summaryrefslogtreecommitdiffstats
path: root/src/boost/libs/numeric/conversion/test/converter_test.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/boost/libs/numeric/conversion/test/converter_test.cpp')
-rw-r--r--src/boost/libs/numeric/conversion/test/converter_test.cpp562
1 files changed, 562 insertions, 0 deletions
diff --git a/src/boost/libs/numeric/conversion/test/converter_test.cpp b/src/boost/libs/numeric/conversion/test/converter_test.cpp
new file mode 100644
index 000000000..a460df158
--- /dev/null
+++ b/src/boost/libs/numeric/conversion/test/converter_test.cpp
@@ -0,0 +1,562 @@
+// (c) Copyright Fernando Luis Cacciola Carballal 2000-2004
+// Use, modification, and distribution is subject to the Boost Software
+// License, Version 1.0. (See accompanying file LICENSE_1_0.txt or copy at
+// http://www.boost.org/LICENSE_1_0.txt)
+
+// See library home page at http://www.boost.org/libs/numeric/conversion
+//
+// Contact the author at: fernando_cacciola@hotmail.com
+//
+#include<cstdlib>
+#include<iostream>
+#include<iomanip>
+#include<string>
+#include<typeinfo>
+#include<vector>
+#include<algorithm>
+
+#include "boost/config.hpp"
+#include "boost/cstdint.hpp"
+#include "boost/utility.hpp"
+
+//
+// Borland 5.5 lacks the following math overloads
+//
+#if BOOST_WORKAROUND(__BORLANDC__, <= 0x551)
+namespace std
+{
+
+inline float ceil (float x) { return std::ceil ( static_cast<double>(x)); }
+inline float floor (float x) { return std::floor ( static_cast<double>(x)); }
+inline long double ceil (long double x) { return std::ceill (x); }
+inline long double floor (long double x) { return std::floorl(x); }
+
+} // namespace std
+#endif
+
+#include "boost/numeric/conversion/converter.hpp"
+#include "boost/numeric/conversion/cast.hpp"
+
+#ifdef __BORLANDC__
+#pragma hdrstop
+#endif
+
+#include "test_helpers.cpp"
+#include "test_helpers2.cpp"
+#include "test_helpers3.cpp"
+
+#include "boost/mpl/alias.hpp"
+
+using std::cout ;
+
+// A generic 'abs' function.
+template<class N> inline N absG ( N v )
+{
+ return v < static_cast<N>(0) ? static_cast<N>(-v) : v ;
+}
+template<> inline unsigned char absG<unsigned char> ( unsigned char v ) { return v ; }
+template<> inline unsigned short absG<unsigned short> ( unsigned short v ) { return v ; }
+template<> inline unsigned int absG<unsigned int> ( unsigned int v ) { return v ; }
+template<> inline unsigned long absG<unsigned long> ( unsigned long v ) { return v ; }
+
+template<class T> inline void unused_variable ( T const& ) {}
+//
+// The following function excersizes specific conversions that cover
+// usual and boundary cases for each relevant combination.
+//
+void test_conversions()
+{
+ using namespace boost ;
+ using namespace numeric ;
+
+ // To help the test found possible bugs a random numbers are used.
