array( 'Group' => 'Mass', 'Unit Name' => 'Gram', 'AllowPrefix' => True ), 'sg' => array( 'Group' => 'Mass', 'Unit Name' => 'Slug', 'AllowPrefix' => False ), 'lbm' => array( 'Group' => 'Mass', 'Unit Name' => 'Pound mass (avoirdupois)', 'AllowPrefix' => False ), 'u' => array( 'Group' => 'Mass', 'Unit Name' => 'U (atomic mass unit)', 'AllowPrefix' => True ), 'ozm' => array( 'Group' => 'Mass', 'Unit Name' => 'Ounce mass (avoirdupois)', 'AllowPrefix' => False ), 'm' => array( 'Group' => 'Distance', 'Unit Name' => 'Meter', 'AllowPrefix' => True ), 'mi' => array( 'Group' => 'Distance', 'Unit Name' => 'Statute mile', 'AllowPrefix' => False ), 'Nmi' => array( 'Group' => 'Distance', 'Unit Name' => 'Nautical mile', 'AllowPrefix' => False ), 'in' => array( 'Group' => 'Distance', 'Unit Name' => 'Inch', 'AllowPrefix' => False ), 'ft' => array( 'Group' => 'Distance', 'Unit Name' => 'Foot', 'AllowPrefix' => False ), 'yd' => array( 'Group' => 'Distance', 'Unit Name' => 'Yard', 'AllowPrefix' => False ), 'ang' => array( 'Group' => 'Distance', 'Unit Name' => 'Angstrom', 'AllowPrefix' => True ), 'Pica' => array( 'Group' => 'Distance', 'Unit Name' => 'Pica (1/72 in)', 'AllowPrefix' => False ), 'yr' => array( 'Group' => 'Time', 'Unit Name' => 'Year', 'AllowPrefix' => False ), 'day' => array( 'Group' => 'Time', 'Unit Name' => 'Day', 'AllowPrefix' => False ), 'hr' => array( 'Group' => 'Time', 'Unit Name' => 'Hour', 'AllowPrefix' => False ), 'mn' => array( 'Group' => 'Time', 'Unit Name' => 'Minute', 'AllowPrefix' => False ), 'sec' => array( 'Group' => 'Time', 'Unit Name' => 'Second', 'AllowPrefix' => True ), 'Pa' => array( 'Group' => 'Pressure', 'Unit Name' => 'Pascal', 'AllowPrefix' => True ), 'p' => array( 'Group' => 'Pressure', 'Unit Name' => 'Pascal', 'AllowPrefix' => True ), 'atm' => array( 'Group' => 'Pressure', 'Unit Name' => 'Atmosphere', 'AllowPrefix' => True ), 'at' => array( 'Group' => 'Pressure', 'Unit Name' => 'Atmosphere', 'AllowPrefix' => True ), 'mmHg' => array( 'Group' => 'Pressure', 'Unit Name' => 'mm of Mercury', 'AllowPrefix' => True ), 'N' => array( 'Group' => 'Force', 'Unit Name' => 'Newton', 'AllowPrefix' => True ), 'dyn' => array( 'Group' => 'Force', 'Unit Name' => 'Dyne', 'AllowPrefix' => True ), 'dy' => array( 'Group' => 'Force', 'Unit Name' => 'Dyne', 'AllowPrefix' => True ), 'lbf' => array( 'Group' => 'Force', 'Unit Name' => 'Pound force', 'AllowPrefix' => False ), 'J' => array( 'Group' => 'Energy', 'Unit Name' => 'Joule', 'AllowPrefix' => True ), 'e' => array( 'Group' => 'Energy', 'Unit Name' => 'Erg', 'AllowPrefix' => True ), 'c' => array( 'Group' => 'Energy', 'Unit Name' => 'Thermodynamic calorie', 'AllowPrefix' => True ), 'cal' => array( 'Group' => 'Energy', 'Unit Name' => 'IT calorie', 'AllowPrefix' => True ), 'eV' => array( 'Group' => 'Energy', 'Unit Name' => 'Electron volt', 'AllowPrefix' => True ), 'ev' => array( 'Group' => 'Energy', 'Unit Name' => 'Electron volt', 'AllowPrefix' => True ), 'HPh' => array( 'Group' => 'Energy', 'Unit Name' => 'Horsepower-hour', 'AllowPrefix' => False ), 'hh' => array( 'Group' => 'Energy', 'Unit Name' => 'Horsepower-hour', 'AllowPrefix' => False ), 'Wh' => array( 'Group' => 'Energy', 'Unit Name' => 'Watt-hour', 'AllowPrefix' => True ), 'wh' => array( 'Group' => 'Energy', 'Unit Name' => 'Watt-hour', 'AllowPrefix' => True ), 'flb' => array( 'Group' => 'Energy', 'Unit Name' => 'Foot-pound', 'AllowPrefix' => False ), 'BTU' => array( 'Group' => 'Energy', 'Unit Name' => 'BTU', 'AllowPrefix' => False ), 'btu' => array( 'Group' => 'Energy', 'Unit Name' => 'BTU', 'AllowPrefix' => False ), 'HP' => array( 'Group' => 'Power', 'Unit Name' => 'Horsepower', 'AllowPrefix' => False ), 'h' => array( 'Group' => 'Power', 'Unit Name' => 'Horsepower', 'AllowPrefix' => False ), 'W' => array( 'Group' => 'Power', 'Unit Name' => 'Watt', 'AllowPrefix' => True ), 'w' => array( 'Group' => 'Power', 'Unit Name' => 'Watt', 'AllowPrefix' => True ), 'T' => array( 'Group' => 'Magnetism', 'Unit Name' => 'Tesla', 'AllowPrefix' => True ), 'ga' => array( 'Group' => 'Magnetism', 'Unit Name' => 'Gauss', 'AllowPrefix' => True ), 'C' => array( 'Group' => 'Temperature', 'Unit Name' => 'Celsius', 'AllowPrefix' => False ), 'cel' => array( 'Group' => 'Temperature', 'Unit Name' => 'Celsius', 'AllowPrefix' => False ), 'F' => array( 'Group' => 'Temperature', 'Unit Name' => 'Fahrenheit', 'AllowPrefix' => False ), 'fah' => array( 'Group' => 'Temperature', 'Unit Name' => 'Fahrenheit', 'AllowPrefix' => False ), 'K' => array( 'Group' => 'Temperature', 'Unit Name' => 'Kelvin', 'AllowPrefix' => False ), 'kel' => array( 'Group' => 'Temperature', 'Unit Name' => 'Kelvin', 'AllowPrefix' => False ), 'tsp' => array( 'Group' => 'Liquid', 'Unit Name' => 'Teaspoon', 'AllowPrefix' => False ), 'tbs' => array( 'Group' => 'Liquid', 'Unit Name' => 'Tablespoon', 'AllowPrefix' => False ), 'oz' => array( 'Group' => 'Liquid', 'Unit Name' => 'Fluid Ounce', 'AllowPrefix' => False ), 'cup' => array( 'Group' => 'Liquid', 'Unit Name' => 'Cup', 'AllowPrefix' => False ), 'pt' => array( 'Group' => 'Liquid', 'Unit Name' => 'U.S. Pint', 'AllowPrefix' => False ), 'us_pt' => array( 'Group' => 'Liquid', 'Unit Name' => 'U.S. Pint', 'AllowPrefix' => False ), 'uk_pt' => array( 'Group' => 'Liquid', 'Unit Name' => 'U.K. Pint', 'AllowPrefix' => False ), 'qt' => array( 'Group' => 'Liquid', 'Unit Name' => 'Quart', 'AllowPrefix' => False ), 'gal' => array( 'Group' => 'Liquid', 'Unit Name' => 'Gallon', 'AllowPrefix' => False ), 'l' => array( 'Group' => 'Liquid', 'Unit Name' => 'Litre', 'AllowPrefix' => True ), 'lt' => array( 'Group' => 'Liquid', 'Unit Name' => 'Litre', 'AllowPrefix' => True ) ); /** * Details of the Multiplier prefixes that can be used with Units of Measure in CONVERTUOM() * * @var mixed[] */ private static $_conversionMultipliers = array( 'Y' => array( 'multiplier' => 1E24, 'name' => 'yotta' ), 'Z' => array( 'multiplier' => 1E21, 'name' => 'zetta' ), 'E' => array( 'multiplier' => 1E18, 'name' => 'exa' ), 'P' => array( 'multiplier' => 1E15, 'name' => 'peta' ), 'T' => array( 'multiplier' => 1E12, 'name' => 'tera' ), 'G' => array( 'multiplier' => 1E9, 'name' => 'giga' ), 'M' => array( 'multiplier' => 1E6, 'name' => 'mega' ), 'k' => array( 'multiplier' => 1E3, 'name' => 'kilo' ), 'h' => array( 'multiplier' => 1E2, 'name' => 'hecto' ), 'e' => array( 'multiplier' => 1E1, 'name' => 'deka' ), 'd' => array( 'multiplier' => 1E-1, 'name' => 'deci' ), 'c' => array( 'multiplier' => 1E-2, 'name' => 'centi' ), 'm' => array( 'multiplier' => 1E-3, 'name' => 'milli' ), 'u' => array( 'multiplier' => 1E-6, 'name' => 'micro' ), 'n' => array( 'multiplier' => 1E-9, 'name' => 'nano' ), 'p' => array( 'multiplier' => 1E-12, 'name' => 'pico' ), 'f' => array( 'multiplier' => 1E-15, 'name' => 'femto' ), 'a' => array( 'multiplier' => 1E-18, 'name' => 'atto' ), 'z' => array( 'multiplier' => 1E-21, 'name' => 'zepto' ), 'y' => array( 'multiplier' => 1E-24, 'name' => 'yocto' ) ); /** * Details of the Units of measure conversion factors, organised by group * * @var mixed[] */ private static $_unitConversions = array( 'Mass' => array( 'g' => array( 'g' => 1.0, 'sg' => 6.85220500053478E-05, 'lbm' => 2.20462291469134E-03, 'u' => 6.02217000000000E+23, 'ozm' => 3.52739718003627E-02 ), 'sg' => array( 'g' => 1.45938424189287E+04, 'sg' => 1.0, 'lbm' => 3.21739194101647E+01, 'u' => 8.78866000000000E+27, 'ozm' => 5.14782785944229E+02 ), 'lbm' => array( 'g' => 4.5359230974881148E+02, 'sg' => 3.10810749306493E-02, 'lbm' => 1.0, 'u' => 2.73161000000000E+26, 'ozm' => 1.60000023429410E+01 ), 'u' => array( 'g' => 1.66053100460465E-24, 'sg' => 1.13782988532950E-28, 'lbm' => 3.66084470330684E-27, 'u' => 1.0, 'ozm' => 5.85735238300524E-26 ), 'ozm' => array( 'g' => 2.83495152079732E+01, 'sg' => 1.94256689870811E-03, 'lbm' => 6.24999908478882E-02, 'u' => 1.70725600000000E+25, 'ozm' => 1.0 ) ), 'Distance' => array( 'm' => array( 'm' => 1.0, 'mi' => 6.21371192237334E-04, 'Nmi' => 5.39956803455724E-04, 'in' => 3.93700787401575E+01, 'ft' => 3.28083989501312E+00, 'yd' => 1.09361329797891E+00, 'ang' => 1.00000000000000E+10, 'Pica' => 2.83464566929116E+03 ), 'mi' => array( 'm' => 1.60934400000000E+03, 'mi' => 1.0, 'Nmi' => 8.68976241900648E-01, 'in' => 6.33600000000000E+04, 'ft' => 5.28000000000000E+03, 'yd' => 1.76000000000000E+03, 'ang' => 1.60934400000000E+13, 'Pica' => 4.56191999999971E+06 ), 'Nmi' => array( 'm' => 1.85200000000000E+03, 'mi' => 1.15077944802354E+00, 'Nmi' => 1.0, 'in' => 7.29133858267717E+04, 'ft' => 6.07611548556430E+03, 'yd' => 2.02537182785694E+03, 'ang' => 1.85200000000000E+13, 'Pica' => 5.24976377952723E+06 ), 'in' => array( 'm' => 2.54000000000000E-02, 'mi' => 1.57828282828283E-05, 'Nmi' => 1.37149028077754E-05, 'in' => 1.0, 'ft' => 8.33333333333333E-02, 'yd' => 2.77777777686643E-02, 'ang' => 2.54000000000000E+08, 'Pica' => 7.19999999999955E+01 ), 'ft' => array( 'm' => 3.04800000000000E-01, 'mi' => 1.89393939393939E-04, 'Nmi' => 1.64578833693305E-04, 'in' => 1.20000000000000E+01, 'ft' => 1.0, 'yd' => 3.33333333223972E-01, 'ang' => 3.04800000000000E+09, 'Pica' => 8.63999999999946E+02 ), 'yd' => array( 'm' => 9.14400000300000E-01, 'mi' => 5.68181818368230E-04, 'Nmi' => 4.93736501241901E-04, 'in' => 3.60000000118110E+01, 'ft' => 3.00000000000000E+00, 'yd' => 1.0, 'ang' => 9.14400000300000E+09, 'Pica' => 2.59200000085023E+03 ), 'ang' => array( 'm' => 1.00000000000000E-10, 'mi' => 6.21371192237334E-14, 'Nmi' => 5.39956803455724E-14, 'in' => 3.93700787401575E-09, 'ft' => 3.28083989501312E-10, 'yd' => 