lumiera_/tests/library/rational-test.cpp
Ichthyostega ce1220ee72 Lib: test coverage for rational-int corner cases and integer-log
- detailed documentation of known problematic behaviour
  when working with rational fractions
- demonstrate the heuristic predicate to detect dangerous numbers

- add extensive coverage and microbenchmarks for the integer-logarithm
  implementation, based on an example on Stackoverflow. Surprising result:
  The std::ilog(double) function is of comparable speed, at least for
  GCC-8 on Debian-Buster.
2022-11-14 05:20:37 +01:00

345 lines
16 KiB
C++

/*
Rational(Test) - verify support for rational arithmetics
Copyright (C) Lumiera.org
2022, Hermann Vosseler <Ichthyostega@web.de>
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 the Free Software Foundation; either version 2 of
the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
* *****************************************************/
/** @file rational-test.cpp
** unit test \ref Rational_test
*/
#include "lib/error.hpp"
#include "lib/test/run.hpp"
#include "lib/format-obj.hpp"
#include "lib/rational.hpp"
#include "lib/format-cout.hpp"//////////////TODO
#define SHOW_TYPE(_TY_) \
cout << "typeof( " << STRINGIFY(_TY_) << " )= " << lib::meta::typeStr<_TY_>() <<endl;
#define SHOW_EXPR(_XX_) \
cout << "#--◆--# " << STRINGIFY(_XX_) << " ? = " << _XX_ <<endl;
//////////////////////////////////////////////////////////////////////////////TODO
#include <chrono>
#include <array>
using std::array;
namespace util {
namespace test {
/***************************************************************************//**
* @test cover some aspects of working with fractional numbers.
* - demonstrate some basics, as provided by `boost::rational`
* - check for possibly dangerous values
* - re-quantise a rational number
* @see rational.hpp
* @see stage::model::test::ZoomWindow_test
*/
class Rational_test : public Test
{
virtual void
run (Arg)
{
demonstrate_basics();
verify_intLog2();
verify_limits();
verify_requant();
}
/**
* @test demonstrate fundamental properties of rational arithmetics
* as provided by boost::rational
* - represent rational fractions precisely
* - convert to other types and then perform the division
* - our typedef `Rat = boost::rational<int64_t>`
* - our user-defined literal "_r" to simplify notation
* - string conversion to reveal numerator and denominator
* - automatic normalisation and reduction
* - some typical fractional calculation examples.
*/
void
demonstrate_basics()
{
CHECK (Rat(10,3) == 10_r/3); // user-defined literal to construct a fraction
CHECK (Rat(10,3) == boost::rational<int64_t>(10,3)); // using Rat = boost::rational<int64_t>
CHECK (rational_cast<float> (10_r/3) == 3.3333333f); // rational_cast calculates division after type conversion
CHECK (2_r/3 + 3_r/4 == 17_r/12);
CHECK (2_r/3 * 3_r/4 == 1_r/2);
CHECK (2_r/3 /(3_r/4)== 8_r/9);
CHECK (2_r/3 / 3 /4 == 1_r/18); // usual precedence and brace rules apply, yielding 2/36 here
CHECK (util::toString(23_r/55) == "23/55sec"); //////////////////////////TICKET #1259 and #1261 : FSecs should really be a distinct (wrapper) type,
//////////////////////////TICKET #1259 and #1261 : ...then this custom conversion with the suffix "sec" would not kick in here
CHECK (util::toString(24_r/56) == "3/7sec" ); // rational numbers are normalised and reduced immediately
CHECK (Rat(10,3).numerator() == int64_t(10));
CHECK (Rat(10,3).denominator() == int64_t(3));
CHECK (boost::rational<uint>(10,3).numerator() == uint(10));
CHECK (boost::rational<uint>(10,3).denominator() == uint(3));
CHECK (boost::rational<uint>(10,3) == rational_cast<boost::rational<uint>> (Rat(10,3)));
CHECK (boost::rational<uint>(11,3) != rational_cast<boost::rational<uint>> (Rat(10,3)));
}
/**
* @test demonstrate the limits and perils of rational fractions
* - largest and smallest number representable
* - numeric overflow due to normalisation
* - predicates to check for possible trouble