+#if !defined(BOOST_NO_STDC_NAMESPACE)
+ using std::rand ;
+#endif
+
+ boost::int16_t v16 ;
+ boost::uint16_t uv16 ;
+ boost::int32_t v32 ;
+ boost::uint32_t uv32 ;
+
+ volatile float fv ; // avoid this to be cached internally in some fpu register
+ volatile double dv ; // avoid this to be cached internally in some fpu register
+
+ //
+ // sample (representative) conversions:
+ //
+ cout << "Testing representative conversions\n";
+
+ // integral to integral
+
+ // signed to signed
+
+ // not subranged
+ v16 = static_cast<boost::int16_t>(rand());
+ TEST_SUCCEEDING_CONVERSION_DEF(boost::int32_t,boost::int16_t,v16,v16);
+
+ // subranged
+ v16 = static_cast<boost::int16_t>(rand());
+ TEST_SUCCEEDING_CONVERSION_DEF(boost::int16_t,boost::int32_t,v16,v16);
+ TEST_POS_OVERFLOW_CONVERSION_DEF(boost::int16_t,boost::int32_t,bounds<boost::int16_t>::highest() + boost::int32_t(1) ) ;
+ TEST_NEG_OVERFLOW_CONVERSION_DEF(boost::int16_t,boost::int32_t,bounds<boost::int16_t>::lowest() - boost::int32_t(1) ) ;
+
+ // signed to unsigned
+
+ // subranged
+ v32 = absG(static_cast<boost::int32_t>(rand()));
+ v16 = absG(static_cast<boost::int16_t>(rand()));
+ TEST_SUCCEEDING_CONVERSION_DEF(boost::uint32_t,boost::int32_t,v32,v32);
+ TEST_SUCCEEDING_CONVERSION_DEF(boost::uint16_t,boost::int32_t,v16,v16);
+ TEST_POS_OVERFLOW_CONVERSION_DEF(boost::uint16_t,boost::int32_t,bounds<boost::uint16_t>::highest() + boost::int32_t(1) ) ;
+ TEST_NEG_OVERFLOW_CONVERSION_DEF(boost::uint32_t,boost::int32_t,boost::int32_t(-1) ) ;
+
+ // unsigned to signed
+
+ // not subranged
+ v32 = absG(static_cast<boost::int32_t>(rand()));
+ TEST_SUCCEEDING_CONVERSION_DEF(boost::int32_t,boost::uint32_t,v32,v32);
+
+ // subranged
+ v16 = absG(static_cast<boost::int16_t>(rand()));
+ TEST_SUCCEEDING_CONVERSION_DEF(boost::int16_t,boost::uint32_t,v16,v16);
+ TEST_POS_OVERFLOW_CONVERSION_DEF(boost::int32_t,boost::uint32_t,bounds<boost::uint32_t>::highest() ) ;
+ TEST_POS_OVERFLOW_CONVERSION_DEF(boost::int16_t,boost::uint32_t,bounds<boost::uint32_t>::highest() ) ;
+
+ // unsigned to unsigned
+
+ // not subranged
+ uv16 = static_cast<boost::uint16_t>(rand());
+ TEST_SUCCEEDING_CONVERSION_DEF(boost::uint32_t,boost::uint16_t,uv16,uv16);
+
+ // subranged
+ uv16 = static_cast<boost::uint16_t>(rand());
+ TEST_SUCCEEDING_CONVERSION_DEF(boost::uint16_t,boost::uint32_t,uv16,uv16);
+ TEST_POS_OVERFLOW_CONVERSION_DEF(boost::uint16_t,boost::uint32_t,bounds<boost::uint32_t>::highest() ) ;
+
+ // integral to float
+
+ // from signed integral
+ v32 = static_cast<boost::int32_t>(rand());
+ TEST_SUCCEEDING_CONVERSION_DEF(double,boost::int32_t,v32,v32);
+
+ // from uint32_tegral
+ uv32 = static_cast<boost::uint32_t>(rand());
+ TEST_SUCCEEDING_CONVERSION_DEF(double,boost::uint32_t,uv32,uv32);
+
+ // float to integral
+
+ // to signed integral
+ v32 = static_cast<boost::int32_t>(rand());
+ TEST_SUCCEEDING_CONVERSION_DEF(boost::int32_t,double,v32,v32);
+
+ dv = static_cast<double>(bounds<boost::uint32_t>::highest()) + 1.0 ;
+ TEST_POS_OVERFLOW_CONVERSION_DEF(boost::int32_t,double,dv) ;
+ TEST_NEG_OVERFLOW_CONVERSION_DEF(boost::int32_t,double,-dv) ;
+
+ // float to float
+
+ // not subranged
+ fv = static_cast<float>(rand()) / static_cast<float>(3) ;
+ TEST_SUCCEEDING_CONVERSION_DEF(double,float,fv,fv);
+
+
+ // subranged
+ fv = static_cast<float>(rand()) / static_cast<float>(3) ;
+ TEST_SUCCEEDING_CONVERSION_DEF(float,double,fv,fv);
+ TEST_POS_OVERFLOW_CONVERSION_DEF(float,double,bounds<double>::highest()) ;
+ TEST_NEG_OVERFLOW_CONVERSION_DEF(float,double,bounds<double>::lowest ()) ;