1.09361329797891E-10, 'ang' => 1.0, 'Pica' => 2.83464566929116E-07 ), 'Pica' => array( 'm' => 3.52777777777800E-04, 'mi' => 2.19205948372629E-07, 'Nmi' => 1.90484761219114E-07, 'in' => 1.38888888888898E-02, 'ft' => 1.15740740740748E-03, 'yd' => 3.85802469009251E-04, 'ang' => 3.52777777777800E+06, 'Pica' => 1.0 ) ), 'Time' => array( 'yr' => array( 'yr' => 1.0, 'day' => 365.25, 'hr' => 8766.0, 'mn' => 525960.0, 'sec' => 31557600.0 ), 'day' => array( 'yr' => 2.73785078713210E-03, 'day' => 1.0, 'hr' => 24.0, 'mn' => 1440.0, 'sec' => 86400.0 ), 'hr' => array( 'yr' => 1.14077116130504E-04, 'day' => 4.16666666666667E-02, 'hr' => 1.0, 'mn' => 60.0, 'sec' => 3600.0 ), 'mn' => array( 'yr' => 1.90128526884174E-06, 'day' => 6.94444444444444E-04, 'hr' => 1.66666666666667E-02, 'mn' => 1.0, 'sec' => 60.0 ), 'sec' => array( 'yr' => 3.16880878140289E-08, 'day' => 1.15740740740741E-05, 'hr' => 2.77777777777778E-04, 'mn' => 1.66666666666667E-02, 'sec' => 1.0 ) ), 'Pressure' => array( 'Pa' => array( 'Pa' => 1.0, 'p' => 1.0, 'atm' => 9.86923299998193E-06, 'at' => 9.86923299998193E-06, 'mmHg' => 7.50061707998627E-03 ), 'p' => array( 'Pa' => 1.0, 'p' => 1.0, 'atm' => 9.86923299998193E-06, 'at' => 9.86923299998193E-06, 'mmHg' => 7.50061707998627E-03 ), 'atm' => array( 'Pa' => 1.01324996583000E+05, 'p' => 1.01324996583000E+05, 'atm' => 1.0, 'at' => 1.0, 'mmHg' => 760.0 ), 'at' => array( 'Pa' => 1.01324996583000E+05, 'p' => 1.01324996583000E+05, 'atm' => 1.0, 'at' => 1.0, 'mmHg' => 760.0 ), 'mmHg' => array( 'Pa' => 1.33322363925000E+02, 'p' => 1.33322363925000E+02, 'atm' => 1.31578947368421E-03, 'at' => 1.31578947368421E-03, 'mmHg' => 1.0 ) ), 'Force' => array( 'N' => array( 'N' => 1.0, 'dyn' => 1.0E+5, 'dy' => 1.0E+5, 'lbf' => 2.24808923655339E-01 ), 'dyn' => array( 'N' => 1.0E-5, 'dyn' => 1.0, 'dy' => 1.0, 'lbf' => 2.24808923655339E-06 ), 'dy' => array( 'N' => 1.0E-5, 'dyn' => 1.0, 'dy' => 1.0, 'lbf' => 2.24808923655339E-06 ), 'lbf' => array( 'N' => 4.448222, 'dyn' => 4.448222E+5, 'dy' => 4.448222E+5, 'lbf' => 1.0 ) ), 'Energy' => array( 'J' => array( 'J' => 1.0, 'e' => 9.99999519343231E+06, 'c' => 2.39006249473467E-01, 'cal' => 2.38846190642017E-01, 'eV' => 6.24145700000000E+18, 'ev' => 6.24145700000000E+18, 'HPh' => 3.72506430801000E-07, 'hh' => 3.72506430801000E-07, 'Wh' => 2.77777916238711E-04, 'wh' => 2.77777916238711E-04, 'flb' => 2.37304222192651E+01, 'BTU' => 9.47815067349015E-04, 'btu' => 9.47815067349015E-04 ), 'e' => array( 'J' => 1.00000048065700E-07, 'e' => 1.0, 'c' => 2.39006364353494E-08, 'cal' => 2.38846305445111E-08, 'eV' => 6.24146000000000E+11, 'ev' => 6.24146000000000E+11, 'HPh' => 3.72506609848824E-14, 'hh' => 3.72506609848824E-14, 'Wh' => 2.77778049754611E-11, 'wh' => 2.77778049754611E-11, 'flb' => 2.37304336254586E-06, 'BTU' => 9.47815522922962E-11, 'btu' => 9.47815522922962E-11 ), 'c' => array( 'J' => 4.18399101363672E+00, 'e' => 4.18398900257312E+07, 'c' => 1.0, 'cal' => 9.99330315287563E-01, 'eV' => 2.61142000000000E+19, 'ev' => 2.61142000000000E+19, 'HPh' => 1.55856355899327E-06, 'hh' => 1.55856355899327E-06, 'Wh' => 1.16222030532950E-03, 'wh' => 1.16222030532950E-03, 'flb' => 9.92878733152102E+01, 'BTU' => 3.96564972437776E-03, 'btu' => 3.96564972437776E-03 ), 'cal' => array( 'J' => 4.18679484613929E+00, 'e' => 4.18679283372801E+07, 'c' => 1.00067013349059E+00, 'cal' => 1.0, 'eV' => 2.61317000000000E+19, 'ev' => 2.61317000000000E+19, 'HPh' => 1.55960800463137E-06, 'hh' => 1.55960800463137E-06, 'Wh' => 1.16299914807955E-03, 'wh' => 1.16299914807955E-03, 'flb' => 9.93544094443283E+01, 'BTU' => 3.96830723907002E-03, 'btu' => 3.96830723907002E-03 ), 'eV' => array( 'J' => 1.60219000146921E-19, 'e' => 1.60218923136574E-12, 'c' => 3.82933423195043E-20, 'cal' => 3.82676978535648E-20, 'eV' => 1.0, 'ev' => 1.0, 'HPh' => 5.96826078912344E-26, 'hh' => 5.96826078912344E-26, 'Wh' => 4.45053000026614E-23, 'wh' => 4.45053000026614E-23, 'flb' => 3.80206452103492E-18, 'BTU' => 1.51857982414846E-22, 'btu' => 1.51857982414846E-22 ), 'ev' => array( 'J' => 1.60219000146921E-19, 'e' => 1.60218923136574E-12, 'c' => 3.82933423195043E-20, 'cal' => 3.82676978535648E-20, 'eV' => 1.0, 'ev' => 1.0, 'HPh' => 5.96826078912344E-26, 'hh' => 5.96826078912344E-26, 'Wh' => 4.45053000026614E-23, 'wh' => 4.45053000026614E-23, 'flb' => 3.80206452103492E-18, 'BTU' => 1.51857982414846E-22, 'btu' => 1.51857982414846E-22 ), 'HPh' => array( 'J' => 2.68451741316170E+06, 'e' => 2.68451612283024E+13, 'c' => 6.41616438565991E+05, 'cal' => 6.41186757845835E+05, 'eV' => 1.67553000000000E+25, 'ev' => 1.67553000000000E+25, 'HPh' => 1.0, 'hh' => 1.0, 'Wh' => 7.45699653134593E+02, 'wh' => 7.45699653134593E+02, 'flb' => 6.37047316692964E+07, 'BTU' => 2.54442605275546E+03, 'btu' => 2.54442605275546E+03 ), 'hh' => array( 'J' => 2.68451741316170E+06, 'e' => 2.68451612283024E+13, 'c' => 6.41616438565991E+05, 'cal' => 6.41186757845835E+05, 'eV' => 1.67553000000000E+25, 'ev' => 1.67553000000000E+25, 'HPh' => 1.0, 'hh' => 1.0, 'Wh' => 7.45699653134593E+02, 'wh' => 7.45699653134593E+02, 'flb' => 6.37047316692964E+07, 'BTU' => 2.54442605275546E+03, 'btu' => 2.54442605275546E+03 ), 'Wh' => array( 'J' => 3.59999820554720E+03, 'e' => 3.59999647518369E+10, 'c' => 8.60422069219046E+02, 'cal' => 8.59845857713046E+02, 'eV' => 2.24692340000000E+22, 'ev' => 2.24692340000000E+22, 'HPh' => 1.34102248243839E-03, 'hh' => 1.34102248243839E-03, 'Wh' => 1.0, 'wh' => 1.0, 'flb' => 8.54294774062316E+04, 'BTU' => 3.41213254164705E+00, 'btu' => 3.41213254164705E+00 ), 'wh' => array( 'J' => 3.59999820554720E+03, 'e' => 3.59999647518369E+10, 'c' => 8.60422069219046E+02, 'cal' => 8.59845857713046E+02, 'eV' => 2.24692340000000E+22, 'ev' => 2.24692340000000E+22, 'HPh' => 1.34102248243839E-03, 'hh' => 1.34102248243839E-03, 'Wh' => 1.0, 'wh' => 1.0, 'flb' => 8.54294774062316E+04, 'BTU' => 3.41213254164705E+00, 'btu' => 3.41213254164705E+00 ), 'flb' => array( 'J' => 4.21400003236424E-02, 'e' => 4.21399800687660E+05, 'c' => 1.00717234301644E-02, 'cal' => 1.00649785509554E-02, 'eV' => 2.63015000000000E+17, 'ev' => 2.63015000000000E+17, 'HPh' => 1.56974211145130E-08, 'hh' => 1.56974211145130E-08, 'Wh' => 1.17055614802000E-05, 'wh' => 1.17055614802000E-05, 'flb' => 1.0, 'BTU' => 3.99409272448406E-05, 'btu' => 3.99409272448406E-05 ), 'BTU' => array( 'J' => 1.05505813786749E+03, 'e' => 1.05505763074665E+10, 'c' => 2.52165488508168E+02, 'cal' => 2.51996617135510E+02, 'eV' => 6.58510000000000E+21, 'ev' => 6.58510000000000E+21, 'HPh' => 3.93015941224568E-04, 'hh' => 3.93015941224568E-04, 'Wh' => 2.93071851047526E-01, 'wh' => 2.93071851047526E-01, 'flb' => 2.50369750774671E+04, 'BTU' => 1.0, 'btu' => 1.0, ), 'btu' => array( 'J' => 1.05505813786749E+03, 'e' => 1.05505763074665E+10, 'c' => 2.52165488508168E+02, 'cal' => 2.51996617135510E+02, 'eV' => 6.58510000000000E+21, 'ev' => 6.58510000000000E+21, 'HPh' => 3.93015941224568E-04, 'hh' => 3.93015941224568E-04, 'Wh' => 2.93071851047526E-01, 'wh' => 2.93071851047526E-01, 'flb' => 2.50369750774671E+04, 'BTU' => 1.0, 'btu' => 1.0, ) ), 'Power' => array( 'HP' => array( 'HP' => 1.0, 'h' => 1.0, 'W' => 7.45701000000000E+02, 'w' => 7.45701000000000E+02 ), 'h' => array( 'HP' => 1.0, 'h' => 1.0, 'W' => 7.45701000000000E+02, 'w' => 7.45701000000000E+02 ), 'W' => array( 'HP' => 1.34102006031908E-03, 'h' => 1.34102006031908E-03, 'W' => 1.0, 'w' => 1.0 ), 'w' => array( 'HP' => 1.34102006031908E-03, 'h' => 1.34102006031908E-03, 'W' => 1.0, 'w' => 1.0 ) ), 'Magnetism' => array( 'T' => array( 'T' => 1.0, 'ga' => 10000.0 ), 'ga' => array( 'T' => 0.0001, 'ga' => 1.0 ) ), 'Liquid' => array( 'tsp' => array( 'tsp' => 1.0, 'tbs' => 3.33333333333333E-01, 'oz' => 1.66666666666667E-01, 'cup' => 2.08333333333333E-02, 'pt' => 1.04166666666667E-02, 'us_pt' => 1.04166666666667E-02, 'uk_pt' => 8.67558516821960E-03, 'qt' => 5.20833333333333E-03, 'gal' => 1.30208333333333E-03, 'l' => 4.92999408400710E-03, 'lt' => 4.92999408400710E-03 ), 'tbs' => array( 'tsp' => 3.00000000000000E+00, 'tbs' => 1.0, 'oz' => 5.00000000000000E-01, 'cup' => 6.25000000000000E-02, 'pt' => 3.12500000000000E-02, 'us_pt' => 3.12500000000000E-02, 'uk_pt' => 2.60267555046588E-02, 'qt' => 1.56250000000000E-02, 'gal' => 3.90625000000000E-03, 'l' => 1.47899822520213E-02, 'lt' => 1.47899822520213E-02 ), 'oz' => array( 'tsp' => 6.00000000000000E+00, 'tbs' => 2.00000000000000E+00, 'oz' => 1.0, 'cup' => 1.25000000000000E-01, 'pt' => 6.25000000000000E-02, 'us_pt' => 6.25000000000000E-02, 'uk_pt' => 5.20535110093176E-02, 'qt' => 3.12500000000000E-02, 'gal' => 7.81250000000000E-03, 'l' => 2.95799645040426E-02, 'lt' => 2.95799645040426E-02 ), 'cup' => array( 'tsp' => 4.80000000000000E+01, 'tbs' => 1.60000000000000E+01, 'oz' => 8.00000000000000E+00, 'cup' => 1.0, 'pt' => 5.00000000000000E-01, 'us_pt' => 5.00000000000000E-01, 'uk_pt' => 4.16428088074541E-01, 'qt' => 2.50000000000000E-01, 'gal' => 6.25000000000000E-02, 'l' => 2.36639716032341E-01, 'lt' => 2.36639716032341E-01 ), 'pt' => array( 'tsp' => 9.60000000000000E+01, 'tbs' => 3.20000000000000E+01, 'oz' => 1.60000000000000E+01, 'cup' => 2.00000000000000E+00, 'pt' => 1.0, 'us_pt' => 1.0, 'uk_pt' => 8.32856176149081E-01, 'qt' => 5.00000000000000E-01, 'gal' => 1.25000000000000E-01, 'l' => 4.73279432064682E-01, 'lt' => 4.73279432064682E-01 ), 'us_pt' => array( 'tsp' => 9.60000000000000E+01, 'tbs' => 3.20000000000000E+01, 'oz' => 1.60000000000000E+01, 'cup' => 2.00000000000000E+00, 'pt' => 1.0, 'us_pt' => 1.0, 'uk_pt' => 8.32856176149081E-01, 'qt' => 5.00000000000000E-01, 'gal' => 1.25000000000000E-01, 'l' => 