*/
void
verify_limits()
{
const Rat MAXI = Rat{std::numeric_limits<int64_t>::max()};
const Rat MINI = Rat{1, std::numeric_limits<int64_t>::max()};
CHECK (rational_cast<int64_t>(MAXI) == std::numeric_limits<int64_t>::max());
CHECK (rational_cast<double> (MAXI) == 9.2233720368547758e+18);
CHECK (MAXI > 0); // so this one still works
CHECK (MAXI+1 < 0); // but one more and we get a wrap-around
CHECK (MAXI+1 < -MAXI);
CHECK (util::toString(MAXI) == "9223372036854775807sec"); /////////TICKET #1259 should be "9223372036854775807/1 -- get rid of the "sec" suffix
CHECK (util::toString(MAXI+1) == "-9223372036854775808sec"); /////////TICKET #1259 should be "-9223372036854775808/1"
CHECK (util::toString(-MAXI) == "-9223372036854775807sec"); /////////TICKET #1259 should be "-9223372036854775807/1"
CHECK (MINI > 0); // smallest representable number above zero
CHECK (1-MINI < 1);
CHECK (0 < 1-MINI); // can be used below 1 just fine
CHECK (0 > 1+MINI); // but above we get a wrap-around in normalised numerator
CHECK (util::toString(MINI) == "1/9223372036854775807sec");
CHECK (util::toString(-MINI) == "-1/9223372036854775807sec");
CHECK (util::toString(1-MINI) == "9223372036854775806/9223372036854775807sec");
CHECK (util::toString(1+MINI) == "-9223372036854775808/9223372036854775807sec");
CHECK ((MAXI-1)/MAXI == 1-MINI);
CHECK (MAXI/(MAXI-1) > 1); // as workaround we have to use a slightly larger ULP
CHECK (MAXI/(MAXI-1) - 1 > MINI); // ...this slightly larger one works without wrap-around
CHECK (1 - MAXI/(MAXI-1) < -MINI);
CHECK (util::toString(MAXI/(MAXI-1)) == "9223372036854775807/9223372036854775806sec");
CHECK (util::toString(MAXI/(MAXI-1) - 1) == "1/9223372036854775806sec");
CHECK (util::toString(1 - MAXI/(MAXI-1)) == "-1/9223372036854775806sec");
// Now entering absolute danger territory....
const Rat MIMI = -MAXI-1; // this is the most extreme negative representable value
CHECK (MIMI < 0);
CHECK (util::toString(MIMI) == "-9223372036854775808sec"); /////////TICKET #1259 should be "-9223372036854775808/1"
CHECK (util::toString(1/MIMI) == "-1/-9223372036854775808sec");
try
{
-1-1/MIMI; // ...but it can't be used for any calculation without blowing up
NOTREACHED("expected boost::rational to flounder");
}
catch (std::exception& tilt)
{
CHECK (tilt.what() == string{"bad rational: non-zero singular denominator"});
}
// yet seemingly harmless values can be poisonous...
Rat poison = MAXI/49 / (MAXI/49-1);
Rat decoy = poison + 5;
CHECK (poison > 0);
CHECK (decoy > 6);
CHECK (rational_cast<double>(decoy) == 6); // looks innocuous...
CHECK (rational_cast<double>(decoy+5) == 11); // ...aaaand...
CHECK (rational_cast<double>(decoy+50) == -42); // ..ultimate answer..
CHECK (rational_cast<double>(decoy+500) == 15.999999999999996); // .dead in the water.
// Heuristics to detect danger zone
CHECK ( can_represent_Sum(decoy,5));
CHECK (not can_represent_Sum(decoy,50));
// alarm is given a bit too early
CHECK ( can_represent_Sum(decoy,15)); // ...because the check is based on bit positions
CHECK (not can_represent_Sum(decoy,16)); // ...and here the highest bit of one partner moved into danger zone
CHECK (decoy+16 > 0);
CHECK (decoy+43 > 0);
CHECK (decoy+44 < 0);
// similar when increasing the denominator...
CHECK (poison + 1_r/10 > 0);
CHECK (poison + 1_r/90 > 0);
CHECK (poison + 1_r/98 < 0); // actually the flip already occurs at 1/91 but also causes an assertion failure
CHECK ( can_represent_Sum(poison,1_r/10));
CHECK ( can_represent_Sum(poison,1_r/15));
CHECK (not can_represent_Sum(poison,1_r/16));
CHECK (not can_represent_Sum(poison,1_r/91));
CHECK (not can_represent_Sum(poison,1_r/100));
}
/**
* @test a slightly optimised implementation of integer binary logarithm
* - basically finds the highest bit which is set
* - can be used with various integral types
* - performs better than using the floating-point solution
* @todo this test (and the implementation) belongs into some basic util header.
