive->seq.end(); ++ai)
+                               ai->coeff = ex_to<numeric>(ai->coeff).div_dyn(c);
+                       
+                       s->push_back(expair(*primitive, _ex1));
+
+                       ++i;
+                       ++j;
+               }
+               if (something_changed) {
+                       while (j!=last) {
+                               s->push_back(*j);
+                               ++j;
+                       }
+                       return (new mul(s, ex_to<numeric>(overall_coeff).mul_dyn(oc))
+                              )->setflag(status_flags::dynallocated);
+               }
        }
+
        return this->hold();
 }
 
@@ -463,6 +546,41 @@ ex mul::evalf(int level) const
        return mul(s, overall_coeff.evalf(level));
 }
 
+void mul::find_real_imag(ex & rp, ex & ip) const
+{
+       rp = overall_coeff.real_part();
+       ip = overall_coeff.imag_part();
+       for (epvector::const_iterator i=seq.begin(); i!=seq.end(); ++i) {
+               ex factor = recombine_pair_to_ex(*i);
+               ex new_rp = factor.real_part();
+               ex new_ip = factor.imag_part();
+               if(new_ip.is_zero()) {
+                       rp *= new_rp;
+                       ip *= new_rp;
+               } else {
+                       ex temp = rp*new_rp - ip*new_ip;
+                       ip = ip*new_rp + rp*new_ip;
+                       rp = temp;
+               }
+       }
+       rp = rp.expand();
+       ip = ip.expand();
+}
+
+ex mul::real_part() const
+{
+       ex rp, ip;
+       find_real_imag(rp, ip);
+       return rp;
+}
+
+ex mul::imag_part() const
+{
+       ex rp, ip;
+       find_real_imag(rp, ip);
+       return ip;
+}
+
 ex mul::evalm() const
 {
        // numeric*matrix
@@ -561,7 +679,7 @@ bool tryfactsubs(const ex & origfactor, const ex & patternfactor, int & nummatch
        return true;
 }
 
-/** Checks wheter e matches to the pattern pat and the (possibly to be updated
+/** Checks wheter e matches to the pattern pat and the (possibly to be updated)
   * list of replacements repls. This matching is in the sense of algebraic
   * substitutions. Matching starts with pat.op(factor) of the pattern because
   * the factors before this one have already been matched. The (possibly
@@ -597,10 +715,28 @@ bool algebraic_match_mul_with_mul(const mul &e, const ex &pat, lst &repls,
        return false;
 }
 
+bool mul::has(const ex & pattern, unsigned options) const
+{
+       if(!(options&has_options::algebraic))
+               return basic::has(pattern,options);
+       if(is_a<mul>(pattern)) {
+               lst repls;
+               int nummatches = std::numeric_limits<int>::max();
+               std::vector<bool> subsed(seq.size(), false);
+               std::vector<bool> matched(seq.size(), false);
+               if(algebraic_match_mul_with_mul(*this, pattern, repls, 0, nummatches,
+                               subsed, matched))
+                       return true;
+       }
+       return basic::has(pattern, options);
+}
+
 ex mul::algebraic_subs_mul(const exmap & m, unsigned options) const
 {      
        std::vector<bool> subsed(seq.size(), false);
        exvector subsresult(seq.size());
+       ex divide_by = 1;
+       ex multiply_by = 1;
 
        for (exmap::const_iterator it = m.begin(); it != m.end(); ++it) {
 
@@ -608,36 +744,35 @@ ex mul::algebraic_subs_mul(const exmap & m, unsigned options) const
 retry1:
                        int nummatches = std::numeric_limits<int>::max();
                        std::vector<bool> currsubsed(seq.size(), false);
-                       bool succeed = true;
                        lst repls;
                        
                        if(!algebraic_match_mul_with_mul(*this, it->first, repls, 0, nummatches, subsed, currsubsed))
                                continue;
 