4.73279432064682E-01, 'lt' => 4.73279432064682E-01 ), 'uk_pt' => array( 'tsp' => 1.15266000000000E+02, 'tbs' => 3.84220000000000E+01, 'oz' => 1.92110000000000E+01, 'cup' => 2.40137500000000E+00, 'pt' => 1.20068750000000E+00, 'us_pt' => 1.20068750000000E+00, 'uk_pt' => 1.0, 'qt' => 6.00343750000000E-01, 'gal' => 1.50085937500000E-01, 'l' => 5.68260698087162E-01, 'lt' => 5.68260698087162E-01 ), 'qt' => array( 'tsp' => 1.92000000000000E+02, 'tbs' => 6.40000000000000E+01, 'oz' => 3.20000000000000E+01, 'cup' => 4.00000000000000E+00, 'pt' => 2.00000000000000E+00, 'us_pt' => 2.00000000000000E+00, 'uk_pt' => 1.66571235229816E+00, 'qt' => 1.0, 'gal' => 2.50000000000000E-01, 'l' => 9.46558864129363E-01, 'lt' => 9.46558864129363E-01 ), 'gal' => array( 'tsp' => 7.68000000000000E+02, 'tbs' => 2.56000000000000E+02, 'oz' => 1.28000000000000E+02, 'cup' => 1.60000000000000E+01, 'pt' => 8.00000000000000E+00, 'us_pt' => 8.00000000000000E+00, 'uk_pt' => 6.66284940919265E+00, 'qt' => 4.00000000000000E+00, 'gal' => 1.0, 'l' => 3.78623545651745E+00, 'lt' => 3.78623545651745E+00 ), 'l' => array( 'tsp' => 2.02840000000000E+02, 'tbs' => 6.76133333333333E+01, 'oz' => 3.38066666666667E+01, 'cup' => 4.22583333333333E+00, 'pt' => 2.11291666666667E+00, 'us_pt' => 2.11291666666667E+00, 'uk_pt' => 1.75975569552166E+00, 'qt' => 1.05645833333333E+00, 'gal' => 2.64114583333333E-01, 'l' => 1.0, 'lt' => 1.0 ), 'lt' => array( 'tsp' => 2.02840000000000E+02, 'tbs' => 6.76133333333333E+01, 'oz' => 3.38066666666667E+01, 'cup' => 4.22583333333333E+00, 'pt' => 2.11291666666667E+00, 'us_pt' => 2.11291666666667E+00, 'uk_pt' => 1.75975569552166E+00, 'qt' => 1.05645833333333E+00, 'gal' => 2.64114583333333E-01, 'l' => 1.0, 'lt' => 1.0 ) ) ); /** * _parseComplex * * Parses a complex number into its real and imaginary parts, and an I or J suffix * * @param string $complexNumber The complex number * @return string[] Indexed on "real", "imaginary" and "suffix" */ public static function _parseComplex($complexNumber) { $workString = (string) $complexNumber; $realNumber = $imaginary = 0; // Extract the suffix, if there is one $suffix = substr($workString,-1); if (!is_numeric($suffix)) { $workString = substr($workString,0,-1); } else { $suffix = ''; } // Split the input into its Real and Imaginary components $leadingSign = 0; if (strlen($workString) > 0) { $leadingSign = (($workString{0} == '+') || ($workString{0} == '-')) ? 1 : 0; } $power = ''; $realNumber = strtok($workString, '+-'); if (strtoupper(substr($realNumber,-1)) == 'E') { $power = strtok('+-'); ++$leadingSign; } $realNumber = substr($workString,0,strlen($realNumber)+strlen($power)+$leadingSign); if ($suffix != '') { $imaginary = substr($workString,strlen($realNumber)); if (($imaginary == '') && (($realNumber == '') || ($realNumber == '+') || ($realNumber == '-'))) { $imaginary = $realNumber.'1'; $realNumber = '0'; } else if ($imaginary == '') { $imaginary = $realNumber; $realNumber = '0'; } elseif (($imaginary == '+') || ($imaginary == '-')) { $imaginary .= '1'; } } return array( 'real' => $realNumber, 'imaginary' => $imaginary, 'suffix' => $suffix ); } // function _parseComplex() /** * _cleanComplex * * Cleans the leading characters in a complex number string * * @param string $complexNumber The complex number to clean * @return string The "cleaned" complex number */ private static function _cleanComplex($complexNumber) { if ($complexNumber{0} == '+') $complexNumber = substr($complexNumber,1); if ($complexNumber{0} == '0') $complexNumber = substr($complexNumber,1); if ($complexNumber{0} == '.') $complexNumber = '0'.$complexNumber; if ($complexNumber{0} == '+') $complexNumber = substr($complexNumber,1); return $complexNumber; } private static function _nbrConversionFormat($xVal,$places) { if (!is_null($places)) { if (strlen($xVal) <= $places) { return substr(str_pad($xVal,$places,'0',STR_PAD_LEFT),-10); } else { return PHPExcel_Calculation_Functions::NaN(); } } return substr($xVal,-10); } // function _nbrConversionFormat() /** * BESSELI * * Returns the modified Bessel function, which is equivalent to the Bessel function evaluated for * purely imaginary arguments * * Excel Function: * BESSELI(x,ord) * * @access public * @category Engineering Functions * @param float $x The value at which to evaluate the function. * If x is nonnumeric, BESSELI returns the #VALUE! error value. * @param integer $ord The order of the Bessel function. If n is not an integer, it is truncated. * If $ord is nonnumeric, BESSELI returns the #VALUE! error value. * If $ord < 0, BESSELI returns the #NUM! error value. * @return float * * @TODO Better handling of the approximation method to support the differences between Excel/Gnumeric and Open/Libre Office * */ public static function BESSELI($x, $ord) { $x = (is_null($x)) ? 0.0 : PHPExcel_Calculation_Functions::flattenSingleValue($x); $ord = (is_null($ord)) ? 0.0 : PHPExcel_Calculation_Functions::flattenSingleValue($ord); if ((is_numeric($x)) && (is_numeric($ord))) { $ord = floor($ord); if ($ord < 0) { return PHPExcel_Calculation_Functions::NaN(); } if (abs($x) <= 30) { $fResult = $fTerm = pow($x / 2, $ord) / PHPExcel_Calculation_MathTrig::FACT($ord); $ordK = 1; $fSqrX = ($x * $x) / 4; do { $fTerm *= $fSqrX; $fTerm /= ($ordK * ($ordK + $ord)); $fResult += $fTerm; } while ((abs($fTerm) > 1e-12) && (++$ordK < 100)); } else { $f_2_PI = 2 * M_PI; $fXAbs = abs($x); $fResult = exp($fXAbs) / sqrt($f_2_PI * $fXAbs); if (($ord & 1) && ($x < 0)) { $fResult = -$fResult; } } return (is_nan($fResult)) ? PHPExcel_Calculation_Functions::NaN() : $fResult; } return PHPExcel_Calculation_Functions::VALUE(); } // function BESSELI() /** * BESSELJ * * Returns the Bessel function * * Excel Function: * BESSELJ(x,ord) * * @access public * @category Engineering Functions * @param float $x The value at which to evaluate the function. * If x is nonnumeric, BESSELJ returns the #VALUE! error value. * @param integer $ord The order of the Bessel function. If n is not an integer, it is truncated. * If $ord is nonnumeric, BESSELJ returns the #VALUE! error value. * If $ord < 0, BESSELJ returns the #NUM! error value. * @return float * * @TODO Better handling of the approximation method to support the differences between Excel/Gnumeric and Open/Libre Office * */ public static function BESSELJ($x, $ord) { $x = (is_null($x)) ? 0.0 : PHPExcel_Calculation_Functions::flattenSingleValue($x); $ord = (is_null($ord)) ? 0.0 : PHPExcel_Calculation_Functions::flattenSingleValue($ord); if ((is_numeric($x)) && (is_numeric($ord))) { $ord = floor($ord); if ($ord < 0) { return PHPExcel_Calculation_Functions::NaN(); } $fResult = 0; if (abs($x) <= 30) { $fResult = $fTerm = pow($x / 2, $ord) / PHPExcel_Calculation_MathTrig::FACT($ord); $ordK = 1; $fSqrX = ($x * $x) / -4; do { $fTerm *= $fSqrX; $fTerm /= ($ordK * ($ordK + $ord)); $fResult += $fTerm; } while ((abs($fTerm) > 1e-12) && (++$ordK < 100)); } else { $f_PI_DIV_2 = M_PI / 2; $f_PI_DIV_4 = M_PI / 4; $fXAbs = abs($x); $fResult = sqrt(M_2DIVPI / $fXAbs) * cos($fXAbs - $ord * $f_PI_DIV_2 - $f_PI_DIV_4); if (($ord & 1) && ($x < 0)) { $fResult = -$fResult; } } return (is_nan($fResult)) ? PHPExcel_Calculation_Functions::NaN() : $fResult; } return PHPExcel_Calculation_Functions::VALUE(); } // function BESSELJ() private static function _Besselk0($fNum) { if ($fNum <= 2) { $fNum2 = $fNum * 0.5; $y = ($fNum2 * $fNum2); $fRet = -log($fNum2) * self::BESSELI($fNum, 0) + (-0.57721566 + $y * (0.42278420 + $y * (0.23069756 + $y * (0.3488590e-1 + $y * (0.262698e-2 + $y * (0.10750e-3 + $y * 0.74e-5)))))); } else { $y = 2 / $fNum; $fRet = exp(-$fNum) / sqrt($fNum) * (1.25331414 + $y * (-0.7832358e-1 + $y * (0.2189568e-1 + $y * (-0.1062446e-1 + $y * (0.587872e-2 + $y * (-0.251540e-2 + $y * 0.53208e-3)))))); } return $fRet; } // function _Besselk0() private static function _Besselk1($fNum) { if ($fNum <= 2) { $fNum2 = $fNum * 0.5; $y = ($fNum2 * $fNum2); $fRet = log($fNum2) * self::BESSELI($fNum, 1) + (1 + $y * (0.15443144 + $y * (-0.67278579 + $y * (-0.18156897 + $y * (-0.1919402e-1 + $y * (-0.110404e-2 + $y * (-0.4686e-4))))))) / $fNum; } else { $y = 2 / $fNum; $fRet = exp(-$fNum) / sqrt($fNum) * (1.25331414 + $y * (0.23498619 + $y * (-0.3655620e-1 + $y * (0.1504268e-1 + $y * (-0.780353e-2 + $y * (0.325614e-2 + $y * (-0.68245e-3))))))); } return $fRet; } // function _Besselk1() /** * BESSELK * * Returns the modified Bessel function, which is equivalent to the Bessel functions evaluated * for purely imaginary arguments. * * Excel Function: * BESSELK(x,ord) * * @access public * @category Engineering Functions * @param float $x The value at which to evaluate the function. * If x is nonnumeric, BESSELK returns the #VALUE! error value. * @param integer $ord The order of the Bessel function. If n is not an integer, it is truncated. * If $ord is nonnumeric, BESSELK returns the #VALUE! error value. * If $ord < 0, BESSELK returns the #NUM! error value. * @return float * * @TODO Better handling of the approximation method to support the differences between Excel/Gnumeric and Open/Libre Office * */ public static function BESSELK($x, $ord) { $x = (is_null($x)) ? 0.0 : PHPExcel_Calculation_Functions::flattenSingleValue($x); $ord = (is_null($ord)) ? 