*/
void
verify_intLog2()
{
CHECK ( 5 == ilog2( int64_t(0b101010)));
CHECK ( 5 == ilog2(uint64_t(0b101010)));
CHECK ( 5 == ilog2( int32_t(0b101010)));
CHECK ( 5 == ilog2(uint32_t(0b101010)));
CHECK ( 5 == ilog2( int16_t(0b101010)));
CHECK ( 5 == ilog2(uint16_t(0b101010)));
CHECK ( 5 == ilog2( int8_t(0b101010)));
CHECK ( 5 == ilog2( uint8_t(0b101010)));
CHECK ( 5 == ilog2( int (0b101010)));
CHECK ( 5 == ilog2( uint (0b101010)));
CHECK ( 5 == ilog2( char (0b101010)));
CHECK ( 5 == ilog2( uchar (0b101010)));
CHECK ( 5 == ilog2( long (0b101010)));
CHECK ( 5 == ilog2( ulong (0b101010)));
CHECK ( 5 == ilog2( short (0b101010)));
CHECK ( 5 == ilog2( ushort (0b101010)));
CHECK (63 == ilog2(std::numeric_limits<uint64_t>::max()));
CHECK (62 == ilog2(std::numeric_limits< int64_t>::max()));
CHECK (31 == ilog2(std::numeric_limits<uint32_t>::max()));
CHECK (30 == ilog2(std::numeric_limits< int32_t>::max()));
CHECK (15 == ilog2(std::numeric_limits<uint16_t>::max()));
CHECK (14 == ilog2(std::numeric_limits< int16_t>::max()));
CHECK ( 7 == ilog2(std::numeric_limits< uint8_t>::max()));
CHECK ( 6 == ilog2(std::numeric_limits< int8_t>::max()));
CHECK ( 5 == ilog2(0b111111));
CHECK ( 5 == ilog2(0b101110));
CHECK ( 5 == ilog2(0b100100));
CHECK ( 5 == ilog2(0b100000));
CHECK ( 2 == ilog2(4));
CHECK ( 1 == ilog2(2));
CHECK ( 0 == ilog2(1));
CHECK (-1 == ilog2(0));
CHECK (-1 == ilog2(-1));
CHECK (-1 == ilog2(std::numeric_limits<uint64_t>::min()));
CHECK (-1 == ilog2(std::numeric_limits< int64_t>::min()));
CHECK (-1 == ilog2(std::numeric_limits<uint32_t>::min()));
CHECK (-1 == ilog2(std::numeric_limits< int32_t>::min()));
CHECK (-1 == ilog2(std::numeric_limits<uint16_t>::min()));
CHECK (-1 == ilog2(std::numeric_limits< int16_t>::min()));
CHECK (-1 == ilog2(std::numeric_limits< uint8_t>::min()));
CHECK (-1 == ilog2(std::numeric_limits< int8_t>::min()));
/* ==== compare with naive implementation ==== */
auto floatLog = [](auto n)
{
return n <=0? -1 : ilogb(n);
};
auto bitshift = [](auto n)
{
if (n <= 0) return -1;
int logB = 0;
while (n >>= 1)
++logB;
return logB;
};
auto do_nothing = [](auto n){ return n; };
array<uint64_t, 1000> numz;
for (auto& n : numz)
{
n = rand() * uint64_t(2147483648);
CHECK (ilog2(n) == floatLog(n));
CHECK (ilog2(n) == bitshift(n));
}
int64_t dump{0}; // throw-away result to prevent optimisation
auto microbenchmark = [&numz,&dump](auto algo)
{
using std::chrono::system_clock;
using Dur = std::chrono::duration<double>;
const double SCALE = 1e9; // Results are in ns
auto start = system_clock::now();
for (uint i=0; i<1000; ++i)
for (auto const& n : numz)
dump += algo(n);
Dur duration = system_clock::now () - start;
return duration.count()/(1000*1000) * SCALE;
};
auto time_ilog2 = microbenchmark(ilog2<int64_t>);
auto time_float = microbenchmark(floatLog);
auto time_shift = microbenchmark(bitshift);
auto time_ident = microbenchmark(do_nothing);
cout << "Microbenchmark integer-log2" <<endl
<< "util::ilog2 :"<<time_ilog2<<"ns"<<endl
<< "std::ilogb :"<<time_float<<"ns"<<endl
<< "bit-shift :"<<time_shift<<"ns"<<endl
<< "identity :"<<time_ident<<"ns"<<endl
<< "(checksum="<<dump<<")" <<endl; // Warning: without outputting `dump`, compiler would optimise away most calls
// the following holds both with -O0 and -O3
CHECK (time_ilog2 < time_shift);
CHECK (time_ident < time_ilog2);
/**** some numbers ****
*
* GCC-8, -O3, Debian-Buster, AMD FX83
*
* with uint64_t...
* - ilog2 : 5.6ns
* - ilogb : 5.0ns
* - shift : 44ns
* - ident : 0.6ns
*
* with uint8_t
* - ilog2 : 5.2ns
* - ilogb : 5.8ns
* - shift : 8.2ns
* - ident : 0.3ns
*/
}
/**
* @test helper to re-quantise a rational fraction
* - recast a number in terms of another denominator
* - this introduces an error of known limited size
* - and is an option to work around "poisonous" fractions
*/
void
verify_requant()
{
}
};
LAUNCHER (Rational_test, "unit common");
}} // namespace util::test