-                       bool foundfirstsubsedfactor = false;
-                       for (size_t j=0; j<subsed.size(); j++) {
-                               if (currsubsed[j]) {
-                                       if (foundfirstsubsedfactor)
-                                               subsresult[j] = op(j);
-                                       else {
-                                               foundfirstsubsedfactor = true;
-                                               subsresult[j] = op(j) * power(it->second.subs(ex(repls), subs_options::no_pattern) / it->first.subs(ex(repls), subs_options::no_pattern), nummatches);
-                                       }
+                       for (size_t j=0; j<subsed.size(); j++)
+                               if (currsubsed[j])
                                        subsed[j] = true;
-                               }
-                       }
+                       ex subsed_pattern
+                               = it->first.subs(ex(repls), subs_options::no_pattern);
+                       divide_by *= power(subsed_pattern, nummatches);
+                       ex subsed_result
+                               = it->second.subs(ex(repls), subs_options::no_pattern);
+                       multiply_by *= power(subsed_result, nummatches);
                        goto retry1;
 
                } else {
-retry2:
-                       int nummatches = std::numeric_limits<int>::max();
-                       lst repls;
 
                        for (size_t j=0; j<this->nops(); j++) {
-                               if (!subsed[j] && tryfactsubs(op(j), it->first, nummatches, repls)) {
+                               int nummatches = std::numeric_limits<int>::max();
+                               lst repls;
+                               if (!subsed[j] && tryfactsubs(op(j), it->first, nummatches, repls)){
                                        subsed[j] = true;
-                                       subsresult[j] = op(j) * power(it->second.subs(ex(repls), subs_options::no_pattern) / it->first.subs(ex(repls), subs_options::no_pattern), nummatches);
-                                       goto retry2;
+                                       ex subsed_pattern
+                                               = it->first.subs(ex(repls), subs_options::no_pattern);
+                                       divide_by *= power(subsed_pattern, nummatches);
+                                       ex subsed_result
+                                               = it->second.subs(ex(repls), subs_options::no_pattern);
+                                       multiply_by *= power(subsed_result, nummatches);
                                }
                        }
                }
@@ -653,15 +788,7 @@ retry2:
        if (!subsfound)
                return subs_one_level(m, options | subs_options::algebraic);
 
-       exvector ev; ev.reserve(nops());
-       for (size_t i=0; i<nops(); i++) {
-               if (subsed[i])
-                       ev.push_back(subsresult[i]);
-               else
-                       ev.push_back(op(i));
-       }
-
-       return (new mul(ev))->setflag(status_flags::dynallocated);
+       return ((*this)/divide_by)*multiply_by;
 }
 
 // protected
@@ -716,21 +843,8 @@ unsigned mul::return_type() const
                }
                if ((rt == return_types::noncommutative) && (!all_commutative)) {
                        // another nc element found, compare type_infos
-                       if (noncommutative_element->rest.return_type_tinfo()->tinfo() == &clifford::tinfo_static) {
-                               if (i->rest.return_type_tinfo()->tinfo() != &clifford::tinfo_static ||
-                                   ((clifford*)(noncommutative_element->rest.return_type_tinfo()))->get_representation_label() !=
-                                   ((clifford*)(i->rest.return_type_tinfo()))->get_representation_label()) {
-                                       // diffent types -> mul is ncc
-                                       return return_types::noncommutative_composite;
-                               }
-                       } else if (noncommutative_element->rest.return_type_tinfo()->tinfo() == &color::tinfo_static) {
-                               if (i->rest.return_type_tinfo()->tinfo() != &color::tinfo_static ||
-                                   ((color*)(noncommutative_element->rest.return_type_tinfo()))->get_representation_label() !=
-                                   ((color*)(i->rest.return_type_tinfo()))->get_representation_label()) {
-                                       // diffent types -> mul is ncc
-                                       return return_types::noncommutative_composite;
-                               }
-                       } else if (noncommutative_element->rest.return_type_tinfo()->tinfo() != i->rest.return_type_tinfo()->tinfo()) {
+                       if (noncommutative_element->rest.return_type_tinfo() != i->rest.return_type_tinfo()) {
+                                       // different types -> mul is ncc
                                        return return_types::noncommutative_composite;
                        }
                }
@@ -740,7 +854,7 @@ unsigned mul::return_type() const
        return all_commutative ? return_types::commutative : return_types::noncommutative;
 }
    