0.0 : PHPExcel_Calculation_Functions::flattenSingleValue($ord); if ((is_numeric($x)) && (is_numeric($ord))) { if (($ord < 0) || ($x == 0.0)) { return PHPExcel_Calculation_Functions::NaN(); } switch(floor($ord)) { case 0 : return self::_Besselk0($x); break; case 1 : return self::_Besselk1($x); break; default : $fTox = 2 / $x; $fBkm = self::_Besselk0($x); $fBk = self::_Besselk1($x); for ($n = 1; $n < $ord; ++$n) { $fBkp = $fBkm + $n * $fTox * $fBk; $fBkm = $fBk; $fBk = $fBkp; } } return (is_nan($fBk)) ? PHPExcel_Calculation_Functions::NaN() : $fBk; } return PHPExcel_Calculation_Functions::VALUE(); } // function BESSELK() private static function _Bessely0($fNum) { if ($fNum < 8.0) { $y = ($fNum * $fNum); $f1 = -2957821389.0 + $y * (7062834065.0 + $y * (-512359803.6 + $y * (10879881.29 + $y * (-86327.92757 + $y * 228.4622733)))); $f2 = 40076544269.0 + $y * (745249964.8 + $y * (7189466.438 + $y * (47447.26470 + $y * (226.1030244 + $y)))); $fRet = $f1 / $f2 + 0.636619772 * self::BESSELJ($fNum, 0) * log($fNum); } else { $z = 8.0 / $fNum; $y = ($z * $z); $xx = $fNum - 0.785398164; $f1 = 1 + $y * (-0.1098628627e-2 + $y * (0.2734510407e-4 + $y * (-0.2073370639e-5 + $y * 0.2093887211e-6))); $f2 = -0.1562499995e-1 + $y * (0.1430488765e-3 + $y * (-0.6911147651e-5 + $y * (0.7621095161e-6 + $y * (-0.934945152e-7)))); $fRet = sqrt(0.636619772 / $fNum) * (sin($xx) * $f1 + $z * cos($xx) * $f2); } return $fRet; } // function _Bessely0() private static function _Bessely1($fNum) { if ($fNum < 8.0) { $y = ($fNum * $fNum); $f1 = $fNum * (-0.4900604943e13 + $y * (0.1275274390e13 + $y * (-0.5153438139e11 + $y * (0.7349264551e9 + $y * (-0.4237922726e7 + $y * 0.8511937935e4))))); $f2 = 0.2499580570e14 + $y * (0.4244419664e12 + $y * (0.3733650367e10 + $y * (0.2245904002e8 + $y * (0.1020426050e6 + $y * (0.3549632885e3 + $y))))); $fRet = $f1 / $f2 + 0.636619772 * ( self::BESSELJ($fNum, 1) * log($fNum) - 1 / $fNum); } else { // $z = 8.0 / $fNum; // $y = ($z * $z); // $xx = $fNum - 2.356194491; // $f1 = 1 + $y * (0.183105e-2 + $y * (-0.3516396496e-4 + $y * (0.2457520174e-5 + $y * (-0.240337019e6)))); // $f2 = 0.04687499995 + $y * (-0.2002690873e-3 + $y * (0.8449199096e-5 + $y * (-0.88228987e-6 + $y * 0.105787412e-6))); // $fRet = sqrt(0.636619772 / $fNum) * (sin($xx) * $f1 + $z * cos($xx) * $f2); #i12430# ...but this seems to work much better. $fRet = sqrt(0.636619772 / $fNum) * sin($fNum - 2.356194491); } return $fRet; } // function _Bessely1() /** * BESSELY * * Returns the Bessel function, which is also called the Weber function or the Neumann function. * * Excel Function: * BESSELY(x,ord) * * @access public * @category Engineering Functions * @param float $x The value at which to evaluate the function. * If x is nonnumeric, BESSELK returns the #VALUE! error value. * @param integer $ord The order of the Bessel function. If n is not an integer, it is truncated. * If $ord is nonnumeric, BESSELK returns the #VALUE! error value. * If $ord < 0, BESSELK returns the #NUM! error value. * * @TODO Better handling of the approximation method to support the differences between Excel/Gnumeric and Open/Libre Office * * @return float */ public static function BESSELY($x, $ord) { $x = (is_null($x)) ? 0.0 : PHPExcel_Calculation_Functions::flattenSingleValue($x); $ord = (is_null($ord)) ? 0.0 : PHPExcel_Calculation_Functions::flattenSingleValue($ord); if ((is_numeric($x)) && (is_numeric($ord))) { if (($ord < 0) || ($x == 0.0)) { return PHPExcel_Calculation_Functions::NaN(); } switch(floor($ord)) { case 0 : return self::_Bessely0($x); break; case 1 : return self::_Bessely1($x); break; default: $fTox = 2 / $x; $fBym = self::_Bessely0($x); $fBy = self::_Bessely1($x); for ($n = 1; $n < $ord; ++$n) { $fByp = $n * $fTox * $fBy - $fBym; $fBym = $fBy; $fBy = $fByp; } } return (is_nan($fBy)) ? PHPExcel_Calculation_Functions::NaN() : $fBy; } return PHPExcel_Calculation_Functions::VALUE(); } // function BESSELY() /** * BINTODEC * * Return a binary value as decimal. * * Excel Function: * BIN2DEC(x) * * @access public * @category Engineering Functions * @param string $x The binary number (as a string) that you want to convert. The number * cannot contain more than 10 characters (10 bits). The most significant * bit of number is the sign bit. The remaining 9 bits are magnitude bits. * Negative numbers are represented using two's-complement notation. * If number is not a valid binary number, or if number contains more than * 10 characters (10 bits), BIN2DEC returns the #NUM! error value. * @return string */ public static function BINTODEC($x) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); if (is_bool($x)) { if (PHPExcel_Calculation_Functions::getCompatibilityMode() == PHPExcel_Calculation_Functions::COMPATIBILITY_OPENOFFICE) { $x = (int) $x; } else { return PHPExcel_Calculation_Functions::VALUE(); } } if (PHPExcel_Calculation_Functions::getCompatibilityMode() == PHPExcel_Calculation_Functions::COMPATIBILITY_GNUMERIC) { $x = floor($x); } $x = (string) $x; if (strlen($x) > preg_match_all('/[01]/',$x,$out)) { return PHPExcel_Calculation_Functions::NaN(); } if (strlen($x) > 10) { return PHPExcel_Calculation_Functions::NaN(); } elseif (strlen($x) == 10) { // Two's Complement $x = substr($x,-9); return '-'.(512-bindec($x)); } return bindec($x); } // function BINTODEC() /** * BINTOHEX * * Return a binary value as hex. * * Excel Function: * BIN2HEX(x[,places]) * * @access public * @category Engineering Functions * @param string $x The binary number (as a string) that you want to convert. The number * cannot contain more than 10 characters (10 bits). The most significant * bit of number is the sign bit. The remaining 9 bits are magnitude bits. * Negative numbers are represented using two's-complement notation. * If number is not a valid binary number, or if number contains more than * 10 characters (10 bits), BIN2HEX returns the #NUM! error value. * @param integer $places The number of characters to use. If places is omitted, BIN2HEX uses the * minimum number of characters necessary. Places is useful for padding the * return value with leading 0s (zeros). * If places is not an integer, it is truncated. * If places is nonnumeric, BIN2HEX returns the #VALUE! error value. * If places is negative, BIN2HEX returns the #NUM! error value. * @return string */ public static function BINTOHEX($x, $places=NULL) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); $places = PHPExcel_Calculation_Functions::flattenSingleValue($places); if (is_bool($x)) { if (PHPExcel_Calculation_Functions::getCompatibilityMode() == PHPExcel_Calculation_Functions::COMPATIBILITY_OPENOFFICE) { $x = (int) $x; } else { return PHPExcel_Calculation_Functions::VALUE(); } } if (PHPExcel_Calculation_Functions::getCompatibilityMode() == PHPExcel_Calculation_Functions::COMPATIBILITY_GNUMERIC) { $x = floor($x); } $x = (string) $x; if (strlen($x) > preg_match_all('/[01]/',$x,$out)) { return PHPExcel_Calculation_Functions::NaN(); } if (strlen($x) > 10) { return PHPExcel_Calculation_Functions::NaN(); } elseif (strlen($x) == 10) { // Two's Complement return str_repeat('F',8).substr(strtoupper(dechex(bindec(substr($x,-9)))),-2); } $hexVal = (string) strtoupper(dechex(bindec($x))); return self::_nbrConversionFormat($hexVal,$places); } // function BINTOHEX() /** * BINTOOCT * * Return a binary value as octal. * * Excel Function: * BIN2OCT(x[,places]) * * @access public * @category Engineering Functions * @param string $x The binary number (as a string) that you want to convert. The number * cannot contain more than 10 characters (10 bits). The most significant * bit of number is the sign bit. The remaining 9 bits are magnitude bits. * Negative numbers are represented using two's-complement notation. * If number is not a valid binary number, or if number contains more than * 10 characters (10 bits), BIN2OCT returns the #NUM! error value. * @param integer $places The number of characters to use. If places is omitted, BIN2OCT uses the * minimum number of characters necessary. Places is useful for padding the * return value with leading 0s (zeros). * If places is not an integer, it is truncated. * If places is nonnumeric, BIN2OCT returns the #VALUE! error value. * If places is negative, BIN2OCT returns the #NUM! error value. * @return string */ public static function BINTOOCT($x, $places=NULL) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); $places = PHPExcel_Calculation_Functions::flattenSingleValue($places); if (is_bool($x)) { if (PHPExcel_Calculation_Functions::getCompatibilityMode() == PHPExcel_Calculation_Functions::COMPATIBILITY_OPENOFFICE) { $x = (int) $x; } else { return PHPExcel_Calculation_Functions::VALUE(); } } if (PHPExcel_Calculation_Functions::getCompatibilityMode() == PHPExcel_Calculation_Functions::COMPATIBILITY_GNUMERIC) { $x = floor($x); } $x = (string) $x; if (strlen($x) > preg_match_all('/[01]/',$x,$out)) { return PHPExcel_Calculation_Functions::NaN(); } if (strlen($x) > 10) { return PHPExcel_Calculation_Functions::NaN(); } elseif (strlen($x) == 10) { // Two's Complement return str_repeat('7',7).substr(strtoupper(decoct(bindec(substr($x,-9)))),-3); } $octVal = (string) decoct(bindec($x)); return self::_nbrConversionFormat($octVal,$places); } // function BINTOOCT() /** * DECTOBIN * * Return a decimal value as binary. * * Excel Function: * DEC2BIN(x[,places]) * * @access public * @category Engineering Functions * @param string $x The decimal integer you want to convert. If number is negative, * valid place values are ignored and DEC2BIN returns a 10-character * (10-bit) binary number in which the most significant bit is the sign * bit. The remaining 9 bits are magnitude bits. Negative numbers are * represented using two's-complement notation. * If number < -512 or if number > 511, DEC2BIN returns the #NUM! error * value. * If number is nonnumeric, DEC2BIN returns the #VALUE! error value. * If DEC2BIN requires more than places characters, it returns the #NUM! * error value. * @param integer $places The number of characters to use. If places is omitted, DEC2BIN uses * the minimum number of characters necessary. Places is useful for * padding the return value with leading 0s (zeros). * If places is not an integer, it is truncated. * If places is nonnumeric, DEC2BIN returns the #VALUE! error value. * If places is zero or negative, DEC2BIN returns the #NUM! error value. * @return string */ public static function DECTOBIN($x, $places=NULL) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); $places = PHPExcel_Calculation_Functions::flattenSingleValue($places); if (is_bool($x)) { if (PHPExcel_Calculation_Functions::getCompatibilityMode() == PHPExcel_Calculation_Functions::COMPATIBILITY_OPENOFFICE) { $x = (int) $x; } else { return PHPExcel_Calculation_Functions::VALUE(); } } $x = (string) $x; if (strlen($x) > preg_match_all('/[-0123456789.]