+}
+
+// Custom OverflowHandler
+struct custom_overflow_handler
+{
+ void operator() ( boost::numeric::range_check_result r )
+ {
+ if ( r == boost::numeric::cNegOverflow )
+ cout << "negative_overflow detected!\n" ;
+ else if ( r == boost::numeric::cPosOverflow )
+ cout << "positive_overflow detected!\n" ;
+ }
+} ;
+
+template<class T, class S,class OverflowHandler>
+void test_overflow_handler( MATCH_FNTPL_ARG(T), MATCH_FNTPL_ARG(S), MATCH_FNTPL_ARG(OverflowHandler),
+ PostCondition pos,
+ PostCondition neg
+ )
+{
+ typedef boost::numeric::conversion_traits<T,S> traits ;
+ typedef boost::numeric::converter<T,S,traits,OverflowHandler> converter ;
+
+ static const S psrc = boost::numeric::bounds<S>::highest();
+ static const S nsrc = boost::numeric::bounds<S>::lowest ();
+
+ static const T pres = static_cast<T>(psrc);
+ static const T nres = static_cast<T>(nsrc);
+
+ test_conv_base ( ConversionInstance<converter>(pres,psrc,pos) ) ;
+ test_conv_base ( ConversionInstance<converter>(nres,nsrc,neg) ) ;
+}
+
+template<class T, class S>
+void test_overflow_handlers( MATCH_FNTPL_ARG(T), MATCH_FNTPL_ARG(S) )
+{
+ cout << "Testing Silent Overflow Handler policy\n";
+
+ test_overflow_handler( SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(S),
+ SET_FNTPL_ARG(boost::numeric::silent_overflow_handler),
+ c_converted,
+ c_converted
+ ) ;
+
+ cout << "Testing Default Overflow Handler policy\n";
+
+ test_overflow_handler( SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(S),
+ SET_FNTPL_ARG(boost::numeric::def_overflow_handler),
+ c_pos_overflow,
+ c_neg_overflow
+ ) ;
+
+ cout << "Testing Custom (User-Defined) Overflow Handler policy\n";
+
+ test_overflow_handler( SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(S),
+ SET_FNTPL_ARG(custom_overflow_handler),
+ c_converted,
+ c_converted
+ ) ;
+}
+
+// For a given float-type number 'n' of integer value (n.0), check the conversions
+// within the range [n-1,n+1] taking values at: (n-1,n-0.5,n,n+0.5,n+1).
+// For each sampled value there is an expected result and a PostCondition according to the
+// specified round_style.
+//
+template<class T, class S, class Float2IntRounder>
+void test_rounding_conversion ( MATCH_FNTPL_ARG(T), MATCH_FNTPL_ARG(Float2IntRounder),
+ S s,
+ PostCondition resl1,
+ PostCondition resl0,
+ PostCondition res,
+ PostCondition resr0,
+ PostCondition resr1
+ )
+{
+ typedef boost::numeric::conversion_traits<T,S> Traits ;
+
+ typedef boost::numeric::converter<T,S, Traits, boost::numeric::def_overflow_handler,Float2IntRounder>
+ Converter ;
+
+ S sl1 = s - static_cast<S>(1);
+ S sl0 = s - static_cast<S>(0.5);
+ S sr0 = s + static_cast<S>(0.5);
+ S sr1 = s + static_cast<S>(1);
+
+ T tl1 = static_cast<T>( Converter::nearbyint(sl1) );
+ T tl0 = static_cast<T>( Converter::nearbyint(sl0) );
+ T t = static_cast<T>( Converter::nearbyint(s) );
+ T tr0 = static_cast<T>( Converter::nearbyint(sr0) );
+ T tr1 = static_cast<T>( Converter::nearbyint(sr1) );
+
+ test_conv_base ( ConversionInstance<Converter>(tl1,sl1,resl1) ) ;
+ test_conv_base ( ConversionInstance<Converter>(tl0,sl0,resl0) ) ;
+ test_conv_base ( ConversionInstance<Converter>(t,s,res) ) ;
+ test_conv_base ( ConversionInstance<Converter>(tr0,sr0,resr0) ) ;
+ test_conv_base ( ConversionInstance<Converter>(tr1,sr1,resr1) ) ;
+}
+
+
+template<class T,class S>
+void test_round_style( MATCH_FNTPL_ARG(T), MATCH_FNTPL_ARG(S) )
+{
+ S min = boost::numeric::bounds<T>::lowest();
+ S max = boost::numeric::bounds<T>::highest();
+
+ cout << "Testing 'Trunc' Float2IntRounder policy\n";