-const basic* mul::return_type_tinfo() const
+tinfo_t mul::return_type_tinfo() const
 {
        if (seq.empty())
                return this;  // mul without factors: should not happen
@@ -756,14 +870,14 @@ const basic* mul::return_type_tinfo() const
        return this;
 }
 
-ex mul::thisexpairseq(const epvector & v, const ex & oc) const
+ex mul::thisexpairseq(const epvector & v, const ex & oc, bool do_index_renaming) const
 {
-       return (new mul(v, oc))->setflag(status_flags::dynallocated);
+       return (new mul(v, oc, do_index_renaming))->setflag(status_flags::dynallocated);
 }
 
-ex mul::thisexpairseq(std::auto_ptr<epvector> vp, const ex & oc) const
+ex mul::thisexpairseq(std::auto_ptr<epvector> vp, const ex & oc, bool do_index_renaming) const
 {
-       return (new mul(vp, oc))->setflag(status_flags::dynallocated);
+       return (new mul(vp, oc, do_index_renaming))->setflag(status_flags::dynallocated);
 }
 
 expair mul::split_ex_to_pair(const ex & e) const
@@ -878,6 +992,7 @@ bool mul::can_be_further_expanded(const ex & e)
 
 ex mul::expand(unsigned options) const
 {
+       const bool skip_idx_rename = ! info(info_flags::has_indices);
        // First, expand the children
        std::auto_ptr<epvector> expanded_seqp = expandchildren(options);
        const epvector & expanded_seq = (expanded_seqp.get() ? *expanded_seqp : seq);
@@ -933,28 +1048,34 @@ ex mul::expand(unsigned options) const
                                ex tmp_accu = (new add(distrseq, add1.overall_coeff*add2.overall_coeff))->setflag(status_flags::dynallocated);
 
                                exvector add1_dummy_indices, add2_dummy_indices, add_indices;
+                               lst dummy_subs;
 
-                               for (epvector::const_iterator i=add1begin; i!=add1end; ++i) {
-                                       add_indices = get_all_dummy_indices(i->rest);
-                                       add1_dummy_indices.insert(add1_dummy_indices.end(), add_indices.begin(), add_indices.end());
-                               }
-                               for (epvector::const_iterator i=add2begin; i!=add2end; ++i) {
-                                       add_indices = get_all_dummy_indices(i->rest);
-                                       add2_dummy_indices.insert(add2_dummy_indices.end(), add_indices.begin(), add_indices.end());
-                               }
+                               if (!skip_idx_rename) {
+                                       for (epvector::const_iterator i=add1begin; i!=add1end; ++i) {
+                                               add_indices = get_all_dummy_indices_safely(i->rest);
+                                               add1_dummy_indices.insert(add1_dummy_indices.end(), add_indices.begin(), add_indices.end());
+                                       }
+                                       for (epvector::const_iterator i=add2begin; i!=add2end; ++i) {
+                                               add_indices = get_all_dummy_indices_safely(i->rest);
+                                               add2_dummy_indices.insert(add2_dummy_indices.end(), add_indices.begin(), add_indices.end());
+                                       }
 