/',$x,$out)) { return PHPExcel_Calculation_Functions::VALUE(); } $x = (string) floor($x); $r = decbin($x); if (strlen($r) == 32) { // Two's Complement $r = substr($r,-10); } elseif (strlen($r) > 11) { return PHPExcel_Calculation_Functions::NaN(); } return self::_nbrConversionFormat($r,$places); } // function DECTOBIN() /** * DECTOHEX * * Return a decimal value as hex. * * Excel Function: * DEC2HEX(x[,places]) * * @access public * @category Engineering Functions * @param string $x The decimal integer you want to convert. If number is negative, * places is ignored and DEC2HEX returns a 10-character (40-bit) * hexadecimal number in which the most significant bit is the sign * bit. The remaining 39 bits are magnitude bits. Negative numbers * are represented using two's-complement notation. * If number < -549,755,813,888 or if number > 549,755,813,887, * DEC2HEX returns the #NUM! error value. * If number is nonnumeric, DEC2HEX returns the #VALUE! error value. * If DEC2HEX requires more than places characters, it returns the * #NUM! error value. * @param integer $places The number of characters to use. If places is omitted, DEC2HEX uses * the minimum number of characters necessary. Places is useful for * padding the return value with leading 0s (zeros). * If places is not an integer, it is truncated. * If places is nonnumeric, DEC2HEX returns the #VALUE! error value. * If places is zero or negative, DEC2HEX returns the #NUM! error value. * @return string */ public static function DECTOHEX($x, $places=null) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); $places = PHPExcel_Calculation_Functions::flattenSingleValue($places); if (is_bool($x)) { if (PHPExcel_Calculation_Functions::getCompatibilityMode() == PHPExcel_Calculation_Functions::COMPATIBILITY_OPENOFFICE) { $x = (int) $x; } else { return PHPExcel_Calculation_Functions::VALUE(); } } $x = (string) $x; if (strlen($x) > preg_match_all('/[-0123456789.]/',$x,$out)) { return PHPExcel_Calculation_Functions::VALUE(); } $x = (string) floor($x); $r = strtoupper(dechex($x)); if (strlen($r) == 8) { // Two's Complement $r = 'FF'.$r; } return self::_nbrConversionFormat($r,$places); } // function DECTOHEX() /** * DECTOOCT * * Return an decimal value as octal. * * Excel Function: * DEC2OCT(x[,places]) * * @access public * @category Engineering Functions * @param string $x The decimal integer you want to convert. If number is negative, * places is ignored and DEC2OCT returns a 10-character (30-bit) * octal number in which the most significant bit is the sign bit. * The remaining 29 bits are magnitude bits. Negative numbers are * represented using two's-complement notation. * If number < -536,870,912 or if number > 536,870,911, DEC2OCT * returns the #NUM! error value. * If number is nonnumeric, DEC2OCT returns the #VALUE! error value. * If DEC2OCT requires more than places characters, it returns the * #NUM! error value. * @param integer $places The number of characters to use. If places is omitted, DEC2OCT uses * the minimum number of characters necessary. Places is useful for * padding the return value with leading 0s (zeros). * If places is not an integer, it is truncated. * If places is nonnumeric, DEC2OCT returns the #VALUE! error value. * If places is zero or negative, DEC2OCT returns the #NUM! error value. * @return string */ public static function DECTOOCT($x, $places=null) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); $places = PHPExcel_Calculation_Functions::flattenSingleValue($places); if (is_bool($x)) { if (PHPExcel_Calculation_Functions::getCompatibilityMode() == PHPExcel_Calculation_Functions::COMPATIBILITY_OPENOFFICE) { $x = (int) $x; } else { return PHPExcel_Calculation_Functions::VALUE(); } } $x = (string) $x; if (strlen($x) > preg_match_all('/[-0123456789.]/',$x,$out)) { return PHPExcel_Calculation_Functions::VALUE(); } $x = (string) floor($x); $r = decoct($x); if (strlen($r) == 11) { // Two's Complement $r = substr($r,-10); } return self::_nbrConversionFormat($r,$places); } // function DECTOOCT() /** * HEXTOBIN * * Return a hex value as binary. * * Excel Function: * HEX2BIN(x[,places]) * * @access public * @category Engineering Functions * @param string $x the hexadecimal number you want to convert. Number cannot * contain more than 10 characters. The most significant bit of * number is the sign bit (40th bit from the right). The remaining * 9 bits are magnitude bits. Negative numbers are represented * using two's-complement notation. * If number is negative, HEX2BIN ignores places and returns a * 10-character binary number. * If number is negative, it cannot be less than FFFFFFFE00, and * if number is positive, it cannot be greater than 1FF. * If number is not a valid hexadecimal number, HEX2BIN returns * the #NUM! error value. * If HEX2BIN requires more than places characters, it returns * the #NUM! error value. * @param integer $places The number of characters to use. If places is omitted, * HEX2BIN uses the minimum number of characters necessary. Places * is useful for padding the return value with leading 0s (zeros). * If places is not an integer, it is truncated. * If places is nonnumeric, HEX2BIN returns the #VALUE! error value. * If places is negative, HEX2BIN returns the #NUM! error value. * @return string */ public static function HEXTOBIN($x, $places=null) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); $places = PHPExcel_Calculation_Functions::flattenSingleValue($places); if (is_bool($x)) { return PHPExcel_Calculation_Functions::VALUE(); } $x = (string) $x; if (strlen($x) > preg_match_all('/[0123456789ABCDEF]/',strtoupper($x),$out)) { return PHPExcel_Calculation_Functions::NaN(); } $binVal = decbin(hexdec($x)); return substr(self::_nbrConversionFormat($binVal,$places),-10); } // function HEXTOBIN() /** * HEXTODEC * * Return a hex value as decimal. * * Excel Function: * HEX2DEC(x) * * @access public * @category Engineering Functions * @param string $x The hexadecimal number you want to convert. This number cannot * contain more than 10 characters (40 bits). The most significant * bit of number is the sign bit. The remaining 39 bits are magnitude * bits. Negative numbers are represented using two's-complement * notation. * If number is not a valid hexadecimal number, HEX2DEC returns the * #NUM! error value. * @return string */ public static function HEXTODEC($x) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); if (is_bool($x)) { return PHPExcel_Calculation_Functions::VALUE(); } $x = (string) $x; if (strlen($x) > preg_match_all('/[0123456789ABCDEF]/',strtoupper($x),$out)) { return PHPExcel_Calculation_Functions::NaN(); } return hexdec($x); } // function HEXTODEC() /** * HEXTOOCT * * Return a hex value as octal. * * Excel Function: * HEX2OCT(x[,places]) * * @access public * @category Engineering Functions * @param string $x The hexadecimal number you want to convert. Number cannot * contain more than 10 characters. The most significant bit of * number is the sign bit. The remaining 39 bits are magnitude * bits. Negative numbers are represented using two's-complement * notation. * If number is negative, HEX2OCT ignores places and returns a * 10-character octal number. * If number is negative, it cannot be less than FFE0000000, and * if number is positive, it cannot be greater than 1FFFFFFF. * If number is not a valid hexadecimal number, HEX2OCT returns * the #NUM! error value. * If HEX2OCT requires more than places characters, it returns * the #NUM! error value. * @param integer $places The number of characters to use. If places is omitted, HEX2OCT * uses the minimum number of characters necessary. Places is * useful for padding the return value with leading 0s (zeros). * If places is not an integer, it is truncated. * If places is nonnumeric, HEX2OCT returns the #VALUE! error * value. * If places is negative, HEX2OCT returns the #NUM! error value. * @return string */ public static function HEXTOOCT($x, $places=null) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); $places = PHPExcel_Calculation_Functions::flattenSingleValue($places); if (is_bool($x)) { return PHPExcel_Calculation_Functions::VALUE(); } $x = (string) $x; if (strlen($x) > preg_match_all('/[0123456789ABCDEF]/',strtoupper($x),$out)) { return PHPExcel_Calculation_Functions::NaN(); } $octVal = decoct(hexdec($x)); return self::_nbrConversionFormat($octVal,$places); } // function HEXTOOCT() /** * OCTTOBIN * * Return an octal value as binary. * * Excel