+
+ test_rounding_conversion(SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(boost::numeric::Trunc<S>),
+ min,
+ c_neg_overflow,
+ c_converted,
+ c_converted,
+ c_converted,
+ c_converted
+ ) ;
+
+ test_rounding_conversion(SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(boost::numeric::Trunc<S>),
+ max,
+ c_converted,
+ c_converted,
+ c_converted,
+ c_converted,
+ c_pos_overflow
+ ) ;
+
+ cout << "Testing 'RoundEven' Float2IntRounder policy\n";
+
+ test_rounding_conversion(SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(boost::numeric::RoundEven<S>),
+ min,
+ c_neg_overflow,
+ c_converted,
+ c_converted,
+ c_converted,
+ c_converted
+ ) ;
+
+ test_rounding_conversion(SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(boost::numeric::RoundEven<S>),
+ max,
+ c_converted,
+ c_converted,
+ c_converted,
+ c_pos_overflow,
+ c_pos_overflow
+ ) ;
+
+ cout << "Testing 'Ceil' Float2IntRounder policy\n";
+
+ test_rounding_conversion(SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(boost::numeric::Ceil<S>),
+ min,
+ c_neg_overflow,
+ c_converted,
+ c_converted,
+ c_converted,
+ c_converted
+ ) ;
+
+ test_rounding_conversion(SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(boost::numeric::Ceil<S>),
+ max,
+ c_converted,
+ c_converted,
+ c_converted,
+ c_pos_overflow,
+ c_pos_overflow
+ ) ;
+
+ cout << "Testing 'Floor' Float2IntRounder policy\n" ;
+
+ test_rounding_conversion(SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(boost::numeric::Floor<S>),
+ min,
+ c_neg_overflow,
+ c_neg_overflow,
+ c_converted,
+ c_converted,
+ c_converted
+ ) ;
+
+ test_rounding_conversion(SET_FNTPL_ARG(T),
+ SET_FNTPL_ARG(boost::numeric::Floor<S>),
+ max,
+ c_converted,
+ c_converted,
+ c_converted,
+ c_converted,
+ c_pos_overflow
+ ) ;
+
+}
+
+void test_round_even( double n, double x )
+{
+ double r = boost::numeric::RoundEven<double>::nearbyint(n);
+ BOOST_CHECK( r == x ) ;
+}
+
+void test_round_even()
+{
+ cout << "Testing 'RoundEven' tie-breaking\n";
+
+ double min = boost::numeric::bounds<double>::lowest();
+ double max = boost::numeric::bounds<double>::highest();
+
+#if !defined(BOOST_NO_STDC_NAMESPACE)
+ using std::floor ;
+ using std::ceil ;
+#endif
+ test_round_even(min, floor(min));
+ test_round_even(max, ceil (max));
+ test_round_even(2.0, 2.0);
+ test_round_even(2.3, 2.0);
+ test_round_even(2.5, 2.0);
+ test_round_even(2.7, 3.0);
+ test_round_even(3.0, 3.0);
+ test_round_even(3.3, 3.0);
+ test_round_even(3.5, 4.0);
+ test_round_even(3.7, 4.0);
+}
+
+int double_to_int ( double n ) { return static_cast<int>(n) ; }
+
+void test_converter_as_function_object()
+{
+ cout << "Testing converter as function object.\n";
+
+ // Create a sample sequence of double values.
+ std::vector<double> S ;
+ for ( int i = 0 ; i < 10 ; ++ i )
+ S.push_back( i * ( 18.0 / 19.0 ) );
+
+ // Create a sequence of int values from 's' using the standard conversion.
+ std::vector<int> W ;
+ std::transform(S.begin(),S.end(),std::back_inserter(W),double_to_int);
+
+ // Create a sequence of int values from s using a default numeric::converter
+ std::vector<int> I ;
+ std::transform(S.begin(),
+ S.end(),
+ std::back_inserter(I),
+ boost::numeric::converter<int,double>()
+ ) ;
+
+ // Match 'w' and 'i' which should be equal.
+ bool double_to_int_OK = std::equal(W.begin(),W.end(),I.begin()) ;
+ BOOST_CHECK_MESSAGE(double_to_int_OK, "converter (int,double) as function object");
+
+ // Create a sequence of double values from s using a default numeric::converter (which should be the trivial conv).