-                               sort(add1_dummy_indices.begin(), add1_dummy_indices.end(), ex_is_less());
-                               sort(add2_dummy_indices.begin(), add2_dummy_indices.end(), ex_is_less());
-                               lst dummy_subs = rename_dummy_indices_uniquely(add1_dummy_indices, add2_dummy_indices);
+                                       sort(add1_dummy_indices.begin(), add1_dummy_indices.end(), ex_is_less());
+                                       sort(add2_dummy_indices.begin(), add2_dummy_indices.end(), ex_is_less());
+                                       dummy_subs = rename_dummy_indices_uniquely(add1_dummy_indices, add2_dummy_indices);
+                               }
 
                                // Multiply explicitly all non-numeric terms of add1 and add2:
                                for (epvector::const_iterator i2=add2begin; i2!=add2end; ++i2) {
                                        // We really have to combine terms here in order to compactify
                                        // the result.  Otherwise it would become waayy tooo bigg.
-                                       numeric oc;
-                                       distrseq.clear();
-                                       ex i2_new = (dummy_subs.op(0).nops()>0? 
-                                                                i2->rest.subs((lst)dummy_subs.op(0), (lst)dummy_subs.op(1), subs_options::no_pattern) : i2->rest);
+                                       numeric oc(*_num0_p);
+                                       epvector distrseq2;
+                                       distrseq2.reserve(add1.seq.size());
+                                       const ex i2_new = (skip_idx_rename || (dummy_subs.op(0).nops() == 0) ?
+                                                       i2->rest :
+                                                       i2->rest.subs(ex_to<lst>(dummy_subs.op(0)), 
+                                                               ex_to<lst>(dummy_subs.op(1)), subs_options::no_pattern));
                                        for (epvector::const_iterator i1=add1begin; i1!=add1end; ++i1) {
                                                // Don't push_back expairs which might have a rest that evaluates to a numeric,
                                                // since that would violate an invariant of expairseq:
@@ -962,13 +1083,12 @@ ex mul::expand(unsigned options) const
                                                if (is_exactly_a<numeric>(rest)) {
                                                        oc += ex_to<numeric>(rest).mul(ex_to<numeric>(i1->coeff).mul(ex_to<numeric>(i2->coeff)));
                                                } else {
-                                                       distrseq.push_back(expair(rest, ex_to<numeric>(i1->coeff).mul_dyn(ex_to<numeric>(i2->coeff))));
+                                                       distrseq2.push_back(expair(rest, ex_to<numeric>(i1->coeff).mul_dyn(ex_to<numeric>(i2->coeff))));
                                                }
                                        }
-                                       tmp_accu += (new add(distrseq, oc))->setflag(status_flags::dynallocated);
-                               }
+                                       tmp_accu += (new add(distrseq2, oc))->setflag(status_flags::dynallocated);
+                               } 
                                last_expanded = tmp_accu;
-
                        } else {
                                if (!last_expanded.is_equal(_ex1))
                                        non_adds.push_back(split_ex_to_pair(last_expanded));
@@ -986,12 +1106,18 @@ ex mul::expand(unsigned options) const
                size_t n = last_expanded.nops();
                exvector distrseq;
                distrseq.reserve(n);
-               exvector va = get_all_dummy_indices(mul(non_adds));
-               sort(va.begin(), va.end(), ex_is_less());
+               exvector va;
+               if (! skip_idx_rename) {
+                       va = get_all_dummy_indices_safely(mul(non_adds));
+                       sort(va.begin(), va.end(), ex_is_less());
+               }
 
                for (size_t i=0; i<n; ++i) {
                        epvector factors = non_adds;
-                       factors.push_back(split_ex_to_pair(rename_dummy_indices_uniquely(va, last_expanded.op(i))));
+                       if (skip_idx_rename)
+                               factors.push_back(split_ex_to_pair(last_expanded.op(i)));
+                       else
+                               factors.push_back(split_ex_to_pair(rename_dummy_indices_uniquely(va, last_expanded.op(i))));
                        ex term = (new mul(factors, overall_coeff))->setflag(status_flags::dynallocated);
                        if (can_be_further_expanded(term)) {
                                distrseq.push_back(term.expand());