Function: * OCT2BIN(x[,places]) * * @access public * @category Engineering Functions * @param string $x The octal number you want to convert. Number may not * contain more than 10 characters. The most significant * bit of number is the sign bit. The remaining 29 bits * are magnitude bits. Negative numbers are represented * using two's-complement notation. * If number is negative, OCT2BIN ignores places and returns * a 10-character binary number. * If number is negative, it cannot be less than 7777777000, * and if number is positive, it cannot be greater than 777. * If number is not a valid octal number, OCT2BIN returns * the #NUM! error value. * If OCT2BIN requires more than places characters, it * returns the #NUM! error value. * @param integer $places The number of characters to use. If places is omitted, * OCT2BIN uses the minimum number of characters necessary. * Places is useful for padding the return value with * leading 0s (zeros). * If places is not an integer, it is truncated. * If places is nonnumeric, OCT2BIN returns the #VALUE! * error value. * If places is negative, OCT2BIN returns the #NUM! error * value. * @return string */ public static function OCTTOBIN($x, $places=null) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); $places = PHPExcel_Calculation_Functions::flattenSingleValue($places); if (is_bool($x)) { return PHPExcel_Calculation_Functions::VALUE(); } $x = (string) $x; if (preg_match_all('/[01234567]/',$x,$out) != strlen($x)) { return PHPExcel_Calculation_Functions::NaN(); } $r = decbin(octdec($x)); return self::_nbrConversionFormat($r,$places); } // function OCTTOBIN() /** * OCTTODEC * * Return an octal value as decimal. * * Excel Function: * OCT2DEC(x) * * @access public * @category Engineering Functions * @param string $x The octal number you want to convert. Number may not contain * more than 10 octal characters (30 bits). The most significant * bit of number is the sign bit. The remaining 29 bits are * magnitude bits. Negative numbers are represented using * two's-complement notation. * If number is not a valid octal number, OCT2DEC returns the * #NUM! error value. * @return string */ public static function OCTTODEC($x) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); if (is_bool($x)) { return PHPExcel_Calculation_Functions::VALUE(); } $x = (string) $x; if (preg_match_all('/[01234567]/',$x,$out) != strlen($x)) { return PHPExcel_Calculation_Functions::NaN(); } return octdec($x); } // function OCTTODEC() /** * OCTTOHEX * * Return an octal value as hex. * * Excel Function: * OCT2HEX(x[,places]) * * @access public * @category Engineering Functions * @param string $x The octal number you want to convert. Number may not contain * more than 10 octal characters (30 bits). The most significant * bit of number is the sign bit. The remaining 29 bits are * magnitude bits. Negative numbers are represented using * two's-complement notation. * If number is negative, OCT2HEX ignores places and returns a * 10-character hexadecimal number. * If number is not a valid octal number, OCT2HEX returns the * #NUM! error value. * If OCT2HEX requires more than places characters, it returns * the #NUM! error value. * @param integer $places The number of characters to use. If places is omitted, OCT2HEX * uses the minimum number of characters necessary. Places is useful * for padding the return value with leading 0s (zeros). * If places is not an integer, it is truncated. * If places is nonnumeric, OCT2HEX returns the #VALUE! error value. * If places is negative, OCT2HEX returns the #NUM! error value. * @return string */ public static function OCTTOHEX($x, $places=null) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); $places = PHPExcel_Calculation_Functions::flattenSingleValue($places); if (is_bool($x)) { return PHPExcel_Calculation_Functions::VALUE(); } $x = (string) $x; if (preg_match_all('/[01234567]/',$x,$out) != strlen($x)) { return PHPExcel_Calculation_Functions::NaN(); } $hexVal = strtoupper(dechex(octdec($x))); return self::_nbrConversionFormat($hexVal,$places); } // function OCTTOHEX() /** * COMPLEX * * Converts real and imaginary coefficients into a complex number of the form x + yi or x + yj. * * Excel Function: * COMPLEX(realNumber,imaginary[,places]) * * @access public * @category Engineering Functions * @param float $realNumber The real coefficient of the complex number. * @param float $imaginary The imaginary coefficient of the complex number. * @param string $suffix The suffix for the imaginary component of the complex number. * If omitted, the suffix is assumed to be "i". * @return string */ public static function COMPLEX($realNumber=0.0, $imaginary=0.0, $suffix='i') { $realNumber = (is_null($realNumber)) ? 0.0 : PHPExcel_Calculation_Functions::flattenSingleValue($realNumber); $imaginary = (is_null($imaginary)) ? 0.0 : PHPExcel_Calculation_Functions::flattenSingleValue($imaginary); $suffix = (is_null($suffix)) ? 'i' : PHPExcel_Calculation_Functions::flattenSingleValue($suffix); if (((is_numeric($realNumber)) && (is_numeric($imaginary))) && (($suffix == 'i') || ($suffix == 'j') || ($suffix == ''))) { $realNumber = (float) $realNumber; $imaginary = (float) $imaginary; if ($suffix == '') $suffix = 'i'; if ($realNumber == 0.0) { if ($imaginary == 0.0) { return (string) '0'; } elseif ($imaginary == 1.0) { return (string) $suffix; } elseif ($imaginary == -1.0) { return (string) '-'.$suffix; } return (string) $imaginary.$suffix; } elseif ($imaginary == 0.0) { return (string) $realNumber; } elseif ($imaginary == 1.0) { return (string) $realNumber.'+'.$suffix; } elseif ($imaginary == -1.0) { return (string) $realNumber.'-'.$suffix; } if ($imaginary > 0) { $imaginary = (string) '+'.$imaginary; } return (string) $realNumber.$imaginary.$suffix; } return PHPExcel_Calculation_Functions::VALUE(); } // function COMPLEX() /** * IMAGINARY * * Returns the imaginary coefficient of a complex number in x + yi or x + yj text format. * * Excel Function: * IMAGINARY(complexNumber) * * @access public * @category Engineering Functions * @param string $complexNumber The complex number for which you want the imaginary * coefficient. * @return float */ public static function IMAGINARY($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); return $parsedComplex['imaginary']; } // function IMAGINARY() /** * IMREAL * * Returns the real coefficient of a complex number in x + yi or x + yj text format. * * Excel Function: * IMREAL(complexNumber) * * @access public * @category Engineering Functions * @param string $complexNumber The complex number for which you want the real coefficient. * @return float */ public static function IMREAL($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); return $parsedComplex['real']; } // function IMREAL() /** * IMABS * * Returns the absolute value (modulus) of a complex number in x + yi or x + yj text format. * * Excel Function: * IMABS(complexNumber) * * @param string $complexNumber The complex number for which you want the absolute value. * @return float */ public static function IMABS($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); return sqrt(($parsedComplex['real'] * $parsedComplex['real']) + ($parsedComplex['imaginary'] * $parsedComplex['imaginary'])); } // function IMABS() /** * IMARGUMENT * * Returns the argument theta of a complex number, i.e. the angle in radians from the real * axis to the representation of the number in polar coordinates. * * Excel Function: * IMARGUMENT(complexNumber) * * @param string $complexNumber The complex number for which you want the argument theta. * @return float */ public static function IMARGUMENT($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); if ($parsedComplex['real'] == 0.0) { if ($parsedComplex['imaginary'] == 0.0) { return 0.0; } elseif($parsedComplex['imaginary'] < 0.0) { return M_PI / -2; } else { return M_PI / 2; } } elseif ($parsedComplex['real'] > 0.0) { return atan($parsedComplex['imaginary'] / $parsedComplex['real']); } elseif ($parsedComplex['imaginary'] < 0.0) { return 0 - (M_PI - atan(abs($parsedComplex['imaginary']) / abs($parsedComplex['real']))); } else { return M_PI - atan($parsedComplex['imaginary'] / abs($parsedComplex['real'])); } } // function IMARGUMENT() /** * IMCONJUGATE * * Returns the complex conjugate of a complex number in x + yi or x + yj text format. * * Excel Function: * IMCONJUGATE(complexNumber) * * @param string $complexNumber The complex number for which you want the conjugate. * @return string */ public static function IMCONJUGATE($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); if ($parsedComplex['imaginary'] == 0.0) { return $parsedComplex['real']; } else { return self::_cleanComplex( self::COMPLEX( $parsedComplex['real'], 0 - $parsedComplex['imaginary'], $parsedComplex['suffix'] ) ); } } // function IMCONJUGATE() /** * IMCOS * * Returns the cosine of a complex number in x + yi or x + yj text format. * * Excel Function: * IMCOS(complexNumber) * * @param string $complexNumber The complex number for which you want the cosine. * @return string|float */ public static function IMCOS($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); if ($parsedComplex['imaginary'] == 0.0) { return cos($parsedComplex['real']); } else { return self::IMCONJUGATE(self::COMPLEX(cos($parsedComplex['real']) * cosh($parsedComplex['imaginary']),sin($parsedComplex['real']) * sinh($parsedComplex['imaginary']),$parsedComplex['suffix'])); } } // function IMCOS() /** * IMSIN * * Returns the sine of a complex number