+ std::vector<double> D ;
+ std::transform(S.begin(),
+ S.end(),
+ std::back_inserter(D),
+ boost::numeric::converter<double,double>()
+ ) ;
+
+ // Match 's' and 'd' which should be equal.
+ bool double_to_double_OK = std::equal(S.begin(),S.end(),D.begin()) ;
+ BOOST_CHECK_MESSAGE(double_to_double_OK, "converter (double,double) as function object");
+}
+
+#if BOOST_WORKAROUND(__IBMCPP__, <= 600 ) // VCAPP6
+# define UNOPTIMIZED
+#else
+# define UNOPTIMIZED volatile
+#endif
+
+void test_optimizations()
+{
+ using namespace boost;
+ using namespace numeric;
+
+ float fv0 = 18.0f / 19.0f ;
+
+ // This code deosn't produce any output.
+ // It is intended to show the optimization of numeric::converter<> by manual inspection
+ // of the generated code.
+ // Each test shows first the equivalent hand-coded version.
+ // The numeric_cast<> code should be the same if full compiler optimization/inlining is used.
+
+ //---------------------------------
+ // trivial conversion.
+ //
+ // equivalent code:
+ UNOPTIMIZED float fv1a = fv0 ;
+
+ float fv1b = numeric_cast<float>(fv0);
+ unused_variable(fv1a);
+ unused_variable(fv1b);
+ //
+ //---------------------------------
+
+ //---------------------------------
+ // nonsubranged conversion.
+ //
+ // equivalent code:
+ UNOPTIMIZED double dv1a = static_cast<double>(fv0);
+
+ double dv1b = numeric_cast<double>(fv0);
+ unused_variable(dv1a);
+ unused_variable(dv1b);
+ //
+ //---------------------------------
+
+ //------------------------------------------------------
+ // subranged conversion with both-sided range checking.
+ //
+
+ // equivalent code:
+
+ {
+ double const& s = dv1b ;
+ // range checking
+ range_check_result r = s < static_cast<double>(bounds<float>::lowest())
+ ? cNegOverflow : cInRange ;
+ if ( r == cInRange )
+ {
+ r = s > static_cast<double>(bounds<float>::highest()) ? cPosOverflow : cInRange ;
+ }
+ if ( r == cNegOverflow )
+ throw negative_overflow() ;
+ else if ( r == cPosOverflow )
+ throw positive_overflow() ;
+ // conversion
+ UNOPTIMIZED float fv2a = static_cast<float>(s);
+ unused_variable(fv2a);
+ }
+
+ float fv2b = numeric_cast<float>(dv1b);
+ unused_variable(fv2b);
+ //
+ //---------------------------------
+
+
+ //---------------------------------
+ // subranged rounding conversion
+ //
+ // equivalent code:
+
+ {
+ double const& s = dv1b ;
+ // range checking
+ range_check_result r = s <= static_cast<double>(bounds<int>::lowest()) - static_cast<double>(1.0)
+ ? cNegOverflow : cInRange ;
+ if ( r == cInRange )
+ {
+ r = s >= static_cast<double>(bounds<int>::highest()) + static_cast<double>(1.0)
+ ? cPosOverflow : cInRange ;
+ }
+ if ( r == cNegOverflow )
+ throw negative_overflow() ;
+ else if ( r == cPosOverflow )
+ throw positive_overflow() ;
+ // rounding
+
+#if !defined(BOOST_NO_STDC_NAMESPACE)
+ using std::floor ;
+#endif
+
+ double s1 = floor(dv1b + 0.5);
+
+ // conversion
+ UNOPTIMIZED int iv1a = static_cast<int>(s1);
+ unused_variable(iv1a);
+ }
+
+ int iv1b = numeric_cast<int>(dv1b);
+ unused_variable(iv1b);
+ //
+ //---------------------------------
+}
+
+int test_main( int, char* argv[] )
+{
+ std::cout << std::setprecision( std::numeric_limits<long double>::digits10 ) ;
+
+ test_conversions();
+ test_overflow_handlers( SET_FNTPL_ARG(boost::int16_t), SET_FNTPL_ARG(boost::int32_t));
+ test_round_style(SET_FNTPL_ARG(boost::int32_t), SET_FNTPL_ARG(double) ) ;
+ test_round_even() ;
+ test_converter_as_function_object();
+ test_optimizations() ;
+
+ return 0;
+}
+//---------------------------------------------------------------------------
+