in x + yi or x + yj text format. * * Excel Function: * IMSIN(complexNumber) * * @param string $complexNumber The complex number for which you want the sine. * @return string|float */ public static function IMSIN($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); if ($parsedComplex['imaginary'] == 0.0) { return sin($parsedComplex['real']); } else { return self::COMPLEX(sin($parsedComplex['real']) * cosh($parsedComplex['imaginary']),cos($parsedComplex['real']) * sinh($parsedComplex['imaginary']),$parsedComplex['suffix']); } } // function IMSIN() /** * IMSQRT * * Returns the square root of a complex number in x + yi or x + yj text format. * * Excel Function: * IMSQRT(complexNumber) * * @param string $complexNumber The complex number for which you want the square root. * @return string */ public static function IMSQRT($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); $theta = self::IMARGUMENT($complexNumber); $d1 = cos($theta / 2); $d2 = sin($theta / 2); $r = sqrt(sqrt(($parsedComplex['real'] * $parsedComplex['real']) + ($parsedComplex['imaginary'] * $parsedComplex['imaginary']))); if ($parsedComplex['suffix'] == '') { return self::COMPLEX($d1 * $r,$d2 * $r); } else { return self::COMPLEX($d1 * $r,$d2 * $r,$parsedComplex['suffix']); } } // function IMSQRT() /** * IMLN * * Returns the natural logarithm of a complex number in x + yi or x + yj text format. * * Excel Function: * IMLN(complexNumber) * * @param string $complexNumber The complex number for which you want the natural logarithm. * @return string */ public static function IMLN($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); if (($parsedComplex['real'] == 0.0) && ($parsedComplex['imaginary'] == 0.0)) { return PHPExcel_Calculation_Functions::NaN(); } $logR = log(sqrt(($parsedComplex['real'] * $parsedComplex['real']) + ($parsedComplex['imaginary'] * $parsedComplex['imaginary']))); $t = self::IMARGUMENT($complexNumber); if ($parsedComplex['suffix'] == '') { return self::COMPLEX($logR,$t); } else { return self::COMPLEX($logR,$t,$parsedComplex['suffix']); } } // function IMLN() /** * IMLOG10 * * Returns the common logarithm (base 10) of a complex number in x + yi or x + yj text format. * * Excel Function: * IMLOG10(complexNumber) * * @param string $complexNumber The complex number for which you want the common logarithm. * @return string */ public static function IMLOG10($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); if (($parsedComplex['real'] == 0.0) && ($parsedComplex['imaginary'] == 0.0)) { return PHPExcel_Calculation_Functions::NaN(); } elseif (($parsedComplex['real'] > 0.0) && ($parsedComplex['imaginary'] == 0.0)) { return log10($parsedComplex['real']); } return self::IMPRODUCT(log10(EULER),self::IMLN($complexNumber)); } // function IMLOG10() /** * IMLOG2 * * Returns the common logarithm (base 10) of a complex number in x + yi or x + yj text format. * * Excel Function: * IMLOG2(complexNumber) * * @param string $complexNumber The complex number for which you want the base-2 logarithm. * @return string */ public static function IMLOG2($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); if (($parsedComplex['real'] == 0.0) && ($parsedComplex['imaginary'] == 0.0)) { return PHPExcel_Calculation_Functions::NaN(); } elseif (($parsedComplex['real'] > 0.0) && ($parsedComplex['imaginary'] == 0.0)) { return log($parsedComplex['real'],2); } return self::IMPRODUCT(log(EULER,2),self::IMLN($complexNumber)); } // function IMLOG2() /** * IMEXP * * Returns the exponential of a complex number in x + yi or x + yj text format. * * Excel Function: * IMEXP(complexNumber) * * @param string $complexNumber The complex number for which you want the exponential. * @return string */ public static function IMEXP($complexNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $parsedComplex = self::_parseComplex($complexNumber); if (($parsedComplex['real'] == 0.0) && ($parsedComplex['imaginary'] == 0.0)) { return '1'; } $e = exp($parsedComplex['real']); $eX = $e * cos($parsedComplex['imaginary']); $eY = $e * sin($parsedComplex['imaginary']); if ($parsedComplex['suffix'] == '') { return self::COMPLEX($eX,$eY); } else { return self::COMPLEX($eX,$eY,$parsedComplex['suffix']); } } // function IMEXP() /** * IMPOWER * * Returns a complex number in x + yi or x + yj text format raised to a power. * * Excel Function: * IMPOWER(complexNumber,realNumber) * * @param string $complexNumber The complex number you want to raise to a power. * @param float $realNumber The power to which you want to raise the complex number. * @return string */ public static function IMPOWER($complexNumber,$realNumber) { $complexNumber = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber); $realNumber = PHPExcel_Calculation_Functions::flattenSingleValue($realNumber); if (!is_numeric($realNumber)) { return PHPExcel_Calculation_Functions::VALUE(); } $parsedComplex = self::_parseComplex($complexNumber); $r = sqrt(($parsedComplex['real'] * $parsedComplex['real']) + ($parsedComplex['imaginary'] * $parsedComplex['imaginary'])); $rPower = pow($r,$realNumber); $theta = self::IMARGUMENT($complexNumber) * $realNumber; if ($theta == 0) { return 1; } elseif ($parsedComplex['imaginary'] == 0.0) { return self::COMPLEX($rPower * cos($theta),$rPower * sin($theta),$parsedComplex['suffix']); } else { return self::COMPLEX($rPower * cos($theta),$rPower * sin($theta),$parsedComplex['suffix']); } } // function IMPOWER() /** * IMDIV * * Returns the quotient of two complex numbers in x + yi or x + yj text format. * * Excel Function: * IMDIV(complexDividend,complexDivisor) * * @param string $complexDividend The complex numerator or dividend. * @param string $complexDivisor The complex denominator or divisor. * @return string */ public static function IMDIV($complexDividend,$complexDivisor) { $complexDividend = PHPExcel_Calculation_Functions::flattenSingleValue($complexDividend); $complexDivisor = PHPExcel_Calculation_Functions::flattenSingleValue($complexDivisor); $parsedComplexDividend = self::_parseComplex($complexDividend); $parsedComplexDivisor = self::_parseComplex($complexDivisor); if (($parsedComplexDividend['suffix'] != '') && ($parsedComplexDivisor['suffix'] != '') && ($parsedComplexDividend['suffix'] != $parsedComplexDivisor['suffix'])) { return PHPExcel_Calculation_Functions::NaN(); } if (($parsedComplexDividend['suffix'] != '') && ($parsedComplexDivisor['suffix'] == '')) { $parsedComplexDivisor['suffix'] = $parsedComplexDividend['suffix']; } $d1 = ($parsedComplexDividend['real'] * $parsedComplexDivisor['real']) + ($parsedComplexDividend['imaginary'] * $parsedComplexDivisor['imaginary']); $d2 = ($parsedComplexDividend['imaginary'] * $parsedComplexDivisor['real']) - ($parsedComplexDividend['real'] * $parsedComplexDivisor['imaginary']); $d3 = ($parsedComplexDivisor['real'] * $parsedComplexDivisor['real']) + ($parsedComplexDivisor['imaginary'] * $parsedComplexDivisor['imaginary']); $r = $d1/$d3; $i = $d2/$d3; if ($i > 0.0) { return self::_cleanComplex($r.'+'.$i.$parsedComplexDivisor['suffix']); } elseif ($i < 0.0) { return self::_cleanComplex($r.$i.$parsedComplexDivisor['suffix']); } else { return $r; } } // function IMDIV() /** * IMSUB * * Returns the difference of two complex numbers in x + yi or x + yj text format. * * Excel Function: * IMSUB(complexNumber1,complexNumber2) * * @param string $complexNumber1 The complex number from which to subtract complexNumber2. * @param string $complexNumber2 The complex number to subtract from complexNumber1. * @return string */ public static function IMSUB($complexNumber1,$complexNumber2) { $complexNumber1 = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber1); $complexNumber2 = PHPExcel_Calculation_Functions::flattenSingleValue($complexNumber2); $parsedComplex1 = self::_parseComplex($complexNumber1); $parsedComplex2 = self::_parseComplex($complexNumber2); if ((($parsedComplex1['suffix'] != '') && ($parsedComplex2['suffix'] != '')) && ($parsedComplex1['suffix'] != $parsedComplex2['suffix'])) { return PHPExcel_Calculation_Functions::NaN(); } elseif (($parsedComplex1['suffix'] == '') && ($parsedComplex2['suffix'] != '')) { $parsedComplex1['suffix'] = $parsedComplex2['suffix']; } $d1 = $parsedComplex1['real'] - $parsedComplex2['real']; $d2 = $parsedComplex1['imaginary'] - $parsedComplex2['imaginary']; return self::COMPLEX($d1,$d2,$parsedComplex1['suffix']); } // function IMSUB() /** * IMSUM * * Returns the sum of two or more complex numbers in x + yi or x + yj text format. * * Excel Function: * IMSUM(complexNumber[,complexNumber[,...]]) * * @param string $complexNumber,... Series of complex numbers to add * @return string */ public static function IMSUM() { // Return value $returnValue = self::_parseComplex('0'); $activeSuffix = ''; // Loop through the arguments $aArgs = PHPExcel_Calculation_Functions::flattenArray(func_get_args()); foreach ($aArgs as $arg) { $parsedComplex = self::_parseComplex($arg); if ($activeSuffix == '') { $activeSuffix = $parsedComplex['suffix']; } elseif (($parsedComplex['suffix'] != '') && ($activeSuffix != $parsedComplex['suffix'])) { return PHPExcel_Calculation_Functions::VALUE(); } $returnValue['real'] += $parsedComplex['real']; $returnValue['imaginary'] += $parsedComplex['imaginary']; } if ($returnValue['imaginary'] == 0.0) { $activeSuffix = ''; } return self::COMPLEX($returnValue['real'],$returnValue['imaginary'],$activeSuffix); } // function IMSUM() /** * IMPRODUCT * * Returns the product of two or more complex numbers in x + yi or x + yj text format. * * Excel Function: * IMPRODUCT(complexNumber[,complexNumber[,...]]) * * @param string $complexNumber,... Series of complex numbers to multiply * @return string */ public static function IMPRODUCT() { // Return value $returnValue = self::_parseComplex('1'); $activeSuffix = ''; // Loop through the arguments $aArgs = PHPExcel_Calculation_Functions::flattenArray(func_get_args()); foreach ($aArgs as $arg) { $parsedComplex = self::_parseComplex($arg); $workValue = $returnValue; if (($parsedComplex['suffix'] != '') && ($activeSuffix == '')) { $activeSuffix = $parsedComplex['suffix']; } elseif (($parsedComplex['suffix'] != '') && ($activeSuffix != $parsedComplex['suffix'])) { return PHPExcel_Calculation_Functions::NaN(); } $returnValue['real'] = ($workValue['real'] * $parsedComplex['real']) - ($workValue['imaginary'] * $parsedComplex['imaginary']); $returnValue['imaginary'] = ($workValue['real'] * $parsedComplex['imaginary']) + ($workValue['imaginary'] * $parsedComplex['real']); } if ($returnValue['imaginary'] == 0.0) { $activeSuffix = ''; } return self::COMPLEX($returnValue['real'],$returnValue['imaginary'],$activeSuffix); } // function IMPRODUCT() /** * DELTA * * Tests whether two values are equal. Returns 1 if number1 = number2; returns 0 otherwise. * Use this function to filter a set of values. For example, by summing several DELTA * functions you calculate the count of equal pairs. This function is also known as the * Kronecker Delta function. * * Excel Function: * DELTA(a[,b]) * * @param float $a The first number. * @param float $b The second number. If omitted, b is assumed to be zero. * @return int */ public static function DELTA($a, $b=0) { $a = PHPExcel_Calculation_Functions::flattenSingleValue($a); $b = PHPExcel_Calculation_Functions::flattenSingleValue($b); return (int) ($a == $b); } // function DELTA() /** * GESTEP * * Excel Function: * GESTEP(number[,step]) * * Returns 1 if number >= step; returns 0 (zero) otherwise * Use this function to filter a set of values. For example, by summing several GESTEP * functions you calculate the count of values that exceed a threshold. * * @param float $number The value to test against step. * @param float $step The threshold value. * If you omit a value for step, GESTEP uses zero. * @return int */ public static function GESTEP($number, $step=0) { $number = PHPExcel_Calculation_Functions::flattenSingleValue($number); $step = PHPExcel_Calculation_Functions::flattenSingleValue($step); return (int) ($number >= $step); } // function GESTEP() // // Private method to calculate the erf value // private static $_two_sqrtpi = 1.128379167095512574; public static function _erfVal($x) { if (abs($x) > 2.2) { return 1 - self::_erfcVal($x); } $sum = $term = $x; $xsqr = ($x * $x); $j = 1; do { $term *= $xsqr / $j; $sum -= $term / (2 * $j + 1); ++$j; $term *= $xsqr / $j; $sum += $term / (2 * $j + 1); ++$j; if ($sum == 0.0) { break; } } while (abs($term / $sum) > PRECISION); return self::$_two_sqrtpi * $sum; } // function _erfVal() /** * ERF * * Returns the error function integrated between lower_limit and upper_limit * * Excel Function: * ERF(lower[,upper]) * * @param float $lower lower bound for integrating ERF * @param float $upper upper bound for integrating ERF. * If omitted, ERF integrates between zero and lower_limit * @return int */ public static function ERF($lower, $upper = null) { $lower = PHPExcel_Calculation_Functions::flattenSingleValue($lower); $upper = PHPExcel_Calculation_Functions::flattenSingleValue($upper); if (is_numeric($lower)) { if ($lower < 0) { return PHPExcel_Calculation_Functions::NaN(); } if (is_null($upper)) { return self::_erfVal($lower); } if (is_numeric($upper)) { if ($upper < 0) { return PHPExcel_Calculation_Functions::NaN(); } return self::_erfVal($upper) - self::_erfVal($lower); } } return PHPExcel_Calculation_Functions::VALUE(); } // function ERF() // // Private method to calculate the erfc value // private static $_one_sqrtpi = 0.564189583547756287; private static function _erfcVal($x) { if (abs($x) < 2.2) { return 1 - self::_erfVal($x); } if ($x < 0) { return 2 - self::ERFC(-$x); } $a = $n = 1; $b = $c = $x; $d = ($x * $x) + 0.5; $q1 = $q2 = $b / $d; $t = 0; do { $t = $a * $n + $b * $x; $a = $b; $b = $t; $t = $c * $n + $d * $x; $c = $d; $d = $t; $n += 0.5; $q1 = $q2; $q2 = $b / $d; } while ((abs($q1 - $q2) / $q2) > PRECISION); return self::$_one_sqrtpi * exp(-$x * $x) * $q2; } // function _erfcVal() /** * ERFC * * Returns the complementary ERF function integrated between x and infinity * * Excel Function: * ERF(x) * * @param float $x The lower bound for integrating ERF * @return int */ public static function ERFC($x) { $x = PHPExcel_Calculation_Functions::flattenSingleValue($x); if (is_numeric($x)) { if ($x < 0) { return PHPExcel_Calculation_Functions::NaN(); } return self::_erfcVal($x); } return PHPExcel_Calculation_Functions::VALUE(); } // function ERFC() /** * getConversionGroups * * @return array */ public static function getConversionGroups() { $conversionGroups = array(); foreach(self::$_conversionUnits as $conversionUnit) { $conversionGroups[] = $conversionUnit['Group']; } return array_merge(array_unique($conversionGroups)); } // function getConversionGroups() /** * getConversionGroupUnits * * @return array */ public static function getConversionGroupUnits($group = NULL) { $conversionGroups = array(); foreach(self::$_conversionUnits as $conversionUnit => $conversionGroup) { if ((is_null($group)) || ($conversionGroup['Group'] == $group)) { $conversionGroups[$conversionGroup['Group']][] = $conversionUnit; } } return $conversionGroups; } // function getConversionGroupUnits() /** * getConversionGroupUnitDetails * * @return array */ public static function getConversionGroupUnitDetails($group = NULL) { $conversionGroups = array(); foreach(self::$_conversionUnits as $conversionUnit => $conversionGroup) { if ((is_null($group)) || ($conversionGroup['Group'] == $group)) { $conversionGroups[$conversionGroup['Group']][] = array( 'unit' => $conversionUnit, 'description' => $conversionGroup['Unit Name'] ); } } return $conversionGroups; } // function getConversionGroupUnitDetails() /** * getConversionGroups * * @return array */ public static function getConversionMultipliers() { return self::$_conversionMultipliers; } // function getConversionGroups() /** * CONVERTUOM * * Converts a number from one measurement system to another. * For example, CONVERT can translate a table of distances in miles to a table of distances * in kilometers. * * Excel Function: * CONVERT(value,fromUOM,toUOM) * * @param float $value The value in fromUOM to convert. * @param string $fromUOM The units for value. * @param string $toUOM The units for the result. * @return float */ public static function CONVERTUOM($value, $fromUOM, $toUOM) { $value = PHPExcel_Calculation_Functions::flattenSingleValue($value); $fromUOM = PHPExcel_Calculation_Functions::flattenSingleValue($fromUOM); $toUOM = PHPExcel_Calculation_Functions::flattenSingleValue($toUOM); if (!is_numeric($value)) { return PHPExcel_Calculation_Functions::VALUE(); } $fromMultiplier = 1.0; if (isset(self::$_conversionUnits[$fromUOM])) { $unitGroup1 = self::$_conversionUnits[$fromUOM]['Group']; } else { $fromMultiplier = substr($fromUOM,0,1); $fromUOM = substr($fromUOM,1); if (isset(self::$_conversionMultipliers[$fromMultiplier])) { $fromMultiplier = self::$_conversionMultipliers[$fromMultiplier]['multiplier']; } else { return PHPExcel_Calculation_Functions::NA(); } if ((isset(self::$_conversionUnits[$fromUOM])) && (self::$_conversionUnits[$fromUOM]['AllowPrefix'])) { $unitGroup1 = self::$_conversionUnits[$fromUOM]['Group']; } else { return PHPExcel_Calculation_Functions::NA(); } } $value *= $fromMultiplier; $toMultiplier = 1.0; if (isset(self::$_conversionUnits[$toUOM])) { $unitGroup2 = self::$_conversionUnits[$toUOM]['Group']; } else { $toMultiplier = substr($toUOM,0,1); $toUOM = substr($toUOM,1); if (isset(self::$_conversionMultipliers[$toMultiplier])) { $toMultiplier = self::$_conversionMultipliers[$toMultiplier]['multiplier']; } else { return PHPExcel_Calculation_Functions::NA(); } if ((isset(self::$_conversionUnits[$toUOM])) && (self::$_conversionUnits[$toUOM]['AllowPrefix'])) { $unitGroup2 = self::$_conversionUnits[$toUOM]['Group']; } else { return PHPExcel_Calculation_Functions::NA(); } } if ($unitGroup1 != $unitGroup2) { return PHPExcel_Calculation_Functions::NA(); } if (($fromUOM == $toUOM) && ($fromMultiplier == $toMultiplier)) { // We've already factored $fromMultiplier into the value, so we need // to reverse it again return $value / $fromMultiplier; } elseif ($unitGroup1 == 'Temperature') { if (($fromUOM == 'F') || ($fromUOM == 'fah')) { if (($toUOM == 'F') || ($toUOM == 'fah')) { return $value; } else { $value = (($value - 32) / 1.8); if (($toUOM == 'K') || ($toUOM == 'kel')) { $value += 273.15; } return $value; } } elseif ((($fromUOM == 'K') || ($fromUOM == 'kel')) && (($toUOM == 'K') || ($toUOM == 'kel'))) { return $value; } elseif ((($fromUOM == 'C') || ($fromUOM == 'cel')) && (($toUOM == 'C') || ($toUOM == 'cel'))) { return $value; } if (($toUOM == 'F') || ($toUOM == 'fah')) { if (($fromUOM == 'K') || ($fromUOM == 'kel')) { $value -= 273.15; } return ($value * 1.8) + 32; } if (($toUOM == 'C') || ($toUOM == 'cel')) { return $value - 273.15; } return $value + 273.15; } return ($value * self::$_unitConversions[$unitGroup1][$fromUOM][$toUOM]) / $toMultiplier; } // function CONVERTUOM() } // class PHPExcel_Calculation_Engineering